Abstract: Multilayer perceptron (MLP) are a method to deal with the classification problems, which can realize nonlinear high latitude classification and has good scalability. However, in the training process of traditional MLP, the quality of the classification results of MLP is closely related to the selection of parameters, and the parameters of traditional algorithms have many shortcomings. Using heuristic algorithm as its trainer is a scheme to overcome these shortcomings. Grey wolf optimizer (GWO) is a new meta-heuristic algorithm based on the predation behavior of grey wolf, which has been proved to be an algorithm with high level of exploration and development capability. In order to improve the exploration ability of the algorithm, the Cauchy variant is introduced into the grey wolf algorithm, and an improved Cauchy variant grey wolf optimizer (IGWO) is proposed. At the same time, the cosine convergence factor is added and the new update formula is used to ensure the robustness of the algorithm. Finally, the IGWO algorithm is used as the trainer of MLP to conduct the classification experiments on three different complexity classification problems and to test the performance of the trainer under diffe- rent structures. The results show that the proposed IGWO outperforms other algorithms in terms of classification accuracy, avoidance of falling into the local optimal, global convergence speed, and robustness.

Key words: grey wolf , optimizer algorithm;Cauchy variant operator;cosine convergence factor;multilayer perceptron;classification problem