Multi-Agent System-Based Real-Time Implementation of Advanced Energy Management in Hybrid Microgrids

Abstract

The growing integration of solar, wind and battery energy storage (BES) of the microgrids (MGs) has increased the necessity of real-time energy management, especially in the multi-microgrid (multi-MG) setting, where the generation and the load change stochastically. This paper presents a Java Agent DEvelopment (JADE)-based Multi-Agent System (MAS) for real-time energy management of a low-voltage hybrid multi-MG system incorporating solar photovoltaic (PV), wind generation, and battery energy storage (BES). The proposed framework’s novelty lies in its physical campus-scale hardware deployment—validated across four operating scenarios (single MG off-grid, single MG on-grid, dual MG off-grid, and dual MG on-grid)—combined with autonomous inter-MG power sharing, which distinguishes it from existing simulation-only MAS-based microgrid studies. The suggested framework facilitates decentralized communication between interconnected MGs and the utility AC grid to facilitate the proper management of power flow, its exchange, and the reliability of the system. The intelligent agents are used to coordinate solar, wind, BES, and load changes in order to adjust to changing demand conditions. The system is physically implemented on a campus rooftop with two 1 kW solar PV arrays and two 1.5 kW wind turbine generators, each paired with a 24 V, 150 Ah battery bank, operating on a 24 V DC bus. Results across 24 h real operational profiles demonstrate effective power balance maintenance, renewable energy maximization, and constraint-compliant battery operation (SOC is bounded within 20–90%). A direct comparison with a conventional centralized JavaScript-based EMS confirms equivalent dispatch accuracy while demonstrating superior scalability, fault tolerance, and modularity of the proposed JADE MAS architecture.

1. Introduction

The increasing electricity demand in the world coupled with the growing concerns about the environment and the decline of conventional fossil fuel has heightened the shift towards renewable energy-based power systems. The hybrid microgrids combining renewable energy sources (RES) like solar photovoltaic (PV) and wind energy and distributed energy storage systems (ESS) have become an effective answer to attain energy efficiency, reliability and sustainability in the grid-connected mode and islanded mode of operation. Microgrids provide a higher level of resilience in the system and less transmission loss and greater use of locally available renewable energy, which makes them especially appropriate in the context of contemporary decentralized power systems [1,2,3]. A hybrid microgrid has been widely studied as a powerful tool towards managing renewable energy generation, power storage and load demand. Various alternative methods of energy management and control have also been proposed to enhance system performance, among them being rule-based, metaheuristic-based, and centralized optimization. A Levy Flight Algorithm (LFA)-optimized hybrid PV-wind microgrid with hydrogen and battery storage was introduced in [4], showing a lower cost of the system and better economic viability than the Particle Swarm optimization (PSO) and Genetic Algorithms (GA) solutions. In the same way, the proposed PSO-driven rule-based demand response framework in [5] was suggested to optimize cost-efficiency and the demand-side flexibility in hybrid microgrids. Further optimization-based energy management schemes, including model predictive control (MPC) were also being studied as they are capable of treating system constraints and multi-time scale goals. In [6], a single-layer move-blocking MPC-based energy management system was suggested to integrate short- and long-term goals in one optimization model and to enhance the economic performance and calculation efficiency. Other important methods are neuro-fuzzy-based islanding detection and power management models [7], adaptive gravity search models of shipboard microgrids [8] and decentralized multi-energy management models with fuzzy optimization and Pareto-based decision-making to minimize the operational costs and carbon emissions [9].

Regardless of these developments, hybrid microgrids remain more complex to operate because of the intermittency of the renewable energy sources, non-deterministic variations in loads, and the coordination needs of various distributed components. Traditional centralized or rule-based control mechanisms tend to lack flexibility, experience excessive communication costs, have a low level of scalability, and are susceptible to a single point of failure. Such constraints greatly limit their practicality when operating in real time especially in systems where renewable penetration is high and interlocked micro grids.

In order to deal with these challenges, smart and decentralized control architectures have gained importance as next-generation microgrid energy management. Multi-Agent Systems (MAS) have become an influential paradigm in this respect in the paradigm of distributed control in microgrids. MAS are made up of several autonomous and intelligent agents who have the ability to sense what is happening around them, make decentralized decisions and collaborate with others in the nearby environment towards the accomplishment of global system goals. Such features influence MAS in the challenging management of the dynamic, uncertain and distributed environment of microgrids powered by renewable energy.

The promising potential of MAS-based energy management frameworks in enhancing the reliability, flexibility and economic performance of microgrids has been realized in recent studies. A MAS-based energy management framework was suggested in [10] where a coordinator agent connects the tertiary and primary control level providing real-time economic dispatch of active and reactive power and enhancing system reliability and operational efficiency. In [11] the authors explored energy management in multi-microgrid (MG) systems with the integration of renewable energy sources (RES) and battery energy storage systems (BESS). Conventional centralized control approaches faced limitations in scalability, flexibility, and real-time adaptability. To address these challenges, agent-based and distributed control strategies were proposed for improved coordination and energy sharing among MGs. In [12], the authors proposed an Industry 4.2 framework for green intelligent manufacturing aimed at improving energy efficiency and reducing carbon emissions. The framework utilized cyber-physical agents for data collection and integrated both energy and carbon management systems. This approach enabled enhanced monitoring and control of manufacturing processes. The results demonstrated significant energy savings and a reduction in carbon emissions.

In [13], the authors proposed an ontology-based energy management system based on MAS to integrate building microgrid environments with renewable energy sources and controllable loads allowing uniform distributed generation and demand response. MAS frameworks are improved by AI, which is used to combine short-term load forecasting with adaptability in grid-connected multi-microgrids [14], and hierarchical MAS structures are also improved by peer-to-peer energy trading [15]. MAS methods built around reinforcement learning enhanced further resilience and dispatch efficiency in the case of uncertainty [16]. JADE based MAS monitoring and control have been tested and validated through practical implementations using MATLAB R2024a, including both simulation and hardware testing [17]. Moreover, smart MAS-based solutions have been demonstrated to be useful to off-grid hybrid home systems [18] and islanded AC- DC microgrids, in which agent-based optimization methods made it possible to reduce costs, manage demand response, and decrease load shedding [19].

