Comparative analysis of Kolmogorov–Arnold and recurrent neural networks for day-ahead photovoltaic forecasting with Metaheuristic optimization for cost and battery management at Green Energy Park, Morocco

• This study proposes a two-stage framework that combines forecasting with cost optimization in microgrids. • Kolmogorv arnold neural network achieves higher PV forecasting accuracy with RMSE of 0.0421. • Battery scheduling uses a Time-of-Use pricing strategy to reduce operational costs and decrease grid reliance. • Scenario testing of three methaheuristic algorithms confirms that the Grey Wolf Optimizer achieves up to 53.37 % cost reduction. • Real-case validation under Moroccan electricity tariffs demonstrates the practical effectiveness of the proposed approach. Microgrid technology plays a crucial role in addressing major challenges in modern power systems. These include energy intermittency, cost management, energy allocation, and battery operation. This study presents an integrated approach that combines advanced solar power forecasting with metaheuristic optimization to improve microgrid efficiency under Moroccan time-of-use tariffs. The framework is structured in two stages. In the first stage, a forecasting model based on the Kolmogorov–Arnold neural network is compared with recurrent neural networks. Results show that the Kolmogorov–Arnold model provides superior predictive accuracy, reaching a root mean square error of 0.0421. The forecasted PV output is integrated into the cost optimization stage for microgrid operation. Three scenarios evaluated to optimize operational costs using metaheuristic algorithms, specifically Particle Swarm Optimization, Genetic Algorithm, and Gray Wolf Optimizer. Furthermore, the simulation, based on a Time-of-Use pricing strategy applied in the Solar Decathlon Africa smart village case study in Benguerir, Morocco, enabled effective demand response, resulting in cost reductions of 53.37 % with GWO, 49.79 % with GA, and 46.20 % with PSO. These findings highlight the importance of combining accurate day ahead PV forecasting with efficient cost optimization to develop scalable, sustainable, and effective microgrid energy management system solutions.