Based Artificial Plants Algorithm For Solving Stochastic Programming

Stochastic programming is one type of mathematical programmingcontaining stochastic data. The crucial difference between stochasticprogramming and deterministic programming is the introduction of randomvariables. In stochastic programming, the well known is the so-called stochasticexpected value model which optimizes some expected objectives under someconstraints. Generally, there are two problems for solving SEVM. Firstly, theselected algorithm influences the optimized objective significantly. Secondly,the computation of objection function is estimated with artificial neural network.Therefore, the estimated precision of artificial neural network is one importantfactor. In this paper, the artificial plant optimization algorithm (APOA) isemployed to solve stochastic expected value model, and the main contributionsare listed as follows:(1) Artificial plant optimization algorithm is a novel population-basedstochastic optimization algorithm inspired by plant growing process. In APOA,each individual represents one branch, all of them obtain the energy withphotosynthesis operator, and adjust the growing direction with phototropismoperator and using apical dominance operator to provide some minor revision.Due to the slow growing speed, APOA is easily to escaping from local optima.Therefore, APOA is applied to solve stochastic expected value model, andsimulation results show it is effective.(2) In the above mention application, the estimation for objective functionis the BP neural network, however, the learning speed of BP neural network issmall, and easily trapped into a local optimum. Therefore, we use APOA to trainartificial neural network, and two famous benchmarks: Cleveland heart diseaseclassification problem and forecasting of the sunspot number are selected tocompare, simulation results show APOA is better.(3) Based on the above work, a new version: APOA-for-APOA is designed to solve stochastic expected value model, in which the first APOA is used totrain the artificial neural network, while the last one to optimize the model.Simulation results show this new algorithm is more suit than the standard APOAand BP neural networks.