Research on Unified Information Modeling and Cross-Protocol Real-Time Interaction Mechanisms for Multi-Energy Supply Systems in Green Buildings

Abstract

Green buildings increasingly couple electrical, thermal, and hydrogen subsystems, yet these assets are typically monitored and controlled through separate standards and protocols. The resulting heterogeneous information models and communication stacks hinder millisecond-level coordination, plug-and-play integration, and resilient operation. To address this gap, we develop a unified information model and a cross-protocol real-time interaction mechanism based on extensions of IEC 61850. At the modeling level, we introduce new logical nodes and standardized data objects that describe electrical, thermal, and hydrogen devices in a single semantic space, supported by a global unit system and knowledge-graph-based semantic checking. At the communication level, we introduce a semantic gateway with adaptive mapping bridges IEC 61850 and legacy building protocols, while fast event messaging and 5G-enabled edge computing support deterministic low-latency control. The approach is validated on a digital-twin platform that couples an RTDS-based multi-energy system with a 5G test network. Experiments show device plug-and-play within 0.8 s, cross-protocol response-time differences below 50 ms, GOOSE latency under 5 ms, and critical-data success rates above 90% at a bit-error rate of 10⁻³. Under grid-fault scenarios, the proposed framework reduces voltage recovery time by about 60% and frequency deviation by about 70%, leading to more than 80% improvement in a composite resilience index compared with a conventional non-unified architecture. These results indicate that the framework provides a practical basis for interoperable, low-carbon, and resilient energy management in green buildings.

1. Introduction

Large public buildings and urban districts are increasingly promoting multi-energy supply systems to enhance energy efficiency and enable low-carbon operation [1]. Green buildings typically integrate multiple energy forms. On the electrical side, they include photovoltaic generation, wind turbines, battery storage, and electric vehicle charging facilities. On the thermal side, they employ Combined Heat and Power (CHP) units, heat pumps, and thermal storage devices [2]. In some cases, hydrogen fuel cells or electrolysis units are introduced for energy storage or emergency power supply [3], as shown in Figure 1. By coordinating electricity, heat, hydrogen, and other energy forms, green buildings can enhance overall efficiency, achieve peak shaving and valley filling, and reduce carbon emissions [4,5]. However, because these devices span different domains—electricity, thermal energy, and gas (hydrogen)—the heterogeneity of equipment and diversity of protocols pose major challenges for integrated energy management and control.

First, different energy subsystems maintain independent information models and communication standards, which hinder interconnection and interoperability [6]. For example, building power distribution equipment commonly uses IEC 61850/MMS or DNP3 protocols [7,8]. Photovoltaic inverters and electric vehicle charging stations often follow IEEE 2030.5 (SEP2.0) [8]. Building Automation Systems (BAS) for HVAC equipment typically employ BACnet or Modbus, while gas boilers and fuel cells usually rely on proprietary interfaces. Without unified semantics, data from different devices remain inconsistent [9], and information models are highly fragmented, which hinders the integrated management of multi-energy systems in green buildings [10,11]. As a result, data from different subsystems cannot directly interface, creating “information islands” that prevent joint optimization and control of electricity–heat–hydrogen systems [12].

Second, the coexistence of multiple communication protocols creates challenges for real-time cross-protocol interaction. Traditional building energy management operates in isolation [13]. For instance, BAS systems monitor HVAC, while power monitoring systems manage distribution equipment, without a direct communication interface between them. Even with integrated energy management platforms, inconsistent protocols [14] often necessitate custom-developed interfaces, leading to high integration costs and complex maintenance. In scenarios requiring real-time coordinated control [15]—for example, simultaneous coordination of CHP units and air-conditioning loads during grid faults—heterogeneous protocols without efficient interaction mechanisms may cause long communication delays [16] and reduced reliability, making timely coordination difficult. Green buildings emphasize high-resilience operation, requiring energy devices to respond within milliseconds. Examples include immediate fuel cell start/stop and rapid air-conditioning load switching during distribution faults, which place stringent demands on cross-protocol communication performance [17].

Existing research and standardization efforts have proposed preliminary solutions to these issues. In information modeling, IEC 61850, the international standard for power system communication [18], has been widely applied to substation automation and distributed energy systems [19]. IEC 61850 provides standardized Logical Node (LN) and Data Object (DO) models that facilitate interoperability among IED devices from different manufacturers. Studies show that extending IEC 61850 data models [20] allows the description of new distributed energy devices such as photovoltaics and electric vehicles. For example, IEC 61850-7-420 introduces models for Distributed Energy Resources (DER), including photovoltaic inverters and battery storage. It supports information integration of renewable devices in microgrids through custom LNs. However, these extensions are largely limited to electrical devices. Modeling for thermal and hydrogen energy remains insufficient, and no unified framework currently covers electrical, thermal, and hydrogen devices simultaneously.

Research [21,22] has proposed using Common Information Mode (CIM) or ontologies to establish information models for integrated energy systems, aiming to achieve multi-energy semantic integration. For example, ref. [23] proposed a universal standardized information model for integrated energy systems, while ref. [24] validated the value of unified semantic models for multi-energy information interaction. Overall, these studies have alleviated heterogeneous data integration problems to some extent. However, in highly coupled, small-scale scenarios such as green buildings, they still face challenges including difficulty in unifying semantic details and complex implementation mechanisms.

In communication and control, recent energy internet architectures advocate deep interconnection of different energy networks [25], promoting protocol fusion and transformation. Studies [26,27] have explored gateway technology, converting Modbus, OPC UA, and BACnet data into IEC 61850 models to enable interaction between distribution networks and building automation systems. Other studies [28,29] propose combining publish–subscribe mechanisms (e.g., IEC 61850 GOOSE) to broadcast critical events for rapid control and using edge computing to locally aggregate and process multi-source data, thereby reducing latency. However, multi-protocol fusion involves complex semantic mapping. If not properly handled, it may lead to risks such as data misinterpretation or control errors [30,31]. In green building scenarios with both wired LANs and wireless IoT devices [32,33], emerging 5G and Time-Sensitive Networking (TSN) technologies provide low-latency, high-reliability communication [34,35]. Applying these technologies to enhance heterogeneous protocol interaction is also a current research hotspot.

Overall, efficient coordination and intelligent control in green-building multi-energy supply systems require both a unified information model and a reliable real-time cross-protocol interaction mechanism. Existing IEC 61850 extensions mainly focus on electrical distributed energy resources and provide only limited device-level models for thermal and hydrogen components. CIM- and ontology-based approaches are primarily used for system-level planning and data integration rather than millisecond-scale control at the building edge. In addition, most multi-protocol gateway studies emphasize syntactic protocol translation and seldom provide a unified semantic layer or quantified end-to-end performance evaluation in building-scale scenarios.

Against this background, this paper makes three main contributions:

(1)

A unified multi-energy information model based on extended IEC 61850 logical nodes and standardized data objects that cover electrical, thermal, and hydrogen devices under a single semantic and unit system.

(2)

A cross-protocol real-time interaction mechanism that combines semantic gateways, GOOSE-based fast event distribution, and 5G/edge-assisted MMS optimization to support deterministic low-latency control over heterogeneous networks.

(3)

A digital-twin-based evaluation framework that quantifies plug-and-play capability, cross-protocol latency, robustness under extreme communication conditions, and resilience improvements under grid faults in a representative green-building multi-energy system. The following sections present the design and implementation of the proposed models and mechanisms and evaluate their performance using digital twin simulation.

2. Methods

2.1. Multi-Energy Unified Information Model Design

We design the unified information model to reuse IEC 61850 mechanisms as much as possible instead of introducing a completely new ontology. The main principle is to extend logical nodes and data objects only at key cross-energy coupling points, impose a global unit system for electrical, thermal, and hydrogen quantities, and keep all extensions fully compatible with standard SCL engineering tools as shown in Appendix A. In this way, the model can be deployed in practice without disrupting existing IEC 61850-based workflows while still providing the additional semantics required by green-building multi-energy systems.

2.1.1. IEC 61850 and IEC 61970 Model Mechanisms

When developing unified information models for multi-energy systems, two standard modeling frameworks can be referenced: IEC 61970 and IEC 61850. IEC 61970 defines the Common Information Model (CIM), which is widely applied in power dispatch and planning analysis. It provides high-level static data modeling but offers limited descriptions of real-time control and device-level details [19,20]. IEC 61850 employs a hierarchical structure of Logical Nodes (LNs) and Data Objects (DOs), designed for field device communication and enriched with real-time measurement and control semantics. Because green building multi-energy systems require fine-grained real-time monitoring and control, this study adopts IEC 61850 as the unified modeling framework. By extending LNs and DOs to incorporate thermal energy, hydrogen energy, and other elements not covered by CIM, it ensures semantic consistency and complementarity with CIM models.

