引入用能行为概率和群智能优化的数据驱动高精度小时尺度建筑能耗预测体系

Data-driven high-precision hourly-scale building energy consumption prediction employing occupant energy-use behavior probability and swarm intelligence optimization

提出了一种改进的用户用能行为概率模型,作为新输入集成入能耗预测中,引入麻雀搜索算法(SSA)用于优化长短期记忆神经网络(LSTM)的超参数选择,建立了高精度小时尺度建筑能耗预测体系。在某建筑中的实际应用显示,相比于传统预测体系,改进的能耗预测体系可以使决定系数R 2平均增大0.201,平均绝对百分比误差(MAPE)平均减小18.10%,均方根误差的变异系数(CV-RMSE)平均减小0.176。

This research proposes an improved probability model of occupant energy-use behavior,which is integrated into energy consumption prediction as a new input.The sparrow search algorithm(SSA)is introduced to optimize the hyperparameter selection of the long short-term memory neural network(LSTM).A high-precision hourly-scale building energy consumption prediction system is established.The practical application in a building demonstrates that compared with the traditional prediction system,the improved energy consumption prediction system can increase the coefficient of determination(R 2)by 0.201 on average,and decrease the average absolute percentage error(MAPE)by 18.10%and the coefficient of variation of the root mean square error(CV-RMSE)by 0.176 on average.