Cooling Load Prediction for Metro Station Based on Dynamic Passenger Flow Model

Abstract:

In order to accurately predict the cooling load of metro stations， firstly， by analyzing the historical data from the metro station monitoring platform， the passenger flow and meteorological parameters are identified as the main influential factors. Then， a dynamic passenger flow neural network prediction model is established， by utilizing the hourly monitored carbon dioxide volume concentration data in the metro station， A comparion of the carbon dioxide data with the automatic fare collection data indicates that the correlation coefficient R² of the prediction model can reach 0.87. After that， based on passenger flow prediction， the cooling load predicting model of the metro station is built. A comparsion of the predicting performance of back propagation neural network and support vector machine algorithm under different time scale training data suggests that the R² of both models can reached 0.95 or higher， and the root mean square error is between 70 kW and 90 kW， which is of a high prediction accuracy. However， the calculation time of the support vector machine algorithm is about 3 to 4 times that of the back propagation neural network algorithm. The neural network model is recommended when the amount of data is large.