使用新预测模型的动态多目标优化算法
A Dynamic Multiobjective Optimization Algorithm with a New Prediction Model

针对实际应用中动态多目标优化算法对快速变化的最优解集跟踪能力不强的问题,提出了一种使用结合中心点预测值和垂直扰动分量的新预测模型的动态多目标优化算法。首先,计算变化前最优解集的中心点作为预测对象,改变了通常使用全部解进行预测的方式,提升了算法效率;其次,结合算法迭代的历史信息,选取位置、速度、加速度作为预测的状态向量,保证了算法对大多数情形下解集整体变化的跟踪预测能力;最后,为预测的新解添加了垂直于预测变化方向的超平面随机扰动,增强了解集的多样性,进而提升了算法收敛速度。实验结果表明,该算法在75%的测试函数集上的性能优于其他3种经典的动态多目标优化算法,其耗时较经典的基于卡尔曼滤波预测的动态多目标优化算法平均减少了39%。
A new dynamic multiobjective optimization algorithm is proposed to solve the problem that the existing dynamic multiobjective optimization algorithms have poor ability to track the rapid changing optimal solutions for practical applications, and the algorithm uses a new prediction model that combines the prediction value of central point and the vertical disturbance component. First, the central point of the optimal solution set before change is calculated as a prediction object, which changes the way that all solutions are usually used for prediction and improves the efficiency of the algorithm. Second, the history information of the algorithmic iterations is combined to select the location, velocity and acceleration as a state vector of prediction. The tracking ability for change in the solution set is ensured in most circumstances. At last, a hyperplane random disturbance that is perpendicular to the predicted direction of change is added to the predicted new solution to enhance the diversity of the knowledge set, and the convergence speed of the algorithm is improved. Experimental results show that the proposed algorithm is superior to the other 3 state？？of？？the？？art dynamic multiobjective evolutionary algorithms in 75% test cases, and the run time of the algorithm is 39% lower than those of dynamic multiobjective optimization algorithms based on Kalman filter