Short-Term Prediction of Building Sub-Item Energy Consumption Based on the CEEMDAN-BiLSTM Method

Abstract

In order to improve the accuracy of the short-term prediction of building energy consumption, this study proposes a short-term prediction model of building energy consumption based on the CEEMDAN-BiLSTM method. In this study, the energy consumption data of an office building in 2019 are selected as a sample, and CEEMDAN is used to decompose the energy consumption data into multiple components, and the strong correlation components are selected and sent to the BiLSTM network. The final energy consumption prediction results are obtained by superimposing the prediction results of each sub-component, and five models are built simultaneously to compare the errors with the proposed models. The results showed that the weather type has a great influence on the accuracy of energy consumption prediction. When the weather fluctuates greatly, the prediction error of energy consumption by a single prediction model is large. When the weather suddenly changes, the EMD-LSTM model has a big error in the prediction of air conditioning energy consumption. After CEEMDAN decomposition of energy consumption data, more detailed components can be extracted, which makes the BiLSTM prediction algorithm more accurate. Compared with the CEEMDAN-LSTM model, the CEEMDAN-BiLSTM model reduces eRMSE, eMAPE, and eTIC by 4.1%, 9.441, and 1.3%, respectively. The proposed model can effectively improve the accuracy of short-term prediction of building energy consumption.