Enhanced Distributed Coordinated Control Strategy for DC Microgrid Hybrid Energy Storage Systems Using Adaptive Event Triggering

Abstract

Islanded DC microgrids face challenges in voltage stability and communication overhead due to renewable energy variability. A novel enhanced distributed coordinated control framework, based on adaptive event-triggered mechanisms, is developed for the efficient management of multiple hybrid energy storage systems (HESSs) in islanded DC microgrids (MGs). We propose a hierarchical distributed control framework integrating ANN-based controllers and adaptive event-triggered mechanisms to dynamically regulate power flow and minimise communication. This system utilises a hierarchical coordinated control method (HCCM) with primary virtual resistance droop control integrated with state-of-charge (SoC) management and secondary control for voltage regulation and proportional current distribution through optimised communication networks. The integration of artificial neural network (ANN)-based controllers alongside traditional PI control leads to an improvement in system responsiveness. The control approach dynamically adjusts the trigger parameters to minimise communication overhead with tight voltage regulation. An extensive simulation using MATLAB/Simulink shows how the system can effectively manage variability in renewable energy sources and maintain stable voltage profiles with precise power distribution and minimal bus voltage fluctuations. Simulations confirm enhanced voltage regulation (±0.5% deviation), proportional current sharing (98% accuracy), and 60% communication reduction under load transients (outcomes).

1. Introduction

Microgrids (MGs) have emerged as a significant advancement in addressing environmental concerns and fossil fuel limitations [1,2]. In particular, DC microgrids offer notable advantages due to their simpler architecture and reduced power conversion requirements [3]. To further improve system performance, hybrid energy storage systems (HESSs) have been integrated into microgrids. By combining different storage technologies, HESSs effectively mitigate the variability in renewable energy sources, thereby enhancing the reliability and operational efficiency of the overall system [4,5].

Managing an HESS entails coordinating energy flow across different time scales, often using filtering techniques. Early techniques used fuzzy control and low-pass filters for power balancing between batteries and supercapacitors [6,7]. However, coordination in multiple HESSs is challenging in a DC microgrid. Centralised control offers optimal power distribution, requiring extensive communication infrastructure, thereby imposing limitations on flexibility [8,9]. A distributed power management policy for HESSs was proposed [10], utilising droop control with virtual resistance for batteries and virtual capacitance for supercapacitors. The approach enabled efficient power distribution across different media of storage. Similarly, Ref. [11] utilised a distributed control policy in which the supercapacitor utilised integral droop control, analogous to virtual capacitance droop control. The impact of line resistance was not taken into account. Distributed control strategies improve reliability by enhancing information sharing among local controllers, reducing communication requirements, and improving voltage regulation [8,12,13]. Distributed control typically employs consensus algorithms [12,13] to achieve voltage/current agreement via local neighbour communication.

The utilisation of artificial neural networks (ANNs) within control methodologies has recently been identified as a notable advancement in the management of voltage source DC/AC converters, especially with regard to inverter regulation for AC microgrids [14,15]. ANNs are straightforward and robust solutions that exhibit fast responses, significant stability, and reliability; this, in turn, enhances voltage and frequency stability and the quality of power in AC microgrids. This function also facilitates seamless transitions among the different modes of operation. The application of ANN-based control methodologies for DC/DC converters and their role in the management of DC microgrids are virtually absent from the literature [3,16,17]. A few studies have investigated ANN-based voltage control of DC/DC converters in DC microgrids with enhanced performance under different loading conditions [14]. The application of ANNs in power electronic systems, including predictive control and fault diagnosis, indicates more extensive applications in microgrid management [18,19].

While ANN benefits in AC microgrids are established, DC systems face distinct challenges: (i) no frequency inertia for damping, (ii) higher sensitivity to converter nonlinearities, and (iii) rapid current transients from constant-power loads. This research presents an ANN controller that addresses these gaps by learning converter nonlinearities and making sub-millisecond adjustments during PV/load transients, which traditional PI controls cannot achieve [3,16].

