Optimizing Neural Network Architectures for Intraday Solar Forecasting Applications

Accurate forecasting of solar irradiance and energy production is crucial for the effective integration of solar energy into the power grid, especially for intraday operational planning. To optimize neural network designs for intraday solar forecasting applications, a novel approach is proposed in this paper. By utilizing Particle Swarm Optimization (PSO) in conjunction with the Sparrow Search Algorithm (SSA), we improve the accuracy and efficiency of forecasting models. SSA-PSO is used to optimize a variety of neural network designs, including feedforward, recurrent, Long-Short Term Memory, and Convolutional networks, to maximize forecasting accuracy and reduce computing costs. Mean Squared Error was significantly decreased using the Sparrow Search Algorithm with Particle Swarm Optimization optimized model, improving the neural network architecture and yielding more precise predictions with F1 score improvement of 1.39%. Real-world solar energy data empirical examination shows that SSA-PSO optimized models outperform conventional methods. The useful insights this study offers into improving solar forecasting systems emphasize how optimization strategies inspired by nature can improve neural network performance for renewable energy applications.