Prediction of Hourly Subentry Energy Consumed in a Typical Public Building Based on Pattern Recognition

During the operation and maintenance management, accurate energy consumption forecasts can provide valuable energy saving strategy for decision makers. This paper proposes a hourly data-driven prediction model of building subentry energy consumption based on pattern recognition. Firstly, the typical energy consumption patterns of air conditioning, lighting and sockets were extracted by K-Means algorithm, and then the random forest algorithm was used to match the corresponding patterns for the prediction day, and finally the neural network algorithm was used to establish a time-by-hour energy consumption prediction model within the corresponding energy consumption pattern. The above model is verified by taking the subentry consumption data of air conditioning, lighting and sockets of an office building in a hot summer and warm winter region as the research object. The result shows that the prediction model based on pattern recognition has greatly improved the prediction accuracy and stability compared with the direct energy consumption prediction model, among which the relative error of the air conditioning increased by 75.26%, the CV value increased by 46.90%, the lighting’s relative error and CV value increased by 40.19% and 31.11%, the socket’s relative error and CV value increased by 28.59% and 39.79% respectively.