2.1.2. Device Modeling with Extended IEC 61850 Logical Nodes

To incorporate electrical–thermal–hydrogen multi-energy devices in green buildings, the model design follows the IEC 61850-7-4 and 7-3 specifications. It extends and defines a set of new Logical Nodes (LNs) and Data Objects (DOs) to describe the monitoring and control functions of these devices. As shown in

Table 1. Examples of Extended Logical Nodes (LNs) and Data Objects (DOs).

Logical Node	Core Function	Key Data Objects (DOs)	CDC Type	Typical Application Scenario	Function and Definition Details
HFCLN	Hydrogen Fuel Cell Operation Monitoring	OutPwr (Output Power), FuelRte (Fuel Rate), EmgStop (Emergency Stop), etc.	MV/SPC	Distributed Power Generation Management	Cross-energy coupling: Supports power interaction with the grid through Elec2H2_Efficiency (electricity-to-hydrogen efficiency).
CHPLN	Integrated Control of Combined Cooling, Heating, and Power Systems	OutHeat (Heat Output), OutCool (Cooling Output), TotEfc (Total Efficiency), etc.	MV/ENG	Multi-energy Collaborative Optimization	Multi-energy coupling: TotEfc comprehensively calculates electricity, heat, and cooling output efficiency. Mode switching: CHPOpMod enumeration defines heat/electricity priority strategies.
ESSLN	Energy Storage System State Management	Voltage (Terminal Voltage), SoC (State of Charge), BatTemp (Temperature), etc.	MV/ASG	Energy Storage Charging/Discharging Strategies	Health monitoring: BatTemp combined with SoC evaluates battery health state. Charge/discharge direction: current positive/negative convention (discharge positive, charge negative) must align with converter logic.
ECLN	Cross-Energy Coupling Efficiency Modeling and Dynamic Interaction	InPwr (Input Power), OutPwr (Output Power), Efficiency (Conversion Efficiency)	MV/ASG/MX	Multi-energy Collaborative Scheduling	Dynamic coupling: Defines multi-directional conversion efficiency models (e.g., electricity-to-hydrogen, electricity-to-heat, heat-to-electricity), supporting real-time efficiency calibration and optimization.

, this study extends two logical nodes: Hydrogen Fuel Cell Monitoring (HFCLN) and Combined Cooling, Heating, and Power Control (CHPLN). These correspond to fuel cell/electrolysis equipment and combined cooling, heating, and power or heat pump systems in buildings. Each new LN defines data object sets tailored to the multi-energy characteristics of its devices, thereby providing a comprehensive representation of operating states and control variables.

In the above LNs, HFCLN denotes the monitoring of hydrogen fuel cells or electrolysis equipment. Key data objects include output power OutPwr (electrical output or consumption), hydrogen flow rate FuelRte (consumption rate in fuel cells or production rate in electrolyzers), and emergency stop status EmgStop (signal indicating emergency shutdown). These objects allow HFCLN to capture both the electrical and hydrogen behavior of devices, thereby incorporating power interactions between hydrogen equipment and the grid into the unified model. For instance, OutPwr may represent either the electrical power output from fuel cells to the grid or the power consumed by electrolyzers from the grid. FuelRte quantifies the hydrogen consumption or production rate. The data object Elec2H2Eff specifies electricity-to-hydrogen conversion efficiency, which is used to evaluate the performance of the conversion process. CHPLN is designed for combined cooling, heating, and power systems or heat pump equipment. Its main data objects include heating power OutHeat, cooling power OutCool, and total efficiency TotEfc. OutHeat and OutCool, respectively, represent the thermal and cooling power delivered to the building, whereas TotEfc indicates the overall efficiency of the device in converting primary energy into electricity, heat, and cooling. In addition, CHPLN defines an operating mode OpMode, which can be configured to modes such as “electricity priority” or “heat priority.” This enables dynamic switching of output strategies according to building electrical load or heating demand. For example, when the heat priority mode is selected during winter, CHPLN primarily ensures heating output [36,37]. Conversely, switching to the cooling priority mode in summer ensures adequate cooling supply. This mode-control modeling allows the energy management system to flexibly schedule and allocate building energy supply.

It is important to emphasize that all new LN and DO names and types strictly comply with IEC 61850 specifications, thereby ensuring compatibility with existing models. On the one hand, the Common Data Classes (CDCs) defined in IEC 61850 are adopted to specify data object types. For example, OutPwr adopts the Measured Value (MV) class, which includes attributes such as rated value and instantaneous value. EmgStop employs the Single Point Status (SPS) or Single Point Control (SPC) classes to represent binary quantities. OpMode utilizes the Status Enumeration (INS) or Control Command (DCS) classes to implement mode switching.

These particular logical nodes were selected because they coincide with the main cross-energy coupling points in typical green buildings. HFCLN captures the bidirectional interaction between hydrogen equipment and the electrical system; CHPLN represents devices that simultaneously provide electricity, heat, and cooling; ESSLN models electrochemical storage that can buffer fast fluctuations; and ECLN explicitly encodes multi-directional conversion efficiencies among electricity, heat, and hydrogen. Modelling these nodes at device level within IEC 61850 allows the proposed architecture to directly bind cross-energy semantics to real-time communication services, rather than relying on higher-level CIM classes that lack native support for millisecond-scale control.

On the other hand, each data object explicitly defines its physical meaning, units of measurement, and value ranges, thereby ensuring the model’s self-descriptive nature and semantic consistency across devices. For example, power-type data are uniformly expressed in kilowatts (kW), temperature in degrees Celsius (°C), and pressure in megapascals (MPa) (see the semantic anchoring section below for further details).

Finally, by extending these LN classes in the System Configuration Description (SCL) file and instantiating specific devices, seamless integration of new hydrogen and thermal energy devices with conventional distribution equipment is achieved within a unified model framework. Because the extensions strictly comply with IEC 61850 specifications, existing distribution equipment LNs (e.g., circuit breaker switch XSWI, relay protection PTRC) remain unchanged. Together with the newly defined LNs, they constitute a complete electrical–thermal–hydrogen unified information model, thereby establishing the semantic foundation for multi-energy device data collection and control.

2.1.3. Cross-Energy Semantic Unification and Data Object Standardization

Building upon the construction of new LNs, the unified information model must also address semantic heterogeneity among different energy-related physical quantities. Typical parameters in green building multi-energy systems include the following: on the electrical side, voltage, current, and active power; on the thermal side, temperature, thermal power, and medium flow rate; and on the hydrogen side, gas pressure, hydrogen concentration, and flow rate [38,39]. Different domains frequently employ different units of measurement and reference standards—for example, electrical power in kW, thermal power in GJ/h or kcal/h, and hydrogen flow in Nm³/h or kg/h. Direct integration of such data would hinder comparison or coordinated control due to dimensional inconsistencies. Therefore, this study proposes a globally unified unit system and semantic anchoring mechanism.

First, the IEC 61850 data class library is extended by defining corresponding standard data object types for each category of energy physical quantity, specifying their semantic interpretation and units of measurement, and providing predefined conversion relationships with other commonly used units. For example, HeatPow is defined to represent thermal power, uniformly adopting [kW_th] (thermal kilowatts) as the unit of measurement, with model annotations specifying that 1 kW_th equals 0.239 kcal of heat per second. Similarly, H2Flow is defined to represent hydrogen volumetric flow rate, uniformly expressed in [Nm³/h] (cubic meters per hour under standard conditions), with conversion coefficients provided for kg/h. Likewise, pressure is uniformly expressed in MPa, temperature in °C, and energy in kWh or MJ.

Subsequently, when establishing device LN models, all DOs involving the above physical quantities reference the corresponding standard data object types, thereby anchoring their physical meaning and units semantically. For example, the OutHeat data object type of CHPLN is defined as HeatPow, and the FuelRte type of HFCLN as H2Flow. Thus, regardless of the originating subsystem, identification as HeatPow explicitly denotes “thermal power” (unit kW_th), while identification as H2Flow denotes “hydrogen flow” (unit Nm³/h). This ensures consistent semantic interpretation by the monitoring center and avoids misjudgments caused by unit confusion or omission. Taking ECLN as an example, the extended key data objects are presented in

Table 2. Example of Key Data Object Definitions (ECLN).

DO Name	Description	Unit	CDC Type	Value Range/Enumeration
InPwr	Input power rate (multi-source type)	kW/kW_th/kg	MV	≥0 (supports electricity, heat, and hydrogen flow input)
OutPwr	Output power rate (multi-source type)	kW/kW_th/kg	MV	≥0 (supports electricity, heat, and hydrogen flow input)
Efficiency	Dynamic conversion efficiency	%	MX	0–100 (real-time calibrated value)
OpMode	Operating mode in multi-source coupling scenarios	—	ENG	{Hydrogen production mode, Power generation mode, Heat priority mode}
InputType	Identifier of input energy source type	—	SPS	{Electricity, Hydrogen, Heat}
OutputType	Identifier of output energy source type	—	SPS	{Electricity, Hydrogen, Heat}
Calibration Time	Timestamp for efficiency calibration	—	TSG	Timestamp (e.g., “2025-01-01 08:00:00”)

.

