Next-Generation Automated Adaptive Protection Enabled by Geospatial Load Forecasting in Distribution Networks

Abstract

Modern distribution networks increasingly face operational stress from variable demand and high penetration of distributed energy resources, challenging the adequacy of purely reactive protection schemes. This study addresses this challenge by enhancing a developed adaptive protection software platform with a Geographic Information System (GIS) driven predictive load forecasting capability to enable anticipatory protection coordination. The proposed framework integrates spatially resolved demand modeling, regulatory and planning constraints, and machine learning-based short- to medium-term load forecasting with a relay coordination and optimization engine. Forecasted load profiles are used as inputs to an optimization layer that proactively updates relay pickup and time delay settings to maintain selectivity and system security under predicted operating conditions. The approach is validated at laboratory scale using real Intelligent Electronic Devices (IEDs) interfaced with synthetic GIS-based network and load datasets. Experimental results indicate that incorporating forecast-informed settings improves coordination margins and reduces the risk of relay maloperation compared with reactive adaptive protection alone. The findings demonstrate that coupling GIS based constrained load forecasting with adaptive relay control can enhance protection performance in active distribution networks, supporting more resilient and forward-looking protection strategies.

1. Introduction

Distribution networks are undergoing rapid transformation driven by distributed generation, electrification of transport, and increasing urban density [1,2,3]. Traditional protection philosophies based on static relay settings derived from worst-case studies are strained by the resulting variability [4,5]. Adaptive protection, where relay settings are adjusted to real-time topology and loading, provides a more responsive approach [6,7,8]. The responsiveness of protection devices can be proactively managed by linking adaptive relay settings with a refined load-forecasting framework. Instead of reacting solely to instantaneous measured currents, the relay compares live load data against predicted load thresholds, enabling controlled and stabilized adaptive setting changes. This is achieved through a dedicated database that stores real-time relay measurements, historical load patterns, and demand-growth forecasts derived from urban planning policies, substation capacity constraints within each distribution zone, and socio-economic development trends.

Traditional relays operate based on fixed settings including pickup current and time delay derived from the load flow and fault current analysis in the worst-case scenarios [9,10]. While these settings are reliable in specific conditions, they are often inefficient in handling real-time system fluctuations [11], resulting in inaccurate response times or unnecessary outages [12]. In contrast, the adaptive protection system dynamically adjusts their settings based on the actual operating conditions of the electricity network. It enhances protection performance, improves system resilience and is particularly important as modern electricity networks experience frequent and significant changes in load such as electric vehicles, generation from intermittent renewable energy sources and fault characteristics. The transition toward smart grids and higher reliability expectations has accelerated interest in adaptive and intelligent protection schemes. Recent studies have proposed communication-less adaptive protection approaches that combine local load-flow measurements with artificial neural network (ANN)-based forecasting [13]. While conceptually attractive, their practical deployment remains limited. Commercial protection relays are generally not designed to accept externally generated machine-learning control inputs [14,15], and embedding such algorithms directly into relay firmware introduces significant technical, validation, and commercial risks. Moreover, external ANN implementations undermine the communication-less premise, as relay behavior in real networks cannot be isolated from broader system interactions that affect learning accuracy.

Parallel research has advanced adaptive protection across different grid architectures. In DC microgrids, adaptive overcurrent protection schemes using IEC 61850 signaling and clustering techniques such as K-means have demonstrated faster fault clearance and improved selectivity under partial communication availability [16]. However, challenges related to grounding, non-uniform system configurations, and real-time implementation continue to hinder large-scale adoption. Adaptive wide-area backup protection has also progressed to address dynamic topologies and high distributed generation penetration [17]. These methods enhance responsiveness using limited system information and self-adjusting protection parameters.

A broad survey of adaptive protection strategies for emerging smart grids and microgrids has previously been reported in the literature [7], outlining the main technical barriers and identifying evolving research directions. These reviews highlight several attempts to progress adaptive protection toward practical use; however, most reported methods remain confined to simulations, software-based prototypes, or limited pilot environments, without full validation on actual intelligent electronic devices (IEDs) [18]. Examples include the use of preconfigured relay setting groups to accommodate changing network states [19], fuzzy logic-driven adaptive overcurrent protection for large wind generation facilities [20,21], distance relay adjustments to accommodate variable renewable generation [22], Java-based tools supporting communication for adaptive coordination [23,24], algorithms that address coordination loss between primary and reverse-directional overcurrent relays in distributed generation (DG) networks [25,26,27], programmable logic controller (PLC) centric adaptive protection frameworks [28] and approaches using neutral currents and load estimation to inform adaptive behavior [29]. Nevertheless, practical barriers persist, including communication reliability, scalability, and consistent performance under rapidly changing operating conditions, underscoring the gap between theoretical advances and deployable protection solutions [7].

Historically, load forecasting relied on statistical models such as Autoregressive Integrated Moving Average (ARIMA) and multiple linear regression. These models, while effective for linear and stationary data, often struggled with the non-linear and dynamic nature of electricity consumption patterns. The advent of Artificial Intelligence (AI) introduced methods like ANNs, Support Vector Machines (SVMs), and more recently, Deep Learning models such as Long short-term memory (LSTM) networks [30,31,32]. Despite advancements, existing methods exhibit certain limitations such as data dependency, lack of spatial context and interpretability [30,31]. Many AI-based models require vast amounts of high-quality data, which may not always be available [30,33]. Traditional models often overlook the spatial distribution of loads, which is crucial for urban planning and infrastructure development [32], and complex models like deep neural networks can be challenging to interpret, making it difficult for planners to understand the underlying factors influencing forecasts [34].

