Solar irradiance exhibits intermittent and stochastic characteristics, posing significant challenges to the operational resilience, stability and reliability of photovoltaic (PV) systems and modern power grids. With the increasing penetration of large‐scale PV installations, accurate solar power forecasting has become essential for maintaining grid balance, optimising energy dispatch and enhancing overall system efficiency. In this paper, a hybrid Bi‐LSTM–GRU architecture is proposed for solar forecasting in renewable‐integrated power systems due to its accuracy. The hybrid architecture takes advantage of the temporal learning power of Bi‐LSTM layers to learn bidirectional relationships in irradiance and generation data, whereas GRU layers are used to improve sequential representations and faster convergence through the alleviation of the vanishing gradient problem. The model is trained and tested on the UNISOLAR dataset at La Trobe University, Australia, which offers time‐series measurements of the PV generation at the large scale, as well as of the meteorological parameters. Comprehensive evaluations are conducted against multiple baseline models. The proposed Bi‐LSTM–GRU hybrid architecture consistently demonstrated superior predictive accuracy and robustness, achieving an RMSE of 0.298, MAE of 0.287 and MAPE of 0.033. These results indicate that the proposed model effectively captures complex nonlinear temporal dependencies and irradiance variability inherent in solar energy data. Consequently, the model enables more accurate short‐term forecasting and supports adaptive grid management strategies.
Accurate short-term prediction of line loss rate is of great significance for improving the operational efficiency and economic performance of distribution networks. To address the limitations of conventional prediction models in nonlinear feature extraction and hyperparameter optimization, this paper proposes a CNN–LSTM prediction model optimized by an improved sparrow search algorithm incorporating logistic chaotic mapping and a linearly decreasing weight strategy, referred to as the LCSSA–CNN–LSTM model. First, a convolutional neural network is employed to extract local features from multidimensional input variables, and a long short-term memory network is then used to capture temporal dependencies in the line loss rate series. Second, logistic chaotic mapping is introduced to improve the initialization quality of the sparrow population, while a linearly decreasing weight is incorporated to enhance the balance between global exploration and local exploitation during iterative optimization. Finally, comparative experiments are conducted on a real line loss rate dataset. The results show that the proposed model significantly outperforms the baseline CNN–LSTM model, with the RMSE reduced by 55.79% and the prediction error of representative peak samples controlled within 3%. These results indicate that the proposed model can effectively improve the accuracy and robustness of short-term line loss rate prediction.
Accurate estimation of fault resistance and reliable discrimination of fault type are foundational to the protection and self-healing capability of medium-voltage distribution networks, yet both tasks are severely confounded by high-impedance faults (HIFs) whose low and variable current signatures evade conventional overcurrent and distance protection. This paper presents a noise-resilient machine learning framework that performs continuous fault resistance regression and multiclass fault-type classification on a simulated 11-kV radial distribution feeder. A high-fidelity electromagnetic model was constructed in MATLAB/Simulink–Simscape, comprising a 35 MVA Thévenin-equivalent source, two cascaded PI-section lines totaling 10 km, a 5-MVA distribution transformer, and an aggregated downstream load. An automated Monte Carlo pipeline generated 6000 single-phase-to-ground (AG) fault cases, with fault resistance log uniformly sampled over 0.01–500 Ω, and a further 3500 cases spanning seven canonical fault types. From pre- and postfault voltage and current windows sampled at 10 kHz, 22 physically interpretable features were extracted, encompassing faulted- and healthy-phase magnitudes, asymmetry indices, zero-sequence components, and derived impedance quantities. Gradient boosting machine (GBM), random forest (RF), and support vector machine/regression (SVM/SVR) models were trained and benchmarked under clean and noise-contaminated measurement conditions with additive white Gaussian noise spanning 10–50 dB SNR. Under clean conditions, the GBM regressor achieved R2 = 0.9961, RMSE = 8.73 Ω, and MAE = 4.21 Ω; the RF classifier attained 99.34% accuracy on fault-severity classification. The framework maintained R2 > 0.97 and classification accuracy > 96% down to 30 dB SNR, which serves as a qualitative reference level representative of metering-class instrument transformer measurement quality, establishing practical sensor requirements for field deployment. Multifault-type extension yielded 98.91% identification accuracy with per-type regression R2 consistently exceeding 0.98. All results are based on synthetic simulation data from a single feeder configuration, one fault location, and AWGN-based noise modeling; field validation remains future work. The results demonstrate that interpretable, ensemble-based learning on physically grounded features can deliver simultaneous resistance estimation and fault typing suitable for integration into intelligent protection and smart-grid monitoring systems.
