Air-Conditioning Load Prediction of Subway Station Based on Clustering and Optimization Algorithm Ensemble Neural Network

Abstract:

Three models were developed to predict the air-conditioning hourly cooling load of a subway station from the aspects of optimization algorithm ensemble back propagation neural network （BPNN） and BPNN with data clustering pre-processing. The results show that the influence of the same physical parameters on the air-conditioning load of the subway station reflects a certain dynamic change characteristic over time. Quantitative analysis of these features based on historical data is of great benefit to precisely selecting the model input parameters and improving model prediction accuracy. In the three given models， the predicted mean absolute percentage error （MAPE） of particle swarm optimization （PSO）-BPNN and the fruit fly optimization algorithm （FOA）-BPNN decreases by 25.87% and 40.08% respectively compared with that of BPNN， while the MAPE of Kmeans-BPNN is reduced by 61.12% and 51.90% respectively compared with that of PSO-BPNN and FOA-BPNN， which means that the performance of optimization algorithm ensemble models is better than that of pure BPNN on even ground. Moreover， BPNN with data clustering is better than optimization algorithm ensemble BPNNs after distinguishing the characteristics of real load changes.