Despite the progress outlined above, the existing body of MAS-based microgrid research is characterized by three significant limitations: (i) the vast majority of studies are validated exclusively through software simulation (e.g., MATLAB/Simulink) with little or no physical hardware deployment; (ii) most frameworks address single-microgrid configurations, leaving the more complex problem of real-time coordinated energy management across multiple interconnected microgrids underexplored; and (iii) systematic performance comparison of MAS-based EMS against conventional centralized control under identical real-world operating conditions is rarely reported. These gaps motivate the present work.

Based on these identified gaps, this paper makes the following original contributions, which collectively distinguish it from the existing MAS-microgrid literature as summarized in

Table 1. Novelty positioning of the proposed work relative to existing MAS-microgrid literature.

Dimension	Existing Literature	This Work
Deployment level	Predominantly simulation-only (MATLAB, NS-2, Java simulation)	Physical campus hardware: rooftop solar PV, wind turbines, battery banks, energy metres
MG configuration	Mostly single-MG MAS frameworks	Dual interconnected hybrid MGs with autonomous inter-MG power exchange
Operating scenarios	Typically 1–2 scenarios per study	Four scenarios: single/dual × off-grid/on-grid
Baseline comparison	Rarely compared to centralized EMS on same real data	Direct comparison with centralized JavaScript EMS on identical 24 h data
JADE implementation	Often conceptual or simulation-level JADE use	Live JADE deployment with Sniffer-verified ACL message traces

below.

This paper has made several key contributions as summarized below:

• Design and physical implementation of a real-time, low-voltage hybrid dual microgrid system on a campus rooftop and laboratory, comprising two interconnected microgrids each with solar PV, wind generation, and battery energy storage.
• Development of a JADE-based multi-agent architecture—encompassing Generation Agents (SA, WA), Battery Agent (BA), Load Agent (LA), Grid Agent (GA), and Control Agent (CA)—for fully decentralized, autonomous energy management with FIPA-compliant ACL inter-agent communication.
• Comprehensive real-time performance evaluation across four distinct operating scenarios (single MG off-grid, single MG on-grid, dual MG off-grid, and dual MG on-grid) using 24 h measured operational data.
• Structured quantitative comparison against a conventional JavaScript-based centralized EMS, demonstrating equivalent dispatch accuracy alongside superior scalability, fault tolerance, and modularity of the proposed MAS framework.

The remainder of this paper follows the following structure. Section 2 gives a review of hybrid microgrid systems. Section 3 explains the modelling of components of microgrids. In Section 4, the topic of the Multi-Agent System (MAS) framework with the implementation of JADE is discussed. In Section 5 the problem formulation is described. Section 6 describes the autonomous energy management approach of the hybrid microgrid based on MAS. Section 7 describes how the hybrid microgrid was going to be implemented in real-time. The real-time simulation results and discussion are discussed in Section 8. Section 9 gives a comparative analysis of the proposed approach. Lastly, the paper is summarized in Section 10.

2. Hybrid Microgrid System

A hybrid microgrid system can be used in both coordinated mode with the main utility grid (grid-connected mode) and in decoupled mode (off-grid or islanded mode). This dual functionality can increase energy resiliency, solar and wind energy sources can be more fully penetrated, and power quality and system stability are also improved [20,21,22,23,24,25,26]. Through smart generation, storage and loads, the microgrids can be used to effectively reduce the challenges encountered by grid instability, increase in energy prices, and the need to have sustainable energy and low-carbon energy sources. Due to these benefits, the hybrid microgrid systems make them especially applicable in remote and isolated locations, industrial and business facilities, military camps, and disaster-prone areas, where consistent, uninterrupted and autonomous electricity is essential. Figure 1 illustrates the architecture of a single hybrid microgrid, and Figure 2 presents the dual microgrid configuration implemented in this work. In the dual-MG topology, each microgrid operates as an independent unit capable of islanded operation while also being interconnected via switch S1 to enable bidirectional inter-MG power exchange, coordinated by the Control Agent (CA) as described in Section 6.

Here, various components of a microgrid system and their arrangement are depicted.

3. Modelling of Microgrid Components

Modelling of a microgrid which comprises components like solar panels, a wind turbine, energy storage devices, power converters, loads, and control systems is essential for the design, analysis, and optimization of microgrid system performance through scheduling and allocation in power-aware applications. The prediction of the system’s behaviour across various operating situations optimizes energy distribution and guarantees grid stability and dependability will be as near the real-time situations when modelling is accurate [27,28,29,30,31]. The models presented in

Table 2. Component specifications and mathematical models.

S. No.	System Component	Technical Specifications	Mathematical Modelling
1	Wind Turbine	Rated power: 1.5 kW Air density: ρ	$P_w={\displaystyle \frac{1}{2}}\rho A{C}_p\left(\lambda ,\beta \right)v^3$ where $C_p$ is the power coefficient, $\lambda$ is tip speed ratio, and $\beta$ is blade pitch angle
2	Solar PV Array	Peak power: 1 kW Open-circuit voltage: $V_{oc}$ Short-circuit current: $I_{sc}$	$I_{pv}=I_{ph}-I_0\left[\mathrm{e}\mathrm{x}\mathrm{p}\left({\displaystyle \frac{V_{pv}+I_{pv}{R}_s}{n{V}_t}}\right)-1\right]$ Single-diode PV model
3	Inverter	Rated power: 1.5 kW Output voltage: 230 V AC Frequency: 50 Hz	$V_{ac}\left(t\right)=M\cdot V_{dc}\mathrm{s}\mathrm{i}\mathrm{n}\left(\omega t\right)$ where M is modulation index and $\omega =2\pi f$
4	Battery Energy Storage	Nominal voltage: 24 V Capacity: 150 Ah	$SOC\left(t\right)=SOC\left(t_0\right)-{\displaystyle \frac{1}{C_b}}{\int }_{t_0}^t{I}_b\left(\tau \right)\hspace{0.17em}d\tau$ Charging/discharging dynamics
5	Load	Rated demand: 1.5 kW AC load	$P_L=V_{rms}{I}_{rms}\mathrm{c}\mathrm{o}\mathrm{s}\varphi$ Steady-state load power model

Note on Table 2: All power quantities are expressed in Watts (W). SOC is expressed as a dimensionless fraction (0–1) in the dynamic model and as a percentage (%) in the results tables for readability. Time step Δt is in hours (h).

are intentionally scoped to capture the essential input-output behaviours of each component at the energy management layer—i.e., the power generation, storage dynamics, and load characteristics relevant to the dispatch decisions made by the MAS agents. These models are consistent with those used in comparable published works on agent-based microgrid EMS and are appropriate for the hourly resolution EMS timescale at which the JADE agents operate. The mathematical modelling of microgrid components is presented in Table 2.