An essential component of modern microgrid operation is the energy management system. This work proposes a stochastic programming–based proximal policy optimisation (PPO) technique for a typical wind–solar–diesel–storage microgrid, accounting for the significant uncertainty associated with wind energy. Using Latin Hypercube Sampling, scenario sets for wind power with different probabilities are generated by simulating potential errors in predicted wind power. A Markov decision process with high-dimensional state and action spaces is used to model the energy management optimisation problem. In this manner, day-ahead power scheduling for each generation/storage unit is optimised using PPO, which maximises economic gain while satisfying security constraints. The proposed method’s day-ahead scheduling, evaluated on a real-world microgrid, demonstrates the energy management system’s high efficiency and robustness. The results show that the proposed framework is competitive in solution quality and constraint satisfaction, and it achieves a positive balance among computational efficiency, robustness, operational feasibility and long-term system reliability under renewable uncertainty.
The large-scale development of electric vehicles (EVs) provides new means for fault defense and rapid handling in distribution networks. This paper proposes a distribution network fault defense strategy for charging stations (CSs) electricity purchase guidance, considering the uncertainty of scheduling potential. First, a generalized energy storage model for CSs is established based on Minkowski sum theory, and the scheduling potential probabilistic characteristics of CSs are analyzed. Subsequently, under the constraints of CS scheduling potential uncertainty and distribution network fault risk, a two-stage stochastic programming strategy for multi-CS collaborative electricity purchase guidance is constructed by integrating Latin hypercube sampling and Stackelberg game theory, incorporating fault risk suppression in distribution networks. Finally, an analytical target cascading (ATC) algorithm is employed to propose a decoupling solution strategy for the multi-CS collaborative electricity purchase guidance model considering the fault risk suppression of distribution networks. Simulation results demonstrate that the proposed method effectively motivates CSs to participate in distribution network load regulation, reducing electricity purchase costs by 13.57%–21.17% while ensuring the distribution network’s fault risk is reduced to a safe level.
This paper presents a capability-driven sequential optimization framework for the robust design of electrical machines under manufacturing uncertainty. The proposed methodology integrates surrogate modeling, NSGA-II-based Pareto front exploration, and Design for Six Sigma (DFSS)-driven refinement to ensure production-level reliability. Surrogate models constructed via Response Surface Methodology (RSM) enable efficient trade-off analysis across key performance metrics, including torque ripple, average torque, efficiency, and power factor. The first optimization stage applies NSGA-II to identify Pareto-optimal trade-offs, while statistical analysis reveals residual variability risks in critical objectives. To address this, the second stage introduces DFSS-based capability constraints within a sequential refinement process, driving capability indices Cpk above 2.0 and reducing probability of failure (PoF) to below 0.01% across all objectives. Compared to baseline designs, the optimized solutions demonstrate significant reductions in variability: 36% in torque ripple, 40% in power factor, and 27% in torque spread. The proposed framework is validated as an effective and scalable approach for robust optimization of electrical machines subjected to real-world manufacturing tolerances, enabling production-ready solutions with high statistical confidence.
This paper develops a cyberphysical resilience model for microgrids to facilitate energy exchange with the upstream electrical power network and participation in the wholesale market, essential for energy management systems. Managing energy across multiple microgrids poses a significant challenge for operators. To address this, a new method is proposed to model the cyberphysical resilience for microgrids to engage in energy buying and selling within the wholesale market. The proposed model consists of three different functions. First, the wholesale market price is forecasted using a two-stage game theory approach. Subsequently, the second function employs a two-stage game theory technique to determine local exchange requests for microgrids. Following this, the intermediary entity determines the necessary exchange amounts on behalf of virtual microgrids in the third function. The intermediary institution establishes price limits for the wholesale market, serving as the basis for competitive pricing. The proposed model is implemented and solved numerically using MATLAB and general algebraic modeling system (GAMS) software. The results highlight the optimal technical and economic connection between the upstream electrical power network and the downstream microgrids.