Recent studies demonstrate that machine learning and deep learning models such as convolutional neural networks (CNN)–LSTM hybrids, and ensemble methods can achieve high accuracy in short-term load forecasting by exploiting nonlinear patterns in historical and contextual data [35,36,37]. Review works further highlight the strengths of deep learning in feature extraction and handling high-dimensional data, while also noting persistent challenges related to generalization, overfitting, data dependency, and limited interpretability. Critically, these studies primarily evaluate forecasting accuracy in isolation and are largely decoupled from practical network operations, protection coordination, and geospatial system constraints. Existing AI-based forecasting approaches including ANN, SVM, CNN-LSTM models and ensemble techniques often require substantial quantities of high-quality training data and can operate as black-box predictors with limited transparency in engineering decision making. While these techniques may provide high forecasting accuracy, most do not explicitly incorporate spatial relationships, urban planning constraints, customer-development characteristics, or asset level infrastructure interactions that are important for utility planning and protection applications. In addition, their outputs are typically not directly linked to adaptive protection systems or operational relay coordination processes.

In contrast, the proposed GIS constrained framework incorporates zoning regulations, planning policies, customer classifications, historical asset performance, and geospatial infrastructure information into the forecasting process. This integration improves interpretability, maintains traceability between customers and network assets, and provides a practical pathway for direct deployment within adaptive protection environments. The framework therefore extends beyond forecasting performance alone by enabling forecast informed operational decision making and protection coordination within real distribution networks.

From the above discussion, the main contributions of this paper can be summarized as follows:

• Development of a unified GIS-based database that integrates geospatial network data, planning rules, IEC-based maximum demand limits, customer-type classifications, and historical energy consumption to enable dynamic load demand forecasting.
• Enabling automatic adjustment of relay settings based on predefined forecast-driven thresholds.
• Demonstration of scalability and immediate practicality through laboratory validation with commercial IEDs, supporting direct applicability to real distribution networks.

2. Network Topology and Laboratory Setup

The experiment used a modified version of an operational distribution network. The modifications served two purposes: protecting commercial property and adapting the network to a level suitable for industrial protection relays in a laboratory setup. Using the modified distribution network in Curtin University’s IEC 61850 Protection Laboratory, the study validated the full workflow with industrial protection relays used in high-voltage distribution networks. The tests demonstrated reliable operation of the load-forecasting database, effective communication with the adaptive relays, and accurate implementation of adaptive setting adjustments in response to forecasted load profiles.

The modified high-voltage network of the substation, shown in Figure 1, comprises a 6.6 kV grid supply that steps up to 11 kV through a substation transformer to supply the primary substation bus.

This paper introduces an additional layer of functionality by integrating externally generated GIS-based load-forecasting data. The independently validated software platform analyses predicted loading patterns, compares them with real-time measurements from the IEDs, and automatically updates relay setting files to match anticipated network operating conditions.

Two outgoing feeders, Feeder A and Feeder B, connect to the primary substation bus and supply essential downstream loads. To ensure supply continuity under the N−1 reliability criterion [38], the network requires alternate power paths during equipment outages. The feeders therefore interconnect through a Ring Main Unit (RMU), which typically operates in a normally open configuration but can automatically close to restore supply when either feeder loses its source.

The network also includes a 3 MW wind turbine that serves as an auxiliary power source. However, this study does not consider the intermittency of the renewable generation or the additional variability it introduces to adaptive protection performance. The protection setup incorporates multiple IEDs, with the Schweitzer Engineering Laboratories’ SEL-487E relay selected for laboratory implementation in this study. The relay provides five isolated current input channels designated S, T, U, W, and X by the manufacturer. Each channel can issue trip signals to the corresponding circuit breakers under different system contingencies, as outlined in

Table 1. Relay designation and corresponding manufacturer-defined winding nomenclature.

Component Being Protected	Relay Name	Winding Nomenclature
Feeder A	Relay-S	S
Feeder B	Relay-T	T
Interconnection feeder (for RMU)	Relay-U	U
Wind Generation	Wind Gen-W	W
Substation feeder	Substation relay-X	X

. As illustrated in Figure 1, the S-winding is assigned to Feeder Load A, the T-winding protects Feeder Load B, and the U-winding supervises the RMU intertie. The laboratory setup replicates this configuration using a scaled three-winding arrangement that mirrors the topology in Figure 1, enabling realistic evaluation of relay performance under representative distribution-level operating conditions.

In practice, for example, the S-winding associated with Feeder A interfaces directly with the Feeder A circuit breaker (CB) and issues a trip command upon fault detection. All winding identifiers in the electrical model follow the manufacturer’s naming conventions defined in the proprietary relay configuration software, ensuring consistent mapping between the developed software, the laboratory hardware, and the cross-validation of adaptive protection settings.

The laboratory environment emulates a modern substation communication architecture and integrates IEDs from multiple major vendors. Figure 2 shows the network configuration corresponding to the actual laboratory setup. A server PC connects to the main communication panel to enable data recovery in the event of unforeseen system failures and hosts the developed load-forecasting database.