4. Multi-Agent System Using JADE

MAS involves several interacting intelligent agents collaborating each of which is capable of making autonomous decisions informed by its surroundings and internal conditions, being adept at managing the complexity and dynamic nature of microgrids [32,33,34,35,36,37,38]. MAS improve the performance of the microgrid system. The MAS algorithm is used for effective energy management. Every agent inside the system possesses the attributes given in

Table 3. Fundamental properties of intelligent agents in distributed systems.

S. No.	Agent Property	Conceptual Definition	System-Level Interpretation
1	Autonomy	The ability of an agent to operate independently without external intervention or centralized supervision	Enables decentralized decision-making and enhances system scalability and fault tolerance
2	Social Ability	Capability to communicate, cooperate, and negotiate with other agents using defined interaction protocols	Facilitates coordinated actions, conflict resolution, and collective optimization in multi-agent environments
3	Reactivity	Capacity to perceive environmental states and respond to changes in real time	Allows adaptive behaviour under dynamic operating conditions, such as load variation or resource uncertainty
4	Proactivity	Ability to initiate actions based on internal objectives rather than solely responding to stimuli	Supports predictive control, goal-driven planning, and long-term performance optimization

.

JADE is a popular, widely accepted open-source framework for constructing Multi-Agent Systems, developed in Java offering a framework for the creation and management of intelligent agents that utilize the FIPA communication (Foundation for Intelligent Physical Agents) standard. Principal attributes of JADE are FIPA-compliant agent communication, distributed architecture with platform independence, agent mobility, dynamic creation, directory and yellow pages services for agent identification, and integrated debugging and monitoring tools. The MAS model is shown in Figure 3.

4.1. Agent Decision-Making and Internal Logic

Each agent in the proposed MAS executes a local decision procedure based on its monitored state variables. The decision logic for each agent type is as follows in

Table 4. Agent decision-making logic in the proposed JADE-based MAS.

Agent	Decision Logic
Load Agent (LA)	Monitors real-time load demand from connected rooms. Initiates a power request cycle at each hourly interval by broadcasting a REQUEST message to the Control Agent (CA) specifying the current power requirement.
Solar Agent (SA)	Monitors solar PV power output from the installed panels. Responds to CA queries with a PROPOSE message indicating available PV power. Surplus power beyond load demand is flagged for battery charging.
Wind Agent (WA)	Monitors wind turbine generator output. Responds to CA queries with available wind power. Operates similarly to SA in terms of priority and surplus handling.
Battery Agent (BA)	Monitors battery SOC continuously. Applies SOC-threshold decision rules: charges when SOC < $SO{C}_{\mathrm{m}\mathrm{a}\mathrm{x}}$ and surplus generation exists; discharges when SOC > $SO{C}_{\mathrm{m}\mathrm{i}\mathrm{n}}$ and a power deficit is present. Communicates SOC status and charge/discharge capacity to the CA.
Grid Agent (GA)	Manages bidirectional power exchange with the utility grid. Imports power from the grid only when battery SOC reaches $SO{C}_{\mathrm{m}\mathrm{i}\mathrm{n}}$ and all local resources are exhausted. Exports surplus power to the grid only when battery SOC reaches $SO{C}_{\mathrm{m}\mathrm{a}\mathrm{x}}.$
Control Agent (CA)	Supervises the overall energy management cycle. Queries all agents, aggregates their reported states, and issues dispatch commands following the priority order: renewable generation → battery storage → inter-MG exchange → grid interaction. In the dual-MG configuration, the CA selects the battery with the higher SOC for discharge and the battery with the lower SOC for charging, implementing implicit cross-MG SOC balancing.

:

4.2. Agent Communication Protocol

All inter-agent communication is implemented using JADE’s FIPA-compliant ACL. The specific ACL performatives and their roles in the energy management cycle are explained in

Table 5. FIPA ACL performatives used in the proposed MAS communication protocol.

ACL Performative	Direction	Purpose
REQUEST	LA → CA	Load Agent notifies CA of current power demand
PROPOSE	SA/WA/BA/GA → CA	Generation/storage agents report available power
ACCEPT-PROPOSAL	CA → Agent	CA confirms power allocation to the agent
INFORM	Agent → CA	Agent reports updated state (SOC, output power)
CONFIRM	CA → LA	CA confirms that load demand has been satisfied

:

The complete message exchange sequence is captured in the Sniffer diagrams in Section 8.

Agent Communication Language (ACL) is used for communication between agents for collaborative actions. All the agents are put into the container and communication happens through ACL. The decentralization and distribution of the decision-making processes among individual autonomous agents, the individual objectives of agents and further the evolution of the global objectives are carried out with the help of communication among all these components.

5. Objective Function and System Constraints

5.1. Objective Function

The main objective is to maintain power balance in both single and dual microgrid configurations to provide operational stability and a reliable energy supply. This is attained by balancing the renewable generation, grid exchange and storage within the system load.

