LSTM based long-term energy consumption prediction with periodicity

Abstract Energy consumption information is a kind of time series with periodicity in many real system, while the general forecasting methods do not concern periodicity. This paper proposes a novel approach based on long short-term memory (LSTM) network for predicting the periodic energy consumption. Firstly, hidden features are extracted by the autocorrelation graph among the real industrial data. The correlation analysis and mechanism analysis contribute to finding the appropriate secondary variables as model input. In addition, the time variable is complemented to precisely capture the periodicity. Then a LSTM network is constructed to model and forecast sequential data. The experimental results on a certain cooling system demonstrate that the proposed method has higher prediction performance compared with several traditional forecasting methods, such as autoregressive moving average model (ARMA), autoregressive fractional integrated moving average model (ARFIMA) and back propagation neural network (BPNN). The RMSE of LSTM is 19.7%, 54.85%, 64.59% lower than BPNN, ARMA, ARFIMA on the May test data. Furthermore, considering the limitation of missing certain measuring equipments, new prediction models with the reduced secondary variables are retrained to explore the relationship between the prediction accuracy and the potential input variables. The experimental results demonstrate that the proposed algorithm has the excellent generalization capability.