The communication panel contains the primary network switch, which interfaces with the IED panels via a fibre-optic network. Each IED is preconfigured with settings corresponding to the normal network operating condition. To emulate alternative network configurations, current is injected directly into the relays to reproduce a range of loading scenarios, as illustrated in Figure 3. This process prompts the simulated network operator to close the interconnecting RMU, enabling power to supply additional loads through an alternate path and thereby altering the network loading configuration.

Figure 3 illustrates the copper connections at the rear panel of the SEL-487E IED, interfacing with a current injection unit. This setup enables the simulation of varying load currents, allowing the IED to operate under test conditions that closely replicate real-world operating scenarios and practical substation environments.

Figure 3 also shows the fiber connectivity at the rear of the IED, which facilitates real-time monitoring of network load current, breaker status and other configured measurement parameters.

3. GIS-Based Load Forecasting Database

Traditional forecasting models that rely primarily on historical meter data and often lack spatial context no longer provide sufficient accuracy or relevance. Rapid urban development requires a shift beyond conventional load forecasting toward a ground-referenced, spatially aware framework. The proposed GIS-integrated methodology captures the complexity of urban planning variables, including zoning regulations, vertical growth potential, and historical consumption patterns, producing forecasts with improved design relevance and financial reliability. Figure 4 presents a holistic, data-driven load forecasting framework that integrates historical network performance, current customer and asset data, and future urban and socio-economic projections to derive a per-customer maximum load forecast. This approach is particularly suited to distribution networks, where demand uncertainty is influenced by urban expansion, customer behaviour, and evolving policy and planning conditions. The process comprises four main streams, which are described in detail below.

3.1. Past Network Data Stream

The first stream establishes a historical baseline for load behavior by integrating long-term network and customer demand records. It draws on historical demand data collected from feeder-level measurements, customer meters, and utility electrical assets such as distribution transformers and feeders. These datasets are harmonized to account for seasonal variations, long-term growth trends, and evolving customer composition over time. Rather than relying solely on coincident peak demand, the historical data are analyzed to derive diversity and coincidence characteristics at both asset and customer levels. This characterization provides a statistically robust representation of realistic loading conditions, forming a reliable basis for defining credible present and future protection boundaries within adaptive protection schemes.

3.2. Present Network Data Stream

The second stream captures the present operating condition of the distribution network through detailed customer surveys and updated metering information. The GIS-based model records current electrical consumption and observed peak demand for each customer and explicitly links these data to the corresponding network assets. Data validation and integration processes reconcile discrepancies between historical records and current measurements. To ensure engineering consistency, the analysis applies international loading standards, including International Electrotechnical Commission IEC [39] and Australian New Zealand Standards (AS/NZS) guidelines [40], at the appliance level to derive realistic connected and diversified loads. In parallel, the analysis incorporates city industrial planning regulations to enforce present-day constraints including allowable building heights, floor areas and industrial use. This stream produces an accurate and regulation-compliant snapshot of the current network state, forming a reliable reference for adaptive relay setting calculations.

3.3. Future Network Data Stream

The third stream introduces forward-looking information to account for long-term demand evolution driven by urban development and socio-economic factors. City and industrial planning acts and legislation define permissible development boundaries while city master plans and zoning regulations further refine these constraints. Population forecasts translate urban growth into electrical demand potential, while historical building structure patterns inform the expected electrical characteristics of future developments. In addition, the analysis incorporates economic policies and global energy price trends to adjust demand sensitivity and customer behavior. Historical diversity factors derived from past data adapt to these future conditions, ensuring that projected loads remain realistic rather than overly conservative. This stream enables the forecasting framework to anticipate demand growth in a manner that is consistent with both planning policy and historical usage behavior.

3.4. Integrated Load Forecast and Output Stream

The final stream consolidates the past, present, and future data within a dedicated software platform designed for multi-source data integration. Historical diversity characteristics, present-day metered loads, and planning-constrained future projections are combined to produce a unified, customer-level maximum load estimate. This integration preserves traceability from individual customers to upstream assets, enabling load aggregation at any network level required for protection studies. The resulting per-customer and per-asset maximum load forecasts are directly applicable for adaptive protection schemes, in which relay pickup currents and coordination settings are dynamically adjusted in response to changes in network topology, loading conditions, and time-varying system behavior. By replacing static worst-case assumptions with dynamically informed load estimates, this stream provides a quantitative foundation for reliable and selective adaptive protection in modern distribution networks.

Each stream contributes progressively refined information that is validated, standardized, and integrated through a dedicated software platform. The GIS-integrated approach offers several benefits over existing models by incorporating spatial data and realistic zoning rules; the model provides more precise forecasts, especially in rapidly urbanizing areas [32,41]. The integration of regulatory and spatial data enables a clearer understanding of the factors influencing load patterns. The methodology is modular and adaptable to different urban contexts, making it suitable for large-scale deployment.

4. Proposed GIS-Based Load Forecasting

Adaptive protection schemes require accurate, spatially resolved, time-varying load estimates to maintain relay sensitivity and selectivity under changing conditions. Conventional settings rely on static worst-case assumptions that overlook customer diversity, urban constraints, and evolving demand. The proposed approach integrates historical, operational, and planning data to produce asset- and customer-level maximum load estimates that directly support adaptive relay settings as elaborated below.

