A Kalman filter-based prediction strategy for multiobjective multitasking optimization

Multiobjective multitasking optimization (MO-MTO) can solve multiple optimization tasks simultaneously through knowledge transfer across tasks. However, how to design an efficient knowledge transfer method is the main challenge. Keeping this in mind, this paper proposes an evolutionary multitasking algorithm based on Kalman filter prediction strategy. Specifically, the incremental support vector machine classifier is used to find valuable solutions. Moreover, the Kalman filter prediction strategy is designed to utilize valuable solutions and historical evolutionary information to estimate the predictive solutions. Finally, the scoring scheme is constructed to adaptively select valuable solutions and predictive solutions as transfer knowledge. Experimental results on three MO-MTO test suites demonstrate that the proposed algorithm can achieve competitive performance.