Self-adaptive Differential Evolution Algorithm Based On Real-time Multi-strategy And Reverse Learning

Differential Evolution(DE), proposed by Storn and Price, is a classic branch of the evolutionary algorithm, with the advantages of broad applicability, simple manipulation, and effective screening. It has excellent performance in numerical optimization problems of the real numbers. However, its single evolutionary strategy could not be flexible enough to address different types of issues. Therefore, scholars have come up with differential evolution algorithm based on multi-strategy and multi-parameter, dynamically selecting strategies for each individual of the population in the evolutionary process. This algorithm has a relatively good speed and optimizing performance. Nevertheless, it may lead to errors and problem of inadequate sampling when individual-level evaluation criteria is used. What's more, in the algorithm, the integrated revolutionary process was replaced by manual subprocess. Although, the change improves its flexibility, it is restrained from timely adjustment to the feedback, and it has no extra measures on the local optimal response.Aiming at the defects of Differential Evolution Algorithm Based on Multi-strategy and Multi-parameter, the present research puts forward the Self-adaptive Differential Evolutionary Algorithm Based on Real-time Multi-strategy and Reverse-learning, and explores its performance and function. Specific work is as follows,With regard to the deficiencies like inadequate sampling, errors in evaluation criteria and weakness in real-time algorithm in multi-strategy and multi-parameter based Differential Evolution Algorithm, the researcher proposes a new real-time and multi-strategy mechanism. This strategy focuses on using the policy strategy pool consisting of multi strategies to select evolution strategy for the entire population,, evaluating the evolution effect based on the trial vector instead of the ancestor vector survival rate, dynamically adjusting the selecting rate of the current strategy according to the criteria. This mechanism accelerates the convergence of algorithm when dealing with single modal problems, reaching the global optimal earlier, and it has a better accuracy in solving the multi-modal problems.Based on the real-time multi-strategy mechanism, to overcome the shortcoming of MspDE, the author introduces the reverse operation by Rahnamayan[49], advances new trigger conditions based on the original ones. The main idea includes: on the original single trigger mechanism adding local optimal state judgement, when the population of contemporary best value is consistent with best value at a certain algebra, triggered mandatory reverse action. It increases the strength of the reverse action when the population entrapped in a local optimum state, and it improves the global search capability of the algorithm in handling the multimodal problems, making it jump out of the early Statess in real-time strategy more easily and frequently, and thus achieving better search accuracy.