Research On Air Conditioning Load Prediction Method Based On Hybrid Data Driven Model

With the rapid development of the economy,building energy consumption is also on the rise,with air conditioning accounting for about 40% of building energy consumption.The optimal operation of air conditioning is conducive to energy conservation and emission reduction of buildings.Accurate air conditioning load prediction can provide favorable data support for the customization of its optimal control strategy,which can reduce the energy consumption and operation cost of air conditioning operation under the premise of guaranteeing the comfort of building living.Therefore,accurate and stable air conditioning load prediction is of great significance for building energy conservation management.Firstly,aiming at the characteristic that air conditioning load data belongs to typical time series data,an air conditioning load prediction method based on Long Short-Term Memory(LSTM)model is proposed in this paper.To reduce the impact of external interference on load prediction,the Spearman is used to filter variables that are highly correlated with air conditioning load as input variables for the model,and then the Locally Weighted Scatterplot Smoothing(LOWESS)is used to denoise the data,improving the authenticity of the data.Finally,input variables are fed into the LSTM model to predict air conditioning load.Secondly,in view of the problems of air conditioning systems in some buildings,such as less historical data of air conditioning load,less sampling points of sensors,and load data of minute level required for optimal control,an air conditioning load prediction method based on CS-LSTM-BP hybrid model is proposed on the basis of LSTM model.The compressed sensing(CS)is used to expand the dimension of air conditioning load data,and the number of training set data is expanded.And the LSTM is improved by combining the optimization of BP neural network with beetle swarm optimization algorithm(BSO),which improves the prediction accuracy of LSTM.Experimental results show that the proposed method has high precision and can predict air conditioning load well.Thirdly,aiming at the problems of single feature and insufficient accuracy and stability of LSTM and CS-LSTM-BP methods,an air conditioning load prediction method based on LBF hybrid data-driven model is proposed in this paper.LSTM is used to make preliminary prediction through the historical load data of air conditioning as the state transition model.BP is used to establish the relationship between the air conditioning load and the operating data of each part of the air conditioning system as the observation model.The particle filter(PF)is introduced to combine the two to form a state space model and perform state estimation.LBF hybrid model is used to predict the air conditioning load,which improves the robustness and accuracy of the prediction.Finally,aiming at the problem of insufficient accuracy of state transition model of LBF model,an air conditioning load prediction method based on HRD-TCN-PF hybrid data-driven model is proposed.Hybrid data decomposition(HRD)combined with time convolutional network(TCN)and LSTM is used to improve LSTM.First,HRD is used to decompose air conditioning load to better extract data features.Then,the decomposed data is predicted by TCN and LSTM to improve the accuracy of the state transition model.The experimental results show that the HRD-TCN-PF hybrid data-driven model has higher prediction accuracy than the above models.