This paper presents a novel maximum power point tracking (MPPT) technique for solar photovoltaic systems based on the enzyme action optimization (EAO) algorithm, inspired by catalytic enzyme behavior in biological processes. The proposed EAO-based MPPT approach effectively balances exploration and exploitation to ensure rapid and accurate tracking of the global maximum power point under both uniform and partial shading conditions. The algorithm is first implemented in MATLAB and evaluated across seven test scenarios representing various irradiance patterns. Its performance is benchmarked against conventional perturb and observe, particle swarm optimization, salp swarm optimization, and hybrid particle swarm–genetic optimization algorithms. To validate real-time applicability, hardware-in-the-loop simulations are performed using the OPAL-RT platform under different operating conditions. The results demonstrate that the EAO algorithm achieves faster convergence, reduced oscillations, and higher tracking efficiency compared to existing methods. Specifically, the proposed EAO-based MPPT achieves an average tracking efficiency of 99.02% across all seven test scenarios, outperforming PSO (98.58%), SSA (93%), and P&O (80.55%) methods. The algorithm successfully tracks the global maximum power point within 0.34 s under partial shading conditions and exhibits negligible steady-state power oscillations. Hardware-in-the-loop validation on the OPAL-RT platform under four different irradiance patterns confirms an average tracking efficiency exceeding 99%, demonstrating the practical applicability and robustness of the proposed approach.
In actual operation, it is difficult to accurately evaluate the comprehensive measurement error of transformers and electricity meters due to the coupling of working and environmental operating conditions, nonlinear response, and other factors. This article proposes a data-driven modeling and optimization strategy that integrates multidimensional operation parameters such as voltage, current, harmonic content, temperature, and load rate collected in the field and constructs a comprehensive error prediction model based on random forest. Then, based on this, three optimization measures are designed: dynamic compensation, equipment matching optimization, and early risk warning. The results show that the root mean squared error (RMSE) of the model is 0.15% and the R2 is 0.93 on the test set, which is obviously superior to the traditional method. After the implementation of the optimization strategy, the average error of the three typical stations decreased from 0.31%–0.58% to 0.14%–0.21%, with the largest decrease of 63.8%, and the annual electricity fee dispute decreased by 89,000 yuan. These results indicate that the data-driven method can effectively capture the coupling error characteristics of the metering link under the tested complex operating conditions. The proposed strategy has good engineering applicability and economic value and provides a feasible way to improve the high-precision measurement level of the smart grid.
Permanent magnet synchronous machines are commonly employed in variable-speed applications because they offer high efficiency, easy control, robust construction, and do not require external field excitation. However, rising costs of permanent magnets have significantly increased the overall price of these machines. Wound-field synchronous machines (WFSMs), which have no permanent magnets and are thus less affected by market price changes, can be a good alternative to permanent magnet synchronous machines. WFSMs use slip rings along with brush assemblies to energize the rotor field winding (RFW) with DC. However, the use of slip rings and brushes introduces sparking and increases maintenance requirements. Consequently, brushless wound-field synchronous machines (BLWFSMs) have recently been developed to overcome these limitations. This paper introduces a novel BLWFSM designed to mitigate the volatility of permanent magnet prices and the maintenance challenges of conventional WFSMs using slip rings and brushes. Unlike existing brushless topologies that necessitate dual inverters, the proposed design utilizes a single three-phase inverter. The proposed machine’s rotor incorporates an excitation winding and a field winding, connected via an embedded rotating bridge rectifier. The stator MMF is dominated by a 30-Hz subharmonic component and a 60-Hz fundamental component. Brushless excitation is achieved by utilizing a subharmonic stator MMF component (30 Hz) that induces an AC voltage in the excitation winding, which is rectified by a rotating rectifier and supplied as DC to the field winding. Two-dimensional finite element analysis validates the performance of an 8-pole, 18-slot configuration. The results demonstrate a rated torque of 8.975 Nm with an exceptionally low torque ripple of 0.501%. These characteristics, combined with a reduced volume and weight profile, position the proposed BLWFSM as a robust alternative for air blowing, aerospace power generation, and electric vehicle applications.
