Ultra Short Term Distributed Photovoltaic Power Prediction Based on Satellite Cloud Images Considering Spatiotemporal Correlation

Abstract

With the advancement of China's carbon peaking and carbon neutrality goals and the development of photovoltaic power generation technology, a large scale of distributed photovoltaics are connected to the rural distribution network in recent years. Photovoltaic power generation features high randomness and uncertainty, Accurate prediction of distributed PV power on ultra-short-term time scale (0-4h) is of great significance to the safe and stable operation of distribution network. This paper proposes a prediction algorithm based on satellite cloud images considering spatiotemporal correlation between solar stations nearby. Firstly, correlation between adjacent power plants are sorted, corresponding prediction models based on LSTM are built using historical power and NWP data, then satellite images are used to choose suitable prediction models for prediction when forecasting. With the actual dataset of photovoltaic power station in northeast China, The proposed algorithm is verified, the test results show that the proposed algorithm proposed in this paper is generally at a better accuracy level compared with other well-established benchmarks in terms of power curve and statistical error.