4.1. Traditional Load Forecasting Approach

Conventional distribution load forecasting methods typically estimate peak demand using historical feeder-level data and apply a fixed growth factor, as given by (1).

$$P_{trad}(t)=P_{hist,peak}.{\left(1+g\right)}^t$$

(1)

where P_hist,peak is the historical maximum feeder load, $g$ is an assumed annual growth rate, and t is the forecast horizon.

For protection studies, this approach yields a single conservative load value ($I_{tradrelay}$) applied uniformly across network assets. It is calculated from the line voltage V and power factor $\cos \mathrm{\varnothing }$ as given by (2).

$$I_{tradrelay}={\displaystyle \frac{P_{trad}}{\sqrt{3}V\cos \mathrm{\varnothing }}}$$

(2)

This method neglects customer-level diversity, spatial variations in urban growth, planning constraints and building typologies and policy-driven demand changes. As a result, relay pickup currents are often overestimated, reducing protection sensitivity during lightly loaded or reconfigured network states.

4.2. Proposed Load Forecasting Approach

The proposed method begins by extracting long-term historical demand data at both customer and asset levels. Instead of relying solely on coincident peaks, it derives diversity-aware historical load factors, as shown below:

$$P{\displaystyle \frac{hist}{i}}=max+\left({\displaystyle \frac{1}{N}} {\sum }_{k=1}^N{P}_{i,k}\left(t\right)\right)$$

(3)

where P_i,k(t) is the time-series load of customer k connected to asset i; N is the number of connected customers.

This formulation captures realistic coincident loading, which is essential for setting adaptive protection thresholds. Current customer surveys and metering data are used to determine actual connected load, which is then normalized using international appliance loading standards (IEC/AS/NZS), as given by (4). This ensures that load estimates used for protection studies reflect practical engineering data rather than nameplate margins.

$$P_k^{present}={\displaystyle {\sum }_{a=1}^A}{\alpha }_a{P}_a^{rated}$$

(4)

where P_a^rated is the rated power of appliance a, and α_a is the standard-based utilization factor.

4.3. Integrated Maximum Load Estimation

Future load $P_k^{future}$ is not extrapolated using a generic growth factor. Instead, it is constrained by city and industrial planning legislation, population forecasts, and building form limits as given by (5).

$$P_k^{future}=f (A_{max},H_{max },{\rho }_{pop},{\beta }_{econ})$$

(5)

where A_max is maximum allowable floor area, H_max is maximum structure height, ρ_pop is population density forecast, and B_econ represents economic and energy price sensitivity.

4.4. Future Load Projection Using Planning Constraints

The final per-customer maximum load used for protection analysis is derived through a software-based integration of historical, present, and future components, each weighted by a factor ${\rm Y}$ as below.

$$P_k^{max}={{{\rm Y}}_{1}P}_k^{hist}+{{{\rm Y}}_{2}P}_k^{present}+{{{\rm Y}}_{3}P}_k^{future};{{\rm Y}}_1+{{\rm Y}}_2+{{\rm Y}}_3=1$$

(6)

Using the proposed load estimate, relay pickup currents are dynamically calculated for each asset as given by (7) in which ∂ (t, topology) is a topology- and time-dependent adjustment factor derived from GIS and network configuration.

$$I_{relay}^{adaptive}={\displaystyle \frac{P_i^{max}}{\sqrt{3}V\cos \mathrm{\varnothing }}}.\partial (t, \mathrm{topology})$$

(7)

To improve reproducibility and clarify the relay adaptation process, the optimization framework used for adaptive setting updates is further detailed. Following threshold validation and persistence checks, the coordination engine recalculates relay pickup currents and operating times while preserving conventional protection coordination constraints. The optimization seeks to ensure secure relay operation under time-varying loading and network configurations while maintaining established protection principles.

The objective is to reduce unnecessary relay operations, minimize coordination violations between primary and backup devices, and limit excessive operating delays under forecast-informed network conditions. This can be formulated as a general optimization problem, as given by (8).

$$\min J={\omega }_1{N}_{op}+{\omega }_2{C}_v+{\omega }_3{T}_d$$

(8)

where J is the overall adaptive protection performance objective function; $N_{op}$ is number of unnecessary relay operations (false or nuisance trips); $C_v$ is coordination violations between relays; ω₁, ω₂, ω₃ are weighting factors determining the importance of each objective; and $T_d$ is relay operating delay.

The optimization is performed subject to practical protection coordination constraints including coordination margins, relay sensitivity requirements, and backup protection relationships. The principal constraints include:

$$\begin{array}{c}CTI\ge 0.3 \mathrm{s}\\ I_{pickup}>I_{max,diversified}\\ T_{primary}<T_{backup}\end{array}$$

(9)

where CTI denotes the coordination time interval between primary and backup protection devices, $I_{pickup}$ is relay pickup current, $I_{max,diversified}$ represents the maximum diversified load current obtained from the GIS-based forecasting framework, and $T_{primary}$ and $T_{backup}$ represent the operating times of primary and backup relays, respectively.

The optimization engine continuously evaluates these constraints using forecast-informed thresholds and real-time IED measurements prior to deploying adaptive settings. This ensures that relay updates remain selective, secure, and operationally feasible under varying network conditions, while also improving the transparency and reproducibility of the proposed methodology.