Modern receiving-end power systems are characterized by concentrated load centers, long-distance power transfer, and increasing penetration of inverter-based renewable resources. To support reproducible voltage-security studies for such systems, this paper develops a load-center–oriented equivalent benchmark based on the classical 3-machine 9-bus framework. The proposed benchmark construction explicitly defines the transformation from the original system, including bus and branch data, generator dispatch, load allocation, corridor ratings, renewable locations, SVG/OLTC/ESS/capacitor parameters, and coordinated-control variables. Beyond steady-state voltage profiles, the validation framework incorporates quantitative load-center representativeness indices, renewable-output sensitivity, load-growth screening, selected N-1 contingency results, and a reduced-Jacobian modal indicator. The benchmark reproduces a dominant transfer corridor with 81.50% loading, a minimum bus voltage of 0.9700 p.u., and a high-renewable overvoltage case reaching 1.0807 p.u. Coordinated multiresource control reduces the maximum target-bus voltage magnitude by 0.0526 p.u. relative to the uncontrolled high-renewable case. The model is intended as a transparent steady-state benchmark for voltage-security, renewable-hosting, and coordinated-control studies. Fast electromagnetic and converter-control phenomena are identified as necessary extensions for dynamic studies.
Accurate battery state-of-health (SOH) prediction is a crucial requirement in advanced battery management, particularly in the expanding domain of electric vehicles (EVs) where safety concerns are of great importance. The challenge in predicting the SOH during the early stages of battery life is predominantly linked to the limited correlation between health indicators and actual SOH. In this context, the efficiency of degradation indicators (DIs) plays a critical role and is emerging as a key determinant that significantly affects the precision of the SOH prediction. This paper proposes a machine learning approach for SOH prediction of high-capacity pouch lithium-ion batteries (LIBs) in the early life cycle using low-frequency real and magnitude parts of impedance spectroscopy, which contains rich information regarding battery degradation, considering the impact of the battery state-of-charge (SOC). The DIs are extracted from electrochemical impedance spectroscopy (EIS) data using Pearson correlation analysis to select the DIs with a strong relationship with the SOH and minimal influence from the SOC to improve the transferability of the selected features under the same cell type and controlled experimental conditions. This approach also led to a dimensionality reduction in the EIS measurements, reducing the time required to perform the EIS measurements by 33.6% while maintaining SOH-relevant information. The proposed methodology is validated using a Gaussian process regression (GPR) model with different kernel functions and a dataset measured at different SOCs. The results show that the proposed DIs, which are from 15 selected frequencies, can accurately predict the SOH with an average maximum RMSE and MAE of 0.073% and 0.0576%, respectively. The proposed methodology provides a practical EIS feature-selection framework for early-life SOH prediction under controlled experimental conditions.
Distributed static series compensators (DSSCs) are modular power-electronic devices that dynamically regulate transmission-line reactance by injecting a controlled quadrature voltage. This paper presents the first comprehensive systematic literature review focused exclusively on DSSCs, covering studies published from 2004 to 2024. Using a PRISMA-based methodology, 60 high-quality publications were identified and synthesized across eight themes: technology development, steady-state constraints, protection, stability enhancement, quality of supply, economic analysis, optimization, and controller interactions. The review reveals a clear research evolution, from early conceptual work, modeling, and prototypes to studies addressing operational challenges such as congestion, fault-current limitation, and voltage or stability support, and more recent advances in optimal siting, coordinated control, and system-level integration. Although DSSCs demonstrate strong technical potential, existing evidence is largely simulation-based, with limited field testing, no standardized benchmarking methods, and an incomplete understanding of multidevice coordination, lifecycle reliability, and deployment constraints. Overall, DSSCs appear technologically mature but require further research to support widespread practical adoption. The review concludes with targeted research questions on protection-aware control design, real-world performance validation, coordinated operation with other FACTS/DFACTS devices, techno-economic assessment under uncertainty, and system-level planning for high-renewable grids.
In this study, a modified chaos game optimization technique (MCGOT) is developed for the optimal allocation of renewable distributed generators (DGs) and capacitors (CRs) with a reconfiguration network (RN) while considering the reconfiguration ability of power distribution systems (RPDS). The CGOT is inspired by fractal geometry and chaos theory, where candidate solutions are iteratively updated based on geometric interactions among the global best, mean group, and current solutions within the search space. To address the high computational burden associated with multiple offspring evaluations, an MCGOT is further developed. In MCGOT, an adaptive offspring selection mechanism is introduced to generate and evaluate only a single candidate solution per iteration, thereby significantly reducing the number of function evaluations. In addition, a crossover operator is incorporated to enhance solution diversity, while a stage-based control strategy dynamically balances exploration and exploitation. These modifications improve the computational efficiency and convergence characteristics of the original CGOT, making it more suitable for large-scale and computationally expensive optimization problems. The proposed MCGOT and the conventional CGOT are applied to a practical case study of the 59-bus Cairo distribution system in Egypt and the large IEEE 135-bus distribution system. The simulation results show that integrating renewable DGs and CRs through MCGOT significantly minimizes active power losses and mitigates voltage fluctuations. When tested on two systems, MCGOT outperformed traditional methods like CGOT, slime mold algorithm (SMA), artificial rabbits optimization (ARO), and honey badger algorithm (HBA) in mean, worst, and standard deviation metrics. Moreover, to enhance the practical applicability and clean renewable integration, a probabilistic multiobjective formulation was developed considering annual operating losses and carbon emission costs amid renewable generation and load uncertainties using the Hong point estimate method (PEM). Results from the 135-bus distribution feeder indicate that the proposed MCGOT improves convergence characteristics, lowering the expected annual operating cost from $91,394.95 to $88,307.80 and simultaneously reducing expected emissions under stochastic conditions regarding renewable resources.