For ECLN, the dynamic conversion efficiency (Efficiency) is defined as the ratio of output power to input power, expressed as:

$$\eta (t)={\displaystyle \frac{\mathrm{OutPwr}(t)}{\mathrm{InPwr}(t)}}\times 100\%$$

(1)

where the range of η(t) is 0–100, corresponding to the real-time calibrated percentage efficiency values presented in Table 2. This Efficiency data object dynamically reflects energy conversion performance under different operating modes and is updated in real time during system operation.

In parallel, a cross-domain standard data object list was established to cover key state and control variables in electrical, thermal, and hydrogen energy systems, assigning each a unique definition across the entire system. These objects include Voltage (Volt), Current (Cur), Power (P), Energy (Energy), Temperature (Temp), Pressure (Pres), Flow (Flow), Valve Position (ValvePos), and Status Switch (Status), each with unique semantics and standardized units in the unified model, as shown in

Table 3. Cross-Domain Standard Data Object List and Unified Units.

Object Name	Unified Identifier	Semantic Definition (Unified Interpretation)	Unified Unit	Applicable Energy Domain	Typical CDC Type	Typical Mapping/Example DO (Including Extensions)
Voltage	Volt	Instantaneous or average measurement of phase/line voltage	V	Electrical	MV/MX	MMXU.PhV (phase voltages)
Current	Cur	Instantaneous or average measurement of phase currents	A	Electrical	MV/MX	MMXU.A (phase currents)
Active Power	P	Electrical active power (source/load)	kW	Electrical (including hydrogen-side electrical output)	MV/MX	MMXU.TotW; HFCLN.OutPwr; CHPLN.OutElec
Energy	Energy	Accumulated energy measurement (electrical/thermal)	kWh (recommended)/MJ (optional)	Electrical/Thermal	MV/MX	Meter accumulation TotWh; ThermalEnergy (extended)
Temperature	Temp	Medium or ambient temperature	°C	Thermal/Hydrogen	MV/MX	CHPLN.TempOut; TempBack (extended)
Pressure	Pres	Pipeline or tank pressure	MPa	Hydrogen/Thermal	MV/MX	HFCLN.TankPres (extended); PipePres (extended)
Flow (Liquid)	FlowL	Liquid medium volumetric flow rate	m3/h	Thermal	MV/MX	CHPLN.Flow (extended)
Flow (Gas, Standard State)	FlowG	Gas volumetric flow rate (at standard state)	Nm3/h	Hydrogen	MV/MX	HFCLN.H2Flow (extended); Electrolyzer.H2ProdRate (extended)
Valve Position	ValvePos	Position percentage of control valves/dampers and actuators	%	Thermal/Hydrogen	MV (measurement)/DCS (control)	CHPLN.ValvePos (extended); BAS.Valve.Pos
Status Switch	Status	Binary equipment/circuit status (open/close, fault, etc.)	0/1 (Boolean)	All domains	SPS (status)/SPC (control)	XSWI.Pos.stVal (circuit breaker); HFCLN.EmgStop

Note. CDC type descriptions: MV/MX—Measurement Value classes; SPS/SPC—Single Point Status/Control; DCS—Discrete Control Selection. The above mappings represent recommended configurations, which can be adjusted according to specific IED models during engineering implementation.

.

For example, P uniformly denotes active power (unit: kW), whether for grid feeder power measurement or fuel cell output power, both expressed as P in kW. Similarly, Temp uniformly denotes temperature (unit: °C), applicable to hot water pipe temperature or indoor temperature sensors, and Pres denotes pressure (unit: MPa), applicable to hydrogen storage tanks as well as hot water pipelines. Through these globally unified data definitions, information reported by different devices can be directly compared and aggregated at the semantic level.

For instance, when the fuel cell HFCLN reports OutPwr = 50 kW and the CHPLN reports OutElec = 30 kW (assuming CHPLN contains an OutElec object representing electrical output), the energy management system identifies both as power sources, yielding a total of 80 kW of electrical power supplied to building loads. Simultaneously, HFCLN’s hydrogen FuelRte can be compared with CHPLN’s gas consumption via calorific value conversion to evaluate the equivalent thermal efficiency of different energy carriers.

To ensure that the above semantic consistency is effectively enforced during system operation, a semantic knowledge graph module was implemented on the digital twin information platform. The specific approach is illustrated in

Table 4. Semantic Consistency Checking Rules and Disposal Strategies (Examples).

Rule Category	Rule Content (Example)	Trigger Condition	Disposal Strategy
Unit Consistency	Measurement points must adopt unified units: P → kW; Temp → °C; Pres → MPa; FlowG → Nm3/h	Unit missing or inconsistent with semantic anchor	Issue alarm; perform automatic unit conversion and re-labeling
Range Verification	Values must remain within engineering-reasonable ranges: Temp ∈ [−40, 120] °C; Pres ∈ [0, 25] MPa	Out of bounds or sudden change exceeding threshold	Issue alarm; apply limiting rules or activate backup sensors
Type Matching	Data type must be consistent with CDC definition: MV/MX for measurements, SPS/SPC for binary status/control	Type mismatch	Reject the item; generate log record
Semantic Anchoring	Measurement point semantics must correspond to ontology nodes (e.g., CHPLN.OutHeat ∈ Thermal Power)	Semantic conflict	Issue alarm; trigger mapping rule review/correction
Time Synchronization	Complete timestamps must satisfy error ≤ 1 μs (PTP standard)	Missing timestamp or exceeding tolerance	Issue alarm; request resend or enforce time alignment

. Each defined data object type serves as a node in the ontology graph, establishing equivalence relationships, unit conversion relationships, and energy-type attribute associations among them. During system operation, real-time data are mapped to instances in the knowledge graph, where semantic consistency is automatically verified using SPARQL queries or inference engines. If data are found to deviate from unified model conventions (e.g., missing units or values exceeding reasonable ranges), the system issues alerts or performs the necessary unit conversion and compensation, thereby ensuring the completeness and correctness of global model semantics. This mechanism enables the energy management system to clearly interpret and accurately utilize data from any subsystem.

Additionally, for complex data, a composite data object design is adopted to optimize transmission efficiency and ensure consistency among associated data. As illustrated in

Table 5. Composite Data Object Definitions and DataSet Packaging (Examples).

Composite Object	Included Attributes (Unified Identifiers)	Unified Units	Bound CDC/DO Types (Examples)	DataSet Packaging Example
ThermalStatus	OutHeat, TempOut, TempBack, Flow	kW_th, °C, °C, m3/h	MV: CHPLN.OutHeat; MV: CHPLN.TempOut; MV: CHPLN.TempBack; MV: CHPLN.Flow	DS_ThermalStatus = {OutHeat, TempOut, TempBack, Flow}
HydrogenStorage	TankPres, TankTemp, H2Level (%)	MPa, °C, %	MV: HFCLN.TankPres; MV: HFCLN.TankTemp; MX: HFCLN.H2Level	DS_H2Storage = {TankPres, TankTemp, H2Level}

, for thermal system states associated with CHPLN, the composite object ThermalStatus is defined, incorporating attributes such as supply water temperature, return water temperature, instantaneous flow rate, and thermal power [39]. This design enables the attributes to be packaged and reported as a whole, thereby avoiding data desynchronization caused by dispersed transmission of individual parameters. For hydrogen storage devices, the composite object HydrogenStorage is defined, including attributes such as storage tank pressure, temperature, and hydrogen content percentage [40].

Through the dataset mechanism of the IEC 61850 MMS service, these composite objects can be transmitted in a single message, thereby reducing additional latency associated with multiple interactions. This hierarchical data organization method is particularly suitable for green building scenarios characterized by large data volumes, diverse categories, and strong interdependencies. It enables the unified model to structurally maintain high efficiency and reliability while simultaneously achieving semantic unification.

2.1.4. Ontology Representation and Comparison with CIM and Energy Ontologies

From an implementation viewpoint, the unified information model can be represented as an ontology-style graph in which each IEC 61850 logical node class corresponds to an ontology class and each data object corresponds to a datatype or object property. Domain tags indicate whether a quantity belongs to the electrical, thermal, or hydrogen subsystem, and relationships such as “equivalent-to”, “part-of”, and “derived-from” are used to capture cross-energy couplings. The semantic knowledge-graph engine described above directly instantiates this ontology: for example, HFCLN. OutPwr is anchored as an instance of the generic active-power concept P with unit kW, while CHPLN.OutHeat and CHPLN. OutCool is anchored as a thermal-power concept with unit kW_th. Unit conversions, valid ranges, and composite objects such as ThermalStatus and HydrogenStorage are likewise encoded as ontology rules and constraints.

