SOM-BP Neural Network Applied To Energy Consumption Prediction Of Residential Buildings Heating System

As a vital method in process of building energy conservation,high accurate prediction for building energy consumption can also provide guidance for building energy consumption management.In this thesis,the research of energy consumption prediction is carried out for residential building heating system located in Beijing,China.A residential building heating system energy consumption prediction model based on the SOM-BP neural network is established by using system operating data,local meteorological data and building energy consumption data.And through outlier processing and feature selection,the prediction performance of proposed model is improved.In the original data set,the system operation data is collected through the residential building heating system,and the local meteorological data is collected through the local meteorological station.In view of there are some unreasonable data samples existed in the original data,the scatter plot and boxplot based on statistical principle are used to detect and eliminate the outliers in the experimental data set.To select the optimal input feature variable set of the model,the correlation between each variable and energy consumption is calculated via the maximum information coefficient analysis method.Subsequently,the modeling data is divided into training and test sets according to the ratio of 4:1.The BP neural network is used to construct residential buildings heating system energy consumption prediction model.According to quantitative evaluation result of the model,result indicate that the heating system energy consumption prediction model based on BP neural network still has a large prediction deviation,and the prediction performance of the model needs to be optimized.Considering the similarity of original data samples,the self-organizing feature mapping(SOM)neural network is used to cluster the data set.Then the energy consumption prediction models are established according to different categories.Finally,the heating system energy consumption prediction model based on SOM-BP neural network.Compared with the single BP neural network energy consumption prediction model,the proportion of samples without the prediction error range of(20%)in the test set is decreased by 36.78%,and the RMSE,MAE and MRE decreased by 18.56%,27.05% and 25.53%,respectively.Besides,the influence of initial weight threshold on prediction performance in BP neural network should be taken into consideration.Accordingly,the thought evolution algorithm is adopted to optimize the initial weight threshold of BP neural network model to obtain the optimal initial weight threshold of the model.After optimization by thought evolutionary algorithm,the proportion of samples without the prediction error range of(20%)in the test set of the BP neural network model and the SOM-BP neural network model are respectively decreased by 7.92% and 11.41%.RMSE,MAE and MRE of the former results are decreased by 6.32%,5.41% and 4.25%,respectively,while RMSE,MAE and MRE of the latter results are decreased by 10.53%,10.14% and 12.38%,respectively.The results show that the prediction performance of the model is further improved and the optimization effect is remarkable.