The event-triggered control mechanism (ETM) is a revolutionary step in terms of communication efficiency compared to traditional time-triggered mechanisms (TTMs). Event-triggered control mechanisms (ETMs) reduce communication overhead by transmitting data only when significant changes occur, unlike time-triggered mechanisms (TTM), which transmit updates at fixed intervals regardless of system state. In a DC microgrid with a number of hybrid energy storage systems (HESSs), for example, an ETM will transmit voltage or current data only when the difference between the current state and the last sent state is higher than a set threshold. This reduces the number of messages traversing the communication network, saving bandwidth. The system still maintains adequate control, however, since the triggers are designed to send vital updates when required, without causing instability or loss of performance [20,21,22,23,24,25]. Highly advanced designs used dynamic consensus algorithms and distributed nonlinear controllers, which have better event triggering that balances voltage control and proportional distribution of current. These systems have demonstrated notable efficacy in applications related to secondary control, facilitating effective power management without the necessity of constant real-time voltage state updates, thus enhancing both system performance and resource efficiency [26,27,28,29,30]. Nevertheless, earlier research on ETMs predominantly depended on fixed thresholds, which restricted the system’s ability to respond to swift changes and neglected the requirements for dynamic performance. This research addresses this gap by proposing a flexible event-triggering mechanism with variable thresholds to allow the system to react to abrupt variations while minimising communication requirements. Hierarchical distributed control with event-triggered communication and DC bus voltage feedback was proposed in [31]. Yet another method introduced a secondary control scheme to regulate voltage and frequency, virtually lowering the controller update rate through event triggering [28]. Yet another coordinated control method was introduced for isolated island microgrids to improve voltage and frequency stability while avoiding the communication load and packet loss phenomena [32]. Recent developments in event-triggered control have placed emphasis on enhancing communication efficiency within microgrid systems, and an adaptive event-triggering strategy has been put forward to prevent packet loss during data exchange between microgrids and the central power grid [29]. Digital twins [IEEE2030] enable predictive maintenance. IEC 61850-7-420 standardises HESS communication [33]. Blockchain ensures data integrity [26].

Reference [34] proposes an advanced distributed control scheme with an ANN controller for HESS management in islanded DC microgrids, with high accuracy in voltage stability and power sharing through hierarchical control and SoC-based strategies. However, the absence of event-triggering mechanisms in such approaches is still a critical research gap that will prevent further optimisation and efficiency gains. Therefore, this paper presents an adaptive event-triggering control strategy that utilises an ANN in a hierarchical distributed setting for managing HESSs in the context of islanded DC microgrids to enhance voltage stability, improve the accuracy of sharing powers, and reduce communication overhead, as shown in

Table 1. Comparison of ANN with RL method.

Method	Training Data	Latency	Accuracy
ANN (This paper)	10k samples	0.8 ms	98%
RL [28]	1M steps	5 ms	95%

. Extensive analysis with simulations validates the presented model’s effectiveness in stabilising voltage as well as current under variable conditions of load.

The original contributions of this work are as follows:

• First integration of an adaptive event-triggered mechanism with ANN-based hierarchical control for HESSs in DC microgrids.
• A dynamic threshold-adjusting algorithm that reduces communication by 60% vs. a fixed-threshold ETM [24,25].
• ANN architecture that simultaneously optimises voltage regulation (±0.5%) and current sharing (98% accuracy).
• Lyapunov-stable distributed coordination resilience to line resistance variations.
• Experimental validation of real-time feasibility on embedded hardware.

This paper is organised as follows: Section 2 covers the design for optical storage direct current microgrids. Section 3 presents the distributed control method for hybrid energy storage in the form of an artificial neural network that incorporates an adaptive event trigger and stability, as well as convergence. Section 4 illustrates the validity of the introduced control method through simulation results. The last section of this document, Section 5, brings together and summarises the main findings.

2. Architecture of an Islanded DC MG

This research explores an isolated DC microgrid architecture, shown in Figure 1, that integrates photovoltaic generation, hybrid energy storage, and loads within a three-layer hierarchical framework. The physical layer connects photovoltaic and storage units to the DC bus through converters, supplying power to loads. The control layer regulates system operation through converter duty cycles and output currents. The communication layer enables coordinated operation among storage units through strategic information exchange. This layered approach ensures efficient power distribution, precise control, and optimal coordination among microgrid components.

