Abstract
Smart Electric Supply Management (SESM) plays a pivotal role in the modernised grid, addressing challenges such as energy poverty and grid instability, particularly in developing nations. These challenges stem from rising peak demand and the integration of Renewable Energy Sources (RES) like solar and wind, which are inherently intermittent. Understanding daily peak load consumption is essential for adequate energy planning, management, and resource allocation, as it forms a core aspect of supply-side management. However, traditional methods struggle to provide accurate predictions due to ever-changing demand patterns, highlighting the need for more advanced Machine Learning (ML) models, especially in hybridised or enhanced forms. Additionally, Battery Energy Storage Systems (BESS) offer a practical solution to grid imbalances caused by RES by storing surplus energy during peak generation and discharging it during high-demand periods. Properly sizing BESS is crucial to balancing cost-effectiveness and grid stability, as improper sizing can result in inefficiencies, elevated costs, and decreased reliability. This research introduces several technical innovations in optimisation and forecasting to address these challenges. First, the Pelican Algorithm Optimised Support Vector Machine (POA-SVM) model integrates a novel bio-inspired Pelican Optimisation Algorithm that mimics pelican hunting behavior for adaptive hyperparameter tuning of SVM, improving convergence speed and accuracy in daily peak load and peak hour forecasting. Second, the Chaotic Pelican Optimisation Algorithm - Support Vector Machine (CPOA-SVM) enhances this approach by incorporating chaotic maps into the optimisation process, introducing deterministic randomness that boosts exploration and prevents premature convergence, leading to superior solar generation forecasts. This model also combines advanced feature selection techniques (MRMR and RReliefF) to intelligently prioritize critical meteorological and temporal variables, improving generalization and reducing dimensionality. Third, the study develops the Roulette Chaotic Pelican Optimisation Algorithm (RCPOA), which integrates a roulette wheel selection mechanism with chaotic dynamics and POA principles to robustly balance exploration and exploitation in complex optimisation landscapes. This algorithm is applied to optimally size BESS and optimise strategic energy operations, accommodating multiple system constraints and variable conditions.
Methodologically, the study employs a qualitative approach, with all models developed and validated in MATLAB. The POA-SVM is benchmarked against traditional SVM, Bayesian-
7
optimised SVM, and PSO-SVM using raw and normalized datasets, and evaluated via RMSE, MSE, MAE, and R², with K-fold cross-validation. The CPOA-SVM uses time-related, historical, and meteorological data across three experimental cases, demonstrating superior performance, especially when combined with feature selection. The RCPOA is rigorously tested on classical benchmark functions, CEC 2019 problems, and real-world engineering challenges to confirm its robustness and adaptability. Sensitivity analysis further reveals the critical impact of operational costs, demand deficits, and grid energy costs on optimal energy system performance.
The findings underscore the potential of these innovative algorithms in enhancing SESM by improving demand response, mitigating outages, and integrating renewable resources effectively. The POA-SVM model’s adaptive optimisation enables accurate peak load forecasting for better grid planning, while the CPOA-SVM’s chaotic mechanisms and feature selection advance solar forecasting accuracy and grid reliability. The RCPOA’s hybrid metaheuristic approach ensures efficient and cost-effective BESS sizing, enhancing grid stability in dynamic, real-world energy systems. Collectively, these technical innovations advance the SESM framework, providing practical, scalable, and resilient solutions for smart grid management in developing countries.