Cyclone separator is a typical dry gas-solid separation device,which uses centrifugal force to separate solid particles from gas flow.Cyclone separator is widely used in chemical industry,pharmaceutical industry and environmental protection because of its simple structure and convenient maintenance.Pressure drop and separation efficiency are two important performance parameters of cyclone separator.Accurate prediction of pressure drop and separation efficiency is very important for the design and performance evaluation of cyclone separator.There are four methods to predict the performance of cyclone separator: theoretical analysis,statistical regression,computational fluid dynamics and artificial intelligence modeling.Among them,the theoretical analysis needs to make some assumptions that are not consistent with the actual situation,resulting in a large deviation from the actual results;Statistical regression requires high sample size;The prediction cost of hydrodynamics model is high and the running time is long.Therefore,it is necessary to establish an artificial intelligence model of cyclone separator quickly and accurately.In order to accurately and quickly obtain the performance model of cyclone separator,the performance model of cyclone separator was established based on extreme learning machine(ELM).However,the accuracy of extreme learning machine is too dependent on the number of hidden layer nodes,and its stability is poor in small sample modeling.In this paper,particle swarm optimization(PSO)is used to optimize the extreme learning machine,and the optimized algorithm is applied to the performance modeling of cyclone separator.The specific work of this paper is as follows:(1)In view of the fact that the prediction accuracy of limit learning machine is too dependent on the number of hidden layer nodes,it is easy to cause over fitting in small sample data.In this paper,the standard particle swarm optimization algorithm is introduced to optimize the connection weight and threshold between input layer and hidden layer of extreme learning machine.Firstly,the prediction accuracy and anti-disturbance capability of PSO-ELM are tested by Sinc function simulation test.The test results show that PSO-ELM has higher accuracy and network structure utilization ratio compared with the original ELM,which reduces the demand of ELM for hidden layer node points and avoids the network redundancy.Secondly,to verify the performance of PSO-ELM model,the pressure drop data in published literatures are extracted and the pressure drop of cyclone separator is modeled by PSO-ELM.Firstly,the data are divided into training set(80%)and prediction set(20%),and the normalization method is used to eliminate the influence of dimension on modeling accuracy.then,PSO-ELM model is trained by training set,and the connection weight and threshold of elm hidden layer are found by particle swarm optimization algorithm.finally,the performance of the model is tested by the prediction set.The mean square error of PSO-ELM model is 2.443e-04,correlation coefficient is 0.9984 and operation time is 15.74 s.Compared with the existing cyclone statistical model and artificial intelligence model,it has higher accuracy and generalization,and the running speed is about half faster.(2)In the model of cyclone separator separation efficiency,the model is often unstable because of the less experimental data and more factors affecting the separation efficiency.In order to improve the accuracy and stability of limit learning machine in the modeling of high-dimensional small sample data,the multi-objective particle swarm optimization(MOPSO)is used to find the relative optimal solution of the structural risk and root mean square error of the implicit layer connection weight matrix.The results show that the mean square error of the MOPSO-ELM model is 4.883e-04,the correlation coefficient is 0.9784 and the operation time is 36.74 s.Compared with the existing AI model,it has higher prediction accuracy and faster running speed. |