Variable Refrigerant Flow Air-conditoning Systems Energy Consumption Prediction Based On Data Partitioning And Swarm Intelligence Algorithm Optimization

In recent years,China’s building energy consumption is increasing year by year,accounting for about half of the country’s total energy consumption.Among them,the air conditioning system is the biggest energy consumption in the building system.As one of the common types of air conditioning,the research on energy saving and optimal operation of variable refrigerant flow air-conditioning(VRF)system is of great significance to the construction of resource-saving and environment-friendly society and the realization of carbon peak and carbon neutral.Accurate energy consumption prediction for VRF system is the basis of energy saving management technology such as VRF system optimal control operation,and also an important means of fault diagnosis.Data-driven black box model has been widely used in energy consumption prediction of buildings or air conditioning systems due to its advantages of high modeling efficiency,real-time response processing and portability.At the same time,data partitioning and swarm intelligence algorithms are two research hotspots in the field of data mining,which are used to deal with pattern recognition and parameter optimization problems.This thesis takes VRF system as the research object,explores the feasibility and actual performance of the application of data partitioning and swarm intelligence algorithm in VRF system energy consumption prediction,and proposes a new energy consumption prediction method based on data partitioning and swarm intelligence algorithm optimization,in an attempt to further improve the performance of energy consumption prediction model.Firstly,the energy consumption performance of the VRF system in the standard enthalpy difference laboratory is tested and the data are collected.Then,data preprocessing is carried out on the collected original data set,including data cleaning,outlier detection and elimination,and data normalization,etc.Among them,the local outlier factor(LOF)method is adopted for outlier detection.Then,the input characteristics of the energy consumption prediction model are determined by the expert knowledge of the air-conditioning system,the maximal information coefficient(MIC)and the pearson correlation coefficient.Respectively by using back propagation(BP)neural network,radical basis function(RBF)neural network,generalized regression neural network(GRNN)and Elman neural network,multiple linear regression(MLR)and support vector machine(SVM)six black box model building energy consumption prediction model,the results showed that the BP neural network prediction effect is best,prediction results of coefficient of variation(CV),mean absolute error(MAE)and root mean square error(RMSE)is 2.602,127.279 and182.640,respectively,fitting coefficient(R~2)is 0.9945.Then,the weight and threshold values of BP neural network were optimized by swarm intelligence algorithm.The results show that the optimization effect of particle swarm optimization(PSO)algorithm is better than that of genetic algorithm(GA).Compared with BP neural network before optimization,the CV,MAE and RMSE of the prediction results of particle swarm optimization algorithm decreased by 17.55%,23.91%and 17.55%,respectively.Furthermore,three typical clustering methods,self-organizing Feature map(SOM)neural network,K-means algorithm and density-based spatial clustering of applications with noise(DBSCAN)algorithm,were used to partition the data,identify the potential energy consumption patterns of VRF system,and try to further improve the prediction performance.In addition,in order to visualize the clustering results and facilitate the analysis and evaluation,principal component analysis(PCA)is used to reduce the dimension.The results show that the three data partitioning methods can improve the prediction accuracy of the model in different degrees,and SOM neural network is the best.Taking PSO-BP model as an example,after SOM neural network data partition,the CV,MAE and RMSE of its prediction results decreased by 13.93%,22.60%and 12.88%,respectively.In conclusion,this thesis proposes a mixed black box model based on SOM-PSO-BP for energy consumption prediction of VRF systems,which can significantly improve the prediction accuracy.The CV,MAE and RMSE of the prediction results are 1.847,74.966and 131.195,respectively,and the fitting coefficient is 0.9970.The model can be used to measure the operating parameters of the VRF system.The model can accurately and stably predict the energy consumption of the VRF system under different cooling and heating modes,different control strategies and various operating conditions.It can provide a reference for various energy consumption monitoring platforms and energy management systems,and help the energy-saving and optimal operation of the VRF system.