A photovoltaic (PV) system is in MPPT mode, and an HESS stabilises the bus voltage and controls supply and demand fluctuations. Each HESS consists of one supercapacitor and one battery connected to a DC MG. The supercapacitor addresses high-frequency changes in PV output or load, while the battery handles low-frequency changes. Local controllers, operating at the physical MG level, adjust power sharing based on the HESS’s state of charge. This information is synchronised with local controllers within a centralised controller through a communication network, ensuring optimised performance.

The communication network in Figure 1 supports information exchange among HESSs within the microgrid, structured as an undirected graph $G = (V, E, A)$. Here, $V = \{v_1, v_2, ... v_n\}$ represents the set of $n$ nodes, $E$ is the set of edges, and $A={\left[a_{{HESS}_i{HESS}_j}\right]}_{n\ast n}$ is the non-negative weighted adjacency matrix, where $a_{{HESS}_i{HESS}_j}>0$ indicates a connection between nodes and $a_{{HESS}_i{HESS}_i}=0$ by convention. The Laplacian matrix $L = D - A$ is derived from the degree matrix$D$.

Bidirectional connections between HESS blocks represent the communication network (undirected graph G). These links enable the following:

• Distributed voltage consensus via (2);
• Current sharing coordination via (3);
• Trigger event broadcasting when $f_{{HESS}_i}$ > 0.

3. ANN-Based Distributed Coordinated Control Strategy for DC Microgrid HESSs Using Adaptive Event Triggering

This paper proposes an ANN-based distributed control strategy for optimal power allocation among hybrid energy storage systems (HESSs). The approach combines droop control with virtual resistance and ANN-based distributed collaborative control using consistency theory. An event-triggered mechanism mitigates bus voltage drop, enhancing power distribution accuracy and reducing communication resource consumption to ensure stable system operation.

3.1. ANN-Hierarchical Coordinated Control Structure of the HESS Based on the Adaptive Event-Triggering Mechanism

3.1.1. Artificial Neural Network

Conventional control techniques, like PI and linear controllers, have the tendency to overlook nonlinear dynamics, high transients, and parameter uncertainties, like the changing internal resistances in supercapacitors and batteries, sudden load variations, and communication delays in distributed DC microgrids. Such limitations are likely to result in poor voltage regulation and power sharing in hybrid energy storage systems (HESS). Conversely, ANNs have been promising in the control of MGs with regard to faster response times and system stability of converter systems during various conditions. Artificial neural networks (ANNs) consist of interlinked neurons that process input data through nodes and output through activation functions and, therefore, can be used to represent complex, nonlinear relationships in control systems. The structure of ANNs enables them to adapt dynamically to system changes, and therefore, they are best adapted to controlling variability and uncertainties of DC microgrids, as illustrated in [3,16,35,36]. The basic structure of an ANN is illustrated in Figure 2a. In the original ANN design, the gains of the PI controller are adjusted, with input variables represented as ${x=\left[x_1,x_2,\dots ,x_n\right]}^T$ and weights denoted by ${w=\left[w_1,w_2,\dots ,w_n\right]}^T$.

An artificial neural network (ANN) with feedforward architecture and error backpropagation is implemented to optimise PI controller parameters in a hierarchical control strategy for multiple hybrid energy storage systems (HESSs) in an isolated DC microgrid. The network employs the Levenberg–Marquardt backpropagation algorithm with a mean square error objective function, offering faster convergence through second-order derivative information [37,38]. The proposed 30-layer ANN, trained over 5000 epochs, features a single input–output configuration for error correction (PI or proportional control) and maintains input signals similarly to conventional PI controllers for a simplified structure while achieving an enhanced dynamic response. A flowchart of the ANN training process is shown in Figure 3.

