| Traditional generalized regression neural network has good nonlinear approximation performance and learning speed,but the network has only one smoothing factor,there will be some deviation in the fitting effect for dynamic data,so this paper put forward generalized regression neural network based on multiple smoothing factors(IGRNN),in order to make the regression surface more suitable for local properties of data.At the same time,the multiple smoothing factors in IGRNN are selected by improving the sparrow search algorithm,and the ISSA-IGRNN integration algorithm is proposed to improve the prediction performance of IGRNN.The specific research work is as follows:(1)To solve the population diversity is not rich,easy to fall into local optimum and slow convergence speed,low search accuracy of these deficiencies in SSA algorithm,in this paper,reverse learning strategy,sine-cosine strategy,Levy flight strategy and-distribution mutation strategy are used to improve the SSA algorithm,proposed to the improved sparrow search algorithm(ISSA),The standard test function is used for simulation test.Experimental results show that ISSA algorithm has higher convergence accuracy and faster search speed,and can improve the ability to escape from local optimum.(2)In order to improve the prediction performance of generalized regression neural network,this paper proposes a generalized regression neural network based on multiple smoothing factors,uses ISSA algorithm to optimize its smoothing factor,finds the optimal value of network structural parameters,constructs ISSA-IGRNN integrated learning algorithm,and verifies it with standard UCI data set.The ISSA-IGRNN algorithm is compared with SVR algorithm and BPNN algorithm,the experimental results show that ISSA-IGRNN algorithm is effective and stable.ISSA-IGRNN integrated learning algorithm was used to predict air quality index.In the experimental analysis results,the RMSE and MAE of ISSA-IGRNN are lower than those of SVR and BPNN,indicating that the prediction accuracy of ISSA-IGRNN algorithm is higher. |
