Evolutionary Transfer Optimization Based Dynamic Multiobjective Optimization Algorithms

Dynamic multiobj ective optimization problems(DMOPs)are optimization problems with multiple conflicting optimization objective functions,and these objective functions change over time.DMOPs are challenging since the changing Pareto optimal set should be tracked quickly and accurately.Knowledge transfer has been proved to be promising to solve DMOPs.However,time-consuming and negative transfer are the issues that should be addressed when employing knowledge transfer.To solve the above issues,the paper proposes a series of dynamic optimization algorithms based on knowledge transfer.Firstly,an individual transfer-based algorithm is designed to improve the transfer speed,and an explicit diversity maintenance strategy is embedded into the algorithm for reducing the negative transfer caused by low population diversity.Further,a knee point imbalanced transfer-based algorithm is proposed,which only transfers high-quality individuals in the population to save computing resources and improve optimization efficiency.Finally,the dynamic optimization problem under the expensive optimization scenario is considered.The presented algorithm alleviates the imbalance and lack of samples under the dynamic expensive problem and transfers the optimization experience from the auxiliary task to accelerate the convergence.The proposed three algorithms combine the both advantages of transfer learning and evolutionary algorithms,which can greatly improve the performance of evolutionary algorithms in solving dynamic problems.Through a large number of experiments and analyses,the effectiveness of the proposed three algorithms can be verified.