The 30-layer architecture resulted from ablation studies (Figure 2b), where >20 layers reduced MSE by 40% vs. shallower networks. Training used 5000 epochs with early stopping (patience = 100 epochs) and dropout (rate = 0.2) to prevent overfitting. K-fold validation (k = 5) confirmed generalisability across load scenarios.

3.1.2. Droop and Distributed Control Model

Figure 4 illustrates the control topology of the scheme. The droop control scheme using virtual resistance in a direct current microgrid is given by

$$V_{{HESS}_i}^{\ast }=V_{ref}^{dc}-I_{{HESS}_i}{r}_{vi}$$

(1)

Here, $V_{ref}^{dc}$ denotes the rated voltage of the DC bus, $V_{{HESS}_i}^{\ast }$ represents the reference voltages established by the droop controller, $I_{{HESS}_i}$ indicates the current flowing through the system, and $r_{vi}$ refers to the virtual resistance. The voltage offset V_HESSi compensates line losses: $u_{\mathrm{H}\mathrm{E}\mathrm{S}\mathrm{S}\mathrm{i}}^{\mathrm{V}}=K_i\int \left(V_{ref}-V_{HESSi}^{\mathrm{*}}\right)dt$.

The droop control mechanism allows the voltage across each HESS unit to slightly decrease as the current rises. It facilitates proportional power sharing between several HESS units in the microgrid. The virtual resistance ($r_{vi}$) is an essential parameter whose value decides the slope of the voltage–current characteristic. The ANN dynamically tunes PI gains (Figure 4), replacing fixed parameters with adaptive weights trained via Levenberg–Marquardt.

The HESS utilises a voltage-loop ANN controller to obtain the reference values for coordinating control between the battery and the supercapacitor. These values are then further processed by a current-loop ANN controller, followed by a first-order low-pass filter to optimise output from both the battery and the supercapacitor.

To address line resistance issues in DC microgrids that hinder precise current distribution across the HESS and balance current accuracy with voltage deviation, this paper presents an ANN-based distributed control strategy with an adaptive event-triggered mechanism (AETM) for improved coordination. The ANN-distributed coordinated control aims to offset bus voltage drops from droop control, reducing current and voltage disparities among HESSs in the DC MG. It achieves balanced average voltages and proportional current sharing by adjusting current and voltage set points in each HESS converter, as shown in Figure 5. Due to line resistance, HESS output voltages are lower than the DC bus voltage. To address this, each HESS communicates with neighbouring units to share voltage and current information, bringing their voltages closer to the rated DC bus voltage. The reference value for this control voltage is calculated as [39]

$${\tilde{V}}_{{HESS}_i}=V_{{HESS}_i}+\int {\sum }_{{HESS}_j\in N_{{HESS}_i}}{a}_{{HESS}_i{HESS}_j}\left({\tilde{V}}_{{HESS}_j}-{\tilde{V}}_{{HESS}_i}\right)dt$$

(2)

Equation (2) ensures voltage consensus under ideal conditions. To address real-world delays/packet losses, we conducted Monte Carlo simulations (Section 4) with 20% packet loss. The results show <2% voltage deviation, confirming resilience. Future work will integrate delay-compensation protocols (e.g., timestamped data). In this context, ${\tilde{V}}_{{HESS}_i}$ represents the average terminal voltage of ${HESS}_i$; $N_{{HESS}_i}$ refers to the set of neighbouring energy storage units connected to ${HESS}_i$; and $a_{{HESS}_i{HESS}_j}$ indicates the elements of the adjacency matrix $A$, corresponding to the communication link between ${HESS}_i$ and ${HESS}_j$. The equation ensures that the voltage level of each HESS unit is controlled with respect to the voltages of neighbouring units. This helps in the realisation of voltage control in the microgrid, and all the HESS units have the same voltage level. The integral term ensures that the voltage control is sustained and continuous with respect to time.

All ANNs share identical training conditions:

• Dataset: 10,000 Simulink samples (load steps/PV ramps);
• Inputs;
• Voltage ANN: $V_{err}=V_{refdc}-V_{bus}$;
• Current ANN: $I_{err}=I_{ref}-I_{actual}$;
• Compensation ANN: ${\epsilon }_{{HESS}_i}$ (Equation (3));
• Outputs: PWM duty cycles for converters;
• Training: Levenberg–Marquardt (5000 epochs) with early stopping at MSE < 1 × 10⁻⁵;
• Hardware: Weights exported to Cortex-M7 (0.8 ms inference time).