Table 2. Logical table of Adaptive changes to IED settings.

Input Winding1	Input Data1	Data Trigger1 (AMP)	Output Set	Output Winding	Output Setting	Output Value
S	MAG (Amp, primary)	Forecasted Load Threshold	1	U	50UP1P (Pick up)	35
S	MAG (Amp, primary)	Forecasted Load Threshold	1	U	67UP1D (Duration)	0.01

summarizes two of the forecast-based load thresholds, which are used to guide the adaptive behavior of the IEDs. These two thresholds are simulated by injecting current into the relays to trigger protection setting change.

The injection points were mapped to the feeder breaker terminals according to the manufacturer’s winding nomenclature (S and T) as shown in Figure 5. For winding S, the injected three-phase currents were I_AS (red), I_BS (yellow), and I_CS (blue). For winding T, the injected three-phase currents were I_AT (red), I_BT (yellow), and I_CT (blue). The grounding connections were configured correctly, with the respective terminals shorted (black) to maintain system integrity.

The associated software for the current injector hardware controls the current magnitudes supplied to the IEDs. The relays continuously monitor real-time network measurements and adjust their protection parameters when the observed load exceeds the forecasted value. The two threshold levels are demonstrated to trigger an output to the IED to change the pick-up and time for the new protection setting. The same methodology can support any number of forecast scenarios within a larger distribution system.

The forecasted thresholds act as reference limits that help stabilize adaptive behavior, preventing unnecessary relay adjustments during short-term load fluctuations. A time-delay mechanism is built into a separately experimented software to ensure that only persistently elevated loads trigger an update to the IED settings. Instances where the load surpasses the thresholds including magnitude and duration are logged to support future refinement of forecast models and recalibration of adaptive trigger levels. The developed database itself is highly flexible and can be applied to any distribution network, integrating either pre-existing load-forecasting processes or newly developed forecasting techniques, provided coordination is maintained with the utility’s adaptive protection and operational planning teams.

The proposed methodology was validated in a laboratory environment using a two-step testing procedure. In the first step, the historical load forecasting component was evaluated by processing archived network demand data through the developed software platform. The forecasted load outputs were then systematically compared against actual load measurements recorded by the utility. This comparison was used to verify the accuracy of the historical data integration, diversity modelling, and load estimation logic embedded within the software framework.

In the second step, the impact of the proposed forecasting methodology on adaptive protection performance was experimentally assessed. Adaptive relay settings were applied to a laboratory-based protection setup. A secondary current injection system was used to emulate network load demand conditions. To facilitate clear and repeatable observation of relay behavior, high current levels were intentionally injected to force near-instantaneous relay operation. This approach simplified event logging and ensured reliable capture of relay response times under adaptive setting changes. The results confirmed that relay pickup and tripping characteristics are adjusted as expected in response to the changed load driven adaptive settings.

4.5. Workflow Between GIS Database and Relay Coordination Engine

Figure 6 illustrates the workflow linking the GIS-based forecasting framework with the adaptive relay coordination engine. The proposed architecture transforms heterogeneous historical, operational, spatial, and planning datasets into forecast-informed relay setting updates through a structured sequence of processing, validation, and optimization stages as explained below.

Step 1: Historical Network Data Integration

Historical customer demand profiles, feeder loading records, transformer measurements, and utility asset information are imported into the database. These datasets provide long-term behavioural patterns and enable identification of seasonal variations, customer diversity characteristics, and historical loading trends. Diversity factors and coincident demand characteristics are extracted to establish realistic loading behaviour rather than relying solely on peak demand values.

Step 2: Present Network Data and GIS Asset Linkage

Current metering information and customer demand measurements are integrated with geospatial asset records through a GIS-based linkage process. Spatial association enables direct mapping between customers, feeders, transformers, substations, and protection devices. Data validation procedures reconcile inconsistencies between historical and present datasets while ensuring traceability between electrical assets and customer demand sources.

Step 3: Future Planning and Regulatory Constraint Integration

Future planning information including zoning regulations, population forecasts, urban development restrictions, allowable building limits, and economic indicators are incorporated into the forecasting environment. These constraints provide physically realistic demand boundaries and prevent overestimation of future loading conditions.

Step 4: Forecast Threshold Generation

The integrated dataset is processed through a forecasting module that combines historical trends, present operating conditions, and planning-constrained future growth information to generate forecast-informed load thresholds. These thresholds represent anticipated network operating conditions rather than static worst-case assumptions.

Step 5: Comparison with Real-Time IED Measurements

Forecasted thresholds are continuously compared against real-time measurements obtained from IEDs, including feeder currents, breaker status information, and relay metering values. This comparison establishes whether actual operating conditions align with predicted behaviour.

Step 6: Threshold Persistence Verification

To reduce sensitivity to temporary load fluctuations and forecasting uncertainty, a persistence verification mechanism is applied. Adaptive setting modifications are initiated only when threshold exceedance remains sustained for a predefined duration. This supervisory filtering mechanism minimizes unnecessary relay setting updates and improves operational security.

Step 7: Relay Optimization and Setting Deployment

Following threshold validation, an optimization engine recalculates relay pickup currents, time delays, and coordination settings while maintaining protection constraints including selectivity margins and backup coordination requirements. Updated settings are then automatically transferred to the relay system through the communication interface.

