Solar Power Generation Prediction using LSTM Deep learning algorithm in Ningxia Province of China

Abstract

The rapid expansion of photovoltaic (PV) power generation faces significant challenges due to the intermittent and stochastic characteristics of solar energy, which affect grid stability and energy management. Accurate forecasting of PV power output is crucial for optimizing grid operations and supporting the transition to clean energy. This paper proposes a deep learning approach based on Long Short-Term Memory (LSTM) networks to predict PV power generation in Ningxia Province, China. The model leverages historical power and meteorological data, which undergo comprehensive preprocessing, including outlier removal, normalization, and feature correlation analysis. The experiment is based on collecting data at 15-minute intervals, totaling 35,000 samples from a 1 MW photovoltaic power station in Ningxia for the entire year of 2023. The data include seven characteristic dimensions such as irradiance, temperature, and humidity. Comparative experiments involving Support Vector Machine (SVM), Convolutional Neural Network (CNN), and LSTM demonstrate that LSTM outperforms other methods with superior accuracy and robustness, achieving a coefficient of determination (R²) of 0.9927. The results confirm LSTM's effectiveness in capturing temporal dependencies and nonlinear patterns in PV power data. This study provides valuable insights for enhancing photovoltaic grid integration and advancing intelligent power systems.

-