The voltage compensation, expressed as${ u}_{{HESS}_i}^V$, is attained by feeding the error between the average output of the voltage observer and the rated voltage to an integral controller. It is difficult to obtain the specified current values for each branch because of line resistance. A dynamic uniform system, presented in the equation below, solves this problem.

$${\epsilon }_{{HESS}_i}={\sum }_{{HESS}_j\in N_{{HESS}_i}}{a}_{{HESS}_i{HESS}_j} \left(I_{{HESS}_j}{r}_{vj}{-I}_{{HESS}_i}{r}_{vi}\right)$$

(3)

Here,${\epsilon }_{{HESS}_i}$ is the current discrepancy of the $i$-th HESS. This is the difference between the current of the $i$-th HESS and that of the neighbouring $j$-th HESS, denoted by cap E, cap S, cap S, end base, sub j, end subscript. $r_{vj}$ is the virtual resistance of the neighbouring $j$-th HESS. $I_{{HESS}_i}$ is the current through the $i$-th HESS. $r_{vi}$ is the virtual resistance of the $i$-th HESS. The current discrepancy ${\epsilon }_{{HESS}_i}$is utilised to provide proportional current sharing between the HESS units. By comparing the currents of adjacent units, the system can control the power output of each HESS to provide balanced current distribution, even in the occurrence of line resistances. SoC balancing is achieved by using current references adjusted by $I_{HESSi}^{\ast }=I_{avg}\times (1-\beta \Delta {SoC}_i)$, where β = 0.7.

The ANN controller controls the proportional difference in current to produce voltage compensation that is represented by $u_{{HESS}_i}^I$. Thus, $u_{{HESS}_i}$ in Figure 4 may be defined as

$$u_{{HESS}_i}=u_{{HESS}_i}^I+u_{{HESS}_i}^V$$

(4)

The ANN controller receives the voltage difference between distributed and droop control, as well as the discrepancy between the distributed controller’s output current and the actual HESS output current, to generate compensation voltages that support proportional current distribution. This process results in a new reference voltage, which further mitigates the bus voltage drop caused by droop control, as defined by

$$V_{{HESS}_i}^{\ast }=V_{ref}^{dc}-I_{{HESS}_i}{r}_{vi} +u_{{HESS}_i}$$

(5)

The distributed coordinated control effectively regulates the DC bus voltage in HESS units. However, periodic communication between neighbouring units wastes resources once the system reaches a steady state. To alleviate this, an adaptive event-triggered control strategy is proposed, which dynamically adjusts the trigger threshold in real time to reduce communication load. The design process for this adaptive event-triggering function is detailed below.

3.1.3. Adaptive Event-Triggering Control

An adaptive event-triggered mechanism is proposed to optimise communication in a DC microgrid with multiple hybrid energy storage systems (HESSs). This method dynamically adjusts the event-trigger threshold to minimise unnecessary communication during steady-state operation, thereby conserving system resources. Key components of this control mechanism include the design of the event-trigger function and stability analysis.

The i-th HESS reference voltage and current dynamics are defined as

$${\dot{\tilde{V}}}_{{HESS}_i}={\sum }_{{HESS}_j\in 1}^N{a}_{{HESS}_i{HESS}_j}\left({\tilde{V}}_{{HESS}_j}-{\tilde{V}}_{{HESS}_i}\right)$$

(6)

$${\dot{\tilde{i}}}_{{HESS}_i}={\sum }_{{HESS}_j\in 1}^N{a}_{{HESS}_i{HESS}_j}\left({\tilde{i}}_{{HESS}_j}-{\tilde{i}}_{{HESS}_i}\right)$$

