Short-Term Photovoltaic Power Forecasting Using Convolution Neural Network

Abstract

This work presents the four types of short-term photovoltaic (PV) power forecasting models using a 1D convolu tion neural network based on 4 types of data handling. Multi Timestamp Multi-Feature-2 (MTMF-2) forecasting model gives satisfactory results compared to Single-Timestamp Multi-Feature (STMF) forecasting, Multi-Timestamp Multi-Feature-1 (MTMF 1) forecasting, and Multi-Timestamp Single-Feature (MTSF) forecasting. Historical data taken from one of the PV stations in China is used as input to these forecasting models. Data contains all 6 features which are Power, Irradiance, Temperature, Pressure, Wind Direction, and Wind Speed for every 15-minute timestamp. TensorFlow libraries are used for model building, and the results are taken from the VS code IDE.