Integration of AI-based load forecasting and hybrid metaheuristic optimization for sustainable capacity expansion planning in decarbonized solar energy systems

Abstract

With progressive integration of renewable generation and energy storage techniques, it is crucial to examine the modelling techniques and assumptions of capacity expansion planning (CEP) for power grids dominated by thermal-based systems. The proposed framework integrates artificial intelligence (AI) based demand prediction and develops a novel hybrid optimization algorithm for a multi-objective CEP (MO-CEP) model designed with high renewable integration. First, we propose a hybrid model combined with the grasshopper optimization algorithm and random forest (GOA-RF) technique for load forecasting, achieving high accuracy with an R2 value of 99.81% for training and 98.76% for testing and low error metrics with a mean absolute error (MAE) of 0.1079 and a root mean square error (RMSE) of 0.1573. Furthermore, the forecasted load values serve as an essential constraint for CEP analysis under various scenarios. Typically, conventional optimization algorithms have drawbacks due to an imbalance between exploration and exploitation, inadequate exploration delays convergence, while excessive exploitation leads to premature convergence at local optima. To overcome this challenge, a novel hybrid Differential Evolution-based Harris Hawks optimization (DEHHO) algorithm is proposed, providing better convergence speed and solution efficacy. The model investigates key trade-offs among cost, reliability, emissions, and flexibility over 12-year and 28-year planning horizons. Comparative analysis demonstrates that the overall system costs are reduced by 8.12% and 16.7%, while the loss of load probability (LOLP) improves by 86.1% and 79.5% and unmet demand decreased by 86.1% and 79.5% from base to high solar scenario for the same planning horizon.