Preferred-learning-based Differential Evolution Algorithm And Application To Beneficiation Process

Differential evolution algorithm is an excellent optimized algorithm,and have excellent global optimized ability.Besides,the structure is simple and easy to implement,differential evolution algorithm draw attention to algorithm researcher in the whole world.Of course,DE can be applied to many different areas,and performs well in different subjects.However,compared to other algorithms,DE also have problem of getting solution is slow and have tendency to be caught into local optimum.On account of solving these problems,supported by the National Natural Science Foundation project "closed-loop optimal decision-making approach of technology index for complex industrial processes under dynamic environment",mutation operation is improved,the main research work include:1)Come up with a DE algorithm with preferred-learning—PL_DE.The algorithm introduces the thought of learning like better individuals,use the information of better individuals selectively,overcoming low efficiency and blindness in the process of evolving.Meantime,for the purpose of prove the rationality and effectiveness of the algorithm,we compared the proposed algorithm with many basic DE with different mutation strategies make use of 12 different type test functions in the same circumstance,the result can prove that,preferred-learning strategy can really improve convergence rate and accuracy of finding the best solution.2)Aim at dealing with traditional CBR's weakness:low accuracy,we propose a new method based on distribution of input conditions.We describe the concept of description,and the model of this method.Meanwhile,we apply the thought of distribution to the beneficiation process,build case library and compare the result of 2 different methods,in accuracy,error and objective values,by this way to prove the proposed method is effective.3)Apply PL_DE algorithm to beneficiation process indexes decision-making,and get operational indexes of the whole process,then input these indexes into function model build base on beneficiation process,compare with other results get by different means.We can testify that PL_DE really can get indexes fast and accurately.4)Introduce multi-population strategy to PL_DE,and let preferred-learning mutation strategy work in one of the sub-populations,evolving base on corresponding method.Besides,we chose 11 different test functions to test the performance of the algorithm in the identical environment to prove the effectiveness of the algorithm.