Research on Shallow Neural Network Evolution Method Based on Improved Coyote Optimization Algorithm

As a kind of Neural network training method which is different from stochastic gradient descent, neuroevolution has become an important branch in the field of machine learning research. How to design a better evolutionary strategy and explore new type of neural network weights of the neuroevolutionary method is one of the hot issues of current research. In this paper, an evolutionary method of shallow layer neural network based on improved coyote optimization algorithm is proposed. By introducing the method of adaptive weighting factor and chaos perturbation, the traditional coyotes optimization algorithm is improved from two aspects of convergence speed and optimization ability. Secondly, the improved coyote optimization algorithm is adopted as the neural evolution strategy, which is integrated into the neural evolution process of the shallow neural network. Taking BP neural network as an example, a new updating method of BP neural network weight and threshold optimization is constructed. Finally, several representative data sets in UCI standard database are used to verify the effectiveness of the algorithm. Experimental results show that: The evolutionary strategy of improved coyote optimization algorithm gives full play to the global optimization ability of heuristic optimization algorithm in the parameter space of BP neural network, approximates the optimal solution quickly. The BP neural network that after neural evolution shows excellent performance in classification tasks, the feasibility and effectiveness of the improved coyote optimization algorithm as a new neural evolutionary strategy are fully verified. The research results in this paper enrich and expand the research content in the field of neuroevolution and provide an important reference for building a new machine learning toolbox with neuroevolution as the main body. © 2021, Science Press. All right reserved.