For a single microgrid, the power balance objective will be represented as follows:

$$\mathrm{m}\mathrm{i}\mathrm{n}\left|P_{pv}\left(t\right)+P_{wind}\left(t\right)+P_{battery}\left(t\right)+P_{exchange}\left(t\right)-P_{Totalloaddemand}\left(t\right)\right|$$

For a dual microgrid system, this is extended as follows:

$$\mathrm{m}\mathrm{i}\mathrm{n}{\displaystyle \sum _{i=1}^2}\left|P_{pv,i}\left(t\right)+P_{wp,i}\left(t\right)+P_{b,i}\left(t\right)+P_{g,i}\left(t\right)+P_{exchangein\mathrm{i}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{c}\mathrm{o}\mathrm{n}\mathrm{n}\mathrm{e}\mathrm{c}\mathrm{t}\mathrm{e}\mathrm{d}\mathrm{g}\mathrm{r}\mathrm{i}\mathrm{d}}\left(t\right)-P_{L,i}\left(t\right)\right|$$

5.2. System Constraints

5.2.1. Power Balance Constraint

At any time instant, the sum of generation should be equal to the sum of demand

$$P_{pv}\left(t\right)+P_{wp}\left(t\right)+P_g\left(t\right)+P_b^{dis}\left(t\right)=P_L\left(t\right)+P_b^{ch}\left(t\right)$$

For dual microgrids at interconnection:

$$P_{ex,1}\left(t\right)+P_{ex,2}\left(t\right)=0$$

The above equation shows the lossless power exchange between microgrids. This constraint is enforced by the Control Agent at every hourly decision cycle, ensuring that inter-MG power transfer is always balanced and that no uncompensated power flow occurs across the interconnection.

5.2.2. Storage Capacity Constraint

The SOC should balance the charging and discharging limits. This will be shown by

$$SO{C}_{\mathrm{m}\mathrm{i}\mathrm{n}}\le SOC\left(t\right)\le SO{C}_{\mathrm{m}\mathrm{a}\mathrm{x}}$$

The SOC dynamics are computed by

$$SOC\left(t+1\right)=SOC\left(t\right)+{\displaystyle \frac{{\eta }_{ch}{P}_b^{ch}\left(t\right)-\frac{1}{{\eta }_{dis}}{P}_b^{dis}\left(t\right)}{E_b^{rated}}}\mathrm{\Delta }t$$

where

${\eta }_{ch}$ and ${\eta }_{dis}$ are charging and discharging efficiencies;

$E_b^{rated}$ is the rated battery energy capacity;

$\Delta t$ is the step time control.

In this work, $SO{C}_{\mathrm{m}\mathrm{i}\mathrm{n}}$ = 20% and $SO{C}_{\mathrm{m}\mathrm{a}\mathrm{x}}$ = 90% chosen to prevent battery over-discharge and overcharge, thereby extending battery cycle life. These bounds are respected in all reported scenarios, as confirmed by the SOC profiles in Section 8.

5.2.3. Battery Power Limits

The battery charging and discharging is limited as follows for extended battery life.

$$0\le P_b^{ch}\left(t\right)\le P_b^{ch,max}\hspace{1em}\hspace{1em}0\le P_b^{dis}\left(t\right)\le P_b^{dis,max}$$

6. Multi-Agent Systems-Based Autonomous Energy Management of Microgrid

The agent autonomously executes the most feasible operations based on hourly monitored parameters of wind power, solar power, load, and battery state of charge (SOC) for dynamic energy management of the hybrid microgrid in a distributed environment.

The agents shown in

Table 6. Agent architecture for hybrid single and dual microgrid energy management systems.

S. No	Agent Type	Functional Role	Microgrid-1 Agent	Microgrid-2 Agent	Interaction Scope
1	Load Agent (LA)	Estimates real-time load demand, initiates power requests, and coordinates with generation and storage agents via the Directory Facilitator	LA1	LA2	Intra-MG and Inter-MG
2	Solar Agent (SA)	Supplies photovoltaic power based on availability and responds to demand signals from the LA	SA1	SA2	Intra-MG
3	Wind Agent (WA)	Provides wind-generated power and updates stochastic availability to the LA	WA1	WA2	Intra-MG
4	Battery Agent (BA)	Monitors state of charge, manages charge/discharge decisions, and communicates storage constraints	BA1	BA2	Intra-MG
5	Grid Agent (GA)	Manages bidirectional power exchange between the microgrid and the utility grid	GA	GA	External Grid Interface
6	Control Agent (CA)	Supervises negotiation, arbitrates conflicts, and enforces coordinated power sharing between interconnected microgrids	CA	CA	Global/Supervisory

simulate a MAS within a JADE environment for a single hybrid microgrid and two hybrid microgrid systems.

6.1. Single Hybrid Microgrid (Grid-Connected/Off-Grid)

Figure 4 shows the flow chart including all potential applications of a single hybrid microgrid and Figure 5 shows the interaction of the agents. The control priority hierarchy—solar availability first, wind second, battery third, and grid last—is embedded in the CA’s decision procedure and ensures maximum renewable utilization before resorting to grid dependency. This priority ordering constitutes the conflict resolution protocol of the MAS: when multiple agents simultaneously propose power, the CA follows this deterministic hierarchy to arbitrate and allocate without ambiguity.

The implemented process is as follows:

For off-grid operation of a single hybrid microgrid, adhere to stages 1–4; for on-grid operation, proceed with steps 5–7.

Step 1: The Load Agent (LA) contacts the Control Agent (CA) with power requirements.

Step 2: The CA verifies the availability of the Solar Agent (SA) and solicits electricity from the SA, which autonomously supplies power to the LA. Surplus electricity is allocated to the Battery Agent (BA).

Step 3: The CA verifies the availability of the wind Agent (WA) and solicits electricity from the WA, subsequently receiving the available power from the WA. Surplus electricity is allocated to BA.

Step 4: If power remains necessary, the CA assesses the SOC of the BA. If the SOC exceeds the minimum state of charge ($SO{C}_{\mathrm{m}\mathrm{i}\mathrm{n}}$), it draws power from the battery array (BA).

Step 5: If the SOC of the BA meets the minimal threshold, then the CA verifies the availability of grid power.

Step 6: If necessary, draw electricity from GA whenever the battery’s SOC hits its minimum and supply it to LA.

Step 7: Excess power is transmitted to the GA when the battery’s SOC reaches its maximum capacity.