This paper introduces a data-driven model-free adaptive controller (DDMFAC) as a powerful, practical, and suitable alternative to improve the limitations, challenges, and inherent problems of conventional model–based methods. The proposed method, using input–output data processing independent of the mathematical and systematic model, acts as an effective solution to overcome the limitations and problems caused by complex system modeling and handling large volumes of data. By employing this controller, the overall design becomes significantly simplified and requires no special modeling efforts. The proposed DDMFAC offers several distinctive advantages, including the absence of system identification, online parameter estimation, or state prediction requirements. It features a simple, stable, robust, and adaptive structure; offers relatively low implementation cost; imposes a lower computational burden compared to conventional controllers; and operates independently of the system’s structure and dynamics. These characteristics ensure stable control performance and make it a highly practical solution for industrial systems, devices, and processes. The proposed DDMFAC was implemented in a real-time testbench DSP-based three-phase inverter configured as a shunt active power filter (SAPF) for current control. Also, to validate its effectiveness, several simulation and experimental results are discussed. Using the compensation of the presented method, the total harmonic distortion (THD) of the grid side current was reduced significantly from 21.8% to 2.7%. Also, various verified results of tests demonstrate the effectiveness and superiority of the proposed approach.
The connection of multiterminal HVDC systems to AC networks, along with growing penetration of renewable energy sources (RES) and vehicle-to-grid (V2G) flexibility, creates new challenges for power system operation. This paper presents a two-stage stochastic unit commitment model that jointly integrates AC and DC network operations with EV aggregator flexibility, minimizing expected system cost under renewable uncertainty. The model is formulated as a Mixed Integer Linear Program (MILP) based on linearized DC power flow with a piecewise-linear HVDC converter loss representation. The first stage determines day-ahead commitment, reserve allocation, and converter scheduling; the second stage optimizes real-time dispatch recourse for each renewable scenario. N-1 security constraints ensure postcontingency feasibility. An enhanced Benders decomposition with Magnanti–Wong cut strengthening and adaptive regularization is developed as a solution framework intended for transmission-scale instances; for the distribution-scale test cases studied here, the deterministic-equivalent MILP is solved directly to a 0.5% gap. The scenario reduction technique based on the Wasserstein distance ensures faithful uncertainty representation. Numerical experiments on modified IEEE 33-bus and 123-bus AC–DC hybrid distribution test systems demonstrate cost reductions of 16.4% and 28.5% relative to a deterministic AC baseline, while an ablation study decomposes contributions from stochastic modeling (4.4%), AC–DC coordination (11.0%), and EV flexibility (1.0%). For the 123-bus system, HVDC coordination dominates with a 22.9% cost reduction. A congested-network sensitivity analysis confirms nonzero curtailment under tight line limits, and a tightened-reserve study isolates the V2G contribution as a genuine reserve substitution mechanism rather than slack absorption. An additional scenario-count sensitivity and a head-to-head comparison against deterministic, robust, and Wasserstein distributionally robust formulations confirm the validation robustness of the proposed approach.
