基于随机森林改进深度门控循环单元的高校食堂用能预测方法

Prediction Model of Building Energy Consumption in University Canteens Based on RF-GRU

高校作为耗电量最高的机构之一,其能耗随时间和气象参数的变化幅度较大。因此,高校建筑的用能水平分析和准确预测对稳定供电和建筑智慧运维具有重要意义。K均值(K-means)聚类算法可以对高校建筑的用能水平进行分析。本文提出1种新的基于随机森林RF改进的深度门限循环单元GRU的高校食堂能耗预测方法RF-GRU,采用RF降低输入特征矩阵维数,GRU算法对食堂能耗进行预测。应用北京市某高校食堂为例对该方法进行验证。对其总用能、炊事用能、空调用能、照明用能、其他用能和电梯用能建立K-means模型,分析该高校食堂的日用能模式。结果表明,该高校食堂能耗可分为高负荷和低负荷2种用能水平。建立该食堂总用能、炊事用能、空调用能和照明用能的日能耗预测模型,对比RF-GRU和GRU模型。结果表明,与GRU模型相比,RF-GRU模型的R2更高,RMSE和MAPE更低。实例证明RF可以提高能耗预测模型的准确度,RF-GRU方法可以满足工程应用对精度和速度的要求,为高校食堂建筑多输入特征情况下的能耗预测提供1种新的解决思路。

As one of the highest power-consuming institutions,the energy consumption of universities varies widely with time and meteorological parameters.Therefore,the analysis and accurate prediction of the energy consumption level of university buildings are crucial for stable power supply and intelligent building operation and maintenance.K-means clustering algorithm can be used to analyze the energy consumption level of university buildings.In this paper,a new energy consumption prediction model RF-CRU for university buildings based on the deep gated recurrent unit(GRU)improved by random forest(RF)was proposed.RF was used to reduce the dimensionality of the input characteristic matrix,and GRU was used to predict the energy consumption of university canteens.The method was validated by applying to a university canteen in Beijing.K-means models were established for its total energy consumption,cooking energy consumption,air conditioning energy consumption,lighting energy consumption,other energy consumption,and elevator energy consumption to analyze its daily energy consumption pattern.The results show that the energy consumption of this university canteen can be divided into two energy consumption levels of high load and low load.The daily energy consumption prediction models for total energy consumption,cooking energy consumption,air conditioning energy consumption,and lighting energy consumption of this canteen were established,and comparison with the RF-GRU and GRU models were made.The results show that the RF-GRU model has higher R^(2)and lower RMSE and MAPE compared with GRU model.The example proved that RF can improve the accuracy of the energy consumption prediction model,and the RF-CRU method can meet the requirements of accuracy and speed for engineering applications,providing a new solution for energy consumption prediction in the case of multiple input features in university canteens.