6.2. Two Hybrid Microgrid Systems

Figure 6 illustrates the flow chart including all potential applications of a two-hybrid microgrid system and Figure 7 shows the interaction of the agents. In the dual-MG configuration, the inter-MG power exchange is governed by the following explicit rules enforced by the Control Agent: (i) when surplus generation exists in either MG, the CA charges the battery with the lower SOC (cross-MG SOC balancing); (ii) when a power deficit exists and local batteries cannot fully satisfy demand, the CA discharges the battery with the higher SOC, regardless of which MG it belongs to; (iii) inter-MG exchange is subject to the lossless balance constraint P_(ex,1) (t) + P_(ex,2) (t) = 0 (Section 5.2.1); and (iv) grid interaction is invoked only when both BA1 and BA2 have reached their respective SOC limits. These rules ensure fair, efficient, and constraint-compliant power sharing between the two microgrids at every decision step.

The agents listed below simulate a MAS within a JADE environment for two hybrid microgrids, both in grid-connected and off-grid configurations.

For off-grid operation of hybrid microgrids, adhere to stages 1–6; for on-grid operation, proceed with steps 7–9.

Step 1: Load Agent 1 (LA1) and Load Agent 2 (LA2) solicit power requirements from the Control Agent (CA).

Step 2: The CA verifies the availability of SA1 and grants authority to LA1.Should extra power be available, assess the SOC of BA1 and BA2, and charge the battery with the lower SOC.

Step 3: CA verifies WA1 availability and allocates power to LA1.In the presence of extra power, assess the SOC of BA1 and BA2, and charge the battery with the lower SOC.

Step 4: The CA verifies the availability of SA2 and grants authority to LA2.In the presence of extra power, assess the SOC of BA1 and BA2, and charge the battery with the lower SOC.

Step 5: The CA verifies the availability of WA2 and grants authority to LA2.In the presence of extra power, assess the SOC of BA1 and BA2, and charge the battery with the lower SOC.

Step 6: If power remains necessary, CA assesses the SOC of BA1 and BA2, procures power from the battery with the higher SOC, and distributes it to LA1 and LA2.

Step 7: If the SOC of BA1 and BA2 attains the minimal threshold, then the Control Agent (CA) assesses the grid availability of power.

Step 8: If power remains necessary, it is sourced from the Grid Agent (GA) and distributed to LA1 and LA2.

Step 9: The surplus electricity is allocated to the Grid Agent (GA) once BA1 and BA2 are completely charged.

7. Realtime Implementation of Hybrid Microgrid

In this paper, the term ‘real-time’ refers to real-time energy management (EMS)—a well-established designation in the microgrid literature [4,5,6,7,8,9] denoting dispatch-level control operating at the minute-to-hour timescale, as opposed to inner voltage/current control loops operating at the millisecond timescale. The JADE agents execute hourly decision cycles driven by actual sensor measurements, which is the standard operating frequency for campus-scale EMS deployments. The physical hardware (solar PV, wind turbines, batteries, metres) provides measured inputs; the JADE software version 4.6.0 processes these inputs and issues control decisions; and the Sniffer diagrams confirm actual live agent execution. Two 1 kW solar PV systems and two 1.5 kW wind-generating systems are erected on the rooftop of the electrical department, with control systems, measuring equipment, and sensors connected to the research laboratory of the department. These are configured as a microgrid system having a 1 kW solar PV system, a 1.5 kW wind turbine system, a 24 V, 150 AH storage system, and a local load.Microgrid-1 supplies power to the connected loads in various rooms such as rooms 303, 304, and 305 of the electrical department, while microgrid-2 provides power to the connected loads in rooms 403 and 404. Figure 8 illustrates the installed configurations. The microgrid control mechanism is illustrated in Figure 9. The connected loads of microgrid-1 and microgrid-2 are presented in

Table 7. Connected load profile of Microgrid-1.

Location	Load Category	Quantity	Unit Rating (W)	Aggregate Load (W)
Room 309	Ceiling Fan	6	70	420
	Tube Light	2	36	72
Room 310	Ceiling Fan	3	70	210
	Tube Light	2	36	72
Room 311	Ceiling Fan	3	70	210
	Tube Light	2	36	72
Corridor	Tube Light	1	36	36
Total Connected Load				1092 W

and

Table 8. Connected load profile of Microgrid-2.

Location	Load Category	Quantity	Unit Rating (W)	Aggregate Load (W)
Room 409	Ceiling Fan	6	70	420
	Tube Light	2	36	72
Room 410	Ceiling Fan	6	70	420
	Tube Light	2	36	72
Corridor	Tube Light	1	36	36
Total Connected Load				1020 W

, respectively.

Table 9. Comparative load profile with priority index for dual microgrid system.

Location	Load Type	Load Class	Priority Index (α)	Quantity	Unit Rating (W)	Microgrid-1 Load (W)	Microgrid-2 Load (W)
Room 309	Ceiling Fan	Critical	0.90	6	70	420	-
	Tube Light	Critical	0.95	2	36	72	-
Room 310	Ceiling Fan	Critical	0.85	3	70	210	-
	Tube Light	Critical	0.90	2	36	72	-
Room 311	Ceiling Fan	Critical	0.85	3	70	210	-
	Tube Light	Critical	0.90	2	36	72	-
Room 409	Ceiling Fan	Critical	0.90	6	70	-	420
	Tube Light	Critical	0.95	2	36	-	72
Room 410	Ceiling Fan	Critical	0.90	6	70	-	420
	Tube Light	Critical	0.95	2	36	-	72
Corridor	Tube Light	Non-Critical	0.40	1	36	36	36
Total Connected Load						1092 W	1020 W

shows the comparative load profile with priority index for the dual microgrid system. The sensors consistently monitor all electrical loads while assessing wind and solar energy on an hourly basis.

Sensor specifications: The energy metres used in the installation are commercial-grade metres conforming to IEC 62053 Class 1 standards [39], with a typical power measurement accuracy of ±1–2%. Current sensors used for battery SOC estimation via coulomb counting have a measurement accuracy of ±0.5% FS. While sensor measurement noise introduces a small cumulative error in the SOC dynamic model, the SOC profiles reported in

Table 10. Optimized operational states of a single hybrid off-grid microgrid under MAS control.

