Studies On Multi-stochastic Variable Programming Methods Based On Scenario Analysis

At present,it is inevitable that there are various kinds of random variables in many fields,which makes stochastic programming have a broad application prospect in decision-making,optimization and scheduling problems.However,random variables have uncertain changes and volatility,especially for multi-stochastic variable programming even exists the interaction between multiple random variables,which all makes the reasonable uncertainty expression method of random variable become a common problem faced by stochastic programming applied in many fields.As the mainstream method to solve stochastic programming problems,scenario analysis method has the unique superiority in the uncertainty representation of random variables.However,it is also faced with the scenario generation problem that takes account of the correlation among multiple random variables and the scenario reduction problem caused by the excessive size of the initial scenario.This dissertation studies these two aspects with a view to scenario analysis method can be better applied to multi-stochastic programming problem,in order to make the decision results better.In view of generating correlated multi-stochastic variable scenario,a Tabu-SALHS based method is proposed.This method firstly gets the scenarios of each random variable based on Latin Hypercube Sampling,and then considers correlation control problem as a combinatorial optimization problem.Combined with the advantages of Tabu algorithm and SA algorithm,the entire optimization process is designed as two parts of inner and outer loop,and finally multi-stochastic variable scenarios which satisfying the target correlation can be obtained.The simulation results show that the proposed method is more accurate in controlling correlation,and it is proved to be effective and feasible in multistochastic variable programming problems.Aiming at solving scenario reduction problem,a new scenario reduction method based on the essential characteristics of multiple random variable is proposed,which also considers the correlation loss before and after reduction.The concept of correlation loss is proposed for the case of severe correlation deviation after multi-stochastic variable scenario reduction.Then,the concept of probability similarity is proposed for high dimentional random vector scenarios where the distances between datas are almost the same.Finally,a new integrated scenario reduction method is designed.The simulation results show that the proposed method can ensure the consistency of the correlation before and after scenario reduction,and the stability of the reduced scenarios can be effectively improved.The objective function can still achieve high precision even with fewer scenarios.Besides,our method can effectively improve the solving efficiency of stochastic programming problems.Finally,an economic operation optimization model of microgrid containing wind power generation and photovoltaic generation was built.The simulation results show that our proposed multi-stochastic variable programming methods can obtain lower microgrid operating cost and make the decision results better under the premise of guaranteeing the micro-grid power supply reliability.The methods proposed in this dissertation have the feasibility and validity in the actual problem.