Compared with CIM (IEC 61970) and existing energy and building ontologies, the proposed model targets a different level of abstraction that is closer to device-level control. CIM focuses on grid-level assets and is primarily used for planning and operation studies in transmission and distribution networks, whereas open energy ontologies emphasize data interoperability across tools and datasets. In contrast, our extended IEC 61850 model provides fine-grained semantics that are natively bound to field communication services (MMS, GOOSE, SV), explicitly model multi-directional cross-energy conversion via ECLN, and can be implemented directly in IED configuration files (SCL). CIM and ontology-based models can still be used at the supervisory or market level; the proposed model complements them by supplying a semantically consistent, real-time information layer at the building edge. Figure 2 illustrates the ontology-level representation of the unified information model and its relation to CIM/IEC 61970 [14] and generic energy ontologies.

2.2. Cross-Protocol Real-Time Information Interaction Mechanism

The communication architecture builds on this unified model by clearly separating semantics from protocol syntax. All heterogeneous field devices first map their measurements and controls into the unified IEC 61850 information space via a semantic gateway, and only then are appropriate transport services selected. MMS is used for routine monitoring and control, GOOSE and SV for fast events and high-rate measurements, and 5G-assisted edge computing for dynamic optimization of traffic patterns. This layered design allows protocol-specific optimizations to evolve without changing the underlying semantic model.

Building upon the unified information model, semantic mapping and fusion communication mechanisms were designed for heterogeneous protocols to enable plug-and-play functionality and high-speed coordinated control of multi-energy devices. These mechanisms consist of two core components: a multi-protocol adaptive semantic gateway and a GOOSE/edge-computing fusion communication strategy. The underlying concept is to translate data from heterogeneous protocol terminals into the semantics of the unified model via semantic gateways at the device access layer, while employing publish–subscribe fast messaging and edge optimization technologies at the communication layer to ensure reliable, low-latency information exchange across heterogeneous networks.

2.2.1. Multi-Protocol Semantic Mapping and Adaptive Gateway

To address the diversity of communication protocols in green buildings, a semantic gateway was developed with preset cross-protocol data point mapping rules, enabling bidirectional conversion between heterogeneous protocol data and the IEC 61850 unified model. Representative protocols and application scenarios are summarized in

Table 6. Protocol–Semantic Mapping Matrix and Interface Key Points (Examples).

Protocol/Interface	Typical Equipment/Scenario	Source Data Points (Examples)	IEC 61850 Target LN.DO (Unified Semantics)	Direction	Service/Call Method	Notes (Units/Time Sync)
IEC 61850 MMS	IED/Protection/Monitoring/Meters	MMXU.TotW, MMXU.PhV, XSWI.Pos	MMXU.TotW → P (kW); MMXU.PhV → Volt (V); XSWI.Pos → Status	Bidirectional	MMS Report/Read-Write Control	Native timestamp; PTP/IRIG-B alignment
BACnet/IP	BAS/HVAC	AnalogInput: ZoneTemperature, BinaryOutput: ValveOpen	CHPLN.TempOut → Temp (°C); CHPLN.ValvePos → ValvePos (%)	Bidirectional	ReadProperty/WriteProperty	Temperature unit unified to °C; valve position ratio linearization
Modbus TCP/RTU	Sensors/Actuators/CHP Controllers	Register 40001 = Active Power, 30005 = Temperature	CHPLN.OutElec → P (kW); CHPLN.TempOut → Temp (°C)	Bidirectional	Read/Write Holding Registers	Register mapping template; scale/offset conversion
IEEE 2030.5 (SEP2.0)	PV Inverters/Charging Stations/DER	DERStatus.ActivePower, Storage.SOC	PV.OutPwr → P (kW); ESS.SOC → SoC (%)	Bidirectional	DERControl/Resource REST	CIM semantics approximation, consistency check before storage
MQTT/REST	IoT Terminals/Vendor Cloud	JSON: {“p”:500, “t”:25, “ts”:…}	Map → P (kW), Temp (°C), ts → Timestamp	Bidirectional	Topic Subscribe/HTTP POST	Requires units and UTC timestamp
OPC UA (Publishing)	Gateway → Upper Digital Twin/EMS	Unified model data nodes	—(Semantic unification completed at gateway)	Uplink	UA Subscription/MonitoredItem	Unified publishing point; data caching/batch push
GOOSE	Emergency Events/Interlocking/Fast Action	EmgStop, Trip, EID	HFCLN.EmgStop → Status; Event EID → Enumeration	Up/Down	Publish-Subscribe	Millisecond level; URLLC slice optional guarantee

[41,42,43,44]. Examples include: IEC 61850 MMS for intelligent distribution terminals and protection devices; Modbus for building automation sensors, actuators, or CHP controllers; BACnet/IP for building HVAC systems; IEEE 2030.5 (SEP2.0) for internet-based communication of distributed photovoltaics and charging stations; and MQTT/REST for selected IoT terminals.

The gateway maintains a mapping table that establishes one-to-one correspondences between IEC 61850 logical nodes/data objects and information elements of other protocols. As illustrated in

Table 7. Typical Semantic Mapping Rule Library Entries (Selected).

Rule ID	Rule Type	Source Example	Target LN.DO (CDC)	Conversion/Verification Logic	Notes
R-001	Unit Conversion	BACnet. ZoneTemperature (°F)	CHPLN.TempOut (MV)	Temp (°C) = (°F − 32) × 5/9; Label unit	Threshold [−40, 120] °C out-of-range alarm
R-002	Scale/Offset	Modbus 40001 (0.1 kW)	MMXU.TotW (MV)	P (kW) = RegValue × 0.1	Negative value anomaly blocking
R-003	Enumeration Mapping	DERControl. Mode = 2	CHPLN.OpMode (ENG)	2 → HeatFirst; Other mappings see Enum_OpMode	Unknown enumerations to blacklist
R-004	Boolean Inversion	BinaryOutput: ValveOpen (1 = Closed)	CHPLN. ValvePos (DCS/MV)	Pos (%) = (1 − raw) × 100	Vendor difference guidance confirmation
R-005	Bit Field Parsing	Modbus 30010 (bit3 = Trip)	XSWI.Pos.stVal (SPS)	Trip = (Reg >> 3) & 1	Record trigger timestamp
R-006	Time Alignment	MQTT JSON ts (local time)	Any measurement point (t)	Convert to UTC uniformly; Reject if missing and request resend	PTP drift > 1 μs alarm
R-007	Semantic Anchoring	2030.5 Storage.SOC	ESS.SOC (MV)	Value range [0, 100]%; CDC consistency check	Out-of-range correction/alarm

, for the BACnet protocol, the gateway maps object properties such as ZoneTemperature from building temperature control systems to corresponding Temp measurement values in the IEC 61850 model. For Modbus devices, register addresses are mapped to corresponding DOs under IEC 61850 LNs, based on preconfigured relationships between register addresses and data identifiers [45]. For inverters or charging stations supporting IEEE 2030.5, which adopt CIM semantics, the gateway maps IEEE 2030.5 distributed energy status fields (e.g., active power, storage SOC [46]) to related objects in the IEC 61850 unified model.

Mapping is bidirectional. When devices upload data, the gateway converts it into the IEC 61850 model format according to predefined rules and updates the unified information base [47,48,49,50]. Conversely, when the energy management system issues control commands based on the unified model (e.g., setting CHPLN’s OpMode to “electricity priority”), the gateway translates them into the corresponding protocol-specific command formats (e.g., Modbus write-register commands, BACnet WriteProperty service calls), thereby enabling devices to execute the required operations.

Through this semantic gateway, upper-level scheduling and control are decoupled from the underlying communication protocols, thereby enabling transparent access and unified control of devices using different protocols.

To support plug-and-play integration of multi-vendor devices, the semantic gateway incorporates adaptive mapping configuration capabilities. When new devices are connected, the gateway automatically identifies their protocol type and basic model information and attempts to match them against existing mapping rule libraries, as summarized in

Table 8. Adaptive Access and Mapping Configuration Process with KPIs (Examples).

