Short-term Photovoltaic Power Generation Prediction Based on LSTM Digital Twins

The prediction of photovoltaic power generation is of great significance to the stability and safe operation of power grid. A digital twin prediction model based on long short term memory (abbr.LSTM) network was proposed, and by means of digital twin technology model the accurate prediction of photovoltaic (abbr. PV) power generation was realized. The digital Twin can be divided into physical space and data space. Firstly, according to the meteorological twin data obtained from the physical space the preliminary predicted power was obtained by LSTM algorithm, meanwhile the historical meteorological database was updated. Secondly, the similar day was found out in meteorological database, and comparing the predicted power on similar days with the actual power the error correction of the preliminary predicted power was conducted to obtain the predicted value of final PV power. Using the proposed digital twin the connection of the physical entity with the data-driven method was realized, simultaneously the self-studying and updating of the physical entity could be carried out. Therefore, compared with traditional PV prediction results the obtained result was more accurate. The accuracy of digital twin prediction is further verified by simulation example.