The growing integration of nondispatchable renewable energy sources (PV, wind) and the need to cut CO2 emissions make energy management crucial. Microgrids provide a framework for RES integration but face challenges from intermittency, fluctuating loads, cost optimization, and uncertainty in real-time balancing. Accurate short-term forecasting of solar generation and demand is vital for reliable and sustainable operation. While stochastic and machine learning methods are used, they struggle with limited data, complex temporal patterns, and scalability. Key challenges include capturing seasonal to weekly variations and modeling sudden fluctuations in generation and consumption. To address these issues, this paper presents a novel three-stage centralized EMS for interconnected microgrids. The first stage involves comprehensive data analysis to extract meaningful patterns. The second stage introduces a hybrid forecasting framework that integrates stochastic (Prophet) with machine learning (BiLSTM) techniques to improve prediction accuracy under uncertainty. In the third stage, a modified linear programming approach leverages the improved short-term forecasts to optimize energy sharing between microgrids, with the aim of reducing operational costs, minimizing carbon emissions, and improving system stability under climate variability. The proposed EMS is designed to accommodate diverse microgrid configurations while maintaining computational efficiency. Four scenarios are considered to evaluate the proposed energy management strategy. The obtained results demonstrate that the proposed EMS significantly improves both forecasting accuracy and operational performance. The combined methods achieve the best performance among all tested models, with an RMSE of 0.0070, MAE of 0.0043, and R² of 0.9988, corresponding to improvements of ΔRMSE = −0.2122 and ΔR² = +0.7126 relative to Prophet. These substantial gains in predictive accuracy translate into more precise battery scheduling, reduced grid dependency, and optimized power dispatching, thereby significantly enhancing system efficiency, reliability, and sustainability. Overall, the results highlight the effectiveness of integrating hybrid forecasting with optimization-based EMS, providing a viable pathway toward high penetration of renewable energy sources in future power systems.

The accurate modeling of PV modules is critical for enhancing performance of electrical grids during simulations. However, modeling PV systems requires dealing with a nonlinear current–voltage curve with unknown parameters, which is frequently due to the restricted data available in cell datasheets. This paper presents a unique optimization technique known as the Human Evolutionary Optimizer Algorithm (HEOA) for correctly predicting the parameters of triple- diode model (TDM). In this study, the HEOA optimizer minimizes the root mean square error function in order to precisely find the TDM's ideal parameters. The simulation outcomes clearly demonstrate the effectiveness of HEOA in accurately extracting model parameters. Throughout the optimization process, the HEOA consistently achieved the lowest ideal fitness values when compared to other state-of-the-art optimization algorithms, indicating superior performance. The results further confirm that the HEOA offers faster convergence rates and higher estimation accuracy, highlighting its efficiency and precision in solving complex parameter identification problems. In addition, comprehensive statistical analysis supports the robustness and consistency of the HEOA across multiple trials, reinforcing its reliability as an optimization tool. For further validation, the proposed approach will be tested using standard benchmark cases, including the RTC France Solar Cell and the KC200GT PV module, to ensure comparability with existing methods. Moreover, real-world validation will be conducted under dynamic climate conditions using the Trina Solar Monocrystalline. Then, the TDM for solar cells/modules is notable for its excellent accuracy, which accounts for both electrical and non-electrical losses. The HEOA-based TDM system achieved parameter extraction performance which matched the benchmark results and practical PV test results. The HEOA method produced the best RMSE result of 8.1307E−04 for the RTC France solar cell, which surpassed multiple recent optimization techniques. The KC200GT PV module achieved an RMSE result of 3.32E−04 which demonstrated that the method accurately reconstructed the nonlinear I–V and P–V characteristics through its extremely small pointwise current and power errors. The experimental validation showed that Trina Solar 290 W monocrystalline modules performed accurately under real operational environments, which included testing single-cell, five-cell, and six-cell configurations to achieve minimum RMSE values of 0.0 and mean RMSE values of 0.0141, 0.0059, and 0.0521. The results demonstrate that HEOA produces precise and dependable parameter estimation for PV cells and modules.