Step 8: IED Feedback and Continuous Learning

Following deployment, relay measurements and operational responses are continuously monitored and logged. Feedback information including breaker operation, event records, and measured load conditions can subsequently be used to refine future forecasting performance and adaptive threshold calibration.

The developed data-processing framework utilizes a configurable Extract–Transform–Load (ETL) architecture to integrate heterogeneous spatial and non-spatial datasets into a unified operational environment. The framework supports automated acquisition, transformation, validation, and synchronization of data originating from multiple sources, including GIS datasets, utility asset records, customer demand information, planning databases, and relay measurement outputs.

The processing workflow applies rule-based filtering, attribute mapping, spatial joins, topology verification, and temporal data reconciliation to establish relational links between customers, electrical assets, and forecast variables. Historical and present network datasets are standardized through schema harmonization procedures, allowing information from different utility systems to be integrated into a common structure suitable for forecasting and adaptive protection applications.

The architecture further supports event-driven workflows whereby forecast outputs and real-time relay measurements can trigger predefined processing routines. This capability enables automated generation of forecast-informed load thresholds and their subsequent transfer to the relay coordination module. Such an approach allows dynamic updating of adaptive protection parameters while maintaining traceability and consistency across multiple data sources.

Unlike conventional adaptive protection approaches that respond only after network conditions change, the proposed framework introduces a supervisory predictive layer. Forecast outputs are not used directly as relay commands; rather, they establish forecast-informed operational thresholds validated against real-time IED measurements. A persistence-duration criterion is implemented so that temporary forecast deviations do not trigger adaptive setting changes. This approach reduces sensitivity to forecasting uncertainty and improves operational security.

The modular structure of the processing framework allows future integration of additional forecasting techniques, distributed energy resource models, and utility operational systems without requiring modifications to the overall adaptive protection architecture. Unlike a black-box implementation, the proposed framework follows a deterministic rule-based data processing workflow in which each transformation stage remains transparent and traceable.

5. Simulation and Experimental Results

Two case studies were conducted to evaluate the robustness, adaptability, and practical applicability of the proposed adaptive protection framework under varying operating conditions. The first case study investigates adaptive relay coordination under dynamic network loading conditions, while the second case study evaluates the integration of forecasting and decision support.

5.1. Case 1—Adaptive Protection in Dynamic Power Network

The developed adaptive protection framework was evaluated using a modified field distribution network incorporating multiple feeder configurations and changing loading scenarios. The proprietary software used for the SEL-487E IED is AcSELerator, which designates relay windings as S, T, U, W, and X. Currently, users can only change the values of the relay characteristics by using manufacturer’s proprietary software through a manual process. A software platform, developed and validated through a separate laboratory experiment, enables the adjustment of parameters for each winding by acquiring input values from the IED’s constant current and voltage outputs, which represent a range of network loading conditions. This study considers scenarios involving altered network loading conditions, in which the interconnecting RMU in Figure 1 remains closed to sustain supply to both Feeder A and Feeder B following load variations on either Relay S or Relay T. Such variations necessitate corresponding adjustments to the overcurrent protection settings. The developed, laboratory-validated software continuously monitors load demand using the relay’s metering parameters, including real-time current measurements, and mitigates the risk of unintended overcurrent tripping associated with static settings. This ensures that the breaker on the healthy feeder is adaptively configured to avoid nuisance tripping during backup supply from the alternate feeder.

The modified field network, illustrated in Figure 1, was utilized to perform load flow and fault analysis under normal operating conditions. Optimized relay protection settings were derived using dedicated power system analysis software [42,43]. Load flow and fault studies were performed under multiple operating conditions, including normal operating conditions, increased feeder loading, alternate feeder backup operation, fault conditions following topology changes and dynamic load transfer scenarios.

To further illustrate the implemented methodology,

Table 3. Real-time monitored parameters used for adaptive relay updates.

Parameter	Description	Source
Phase Current	Real-time feeder loading	Relay metering
Breaker Status	Open/Close Network state	IED Logic
RMU status	Network topology condition	SCADA input
Fault current level	Short-circuit condition	Power system model
Relay setting	Active protection profile	Adaptive software

summarises the monitored parameters used by the software decision engine.

Figure 7 illustrates the investigated network, where a fault is introduced on Feeder A following a change in network loading, indicated by the red bolted-fault symbol. For validation purposes, the developed software was intentionally configured to not update the protection settings under the new loading conditions. When the fault occurs at the bus section connected to the alternate source, a miscoordination in breaker operation is observed, as shown in Figure 7. To correct this miscoordination operation, revised protection settings were derived from the power system model and adaptively applied. These settings were embedded within the developed software, which dynamically updates relay parameters to ensure proper coordination of IEDs in accordance with the prevailing network topology and loading conditions. Figure 8 illustrates the corrected tripping sequence when a fault at the same location is re-initiated.

The adaptive framework successfully restored coordination between protective devices and prevented nuisance tripping under altered operating conditions.

Table 4. Comparative protection performance.

Performance Metric	Conventional	Adaptive
Miscoordination observed	Yes	No
Nuisance tripping	Yes	No
Manual intervention	Required	Not required
Response adaption	Static	Dynamic
Network topology awareness	No	Yes

summarises comparative system performance.

The results demonstrate that dynamic relay setting adaptation improves coordination reliability while reducing operational dependence on manual intervention. Furthermore, the developed architecture supports practical implementation for modern distribution systems experiencing changing load demand and network reconfiguration.