Time (h)	Load Demand (W)	Total Generation (W)	Solar (W)	Wind (W)	Power Balance* (W)	Battery SOC (%)	Battery Mode
0	180	120	0	120	−60	75.0000	Discharge
1	180	180	0	180	0	74.1071	Discharge
2	180	250	0	250	+70	74.1071	Charge
3	180	160	0	160	−20	75.0538	Discharge
4	180	90	0	90	−90	74.7562	Discharge
5	180	60	0	60	−120	73.4166	Discharge
6	180	120	50	70	−60	71.6281	Discharge
7	300	380	260	120	+80	70.7322	Charge
8	700	720	520	200	+20	71.8128	Charge
9	900	910	670	240	+10	72.0830	Charge
10	1200	1100	750	350	−100	72.2182	Discharge
11	1150	1350	800	550	+200	70.7260	Charge
12	1200	1240	830	410	+40	73.4274	Charge
13	900	1170	860	310	+270	73.9682	Charge
14	900	1100	880	220	+200	77.6195	Charge
15	1200	1200	910	290	0	80.3275	Charge
16	1150	1230	880	350	+80	80.3275	Charge
17	900	750	230	520	−150	81.4116	Discharge
18	450	710	50	660	+260	79.1919	Charge
19	400	420	0	420	+20	82.7141	Charge
20	180	310	0	310	+130	82.9853	Charge
21	180	200	0	200	+20	84.7484	Charge
22	180	120	0	120	−60	85.0197	Discharge
23	180	110	0	110	−70	84.1343	Discharge

* (+ surplus → battery charging, − deficit → battery discharging).

,

Table 11. Optimized energy management of a single hybrid on-grid microgrid under MAS control.

Hour	Load Demand (W)	Wind Power (W)	Solar Power (W)	Battery Discharge (W)	Battery Charge (W)	Battery SOC (%)	Grid Import (W)	Grid Export (W)
0	300	300	0	0	0	75.0	0	0
1	300	200	0	100	0	72.2	0	0
2	300	240	0	60	0	70.6	0	0
3	300	420	0	0	120	73.9	0	0
4	350	450	0	0	100	76.7	0	0
5	500	460	0	40	0	75.6	0	0
6	600	120	50	430	0	63.6	0	0
7	800	380	260	160	0	59.2	0	0
8	900	660	460	0	220	65.3	0	0
9	1000	750	610	0	360	75.3	0	0
10	1100	1000	700	0	529.2	90.0	0	70.8
11	1300	690	740	0	0	90.0	0	130
12	1400	430	730	240	0	83.3	0	0
13	1400	390	670	340	0	73.9	0	0
14	1300	230	550	520	0	59.4	0	0
15	1300	360	380	560	0	43.9	0	0
16	1250	490	170	590	0	27.5	0	0
17	1100	600	0	270	0	20.0	230	0
18	400	740	0	0	340	29.4	0	0
19	400	630	0	0	230	35.8	0	0
20	300	580	0	0	280	43.6	0	0
21	300	440	0	0	140	47.5	0	0
22	300	620	0	0	320	56.4	0	0
23	200	530	0	0	330	65.5	0	0

,

Table 12. Optimized operational performance of dual off-grid hybrid microgrids under MAS control.

Hour	Load at Grid 1 (W)	Solar PV (W)	Wind (W)	Power Balance (W)	BA1 SOC (%)	Load at Grid 2 (W)	Solar PV (W)	Wind (W)	Power Balance (W)	BA2 SOC (%)	Battery Action
0	300	0	110	−190	85.0	250	0	80	−170	80.0	Discharge
1	300	0	140	−160	75.0	250	0	100	−150	80.0	Discharge
2	300	0	100	−200	75.0	260	0	110	−150	71.39	Discharge
3	300	0	120	−180	65.28	250	0	70	−180	71.39	Discharge
4	270	0	90	−180	65.28	250	0	80	−170	61.39	Discharge
5	280	0	60	−220	55.56	280	0	70	−190	61.39	Discharge
6	280	95	150	−35	55.56	300	90	100	−110	50.0	Discharge
7	300	260	120	+80	51.53	290	260	120	+90	50.0	Charge
8	500	450	200	+150	51.53	280	450	130	+300	54.72	Charge
9	550	650	140	+240	64.03	360	607	150	+397	54.72	Charge
10	700	699	300	+299	64.03	400	699	300	+599	72.42	Charge
11	700	741	250	+291	88.97	500	741	350	+591	72.42	Charge
12	650	772	350	+472	88.97	570	770	320	+520	95.0	Charge
13	550	790	400	+640	95.0	620	810	320	+510	95.0	Charge
14	600	799	370	+569	95.0	600	760	340	+500	95.0	Charge
15	550	722	300	+472	95.0	590	650	350	+410	95.0	Charge
16	600	579	250	+229	95.0	520	420	250	+150	95.0	Charge
17	550	235	180	+135	95.0	390	180	200	−10	95.0	Charge
18	400	55	140	−175	95.0	280	40	180	−60	95.0	Discharge
19	300	0	130	−170	88.47	270	0	170	−100	95.0	Discharge
20	300	0	180	−120	88.47	250	0	160	−90	87.5	Discharge
21	280	0	140	−140	82.64	250	0	140	−110	87.5	Discharge
22	270	0	120	−150	82.64	260	0	140	−120	80.56	Discharge
23	280	0	130	−150	75.14	250	0	120	−130	80.56	Discharge

and

Table 13. Optimized values of two hybrid microgrid systems (on-grid) with MAS.

