Short-Term Solar PV Power Forecasting: A Comparative Analysis of Neural Network Optimization Techniques

Accurate solar photovoltaic (PV) power forecasting is essential for optimizing energy management and ensuring reliable integration of renewable energy into power grids. This study investigates the influence of various optimization algorithms on the performance of neural network architectures, including Convolutional Neural Networks (CNNs), Multi-Layer Perceptrons (MLPs), Long Short-Term Memory networks (LSTMs), and Gated Recurrent Units (GRUs), for solar PV power prediction. Using real-world datasets, the paper evaluates these models across multiple metrics. Advanced optimizers such as Adam, Lookahead, AdaBelief, and Adaptive Gradient Clipping (AGC) are compared to highlight their suitability for specific architectures and their ability to handle the unique challenges of PV power data, including temporal dependencies and noisy inputs. The findings reveal model-specific optimization strategies that maximize the accuracy of the forecast, offering practical approaches to improve renewable energy predictions and advancing the reliability of smart grid systems.