Influence of Data Preprocessing Methods for Energy Consumption Prediction Models in Air Conditioning Systems
According to annual operating data of a heat pump system in northern China,the effects of three denoising methods and three feature selection schemes on the performance of energy consumption prediction models based on radial basis function neural networks are compared.After using empirical mode decomposition,Savitzky Golay,and wavelet transform for denoising,the performance of the energy consumption model and the influence of data satisfying the thermal balance of heat pump units on the model are compared with the model without denoising.Simultaneously the different model performance produced by the feature variables selected by three feature selection methods including filter,wrapper and embedded are compared.The results show that the model accuracy is the highest after denoising using wavelet transform method,and the accuracy of the model can be further promoted by training the model with data that meets the thermal balance of the heat pump unit.Compared to the flitter and wrapper methods,using the embedded method to select the feature variables of the model results in the highest accuracy of the energy consumption prediction model.And the lower the accuracy of the original model,the greater the improvement of the model after wavelet transform denoising and embedding feature selection.The same pattern is observed after training the energy consumption model using data that satisfies the thermal balance of the heat pump unit.
air conditioning systemenergy consumption predictiondata preprocessingfeature selection
万方数据
维普