Step	Trigger/Input	System Operation (Automatic/Semi-Automatic/Manual)	Matching Rules/Heuristics	Output/Product	Target KPI (Typical)
1. Device Identification	New device online (message/scan)	Automatic identification of protocol and basic model (SCL/BACnet objects/register range)	Protocol fingerprint; SCL parsing; Object browsing	Initial device profile	≤100 ms for protocol identification
2. Rule Matching	Device metadata, point list	Automatic matching with unified model mapping rule library	Name similarity, units/dimensions, context (device type)	Candidate mapping scheme	≤300 ms to generate mapping suggestions
3. Semantic Verification	Candidate mapping	Automatic consistency check (units, types, ranges)	Semantic anchoring/unit conversion/CDC consistency check	Verification results and alerts	≤200 ms to complete verification
4. Wizard Confirmation	Unknown/conflicting entries	Manual selection of target DO or creation of extended DO (one-time)	Template-guided/blacklist/whitelist	Final mapping rules	3–5 min (only for few new device types)
5. Activation and Publishing	Mapping rule solidification	Hot loading and start data publishing/command translation	Seamless switching/no interruption	Unified model data flow	<1 s to go online
6. Iterative Refinement	Actual operation data	Automatic recording and rule optimization (learning common mappings)	Closed-loop learning/statistical analysis	Rule library evolution	Weekly/monthly version updates

. For IEC 61850 devices, SCL files are directly parsed to extract LN and dataset definitions. For BACnet devices, object property lists are discovered through network browsing. For Modbus devices, device types are initially inferred from register address ranges or user-provided templates.

Subsequently, the gateway performs semantic matching between new device data points and the unified model. If corresponding standard objects are identified, mapping associations are established automatically. If unknown data items are detected, a mapping configuration wizard is triggered, prompting operations personnel to assign the data item to appropriate objects in the unified model (either by selecting existing DOs or creating new ones), and to add the configuration to the rule library for future reuse.

The entire process is designed to be highly efficient and flexible: automatic matching is typically completed within milliseconds, manual intervention is required only in rare cases involving entirely new devices, and configuration can usually be finalized within minutes.

In simulation testing, when a new hydrogen fuel cell HFCLN device was added during operation, the gateway completed model loading and began publishing data in less than one second. After the removal of a CHP device, communication interruption was detected within approximately one second, and the upper model was notified to mark it as invalid. These results verify the gateway’s rapid response capability for device access and exit.

Notably, the semantic gateway is responsible not only for protocol conversion but also for field-level data aggregation and distribution. All data from heterogeneous protocol devices are first converted into the unified model format and cached in the gateway and are then uniformly published to upper-layer digital twin management servers via the OPC UA server or MMS reports. Conversely, control commands from upper layers are translated by the gateway before being delivered to devices.

This architecture enables single-point aggregation and unified publication of multi-source data, thereby reducing the number of communication links and creating favorable conditions for subsequent centralized optimization via edge computing.

2.2.2. Cross-Domain Fast Communication Strategy Integrating GOOSE

Generic Object-Oriented Substation Event (GOOSE) messages, as defined by IEC 61850, employ a publish–subscribe paradigm to achieve millisecond-level event transmission over connectionless Ethernet and are widely applied for rapid blocking of substation protection actions. This study extends the GOOSE mechanism to cross-domain real-time control in green building multi-energy systems, enabling the transmission of emergency control commands and critical event triggers to compensate for the limitations of conventional request–response communication.

The proposed strategy is as follows: GOOSE clients and servers are deployed at semantic gateways and critical control nodes. When events requiring multi-subsystem coordination occur (e.g., grid faults or voltage drops), event triggers immediately generate GOOSE messages within the gateway, which are then transmitted to all subscribing devices via LAN or 5G wireless networks. First, unified event encoding is defined for critical cross-energy domain events—for example, EID_H2Fault represents hydrogen system faults and EID_PWRDrop represents grid power drops. These identifiers, along with necessary parameters, are attached to the GOOSE message dataset. Upon parsing the event identifiers, receiving devices trigger corresponding control measures according to pre-agreed protocols. For instance, after receiving EID_PWRDrop, the air-conditioning BAS system immediately enters energy-saving mode to reduce load.

Second, by integrating 5G Ultra-Reliable Low-Latency Communication (URLLC) technology, GOOSE messages are transmitted with high reliability through 5G network slicing. GOOSE employs a repetitive sending mechanism, achieving extremely low packet loss rates even in local LANs. However, when transmission over wireless wide-area networks is required, dedicated 5G slicing provides QoS guarantees, ensuring that the first GOOSE frame is delivered to subscribing terminals within 10 ms.

In addition to GOOSE, SV (Sampled Values) high-speed measurement streams are introduced for data acquisition in multi-energy systems. SV messages can transmit instantaneous sampled values of current, voltage, and other quantities at a frequency of 4 kHz. For data requiring high-frequency monitoring (e.g., voltage waveforms or gas pipe pressure fluctuations), the IEC 61850-9-2 SV service is enabled at critical sensors. Synchronized clocks trigger simultaneous sampling across devices from different domains, generating measurement sequences with unified timestamps that are transmitted to the gateway. This approach enables precise alignment of dynamic data from electrical, thermal, and hydrogen domains on the time axis, thereby providing a consistent foundation for comprehensive state perception, analysis, and coordinated control.

If abnormal events occur, an adaptive sampling frequency method is employed. Under normal conditions, SV data are transmitted at lower frequencies (e.g., 1 kHz). Once sudden events are detected, the sampling rate and transmission priority of critical quantities are automatically increased to provide more detailed data in the short term, before returning to normal frequency. This SV scheduling strategy ensures the real-time availability of critical information while keeping routine communication volumes within acceptable ranges.

2.2.3. MMS Communication Optimization Based on 5G and Edge Computing

Large volumes of routine monitoring data and control commands must still be transmitted through IEC 61850’s MMS (Manufacturing Message Specification) service. MMS follows a client–server model and undertakes communication tasks such as periodic reporting of device measurements, event reporting, and remote control in integrated energy management systems. However, as green buildings integrate multiple types of energy devices, the volume and complexity of MMS communication increase significantly. When subsystems access the network via wireless links, problems such as insufficient bandwidth and latency fluctuations may arise without optimization. Therefore, this study proposes dynamic MMS optimization strategies supported by 5G and edge computing.

First, network slicing technology is enabled on the 5G network side to differentiate traffic flows by priority. MMS communication is divided into critical control flows and general data flows. Critical MMS messages involving protection or coordinated control (e.g., emergency commands or important status reports) are allocated to URLLC ultra-reliable low-latency slices, while routine monitoring data and logs are transmitted through eMBB enhanced-bandwidth slices. Slice isolation ensures that critical traffic maintains stable low latency and low packet loss rates even under network congestion, while large volumes of non-critical data do not occupy critical resources.

Second, MEC (Multi-access Edge Computing) servers are deployed near the field as MMS proxies for intelligent communication regulation. The edge proxy aggregates field device data for caching and preprocessing, while simultaneously sensing real-time 5G link status to dynamically adjust communication strategies. For example, intelligent filtering and compression may be applied to uploaded data: reducing the reporting frequency for steadily changing measurements and transmitting only when thresholds are exceeded; or applying batch packaging and fast Fourier transform–based feature extraction for frequently fluctuating data before reporting, thereby reducing communication volume. Meanwhile, the edge proxy monitors 5G signal quality in real time and adaptively adjusts message transmission intervals and redundancy mechanisms if latency increases or packet loss rates are detected—for example, by extending reporting intervals for non-critical data, introducing Forward Error Correction (FEC) encoding, or increasing retransmission attempts for critical commands.

MEC proxies can also directly execute partial coordinated control logic to enhance response speed. For example, when edge nodes detect distribution faults with excessive air-conditioning loads, they can issue temporary load-reduction commands prior to cloud commands, thereby preventing local instability. This cloud–edge collaborative communication architecture significantly enhances the real-time performance and robustness of system communication in heterogeneous networks.

In summary, the proposed cross-protocol interaction mechanism achieves unified access to multi-protocol devices through semantic gateways, ensures high-speed transmission of critical information via GOOSE/SV combined with 5G network slicing, and optimizes routine MMS communication through edge computing. The resulting communication system provides green building multi-energy systems with fast and reliable information interaction under both normal and abnormal conditions, thereby offering strong support for coordinated control of diverse energy subsystems.

3. Simulation Platform and Test Scheme

3.1. Simulation Platform Construction

To verify the effectiveness of the proposed unified information model and cross-protocol interaction mechanism in practical applications, this study constructed a comprehensive digital twin simulation platform for green building multi-energy systems. The platform employs an architecture that integrates physical-layer real-time simulation with information-layer communication control, enabling realistic simulation of the dynamic coupling of electrical, thermal, and hydrogen energy in buildings and their information interaction characteristics, as illustrated in Figure 2.

The physical-layer simulation environment is developed on the RTDS (Real-Time Digital Simulator), which accurately simulates electromagnetic transients and electromechanical dynamics of building energy systems with a simulation time step of 0.05 ms. The system comprises a 10 kV distribution bus supplying the building’s 0.4 kV low-voltage side through a 500 kVA transformer.

