Fault Diagnosis On Opposite-insertion For Gasliquid Separator In Variable Refrigerant Flow System Based On Particle Swarm Optimization And Support Vector Machine Method

The application of air conditioning systems is mainly to improve indoor air quality and indoor environment comfort.VRF(Variable refrigerant flow)system is one of the common forms of air-conditioning system,which is widely used due to its convenient installation,high efficiency and energy saving.Under normal operating conditions,because the system performs well under partial load operation,it can reduce operating energy consumption and achieve the goal of energy saving and emission reduction.However,if there is a failure,not only will increase the energy consumption,but also the equipment wear will be accelerated and the indoor comfort will be reduced.Therefore,it is very important to carry out VRF system fault detection and diagnosis,which can make the equipment run efficiently and reduce energy waste.This thesis mainly studies the gas-liquid separator opposite-insertion fault detection and diagnosis in VRF system.In this study,the structure and data source of the VRF experimental system are introduced firstly.After the initial selection,the original data set contains 14 variables.Considering that high correlation between variables may lead to variable redundancy,the univariate feature selection method is adopted.According to the importance ranking and correlation analysis of variables,the feature subset with high importance and small redundancy to the model is finally selected.This feature subset contains 12 feature variables,and the support vector machine model is established by using the feature subset.The fault diagnosis accuracy of the test set is 87.1% under refrigeration conditions and 85.6% under heating conditions.The grid search and cross validation method were used to optimize the model parameters,and the fault diagnosis accuracy of the optimized model was 93.6% and 91.7% respectively under refrigeration and heating conditions.In order to prove the effect of parameter optimization,another parameter optimization method,particle swarm optimization(PSO)algorithm,is adopted for comparison.The fault diagnosis accuracy of the model optimized by particle swarm optimization algorithm is 97.9% and 96.7% respectively under the refrigeration and heating conditions.It can be seen that the parameter optimization effect using the particle swarm optimization algorithm is better,the fault diagnosis accuracy of the model after optimization is higher,and the model training time is shortened by more than 1/3 compared with the grid search method.Therefore,for this study,the particle swarm optimization algorithm is better for parameter optimization.Finally,the conclusions of this study is summarized and a more suitable method for parameter optimization is obtained.The accuracy of the fault detection and diagnosis model after optimized by particle swarm optimization is very high.At the same time,it also points out the shortcomings of the research,that is,the data collection cycle is not long enough and a complete fault detection and diagnosis system has not been formed,which also points out the direction for future research.