Forecasting building plug load electricity consumption employing occupant-building interaction input features and bidirectional LSTM with improved swarm intelligent algorithms

Building energy consumption prediction is an essential foundation for energy supply-demand regulation. Among them, plug-load energy consumption in buildings accounts for approximately 12–50 % of the total energy consumption, making plug-load energy consumption prediction crucial. However, accurately predicting plug-load electricity consumption is challenging due to the influence of random human behaviors. This study presents a comprehensive plug-load electricity consumption prediction system. First, the conventional input system based on influence factors and the novel input system based on occupant behavior probability were proposed. Second, long short-term memory (LSTM) and its improvement (Bi-LSTM) are used as the fundamental algorithm. Finally, the whale algorithm (WO), a swarm intelligent algorithm, is utilized to improve the prediction accuracy. The results show that the prediction system proposed performs better with R increased by 0.70%–23.97 %, MAPE decreased by 5.33%–40.92 %, and CV-RMSE decreased by 1.10%–21.08 %, compared to the traditional prediction system. The combination of two input systems and four algorithms can accommodate different prediction accuracy requirements, data collection conditions, building functions, and time requirements.