Multiple distributed energy devices are configured within the building: a photovoltaic system with a rated power of 20 kW adopting an MPPT control algorithm with 98% efficiency; a lithium battery storage system with a capacity of 100 kWh and a rated charge/discharge power of 20 kW, equipped with bidirectional converters; a hydrogen fuel cell with a rated power of 10 kW, consuming hydrogen at 2.0 m³/h under full load; an electrolysis unit with a rated power of 15 kW, producing hydrogen at 3.2 Nm³/h under rated conditions; and a combined cooling, heating, and power (CCHP) unit with 30 kW electrical output, 40 kW thermal output, and 85% overall efficiency. The building’s base electrical load is set to 20 kW, with an additional adjustable load of 0–30 kW for simulating demand fluctuations. RTDS achieves hardware-in-the-loop communication with external systems via GTNET interface cards, supporting IEC 61850 GOOSE and MMS protocols. Message transmission periods are flexibly configurable from 1 to 1000 ms (1–4 ms for GOOSE and 100–1000 ms for MMS).

The thermal system adopts lumped-parameter modeling, with a DN100 heating pipeline designed for a rated flow of 15 m³/h and a supply–return temperature difference of 20 °C. The heat pump system provides 30 kW of heating power with a coefficient of performance (COP) of 3.5. The thermal storage tank has a volume of 10 m³ and adopts temperature stratification models to accurately represent storage characteristics. Thermal loads include a 20 kW base load and a 0–40 kW peak load, with time constants ranging from 10 to 300 s to represent thermal inertia characteristics.

For the hydrogen system, the storage tank has a volume of 5 m³ with a working pressure range of 0–25 MPa and is equipped with a temperature monitoring system covering −40 to 85 °C. The hydrogen pipeline adopts a DN25 specification with a design flow of 0–10 Nm³/h. Safety systems include a pressure relief valve (triggered at 25 MPa) and an emergency cutoff valve (response time < 10 ms), ensuring safe operation of the system.

The information-layer communication architecture is centered on a multi-protocol semantic gateway deployed on an industrial control computer equipped with an Intel Xeon E-2288G processor and 32 GB ECC memory, running a real-time Linux operating system with kernel scheduling latency controlled within 100 μs. The semantic gateway integrates a complete protocol stack, including: an IEC 61850 server supporting MMS, GOOSE, and SV communication; a BACnet/IP stack implementing 20 standard device object types; a Modbus TCP/RTU master capable of accessing up to 65,535 register addresses with a 100 ms polling cycle; an IEEE 2030.5 client supporting distributed energy management, power control, and event subscription; an MQTT client supporting three quality-of-service levels; and an OPC UA server capable of managing 10,000 data nodes with sampling periods configurable from 10 ms to 10 s. The gateway’s built-in mapping rule library preconfigures 500 standard mappings covering mainstream device types, achieving an automatic matching success rate of over 85% for new device access and significantly reducing system integration complexity.

The communication network is based on a laboratory-built 5G testbed supporting NSA and SA dual-mode architectures, with its core network built on the OpenAirInterface open-source platform. The network operates in the n78 band (3.5 GHz) with a 100 MHz bandwidth and is configured with two slices: a URLLC slice allocated 1 Mbps guaranteed bandwidth with target latency <5 ms and 99.999% reliability, and an eMBB slice sharing the remaining 99 Mbps bandwidth with average latency controlled within 50 ms. An edge computing server is deployed at the gNodeB base station, equipped with a 4-core ARM processor and 8 GB memory, achieving a round-trip latency to the base station of less than 2 ms. The entire system employs the IEEE 1588v2 [6,7] Precision Time Protocol for clock synchronization, achieving ±1 μs accuracy and ensuring time consistency in distributed data collection. The cloud digital twin server is configured with a 48-core processor and 256 GB memory and runs a Neo4j-based knowledge graph engine containing 1200 semantic nodes and 3500 relationship edges for real-time verification of data semantic consistency.

To ensure accuracy and repeatability, rigorous baseline testing and calibration were conducted on the simulation environment. Local Ethernet communication latency averaged 0.2 ms with a standard deviation of 0.05 ms over 1000 ping tests. The 5G URLLC slice latency was 4.8 ± 0.8 ms under no-load conditions. Protocol conversion overhead tests showed that IEC 61850–to–BACnet conversion required 1.2 ms, while Modbus–to–IEC 61850 conversion required 0.8 ms. For power measurements, electrical accuracy reached ±0.5% with 0.1 kW resolution; thermal power, calculated from flow and temperature difference, achieved ±2% accuracy; and hydrogen flow, measured under standard conditions, achieved ±1% accuracy. These calibration results provide reliable baselines for subsequent performance evaluation. In all tests, IEEE 1588v2 Precision Time Protocol with ±1 μs nominal accuracy is used for time synchronization; electrical quantities are sampled at 1 kHz, thermal quantities at 10 Hz, and communication events are logged with a 1 ms time resolution.

3.2. Test Scheme Design and Implementation

To comprehensively validate the proposed approach, we designed four complementary test sets. Test 1 examines device plug-and-play behavior and the semantic consistency of the unified information model. Test 2 evaluates cross-protocol coordinated control under typical load disturbances. Test 3 stresses the communication stack under extreme network conditions, including high channel load, interference, and link failures. Test 4 assesses the multi-energy system’s resilience under grid faults. Together, these tests form a hierarchical validation workflow that progresses from data-model correctness (Test 1) through communication robustness (Tests 2 and 3) to system-level resilience (Test 4).

3.2.1. Test 1: Device Plug-and-Play and Semantic Consistency Testing

This test aims to verify the capability of the unified information model to support dynamic management of heterogeneous devices. The test schematic is illustrated in Figure 3. Initially, the system operates in a steady state with photovoltaic output of 12 kW, storage discharging at 10 kW, and the CHP unit producing 25 kW. The total supply of 45 kW meets the total load demand of 44.5 kW (including 0.5 kW of line losses).

At t = 0.8 s, a new hydrogen fuel cell (HFCLN) device with a rated power of 10 kW is simulated to join the system, operating at 8.5 kW with a hydrogen consumption rate of 2.1 Nm³/h. The system is required to complete device discovery within 0.8 s after access, load the SCL configuration file within 0.3 s, and begin publishing device data within a total time not exceeding 1.6 s.

During the access process, the knowledge graph engine continuously verifies semantic consistency, checking 120 data points over a 3.2 s test period, including 48 from the electrical domain, 42 from the thermal domain, and 30 from the hydrogen domain. Verification items include unit consistency (all power data in kW, temperature in °C, pressure in MPa, and gas flow in Nm³/h), range validity (temperature between −40 and 120 °C, pressure between 0 and 25 MPa), and timestamp alignment accuracy (deviation < 1 μs). The expected semantic consistency pass rate is above 99%, with power balance calculation error controlled within 0.5%.

At t = 2.5 s, a CHP device fault is simulated, resulting in termination of its MMS communication. The test evaluates whether the system can detect the failure within 0.1 s, update the model status, and ensure that data flows from other devices remain unaffected.

3.2.2. Test 2: Cross-Protocol Coordinated Control Effect Testing

This test evaluates the capability of the unified model framework to coordinate control among devices with different communication protocols. The test schematic is illustrated in Figure 4. The test scenario is configured such that a chiller unit starts suddenly at t = 0 s, producing a 20 kW load step. The system is required to coordinate multiple heterogeneous devices to maintain power balance.

The scheduling strategy specifies the following actions: the photovoltaic inverter (via IEEE 2030.5) increases from 12 kW to its full output of 20 kW, adding 8 kW; the battery storage system (IEC 61850) switches from standby to discharge mode, providing 20 kW; the CHP unit (Modbus) increases from 25 kW to 30 kW and switches its operating mode from “heat priority” to “electricity priority,” adding 5 kW; and the BAS (BACnet) temporarily shuts down air conditioning in selected non-critical areas, reducing load by 5 kW.

The test records the delay from control command issuance to actual device execution, along with the dynamic power output curves of each device. During steady-state operation, the system maintains frequency at 50.00 ± 0.01 Hz and voltage at 1.0 ± 0.01 pu. After the disturbance, the system’s dynamic frequency and voltage responses are monitored, recording maximum deviations and recovery times. To ensure statistical reliability, the test is repeated 100 times to analyze delay distribution characteristics of each protocol path.

Performance evaluation requires that the response-time difference between the fastest and slowest devices does not exceed 0.06 s, power balance error remains below 0.5 kW, maximum frequency deviation does not exceed 0.15 Hz, and the system returns to stable operation within 0.15 s.

3.2.3. Test 3: Extreme Communication Condition Robustness Testing

This test evaluates the robustness of the system’s communication mechanisms by simulating extreme conditions, including high load, strong interference, and link failures. The test schematic is illustrated in Figure 5.

In the high-load test, background data flow is gradually increased from 20% to 95% channel occupancy, with each load point maintained for 60 s. Latency variations are recorded separately for ordinary MMS messages, critical MMS messages with URLLC slice guarantees, and GOOSE first-frame transmission times.