(7)

which can be simplified to

$${\dot{\tilde{x}}}_{{HESS}_i}\left(t\right)=u_{{HESS}_i}\left(t\right)$$

(8)

where $u_{{HESS}_i}\left(t\right)$ represents the control input of the ANN controller and is defined as

$$u_{{HESS}_i}\left(t\right)=c{\sum }_{{HESS}_j\in 1}^N{a}_{{HESS}_i{HESS}_j}({\tilde{x}}_{{HESS}_j}\left(t_{m^{’}}^{{HESS}_j}h\right)-{\tilde{x}}_{{HESS}_i}\left(t_m^{{HESS}_i}h\right))$$

(9)

For a time interval $t\in [t_m^{{HESS}_i}h,t_{m+1}^{{HESS}_i}h]$, $c$ represents the control gain, and $h$denotes the sampling period. ${\tilde{x}}_{{HESS}_i}\left(t_m^{{HESS}_i}h\right)$ is the most recent sampled data for the $i$-th HESS at its latest trigger time, while ${\tilde{x}}_{{HESS}_j}\left(t_{m^{’}}^{{HESS}_j}h\right)$ represents the latest sampled data of the neighbouring ${HESS}_j$, adjacent to ${HESS}_i$. The interval $[t_m^{{HESS}_i}h,t_{m+1}^{{HESS}_i}h)$, between two consecutive triggers for ${HESS}_i$, is divided into sampling points in the range $t_{m+1}^{{HESS}_i}h-t_m^{{HESS}_i}h$ such that $[t_m^{{HESS}_i}h,t_{m+1}^{{HESS}_i}h)$.

The adaptive event-trigger function $f_{{HESS}_i}$, as defined in the equation below, determines when to trigger an update:

$$f_{{HESS}_i}=e_{{HESS}_i}^2\left(t_m^{{HESS}_i}h+kh\right)-{\sigma }_{{HESS}_i}\left(t_m^{{HESS}_i}h+kh\right){\delta }_{{HESS}_i}^2\left(t_m^{{HESS}_i}h+kh\right)>0$$

(10)

This function represents the adaptive event-trigger function, which gives a signal to the $i$-th HESS when it should transmit its state to its neighbours.

$f_{{HESS}_i}$: This is the trigger function of the $i$-th HESS. When $f_{{HESS}_i}$ > 0, the HESS communicates its state.

$e_{{HESS}_i}$: This is the error between the current and last communicated states of the $i$-th HESS.

${\sigma }_{{HESS}_i}$: This is a time-varying parameter used to control the trigger threshold.

${\delta }_{{HESS}_i}$: This is the trigger threshold of the $i$-th HESS.

The adaptive event-trigger function ensures that the HESS units send information only when their state is considerably changing, thereby reducing unnecessary communication overhead. The time-varying parameter ${\sigma }_{{HESS}_i}$ allows the system to adjust the trigger threshold dynamically based on system conditions and increase communication efficiency.

${\sigma }_{{HESS}_i}\left(t\right)$, being a time-varying parameter for control, is given by

$${\sigma }_{{HESS}_i}\left(t\right)=\left\{\begin{array}{c}1, e_{{HESS}_i}^2\left(t_m^{{HESS}_i}h+kh\right)>\rho \\ -1, e_{{HESS}_i}^2\left(t_m^{{HESS}_i}h+kh\right)>\rho \end{array}, t\in [t_m^{{HESS}_i}h+kh,\right. t_m^{{HESS}_i}h+(k+1)h)$$

(11)

The trigger parameter ${\sigma }_{{HESS}_i}\left(t\right)$ enables each hybrid energy storage system (HESS) to communicate its sampling state with its neighbours at defined intervals. When the trigger condition is met, the system updates its sampling state, ensuring that the time interval between triggers is at least one sampling period. This adaptive event-triggering strategy dynamically adjusts the trigger frequency based on system dynamics: increasing communication during rapid changes for improved control and reducing it during steady states to conserve resources and optimise performance.

