Random Forest Algorithm Based Dynamictraining Set for Cold Load Prediction

Under the background of double carbon development, building energy consumption will gradually tend to be electric, and the energy consumption of building air conditioning system, as an electric power consumption system, accounts for 60% of the total building energy consumption. Achieving accurate prediction of building load is one of the key technologies to optimize the energy consumption of building air conditioning systems. However, traditional building load prediction methods often use historical load data as the training set for prediction, which cannot be done without sufficient historical load data. The paper proposes a building cold load prediction model based on dynamic training set using random forest algorithm. In the initial stage of constructing the prediction model, the random forest is used as the basic algorithm to build the model, and the short-term cold load of the future day is obtained from the building simulation data; in the subsequent stage, the actual data of the building operation is obtained by testing, and the test data is gradually used to replace the earlier data of the original training set, and the algorithm is trained again using the newly obtained dynamic training sample set to predict the short-term cold load of the subsequent day. The results verify that the accuracy of this prediction model is gradually improving as time advances and the training set is continuously updated.