Hour of the Day	Microgrid-I Power Credit/Debit	Microgrid-I Battery SOC	Microgrid-II Power Credit/Debit	Micro Grid-II Battery SOC	Battery Action	Overall Power Surplus/Deficit	Grid Action
0	−190	85	−170	80	discharge	-	-
1	−160	75	−150	80	discharge	-	-
2	−200	75	−150	71.389	discharge	-	-
3	−180	65.278	−180	71.389	discharge	-	-
4	−180	65.278	−170	61.389	discharge	-	-
5	−220	55.556	−190	61.389	discharge	-	-
6	−35	55.556	−110	50	discharge	-	-
7	80	51.528	90	50	charge	-	-
8	150	51.528	300	54.722	charge	-	-
9	240	64.028	397	54.722	charge	-	-
10	299	64.028	599	72.416	charge	-	-
11	291	88.972	591	72.416	charge	+69	charge
12	472	88.972	520	95	charge	+775	charge
13	640	95	510	95	charge	+1150	charge
14	569	95	500	95	charge	+1069	charge
15	472	95	410	95	charge	+882	charge
16	229	95	150	95	charge	+379	charge
17	135	95	−10	95	charge	+125	charge
18	−175	95	−60	95	discharge	-	-
19	−170	88.472	−100	95	discharge	-	-
20	−120	88.472	−90	87.5	discharge	-	-
21	−140	82.639	−110	87.5	discharge	-	-
22	−150	82.639	−120	80.556	discharge	-	-
23	−150	75.139	−130	80.556	discharge	-	-

are consistent with the observed charge/discharge behaviour across the 24 h operational periods. The natural day-to-day variability in the recorded wind (60 W–520 W) and solar (0 W–910 W) profiles demonstrates the system’s adaptive response to stochastic renewable generation, providing an implicit sensitivity validation across the observed input range.

Load Characteristics of Microgrids

The variation in load in the classrooms supplied by microgrids 1 and 2 is analyzed for 24 h along with the power generated from solar and wind units. At any point of time the load requirement of the classrooms is managed by the solar, wind and power available in the battery. Any further deficiency is managed by taking power from the other microgrid through distributed energy management. Power deficit analysis is shown in Figure 10. The battery SOC and charge–discharge profile are shown in Figure 11. Renewable utilization to meet load and grid dependency over 24 hrs shown in Figure 12. Cumulative energy vs. load is shown in Figure 13.

8. Real-Time Simulation Results and Discussion

This section presents the results of the JADE-based MAS across all four operating scenarios, based on 24 h of real operational data collected from the campus microgrid installation. Results are discussed in Section 8.1, Section 8.2, Section 8.3 and Section 8.4. Figure captions have been updated to include physical units for all axes; all power quantities are in Watts (W), SOC in percentage (%), and time in hours (h).

8.1. Single Off-Grid Hybrid Microgrid System

The power generated by solar and wind is monitored and the corresponding load variation in the specified classrooms where microgrid-1 is supplying power is recorded for 24 h duration. Table 10 presents the optimized values for the operation of a single hybrid microgrid in islanded (off-grid) mode. The nature of available power and surplus, scarcity in every hour is considered and the Control Agent takes the appropriate action for stable operation of the microgrid.

As can be observed, between the 0th and 6th hour, the load exceeds power generation, causing the battery to discharge and the SOC to drop from 75% to 70.7322% as shown in Figure 14. Between the 7th and 16th hours, solar power increases, reaching a high of 910 W at the 15th hour. This surplus enables the battery to switch to charging mode, with the SOC recovering to 80.3275% by the 15th hour. During the 17th to the 23rd hour, i.e., evening, a change in wind power is noted, causing the battery to alternate between charging and discharging. This agent operation is implemented in the MAS execution platform JADE and the agent communication diagram called the Sniffer diagram is shown in Figure 15. This diagram shows all the agents registered in the Agent Management System and the load agent requests the Control Agent for power. The Control Agent can see the nature of agents through the Directory Facilitator Agent and communicate with agents according to the requirement at that point in time. The intercommunication between agents at every instant and based on the strategic decision by the Control Agent and the outcome action is reflected in the form of communication.

8.2. Single On-Grid Hybrid Microgrid System

Table 11 presents the optimized values for the operation of a single hybrid microgrid in grid-connected (on-grid) mode during 24 h duration.

Table 11 demonstrates clearly that the overall power generation and battery storage will adequately meet the demand during the 0th to 9th, 12th to 16th, and 18th to 23rd hours. As the generation is surplus and the battery is fully charged, the microgrid delivers surplus power to the grid during the 10th and 11th hours. At the 17th hour, with minimal wind and solar power output and exhausted battery reserves, the microgrid draws power from the grid as shown in Figure 16. The energy flow diagram is shown in Figure 17 and the Sniffer diagram is shown in Figure 18. Here, even after receiving power from the available solar wind and the battery, if the load requires further power, it can be supplied from the Grid Agent.

8.3. Two Off-Grid Hybrid Microgrid Systems

Table 12 presents the optimized values for the operation of a two-hybrid microgrid system in islanded (off-grid) mode during a 24 h duration.

It is clearly evident from Table 12 that between the 0th and6th hour, power generation is inadequate to satisfy the load in both microgrids; thus, power from the battery with a greater SOC will be discharged. Between the 7th and17th hours, as solar power generation increases, the overall power output exceeds the load demand causing the excess power to charge the battery. Between the 18th and23rd hours, when the output of solar power reduces, battery power must be utilized to satisfy load demands; hence, the battery with a greater state of charge will be discharged. The sniffer diagram is shown in Figure 19. Here, both the microgrids collectively manage the deficiency of power. The load at microgrid-1 can access the power from the solar in microgrid 2 or vice versa.

8.4. Two On-Grid Hybrid Microgrid Systems

Table 13 presents the optimized values for the operation of a two-hybrid microgrid system in grid-connected (on-grid) mode during a 24 h duration.

From Table 13 it is evident that between the 0th and10th hours and between the 18th and 23rd hours, energy balance is attained in the microgrids, with the batteries operating in charging and discharging modes. From the 11th to the 17th hour, the excess power in the microgrids is sent to the grid as shown in Figure 20. Thus, the agent takes autonomous action and if it is not able to take autonomous action, it gets help from the control agent to take action towards global objectives and thereby evolves to participate in larger goals.

9. Comparative Analysis

This section compares the proposed JADE-based MAS against a conventional JavaScript-based centralized EMS.

Table 14. Structured quantitative performance comparison between the proposed JADE-based MAS and the centralized JavaScript EMS baseline.