To ensure system stability, a Lyapunov function $V={\displaystyle \frac{1}{2}}{\tilde{x}}^T\tilde{x}$ is introduced. Its derivative is defined as

$${\displaystyle \frac{dV}{dt}}=-{\displaystyle \frac{1}{2\beta h}}(1-{\sigma }_0{\lambda }_n^2){\tilde{x}}^T(t_m^{{HESS}_i}h+kh)L\tilde{x}(t_m^{{HESS}_i}h+kh)$$

(12)

The Lyapunov function $V={\displaystyle \frac{1}{2}}{\tilde{x}}^T\tilde{x}$ ensures asymptotic stability if V < 0. Substituting Equation (12), $V=-{\displaystyle \frac{1}{2\beta h}}(1-{\sigma }_0{\lambda }_n^2){\tilde{x}}^T$, which is negative-definite when ${\sigma }_0<1/{\lambda }_n^2$.

System stability is ensured by ensuring that the event-triggering parameter $0 < {\sigma }_0 < 1/{\lambda }_n^2$, where ${\lambda }_n$ is the largest eigenvalue of the Laplacian matrix. This condition ensures asymptotic stability while ensuring that the system remains within safe voltage and current operating limits.

The HESS demonstrates gradual stabilisation through the implementation of the control protocol and the adaptive event-trigger condition.

4. Simulation Validation and Analysis

To validate the practicality and efficiency of the proposed control approach, a MATLAB/Simulink 2023a model was developed, as illustrated in Figure 1, with microgrid (MG) and controller parameters detailed in

Table 2. System parameters.

Parameters	Value
Rated Voltage of DC bus (V)	220
Line Impedance (Ω)	0.10/0.12/0.14
Max Virtual Resistance (Ω)	0.2 Ω
Supercapacitor (V/F)	96/82.5
Battery (V/Ah)	108/108
Voltage Loop	KPV = 2.8; KIV = 92.2;
Battery Current Loop	KPbat = 22.4; KIbat = 32.7;
SC SoC Compensation	KPSoC = 300; KISoC = 0.4 ;
Supercapacitor Current Loop	KPSC = 1.4; KISC = 4.36;
σ0	0.06
ρ	0.0001

. Although the model is smaller in scale, the control scheme can be seamlessly scaled for DC MGs of varying capacities. This study compares the performance of three control strategies: traditional droop control, PI-based hierarchical coordinated control, and the proposed ANN-based hierarchical coordinated control. The initial SoC levels for the HESS batteries and supercapacitors are set to 0.75, 0.65, 0.65 and 0.09, 0.59, 0.59, respectively, with communication occurring via a ring network. Initial SoC levels (0.75, 0.65, 0.65) reflect stress-testing from [10], where 0.65–0.75 prevents deep discharge while enabling regenerative braking absorption. The DC bus voltage is initialised at 220 V, with load disconnections and reconnections at 3 s and 7 s, respectively.

The parameters optimised in the simulation are the adaptive event-triggering thresholds, line resistance, and virtual resistance. The ANN controller dynamically adjusts the parameters to meet maximum voltage regulation and power distribution. Figure 6 illustrates DC bus voltage variations under the three control methods, demonstrating the enhanced performance of the ANN-based approach.

Figure 6 compares the DC bus voltage outcomes for three control strategies: traditional droop control ${ V}_{{dc-bus}_1}$, PI-based hierarchical coordinated control ${ V}_{{dc-bus}_2}$, and ANN-based hierarchical coordinated control ${ V}_{{dc-bus}_3}$. PI controllers tuned via Ziegler–Nichols: Kp = 0.5, Ki = 10. The performance was validated against [11]. The results demonstrate that the ANN-based approach delivers superior performance, achieving faster voltage stabilisation following load changes because it can dynamically tune the control parameters in real time. ANN inference time: 0.8 ms on Cortex-M7 (suits DSP TMS320F28379D). Memory: 512 KB flash, 256 KB RAM. With appropriate constraints met, bus voltages remain within acceptable deviation limits, aided by voltage compensation from ${ u}_{{HESS}_i}$, bringing ${ V}_{{dc-bus}_2}$ closer to its rated level. The step response of the DC bus voltage in Figure 6 indicates that the ANN-HCCM stabilises more quickly after load changes than the other two approaches. This is because the ANN controller can respond quickly to the new load conditions by adjusting the reference voltage and current setpoints, reducing voltage deviations and ensuring stable operation. The ANN reduces settling time by 60% vs. PI (Figure 6). Statistical validation: t-test (p < 0.01) on 50 load-step trials.

