Research on short-term wind power Prediction of GRU based on similar days

Abstract

Abstract As the proportion of wind power generation continues to increase, accurate forecasting of wind power output is of great significance to the smooth operation of the entire power grid. However, due to the greater impact of environmental factors, wind power generation has strong randomness, and it becomes difficult to accurately predict the power generation. Thus, a new hybrid model for wind power generation prediction combining GRU neural networks and similar days’ characters analysis is proposed to address solve this problem. The prediction method employs grey relation analysis to screen similar days, which not only reduces the amount of data required to train the model, reduces the computational complexity, and improves the training speed, but also improves the prediction accuracy based on the selected datasets. In addition, this method also filters and processes the data through box-plot analysis and linear smoothing, which further improves the prediction accuracy of the model. The results show that compared with a single GRU network, the MAE of this method has dropped by 1.89, RMSE has dropped by 1.9, and MAPE has dropped by 11.07%. Obviously, the prediction model based on similar days extraction has obvious advantages.