In the strong-interference test, the channel bit error rate is gradually increased from 10⁻⁶ to 10⁻³. Performance is compared between an unoptimized standard transmission scheme and an optimized scheme employing Reed–Solomon forward error correction together with an adaptive retransmission mechanism (up to three retransmissions).

In the fault-recovery test, the main communication link is deliberately disconnected at t = 10 s, after which the MEC edge server is restarted within 100 ms. The test monitors GOOSE message switching time through the backup link, MMS reconnection time, and data loss during the outage.

Performance benchmarks require that GOOSE message latency does not exceed 0.01 s at the 95% confidence level and that critical MMS messages with URLLC slicing remain below 0.03 s. Even under extreme interference conditions with bit error rates up to 10⁻³, the successful transmission rate of critical data should remain above 90%. Link-failure recovery time should be within 0.1 s, ensuring that control continuity is maintained.

3.2.4. Test 4: Grid Fault Condition Resilience Testing

This test simulates three-phase short-circuit faults in the distribution network to evaluate the coordinated self-healing capability of the multi-energy system. As illustrated in Figure 6, the fault is initiated at t = 0 s on the 10 kV bus, lasting 0.1 s before protection devices clear it. However, the mains supply is completely interrupted, requiring the building to transition to islanded operation mode.

The system’s preset resilience control strategy operates in three phases. (i) Rapid isolation: GOOSE messages broadcast fault signals to all relevant devices, and tie switches trip immediately to achieve physical isolation between the building microgrid and the faulted grid. (ii) Emergency response: Hydrogen fuel cells execute an emergency shutdown to prevent equipment damage, the storage system switches to maximum 20 kW discharge mode, and the BAS reduces 8 kW of non-critical load. (iii) Stable recovery: The CHP unit ramps up to 15 kW full-load operation, hydrogen fuel cells complete voltage stabilization checks before reconnecting to provide 10 kW power support, gradually restoring a new power balance.

For control comparison, traditional schemes in which subsystems operate independently without cross-domain coordination mechanisms are also tested. Resilience evaluation spans multiple dimensions: Voltage resilience is assessed by recording minimum voltage values and the recovery time to 0.95 pu; frequency resilience is measured by the maximum frequency deviation and stabilization time; power balance is evaluated by the percentage deviation between supply and demand; response speed is analyzed through the action sequences of various devices; and a comprehensive resilience index is obtained through weighted averaging.

Testing requires electrical quantities to be recorded at a 1 kHz sampling rate and thermal quantities at 10 Hz, covering the complete process from 1 s before the fault to 5 s after the fault. Each scenario is repeated 20 times to ensure statistical significance.

Data analysis employs rigorous statistical methods, including calculation of mean values, standard deviations, and 95% confidence intervals; normality verification using Shapiro–Wilk tests; and correlation analysis with Pearson or Spearman coefficients. Performance comparisons are conducted using paired t-tests to evaluate the significance of differences before and after optimization, with Cohen’s d values calculated to quantify effect sizes. All improvement rates are calculated using the formula: (after optimization − before optimization)/before optimization × 100%.

Test results are presented through time-series curves of system dynamics, statistical charts of latency distribution and packet loss comparisons, and radar charts evaluating multi-dimensional resilience indicators, thereby providing comprehensive data support for validating the effectiveness of the proposed methods.

3.3. Conceptual Multi-Building Extension Scenario

To illustrate how the proposed framework scales beyond a single building, we outline a conceptual multi-building scenario. Multiple green buildings, each equipped with photovoltaic generation, storage, CHP units, and hydrogen equipment, are connected to a common 10 kV feeder. Every building hosts a local semantic gateway that implements the unified IEC 61850-based information model and semantic mapping towards its internal BAS and field devices, while a campus-level coordination layer subscribes to selected data points via OPC UA and IEC 61850 reports.

In this setting, the unified information model ensures that key quantities such as net active power, available flexibility, state of charge, and hydrogen storage levels are represented in the same semantic and unit space for all buildings. The cross-protocol interaction mechanism remains unchanged: local gateways handle legacy protocols, and campus-level applications interact only with the unified semantic model. Typical use cases include campus-wide demand response, shared hydrogen storage management, and coordinated islanding of several buildings under grid-fault conditions.

Although a full quantitative evaluation of campus-scale performance is beyond the scope of this study, the architecture shows that no changes to the underlying model or gateway logic are required when moving from one building to multiple buildings. Scalability is mainly limited by communication and computation capacity at the semantic gateways and edge nodes, which can be increased by deploying hierarchical gateway clusters and distributing coordination tasks across multiple MEC servers.

4. Simulation Results and Discussion

We collected extensive experimental data on the digital twin simulation platform and analyzed the key results below.

Unless otherwise noted, each test compares the proposed unified-model-based architecture with a conventional baseline in which the electrical, thermal, and hydrogen subsystems are monitored and controlled by separate domain-specific systems without a shared semantic model or semantic gateway. In the baseline, any protocol conversion that is required is limited to simple field-to-field mappings without unit unification or knowledge-graph checks; MMS traffic is carried on a best-effort network without slicing or edge optimization; and no GOOSE-based cross-domain events or fault-ride-through strategies are configured. This baseline reflects common practice in existing building energy management systems and provides a meaningful reference for quantifying the benefits of the proposed framework.

4.1. Test 1 Results

Test 1 validated the effectiveness of the unified information model in supporting dynamic device access. As illustrated in Figure 7a, when a 10 kW hydrogen fuel cell (HFCLN) device was connected at t = 0.8 s, the semantic gateway automatically loaded the device’s LN and began publishing data within ~0.8 s. The total response time to complete detection and configuration was 1.6 s. The digital twin server synchronously updated the model view, with all measurement points of the new device appearing on the monitoring interface at 1.6 s.

As illustrated in Figure 7b, the device’s output power (OutPwr) rose rapidly from 0 to 10 kW (t = 0.8–1.2 s), while hydrogen flow rate (H2Flow) stabilized at 2.0 Nm³/h. These curves refreshed normally and were consistent with global variables such as the building’s total power balance in correlation calculations.

As shown in Figure 7c, semantic consistency verification by the knowledge graph module indicated that during the 3.2 s test period, the system checked 120 data points, including 48 from the electrical domain, 42 from the thermal domain, and 30 from the hydrogen domain. Semantic consistency remained above 99% for all data points, briefly dropping to 95% at t = 0.8 s (due to new device access) before quickly recovering to 100%. Unit consistency was maintained at 100% throughout. Figure 7d shows the building’s total power balance. After HFCLN access, Total Supply increased from 85 kW to 100 kW, Total Demand remained within 85–90 kW, and Power Deviation was controlled within ±10 kW, verifying the accuracy of the model’s power-conservation calculations.

When simulating CHP device failure at t = 2.0 s (dashed line in Figure 7a), the server detected the LN failure after 0.1 s (t = 2.1 s) and automatically marked it as offline, while data flows from other devices remained unaffected. These results demonstrate that the proposed method achieves device plug-and-play, with new device access latency of only 0.8 s and retired device detection time of 0.1 s, both meeting the performance requirements of <1 s and <2 s.

Thus, the proposed method enables device plug-and-play: new devices integrate into the unified model without manual intervention, and retired devices are promptly removed without leaving invalid data. This capability is crucial for practical engineering applications, as energy devices in green buildings may undergo gradual expansion or replacement. The solution ensures that new devices integrate quickly into existing systems without creating new information islands or additional integration overhead.

4.2. Test 2 Results

Test 2 verified the capability of the unified model framework to coordinate control among devices using different communication protocols. As illustrated in Figure 8a, after a 20 kW load-surge event was triggered at t = 0 ms, heterogeneous devices employing different protocols achieved close coordination under the unified model. The storage battery ESS (IEC 61850) responded first, increasing output about 20 ms after command issuance, with a 100% response level. The CHP unit (Modbus) elevated its electrical output after 40 ms, reaching ~70%. The photovoltaic inverter (IEEE 2030.5) increased its output after 60 ms, reaching ~35%. The BAS air-conditioning system (BACnet) reduced load within 80 ms, reaching ~65%.

As illustrated in Figure 8b, the power coordination curves of heterogeneous sources demonstrate the coordination effect: photovoltaic output stabilized at 20 kW, storage discharged 20 kW, CHP output reached 30 kW, and air-conditioning load reduction was represented as a negative value (~−5 kW). Total demand was maintained at 20 kW.

Protocol communication latency distributions, shown in Figure 8c, indicate that in 100 repeated tests: IEC 61850 latencies were approximately 3, 9, 15, 20, and 27 ms for Min, Q1, Median, Q3, and Max, respectively. The corresponding values for Modbus RTU were 75, 125, 148, 170, and 250 ms; for BACnet/IP, 100, 147, 153, 175, and 247 ms; and for IEEE 2030.5, 80, 150, 175, 210, and 350 ms.