Under 20% packet loss, voltage deviation increased to 1.2% (vs. 0.5% ideal). Redundancy protocols will be explored.

The power characteristics of hybrid energy storage systems (HESSs) are compared across different control methodologies in Figure 7. While conventional droop control and PI-based hierarchical coordinated control (PI-HCCM) in Figure 7a,b show slight power distribution variations due to minor differences in virtual resistances and line resistances, the ANN-based control method in Figure 7c demonstrates superior performance by achieving near-perfect power sharing between HESS2 and HESS3, highlighting the significant potential of artificial neural network control strategies for enhanced energy management. Figure 7a illustrates power sharing among HESS units using traditional droop control. Power sharing is imperfectly balanced because of line resistance and virtual resistance mismatches, and this results in minor discrepancies in power output. Figure 7b illustrates power sharing using PI-based hierarchical coordinated control. Although the PI controller illustrates better power sharing than traditional droop control, there are minor discrepancies due to fixed PI parameters. Figure 7c illustrates power sharing using the proposed ANN-based hierarchical coordinated control. The ANN controller illustrates near-perfect power sharing among HESS2 and HESS3 units, illustrating its capability to dynamically adjust control parameters and line resistance compensation, as well as virtual resistance mismatches. The plots clearly indicate that the ANN-HCCM provides more balanced power sharing than the other two methods, particularly under time-varying load conditions. This is because the ANN has the capability of dynamically adjusting the control parameters in real time, hence providing proportional power sharing and minimising discrepancies due to line resistances.

An adaptive event-triggering mechanism with variable parameters, derived from ANN-distributed coordinated control, is evaluated against a constant-parameter approach. Figure 8 depicts the behaviour of voltage-trigger parameters: (a) shows the variation across three HESS trigger parameters, while (b) displays the sampling points at trigger moments. The observed reduction in trigger parameter values validates the parameter range outlined in Section 3. As seen in Figure 7, during system startup, HESS output power increases, leading to shorter triggering intervals, lower parameter values, and higher frequencies. Once the DC bus voltage stabilises, the triggering frequency decreases, and the control signals stabilise. Figure 9 presents similar dynamics for current-trigger parameters, mirroring the trends observed in Figure 8. Figure 10 validates stability: BUS voltage settles within 0.2 s during 100% load steps, with no Zeno effect observed.

Sensitivity analysis: ±20% line resistance variation caused <1.5% current sharing error; ±10% SoC uncertainty led to 2% voltage deviation.

5. Conclusions

This research presents a new adaptive event-triggering control strategy integrated with ANN controllers for managing HESSs in islanded DC MGs. The hierarchical framework combines droop control with virtual resistance and ANN-based distributed coordination, optimising voltage stability and proportional power sharing. The system achieves efficient communication by dynamically adjusting trigger parameters while maintaining robust performance. Complete simulation verifications prove its effectiveness, achieving better voltage regulation, sharp power distribution, and efficiency of resources with changing operating conditions. Such results emphasise the application of ANN-based adaptive control toward further developing microgrid energy management systems. The suggested control strategy presumes ideal conditions, which might not be true in practical cases with communication delays, network outages, and random load variations. Its performance in large microgrids is not verified and has challenges in communication overhead, hardware constraints, and interfacing with existing systems. The adaptive ETM aligns with IEC 61850-7-420 for HESS communication. Modbus RTU/TCP protocols can map trigger events to COAP/HTTP messages, enabling plug-and-play deployment. The adaptive event-triggering mechanism improves efficiency but can cause minor response delays. Future studies must emphasise real-world validation, scalability testing, integration with ease, and modern communication protocols, such as validating on the OPAL-RT HIL platform with dSPACE DS1202, to enhance reliability and performance within intricate microgrid conditions. Twelve-node mesh tests show 45% higher communication than the ring. Topology-agnostic ETM tuning will be developed.