Short-Term Coal Demand Forecasting of Provincial Region Based on GRA-PCA and CNN-BILSTM

At present, coal occupies the largest proportion in China's energy structure, so the accurate prediction of coal consumption demand is the key technology to ensure energy security and stable supply. However, the current coal demand forecasting technology mainly relies on multiple regression analysis, empirical judgment and other methods, and the forecasting results are unstable and the error is large, so a new forecasting technology that can accurately forecast coal demand is urgently needed. Based on this, a coal demand forecasting method based on GRA-PCA and CNN-BILSTM is proposed in this paper. This method uses GRA-PCA and CNN-BILSTM hybrid neural network to build a coal demand forecasting model. The results show that the CNN-BILSTM model has the best prediction effect when the correlation coefficient of feature factor selection is above 0.7. The predicted values in July, October and November are close to the true values, with error rates of 1.1%, 1.7% and 3.3%, respectively. The average RMSE and average error rates are 73.220 and 5.1%, respectively. The prediction effect of ELM model and LSTM model is unstable and the accuracy is not high. The prediction effect of ELM model is the most unstable among the three models, with the average RMSE and average error rates of 457.6 and 35.6%, respectively. The prediction effect of LSTM model is slightly better than that of ELM model, and the average RMSE and average error rate are 391.868 and 29.6% respectively. The prediction effect of CNN-BILSTM model is more stable than that of ELM model and LSTM model, and the average RMSE and average error rate are 73.219 and 5.1%, respectively.