Air Conditioning Load Forecasting Based On LSTM And The Optimization Control Of Air Conditioning Energy Consumption

Air conditioning has become a necessity in People’s Daily life.Air conditioning will cause a lot of energy consumption while giving people a comfortable environment.In order to reduce the waste of electric energy during the operation of air conditioning,this paper predicts the air conditioning load at the next moment,using the long-short Term Memory(LSTM)model.Through load forecasting,the optimal working condition of air conditioning equipment can be determined in advance,so as to reduce the overload working time of air conditioning and achieve the purpose of energy saving.A good controller can make the system respond quickly and reach the stable state faster,but also can effectively reduce the energy consumption of air conditioning.So in this paper,fractional order internal model controller is used to adjust the temperature of the air conditioning room,and neural network is used to modify the controller parameters in real time.For the optimal control of chilled water system,this paper uses the improved sparrow search algorithm to find the optimal parameters of the controller.The research contents of this paper are as follows:(1)The types of parameters used for load prediction in this paper include load,outdoor temperature and outdoor humidity.Grey correlation degree is used to analyze the data and find out the load influencing factors with large correlation degree.One part of the data is used for off-line training,and the other part is used to predict the value at the next moment.Put the data into the trained LSTM model for load prediction.The prediction effects of Back Propagation Neural Networks(BPNN)model and Support Vector Machine(SVM)model were compared.The LSTM model had the best prediction effect,and the values of MAE and RMSE were 24.34 and 33.88 respectively.Moreover,the prediction time of this model is 120 S,which is less than that of SVM model and a little longer than that of BP model.Comparing the prediction results before and after using the grey relational degree analysis method,it is found that the two evaluation indexes are 24.34 and 33.88 respectively after using the grey relational degree analysis method,and the two evaluation indexes are 27.35 and 35.60 respectively without using the grey relational degree analysis method.The prediction accuracy of the model is effectively improved by using the grey relational degree analysis method.(2)LSTM model can be deeply mined for load,but the convergence speed is relatively slow.Gated Recurrent Unit(GRU)has a simple structure and fast convergence speed,but its ability to handle big data is not as good as that of LSTM.Based on their advantages,GRU and LSTM are combined in this paper.First,the data is passed through the LSTM layer to find the connection between the data,and then the data is passed through the GRU layer to accelerate the convergence.Bidirectional LSTM and bidirectional GRU are built and combined,and the combined model is used to predict the load.Through simulation comparison,it is found that the model proposed in this paper has the best prediction effect.The value of the evaluation index on the first day is 18.725807 and 23.796540 respectively,and the training time is 251s;the value of the evaluation index on the second day is 11.0212 and 14.2416 respectively,and the training time is 135s.(3)The parameters of PID controller will not change during the control process,which will lead to higher overshoot and longer stability time in the control process.In order to improve the temperature control effect of air-conditioned room and save energy consumption,a fractional internal model controller based on neural network self-tuning is proposed in this paper to adjust the temperature of air-conditioned room.Firstly,the PID controller is modified and the parameters in the transfer function of the controller are expressed only by the filtering time constant through the inner membrane principle.Then the fractional order is added to the integral and differential terms of the controller to construct the fractional internal model controller.Finally,the parameters of the controller are modified in real time by using neural network.After tuning,the controller makes the system reach the steady state in the least time,which is 6.1s,and the maximum peak value generated by the system response is also the smallest,which is29.3223.The control effect of this controller is better than that of other controllers.(4)Using the classical PID controller can not effectively control the chilled water system.In this paper,an improved sparrow search algorithm is proposed to optimize the control of chilled water system.Firstly,the random walk strategy is used to randomly change the position of the sparrow,so as to improve the ability of the sparrow group to find a better position.Secondly,Gaussian mutation is used to carry out position mutation operation on the sparrow after updating its position,so as to enhance the individual’s ability to fully search in its surrounding area.Finally,modeling is carried out in Matlab and compared with other swarm optimization algorithms.The PID parameter values found by the method in this paper are 3.6982,0.034355 and 3.3944,respectively.The time required for the system to reach the steady state is at least 12.75s,and the control effect is the best.By comparison with the literature[87],the overshoot was 0,the rise time was 2.713s,and the adjustment time was 4.95s,which were better than the indexes in the literature.