Load forecasting method for CCHP system based on deep learning strategy using LSTM-RNN

Abstract

The load forecasting plays an important role in the controlling and optimizing operations for combined cooling heating and power (CCHP) system, and the forecast precision affects the control strategy and system comprehensive energy efficiency directly. Pearson correlation coefficient is used to indicate that the multivariate time series construed by cooling load, heating load and electrical load of the CCHP system are typical chaotic time series which are affected by not only themselves but also by each other. In this study, a novel forecasting method based on long short term memory (LSTM) neural network considered the coupling relationship of three kinds of load in the CCHP system is proposed. The LSTM prediction method can model long-term dependencies effectively by extracting inherent important features from historical data automatically. Compared with the univariate method and some statistical methods, the proposed method has better load prediction preciseness with lower root mean square error (RMSE), lower mean absolute error (MAE), and higher goodness of fit. This method provides an effective way for the load forecasting of CCHP system.