Ultra-short-term power prediction method of photovoltaic power stations based on ConvLSTM and multimodule fusion algorithm

This paper proposes an ultra-short-term power prediction algorithm for new PV stations using AI technology. Its goal is to accurately predict PV station output power during peak sunlight hours. The algorithm includes small model, residual, Kalman, and neural network fusion modules. Compared to previous methods, this scheme reduces data dependence, deployment time, and improves accuracy, it is helpful to solve the phenomenon of light abandonment of photovoltaic power stations, reduce the peak shaving pressure of power grid, and promote the sustainable development of renewable energy.