Evaluation Criterion	JADE-Based MAS (Proposed)	Centralized JavaScript EMS (Baseline)
Power balance accuracy	Maintained at every hour; zero unmet load across all four scenarios	Equivalent dispatch for same input—correctness validated
Battery SOC constraint compliance	100%—SOC bounded within [20%, 90%] in all scenarios	Compliant within programmed thresholds only
Renewable utilization (on-grid, 24 h)	~107% avg (surplus exported at hrs 10–11)	Identical—no autonomous surplus routing
Grid import events	1 event (hour 17, 230 W import)	Same—no autonomous grid prioritization
Inter-MG power exchange	Autonomous SOC-based negotiation between BA1 and BA2	Not supported—single MG architecture
Fault tolerance	Agent failure does not collapse system (JADE AMS/DF recovery)	Single point of failure—centralized controller
Scalability	Agent plug-and-play via JADE DF—new MGs added without redesign	Requires full code re-architecture per additional MG
Platform independence	JADE runs on any JVM-compatible platform	Node.js/browser-dependent

below provides a structured quantitative comparison across eight evaluation dimensions. Both systems are evaluated on identical 24 h input data—the same hourly load, solar, and wind profiles—ensuring a fair, unbiased comparison.

Table 15. Optimized hourly energy management of a single hybrid on-grid microgrid.

Hr	Load (kW)	Solar (kW)	Wind (kW)	Renewable Share (%)	Battery SOC (%)	Battery Action (kW)	Grid Interaction (kW)
0	300	0	300	100	75.0	0	0
1	300	0	200	67	72.2	−100	0
2	300	0	240	80	70.6	−60	0
3	300	0	420	140	73.9	+120	0
4	350	0	450	129	76.7	+100	0
5	500	0	460	92	75.6	−40	0
6	600	50	120	28	63.6	−430	0
7	800	260	380	80	59.2	−160	0
8	900	460	660	124	65.3	+220	0
9	1000	610	750	136	75.3	+360	0
10	1100	700	1000	155	90.0	+529.2	+70.8
11	1300	740	690	110	90.0	0	+130
12	1400	730	430	83	83.3	−240	0
13	1400	670	390	75	73.9	−340	0
14	1300	550	230	60	59.4	−520	0
15	1300	380	360	57	43.9	−560	0
16	1250	170	490	53	27.5	−590	0
17	1100	0	600	55	20.0	−270	−230
18	400	0	740	185	29.4	+340	0
19	400	0	630	158	35.8	+230	0
20	300	0	580	193	43.6	+280	0
21	300	0	440	147	47.5	+140	0
22	300	0	620	207	56.4	+320	0
23	200	0	530	265	65.5	+330	0

presents the JavaScript EMS results.

As Table 14, Table 15,

Table 16. Energy source utilization.

Time Block	Dominant Source	Battery Role	Grid Dependency	Operational Interpretation
00–05	Wind	Mild discharge/charge	None	Wind-driven autonomous operation
06–08	Battery	Heavy discharge	None	Morning load ramp support
09–11	Solar + Wind	Aggressive charging	Import at the SOC limit	Renewable surplus saturation
12–16	Battery	Deep discharge	None	Peak demand shaving
17	Wind + Grid	Discharge + export	Active	SOC floor protection
18–23	Wind	Progressive charging	None	Night-time recovery phase

,

Table 17. Battery utilization and stress assessment.

Metric	Value	Interpretation
Maximum SOC	90%	Upper safety bound respected
Minimum SOC	20%	Battery degradation avoided
Peak Discharge	−590 kW (Hr 16)	Worst-case support during load peak
Peak Charge	+529.2 kW (Hr 10)	Renewable over-utilization
SOC Violations	None	Constraint-compliant optimization

and

Table 18. Grid interaction summary.

Mode	Hours	Energy Role
Grid Import	10–11	Backup during SOC saturation
Grid Export	17	Excess wind export
Islanded Operation	21 h	High resilience and autonomy

demonstrates, while both approaches yield numerically equivalent energy dispatch results for the same input data—thereby validating the correctness of the MAS agent decisions against a known baseline—the JADE-based MAS offers substantial qualitative and architectural advantages. These advantages become decisive in larger, more complex multi-MG deployments where centralized EMS architectures fail due to scalability limitations, single-point-of-failure vulnerabilities, and inability to support autonomous inter-MG coordination. The equivalence in energy outcomes is itself a meaningful result: it confirms that decentralized multi-agent decision-making does not compromise dispatch optimality relative to centralized control, while delivering superior architectural properties.

10. Conclusions

This paper has presented a JADE-based Multi-Agent System (MAS) for real-time energy management of a low-voltage hybrid dual microgrid system, physically implemented on a campus rooftop and in a laboratory. The proposed system comprises two interconnected microgrids, each rated at 2.5 kW (1 kW solar PV + 1.5 kW wind), operating on a 24 V DC bus with 150 Ah battery storage and validated across four operating scenarios: single MG off-grid, single MG on-grid, dual MG off-grid, and dual MG on-grid.

The key findings of this work are as follows: (i) The JADE-based MAS successfully maintains power balance at every hourly decision step across all four scenarios, with zero unmet load demand; (ii) battery SOC is maintained within the safe operating bounds of 20–90% in all cases, confirming constraint-compliant autonomous operation; (iii) the inter-MG SOC-balancing mechanism effectively enables cross-MG power sharing, reducing grid dependency in the dual-MG configuration; and (iv) a direct comparison with a centralized JavaScript-based EMS confirms equivalent dispatch accuracy while demonstrating the superior scalability, fault tolerance, and modularity of the proposed decentralized MAS architecture.

The limitations of the current study that should be noted include: the experimental platform is a campus-scale, low-voltage system (2.5 kW per MG, 24 V DC bus), and extrapolation of conclusions to higher-voltage, larger-scale networks requires further validation; the EMS operates at an hourly scheduling timescale and does not address inner voltage/current control loops; formal fault injection testing (e.g., agent failure, communication disruption, generation dropout) has not been conducted, though the JADE platform’s AMS and DF services provide inherent recovery mechanisms; and the battery SOC model relies on coulomb counting, which accumulates measurement error over time.

Inherent features of MAS are leveraged for dynamic adaptation and distributed optimization to get the best of the service to the consumers across the network all the time, leading to economic and environmental optimization.

Future endeavours will concentrate on incorporating sophisticated machine learning methodologies and evaluating the framework within extensive, linked microgrid networks.