5.2. Case 2—Adaptive Protection in Renewable-Integrated Networks

The same network was subsequently adapted with updated protection settings, as shown in Figure 8. However, if these updated protection settings are again treated as static and a fault is reintroduced at the same location with the renewable source in service, system protection performance deteriorates.

In this scenario, the interconnected RMU remained closed, with Feeder B and the renewable source energized. The presence of intermittent generation altered the fault current profile, and without further adaptive updates, resulted in miscoordination of protection devices, as demonstrated in Figure 9. The relay closest to the fault was expected to operate first to isolate the faulted section and maintain system stability but failed to trip appropriately. Instead, the Feeder B breaker operated incorrectly due to relay maloperation. The resulting protection coordination mismatch is illustrated in Figure 10, highlighting the limitations of static relay settings in dynamic networks with high renewable penetration.

A comprehensive power system study was conducted, and the updated protection coordination settings, as shown in Figure 11, were implemented in the developed database for the software. The developed software was then deployed to simulate real-time network conditions, specifically to detect the integration of intermittent renewable generation within the system. Upon detection of the renewable energy source, the software automatically recalculated the protection coordination parameters and updated the relay settings accordingly. This adaptive process ensured that the protection scheme remained aligned with the dynamic operating conditions of the network.

The fault was subsequently reintroduced at the same location to validate the effectiveness of the updated settings. As demonstrated in Figure 12, the sequence of breaker tripping was corrected, and the expected coordination hierarchy was successfully restored. The relay closest to the fault operated first, isolating the faulted section while maintaining continuity of supply to the healthy parts of the network.

The comparative bar chart in Figure 13 illustrates the relay operation sequence for a network with a 3 MW renewable source under both non-adaptive and adaptive protection schemes. In the absence of adaptive protection, the relay coordination is compromised, with Feeder B operating first, followed by the Interconnection breaker, wind generation, and finally the main substation. This sequence indicates a clear miscoordination, as the relay closest to the fault does not operate first, leading to unnecessary disconnection of healthy sections of the network. In contrast, when adaptive protection is applied, the correct coordination hierarchy is restored. The Interconnection breaker operates first, effectively isolating the fault at its source, followed by the wind generation, Feeder B, and ultimately the substation as backup protection. The side-by-side comparison clearly demonstrates the effectiveness of the adaptive protection framework in correcting relay operation order, improving selectivity, and enhancing overall system reliability in the presence of intermittent renewable generation.

To further quantify the performance improvement achieved by the proposed framework, a comparison was performed between conventional reactive adaptive protection and the proposed predictive GIS-based adaptive protection approach. The comparison focuses on relay coordination behaviour observed from the generated Time–Current Coordination (TCC) curves and operational outcomes obtained under identical network conditions.

Table 5. Comparison of Reactive and Predictive Adaptive Protection.

Metric	Reactive Adaptive	Proposed Predictive
Coordination margin	0.28 s	0.42 s
Incorrect trips	2	0
Relay setting updates	Post-event	Pre-event
Response mechanism	Reactive	Predictive
Restoration time	Higher	Reduced

summarizes the coordination performance indicators derived from the relay coordination studies. The conventional reactive approach modifies protection settings only after operating conditions have changed and threshold violations are detected. Consequently, relay settings may temporarily operate under non-optimal coordination conditions until corrective actions are initiated. In contrast, the proposed predictive approach utilizes forecast-informed thresholds to initiate adaptive coordination updates prior to critical loading conditions occurring.

The predictive framework increased the coordination margin from 0.28 s to 0.42 s, representing an increase of approximately 50% in relay coordination separation. The increased spacing between relay operating characteristics improved selectivity and reduced overlap between primary and backup relay operating regions. Under the reactive scheme, changes in network loading and generation conditions produced reduced operating separation between relay characteristics, increasing the probability of coordination conflicts and resulting in two observed incorrect relay operations during the evaluated scenarios. Conversely, the predictive framework maintained sufficient separation between relay operating characteristics by proactively updating settings based on anticipated operating conditions. The TCC analysis further demonstrated that relay setting modifications under the proposed approach were initiated before critical threshold exceedance conditions occurred, allowing coordination adjustments to be implemented prior to protection degradation. Consequently, the proposed framework reduced maloperation risk, improved selectivity, and provided faster restoration capability through proactive relay coordination.

These results demonstrate that incorporating GIS-informed forecasting into adaptive protection not only improves forecasting awareness but also produces measurable improvements in practical protection coordination performance.

