Study On MADS And CSA Algorithm For Pwr Loading Pattern Optimization

Reload design is directly related to the economy and safety of a Pressurized Water Reactor (PWR). Because of the large search space and large number of optimization variables, PWR LP search is a very complicated combinatorial problem. In this thesis, the performance of Mesh Adaptive Direct Search Algorithm (MADS) and Characteristic Statistic Algorithm (CSA) for their applications to the LP optimization problem is studied. MADS is one of the most noticeable research achievement in the optimization method research area and is reported to have a good performance when applied to general continuous variable optimization problem, but so far it has not been attempted to solve the LP optimization problem yet. While for CSA, although its application to the LP search problem has been tested and it is reported to have a good performance, there is still no comprehensive and quantitative assessment of its ability to find the global optimum and its efficiency to find the feasible solutions, due to the lack of an LP search benchmark problem. In this thesis, MADS and CSA are evaluated quantificationally by using PWR LP searching benchmark problems with all LPs enumerated and pre-evaluated, and with the introduction of parameters of search efficiency and quality of the searched optimal solution. In the end, refueling optimization problem of Qinshan nuclear plant is solved by using CSA with optimal solution analysed.Firstly, MADS is used to solve the LP optimization by transforming discrete variables into continuous variables and using Latin Hypercube Sampling (LHS). Studies on the feasibility of MADS method are carried out by use of the benchmark problem. Investigations show that when the unconstrained problem of maximizing the effective multiplication factor (Keff) or minimizing the peak pin power (PPP) is solved, MADS is able to find high quality solution effectively. While for the constrained optimization problem of maximizing the effective multiplication factor under PPP limited, the quality of the optimal solution and the search efficiency of MADS is not better than that of normal combinatorial optimization algorithms. Study finds that this is because the next generation LPs MADS generated from the current mesh center LP are too scattered and do not behave a certain regularity.Secondly, the quality of optimal solution and the search effienciey of CSA are evaluated quantificationally by applying the algorithm to the two refuelling optimization benchmark problems. While solving LP optimization problem, the assembly power is set to be the statistic characteristic and next generation LPs are generated by performing all the possible one-to-one fuel shuffling for a randomly generated LP. Investigations show that no matter dealing with the unconstrained problem of maximizing the effective multiplication factor or minimizing the peak pin power or the constrained problem of maximizing the effective multiplication factor under PPP limit, CSA possesses a good performance for both the quality of the optimal solution and the search efficiency. Studies on the mechanism of CSA LP optimization are also conducted, which reveals the stability of the algorithm.Finally, CSA is applied to Qinshan NPP loading optimization problem. The optimal solutions have good properties, including the effective multiplication factor, peak pin power and the power distribution. From the result it is concluded that CSA has a good application prospect.