Optimization And Its Application Of Stochastic Neural Networks Based On Evolutionary Computation Research

Evolutionary computation is a class of search method based on biological natural selection and natural genetic mechanisms, including genetic algorithms, evolution strategies and evolutionary planning algorithm. Evolutionary computation is an effective optimization technique to optimize weights and select the appropriate topology, and does not require a continuous objective function, with a good global search property.Random neural network is a class of powerful artificial neural network, playing an important role in the field of optimization and simulation learning. People often say there are two artificial neural networks: one is the random neural network which neurons using random activation function, the other is weighted by random access, that is, the Boltzmann machine which adding appropriate random noise in general artificial neural network.Although the random neural network powerful and widely used, the complex learning process and the slower convergence speed limited the further promotion and application of random neural network. How to optimize the learning process and increase the convergence speed of random neural network is becoming a study difficulty. Therefore, making a combination of evolutionary computation and random neural networks, using the randomness, global, implicit parallelism and robustness of evolution computation, gradually make up the learning process complexity, training slow and easy to fall into local minimum point and poor global search capability, optimize the random neural network learning process is known as an effective solution. Evolutionary Computation and Neural Networks combination make the "evolutionary artificial neural networks" development.In this paper, we make a study of evolutionary algorithm and random neural network combination situation, try to conducted optimize the random neural network by the evolutionary algorithms'powerful computing ability, simplify the random Neural network learning process through the evolutionary algorithm's competition for optimization and implicit parallel computing capabilities. Confirmed by experiments using the optimized random neural network solution can achieve optimum results. This paper conducted the following research:First, this paper describes the background and significance of writing, the subject of current research results and the main contents of this article.Second, introducing the principle of genetic algorithms, mathematical mechanism and the genetic algorithms solving process, making system described on the random neural network model, the main features and learning process. And the current evolutionary algorithm and random neural network research was reviewed.Third, for the Boltzmann machine learning characteristics, adding the genetic algorithm to Boltzmann machine learning process, using genetic algorithms to optimize the Boltzmann machine learning process, expected through the Boltzmann machine learning process probability and statistics didn't need to done, designed the genetic algorithm to optimize the learning process according to the Boltzmann machine features. And to apply this method to forecast the weather instance, achieved good results.Fourth, making research on random neural network reinforcement learning model and the weight updating the strategy, for random neural network reinforcement learning model weight update complexity and algorithm insensitive to environmental changes Limitations, proposed genetic algorithm applied to strengthen the process of learning random neural network to optimize random neural network weight update strategy approach. And designed the network'computing genetic algorithm which is based on the characteristics of random neural network.Although this article research on the evolutionary algorithm to optimize a lot of random neural network, and made a number of algorithms, but the algorithm theoretical basis and application needing to do further research.