Study And Application On The Intelligent Algorithm For A Kind Of Stochastic Dynamic Programming

Random factors are common in our daily life by which it is not convinced to solve the problems effectively. Nowadays, the optimization problems under random conditions are solved by the way which is based on the view of expected value or the chance measure or the other methods, the hybrid algorithm which is combined with stochastic simulation, neural networks and intelligent algorithm is used to find the best value of problems.On the basis of predecessors, in this paper, according to the characteristics of the dynamic programming model under random conditions, we design intelligent algorithm for a kind of stochastic dynamic programming to solve the multi-stage problem. At each stage, the hybrid algorithm of stochastic expected value which is based on the improved differential evolution algorithm is used to compute the best expected value. The main work is as follows:(1) To improve the differential evolution algorithm. The algorithm based on the multi-population is proposed. In the algorithm, the two populations will not overlap by the way the peak of each one can not be the same as other one and we increase the diversity of the population to avoid falling into local optimal solution, the new individuals are distributed inside ball body in the centre of selected individual.(2) The hybrid algorithm to solve stochastic expectation value model. Stochastic simulation is used to compute the expectations, the RBF network is used to simulate the expected value function, and the improved differential algorithm is used to get the best results. Finally, the comparison examples show that the hybrid algorithm can solve the problems effectively.(3) The intelligent algorithm for a kind of stochastic dynamic programming. The algorithm solves the problem by dividing the multi-stage decision into a single stage problem. In a single stage, hybrid algorithm to solve stochastic expectation value model is used to get the optimal value of the excepted value; the set of optimal value is used to train the RBF network. Finally, reverse the neural network of each stage to get the sequence of optimal value.(4) The intelligent algorithm for a kind of stochastic dynamic programming applies to the optimal scheduling of a single reservoir. The decision set will get by each stage.