The Improved Stochastic Optimization Algorithm And Its Application

Stochastic optimization approach is a kind of intelligent algorithm and it’s different from the traditional optimization algorithm. It is usually the simulation of nature, which searches for the optimal solution through iterative random operator. By learning from the random exploratory, reinforcement learning is a simulation of bio-learning, which is inextricably linked with stochastic optimization algorithms.This thesis briefly introduces the origin and development of stochastic optimization algorithm and reinforcement learning. Global convergence conditions are given based on the generalized model of stochastic optimization algorithm. These conditions are keeping the better solutions and randomly initializing some more solutions.Combined with reinforcement learning thought, the reinforcement learning optimization algorithm is constructed. Inspired by the growth and reproduction of plants, the plant spread optimization algorithm is proposed. The simulation demonstrates the global convergence ability of these two algorithms. The comparison of the typical stochastic optimization algorithms shows that these two algorithms are effective.The reinforcement learning optimization algorithm and neural network are combined to handle the prediction problem. To solve the parameter identification problem, the stochastic optimization algorithms and traditional identification methods are combined with each other. The simulation shows the effectiveness of these methods in these areas.Inspired by the thought of reinforcement learning and optimization, a self-learning control algorithm based on relationship network is designed, and the convergence of the algorithm is discussed. The simulation shows the effectiveness and limitation of the algorithm.