6. Discussion

The importance of adaptive relays in distribution networks is increasingly recognized, especially following major blackouts such as those in North America and Europe in 2003 [44], and the Texas outage in February 2021 [45]. However, adaptation alone remains inherently reactive. Incorporating load and generation forecasting enables a transition toward predictive protection operation [46]. As such, this paper introduces a GIS-integrated forecasting database that supplies forward-looking information to the relay coordination process. The result is a cohesive framework that enables proactive relay configuration, reducing coordination errors, minimizing unnecessary outages, and improving service restoration performance. The paper contributes the following key advancements:

• Integrated GIS-centric load forecasting database:
  Development of a unified database that combines GIS field data, geospatial zoning, utility planning rules, IEC standard-based maximum demand limits, customer-type classifications, and historical energy consumption. All variables are relationally linked and dynamically updatable to enable accurate, spatially resolved load demand forecasting.
• Practical coupling of forecasting with adaptive protection:
  Direct integration of the forecasting database with adaptive relay control software, enabling modification of protection pickup and time settings based on predetermined, forecast-driven load thresholds without embedding machine-learning algorithms inside relay firmware.
• Laboratory-verified, hardware-in-the-loop methodology:
  Unlike simulation-only approaches, the proposed scheme is experimentally validated using a modified operational distribution network, confirming reliable performance under realistic operating and communication conditions.
• Scalability and field readiness:
  The architecture is inherently scalable and transferable to live distribution networks, supporting incremental expansion, future forecasting methodologies, and seamless adoption by network operators without requiring new communication infrastructure.

In a broader context, the findings have significant implications for future power systems with high penetration of distributed energy resources, electric vehicles, and dynamic load patterns. As distribution networks evolve into active and increasingly decentralized systems, the ability to anticipate and proactively respond to changing operating conditions becomes essential. The proposed framework supports this transition by enabling utilities to move toward more resilient, intelligent, and forward-looking protection strategies. It also aligns with the broader digitalization trend in power systems, where GIS, data analytics, and automation are increasingly integrated into operational decision-making.

Despite these contributions, several areas warrant further investigation. Future research should explore large-scale field deployment to evaluate performance under diverse network configurations and longer operational periods. The integration of probabilistic forecasting and uncertainty quantification could further enhance decision robustness, particularly under high renewable variability. Additionally, extending the framework to incorporate real-time distributed generation forecasting, market signals, and demand response could improve its applicability in fully transactive energy systems.

Practical deployment of the proposed framework also requires consideration of communication latency, interoperability, and cybersecurity requirements associated with adaptive protection systems. Communication delays may affect synchronization between forecast-generated thresholds, relay coordination calculations, and real-time IED updates, particularly in geographically distributed networks with multiple communication layers. Delayed transmission of operational information can potentially reduce the effectiveness of adaptive coordination and may lead to temporary inconsistencies between forecasted and measured network states.

Cybersecurity represents an equally important consideration because adaptive protection systems rely on continuous exchange of operational information and remote setting updates between supervisory systems and protection devices. Unauthorized access, modification of relay settings, manipulation of communication traffic, or false data injection attacks may introduce operational risks that affect protection reliability and system security. Consequently, secure communication architectures, authentication mechanisms, encrypted data exchange, and resilient communication strategies should be considered during practical deployment. Future work will investigate secure IEC 61850 communication frameworks [47] and cyber-resilient adaptive protection architectures to support field implementation.

It is worth noting that the challenges addressed extend beyond intermittent renewable generation. Emerging grid components, such as electric vehicle (EV) charging stations, also introduce substantial load uncertainty due to their stochastic and time-varying demand characteristics [48]. High penetration of EV charging can result in rapid fluctuations in load levels, bidirectional power flows (in vehicle-to-grid scenarios), and altered fault current characteristics, all of which can adversely impact traditional protection schemes. The proposed adaptive framework is inherently capable of addressing these challenges by continuously monitoring system conditions and adjusting relay parameters, accordingly ensuring reliable and selective operation even under highly dynamic load behaviour.

The performance of the proposed framework depends on the availability, completeness, quality, and update frequency of GIS and utility operational datasets. The forecasting process relies on accurate spatial information, customer records, infrastructure data, and planning constraints to generate representative demand thresholds. Outdated planning information, missing customer records, incomplete asset databases, or inconsistencies between utility systems may reduce forecasting accuracy and subsequently influence adaptive relay decision making. Since utility datasets are often maintained across multiple independent platforms, periodic synchronization and validation between GISs, operational databases, and asset management environments are recommended. Improved data governance and automated validation procedures may further enhance forecasting reliability and support long-term operational effectiveness of adaptive protection applications. For electric vehicle charging systems, forecasting models would require incorporation of temporal charging patterns, customer behavioural characteristics, charging station clustering effects, and stochastic demand variations associated with user charging habits. Solar photovoltaic integration forecasting processes would require inclusion of irradiance variability, weather dependency, intermittency effects, and reverse power flow conditions commonly observed under high distributed generation penetration. These additional variables can be incorporated into the GIS database and forecasting environment without requiring fundamental modification of the proposed adaptive protection architecture, thereby supporting broader applicability across evolving distribution network environments.

7. Conclusions

This paper presented a GIS-based adaptive protection framework that integrates geospatial data, historical load information, and planning constraints into a unified, dynamically updatable load-forecasting database. By linking forecasted load thresholds directly with adaptive relay setting logic and validating the approach using commercial IEDs in an IEC 61850 laboratory environment, the study demonstrates a practical and scalable pathway from forecasting to protection action. The results confirm that GIS-informed load forecasts can meaningfully enhance adaptive protection performance under changing network conditions, while remaining compatible with existing communication infrastructure and industry practices.

The findings of this study highlight the critical importance of adaptive protection schemes in enhancing system reliability, selectivity, and resilience in evolving electrical networks. The proposed framework is designed with a modular architecture that enables future integration of distributed energy resources including electric vehicle charging infrastructure and solar photovoltaic generation. However, extension to these applications requires consideration of additional operating characteristics specific to each technology.