This study presents a safety-aware conditional reinforcement learning (RL)–based supervisory proportional–integral (PI) gain scheduling scheme for robust load frequency control (LFC) of a three-area interconnected power system. Different from most RL-based LFC schemes that either replace the existing controllers or provide direct control action to actuators, the proposed scheme keeps the conventional PI controller as the only actuator-driving unit and utilizes RL as a steady-state supervisory gain tuner. A detailed nonlinear three-area LFC model, which includes a full tie-line coupling scheme, is considered. Multistep load disturbances are applied to one control area, and robustness is examined against simultaneous ±30% uncertainty in inertia, damping, governor, and turbine parameters. The proposed supervisory RL policy is implemented using an actor–critic RL algorithm, and the policy is activated only when both area control error (ACE) and its derivative are below a certain prespecified threshold. All gain changes are bounded and low-pass filtered to ensure the nonintrusive nature of the controller. Simulation results, including single-run and Monte Carlo analysis, demonstrate that the proposed controller retains the transient response and control effort characteristics of the particle swarm optimization (PSO)–tuned PI baseline while achieving an improved steady-state frequency distribution with a lower spread under uncertainty. It is demonstrated that RL can provide useful benefits in LFC when it is used as a conditional steady-state supervisory tool instead of a direct transient controller.
Transformer faults can cause power outages, equipment damage, and costly repairs, making early detection critical. Traditional diagnostic methods like dissolved gas analysis (DGA) and frequency response analysis (FRA) are useful, but they often struggle with noisy, overlapping signals, and varying operating conditions. Vibration analysis provides a noninvasive, real-time alternative for monitoring transformer health although analyzing these signals is difficult because of external noise and complex fault patterns. The present study explores using deep learning models to predict faults from vibration time-series data, with a focus on a hybrid architecture that combines convolutional neural networks (CNNs) and multihead self-attention (MHSA). CNNs effectively extract local signal features, while MHSA captures long-range dependencies and global patterns. This combination enables the model to attend to the most relevant parts of the signal and suppress noise. Interestingly, our results also reveal that when CNNs are used as feature extractors, adding the Transformer Encoder’s feedforward network (FFN) does not improve performance and can even slightly degrade it. Overall, our experiments show that the CNN–MHSA model consistently outperforms other architectures, including standalone CNNs, MHSA, Transformer Encoders, and CNN–Transformer Encoder hybrids. In terms of relative absolute error (RAE), the CNN–MHSA model achieves consistent improvements over the baseline models. In particular, the proposed model achieved an average reduction in RAE across all evaluated CNN configurations within the considered hyperparameter space, ranging from 0.5% to 1.3% for excitation voltage prediction and up to 3.7% for load current prediction—these improvements are statistically significant p<0.05, as determined by a one-sided paired t-test. In addition to model development, this study provides a systematic and comprehensive comparative evaluation of multiple architectures, including standalone CNNs, MHSA, Transformer Encoders, and hybrid models under consistent experimental settings. Overall, our findings demonstrate that integrating CNNs with attention mechanisms provides a powerful approach for accurate and reliable transformer fault prediction.
The integration of renewable energy into microgrids is crucial for sustainability and resilience, but its variability requires robust energy storage. This study investigates the optimal sizing of vanadium redox flow battery (VRFB) systems, simultaneously considering both energy and power ratings. Our analysis demonstrates that an optimally sized VRFB (800 kW/6.5 MWh) reduces annual microgrid costs from $224,528 to $221,450 through enhanced energy arbitrage and better renewable utilization. Furthermore, we examine how varying storage sizes impact costs. The results show a trade-off: While investment costs rise linearly with storage size, operational costs decrease due to improved efficiency. However, these operational savings saturate when generation units are power-limited, underscoring the necessity of a balanced power-energy capacity. This research provides key insights for designing cost-effective and resilient VRFB-supported microgrids.
This paper investigates parameter identification for second-order equivalent circuit models of lithium-ion batteries using test protocols of different durations. A commercially available INR21700-45E (NMC) cell is characterized using two segments from hybrid pulse power characterization (HPPC) testing: a 1 h relaxation window and a 6 min discharge window. Parameters are extracted through particle swarm optimization combined with direct resistance calculation. Validation is performed under independent operating conditions, including 2C HPPC pulses and C/2 constant-current discharge. Comparative results indicate that the 1 h relaxation window yields parameters with significantly lower validation errors, achieving RMSE values of 21.2 mV (2C HPPC) and 29.9 mV (C/2 discharge), with R2>0.98. In contrast, the 6 min discharge window produces parameters with substantially higher errors, particularly under continuous discharge (RMSE = 100.8 mV, R2=0.8045), revealing condition-dependent limitations. The findings highlight the role of test duration in influencing parameter estimation and model performance, offering quantitative insights for battery test protocol design. This is an open access article under the terms of the Creative Commons Attribution-Noncommercial License, which permits use, distribution, and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.