As illustrated in Figure 8d, system Power Imbalance reached 20 kW at the disturbance but decreased to nearly 0 within 100 ms. The maximum Frequency Deviation was 0.15 Hz, recovering to within 0.02 Hz in 150 ms, demonstrating the effectiveness of cross-protocol coordinated control.

Moreover, throughout the scheduling process, upper-level control logic operated entirely on the unified model, without needing to account for differences in the underlying device protocols. This demonstrates that the designed semantic gateway effectively shields protocol heterogeneity: upper-level systems do not need to consider which communication interfaces specific devices use, relying solely on the unified information model to drive all devices. Compared with traditional methods that require separate commands for each subsystem with manual coordination, this approach significantly reduces operational complexity and latency while improving control reliability and consistency.

4.3. Test 3 Results

The high-load scenario of Test 3 yielded comparative curves of system performance under different communication strategies. As illustrated in Figure 9a, when 5G channel utilization increased from 5% to 95%, ordinary MMS message latency without network slicing grew exponentially. Latency remained stable at 30–40 ms when channel occupancy was below 60% but deteriorated sharply after exceeding 80%, with average latency reaching 180 ms at 90% and exceeding 200 ms at 95%. In contrast, MMS latency with URLLC slice guarantees remained stable at ~30 ms throughout, while GOOSE messages maintained an extremely low latency of ~5 ms. The critical threshold (50 ms) is clearly marked in the figure.

Figure 9b compares packet loss rates under strong interference scenarios. As the channel bit error rate (BER) increased from 10⁻⁶ to 10⁻³, the unoptimized scheme (MMS Baseline) exhibited significant loss starting at BER = 10⁻⁴, reaching 45% at BER = 10⁻³. In contrast, the optimized scheme (MMS Edge FEC + Retransmission) controlled packet loss within 10% even at BER = 10⁻³. GOOSE critical messages (green) consistently maintained a loss rate below 1%, with redundancy mechanisms (GOOSE Redundancy) further improving reliability.

Figure 9c illustrates the network fault recovery timeline. After the main link was interrupted at t = 0 ms (Fault Event), the GOOSE service completed switchover within 2 ms, MMS reconnected within 50 ms, and control commands resumed normal transmission within 100 ms. Overall service availability was fully restored within 150 ms.

The performance indicator radar chart in Figure 9d demonstrates the effects of optimization. Compared with the unoptimized scheme (red), the “GOOSE/MMS Critical” optimization (green) reduced latency by ~60%, improved determinism by ~70%, increased throughput by ~40%, and raised reliability to nearly 100%, with significant improvements in both fault tolerance and scalability.

These results demonstrate that the proposed communication fusion strategy significantly enhances the robustness of multi-energy system communication under extreme conditions. The success rate and timeliness of critical control information transmission improved by an order of magnitude compared with traditional schemes, thereby ensuring safe and stable system operation.

4.4. Test 4 Results

The three-phase short-circuit fault simulations in Test 4 indicate that the multi-energy system achieves excellent coordinated self-healing performance. As illustrated in Figure 10a, when the fault occurred at t = 0 s, the system without the proposed scheme (red line) experienced a voltage drop to 0.62 pu lasting ~0.1 s before recovery began, not exceeding 0.9 pu until t = 0.8 s. With the proposed scheme (green line), voltage dropped only to 0.85 pu, recovered to 0.95 pu in 200 ms (t = 0.2 s), and stabilized above 0.98 pu within 400 ms (t = 0.4 s). Voltage recovery time was shortened by 60%, and minimum voltage improved by 0.23 pu.

The frequency response comparison in Figure 10b is even more pronounced. The unoptimized system (red line) dropped from 50 Hz to 49.0 Hz (1.0 Hz deviation), requiring 3 s to recover above 49.5 Hz. With the proposed scheme (green line), frequency dropped only to 49.7 Hz (0.3 Hz deviation) and recovered fully to within the 50 ± 0.02 Hz range within 2 s. Frequency deviation was reduced by 70%, and recovery time shortened by 33%.

Figure 10c shows the cascade response timing of the multi-energy system. After the fault, GOOSE isolation triggered within 5 ms; HFCLN Stop/Restart executed an emergency stop within 100 ms and began restarting at 800 ms; ESS switched to maximum discharge mode (100% response) within 200 ms; BAS load shedding reduced load to 50% within 400 ms; and CHP ramping began at 600 ms, reaching full output at 1200 ms. The entire response chain was tightly linked with precise timing.

The resilience indicator radar chart in Figure 10d quantitatively evaluates the improvements. Compared with the unoptimized scheme (red), the optimized scheme (green) increased the Voltage Quality Index from 45% to 82%, Stability from 35% to 95%, Load Survival Rate from 60% to 92%, Temperature Variation from 70% to 92%, Recovery Time (reverse index) from 40% to 88%, and Energy Availability from 55% to 94%. Overall resilience indicators improved by more than 85%.

The comparison shows that with the proposed unified model and cross-protocol real-time interaction mechanism, electrical–thermal–hydrogen subsystems can share critical information and perform coordinated control at millisecond levels. This constrains fault impacts to a minimum and significantly enhances the self-healing capability and operational resilience of green building multi-energy systems under extreme conditions.

Overall, the simulation results fully validate the effectiveness of the proposed methods: device plug-and-play latency was controlled within 800 ms; cross-protocol control response time differences were <50 ms; GOOSE event transmission remained below 5 ms; critical data success rates exceeded 90% under extreme interference; system recovery time was shortened by >60% under fault conditions; and comprehensive resilience indicators improved by >80%. These quantitative results demonstrate the critical role of unified information models and cross-protocol real-time interaction mechanisms in enhancing the intelligence and operational reliability of green building multi-energy systems.

4.5. Limitations and Future Work

Although the simulation results demonstrate clear benefits of the proposed approach, several limitations should still be acknowledged and motivate future work.

Second, the communication experiments mainly focus on a single building connected to a 5G test network, with the multi-building scenario discussed conceptually in Section 3.3. While the results indicate that URLLC slicing, edge-assisted MMS optimization, and GOOSE-based events can substantially improve performance under extreme network conditions, the scalability of these mechanisms to very large campuses or city-scale deployments has not yet been quantified. In particular, the impact of thousands of devices and hundreds of buildings on gateway CPU load, memory consumption, and configuration complexity requires further investigation.

Third, the integration of multiple protocols and 5G connectivity inevitably enlarges the cyber-physical attack surface. In the present study, security is assumed to be provided by standard best-practice measures at the network and application layers, and potential vulnerabilities such as spoofed GOOSE messages, misconfigured mapping rules, or denial-of-service attacks on edge servers are not explicitly modeled. Future work will combine the unified information model with defence-in-depth security architectures and intrusion-detection mechanisms tailored to multi-protocol gateways and 5G edge nodes.

Fourth, although the semantic gateway achieves a high degree of automation for device onboarding, legacy devices that expose only minimal or vendor-specific interfaces still require manual mapping and validation. Extending the rule library to cover a broader range of devices and applying machine-learning methods to infer likely mappings from historical configuration data, are promising directions. In addition, large-scale field trials in real green-building projects will be necessary to validate the proposed framework under long-term operational and maintenance conditions.

5. Conclusions

These specific quantitative gains indicate that the proposed unified model and interaction mechanism provide a practical foundation for interoperable, low-carbon, and resilient operation of green buildings. The same design principles can be extended to zero-carbon parks and multi-energy microgrids, and further combined with standardization, AI-based semantic adaptation, and cyber-physical security measures as outlined in the limitations and future work.

Simulation results show that the proposed scheme reduces device plug-and-play time to about 0.8 s, limits cross-protocol response-time differences to less than 50 ms, and maintains GOOSE latency below 5 ms with critical-data success rates above 90% even at a bit-error rate of 10⁻³. Under three-phase short-circuit faults, the unified architecture shortens voltage recovery time by roughly 60%, reduces frequency deviation by around 70%, and improves a composite resilience index by more than 80% compared with a conventional non-unified baseline.

First, by extending IEC 61850 with new logical nodes (HFCLN, CHPLN, ECLN and others) and a global unit system, we built an information model that represents key electrical, thermal, and hydrogen devices in a single semantic space and can be implemented directly in SCL files. Second, we designed a semantic gateway and GOOSE/SV/5G/edge-computing fusion communication strategy that enables plug-and-play integration of heterogeneous BACnet, Modbus, IEEE 2030.5 and IoT devices while preserving millisecond-level response for critical events. Third, we constructed an RTDS-based digital twin and 5G testbed to quantify the end-to-end behavior of the architecture, including device onboarding, cross-protocol coordination, communication robustness, and grid-fault resilience.

This paper proposed a unified information modeling framework and a cross-protocol real-time interaction mechanism for green-building multi-energy systems that couple electrical, thermal, and hydrogen subsystems.