Linear Distribution-based SHADE with Variable Population Size

In adaptive DE algorithms, the parameter p is used to strike a balance between the algorithm’s exploration and exploitation abilities. In this paper, an effective method for generating a suitable p value is proposed. The probability distribution function (PDF) of p is viewed as a linear function. At the start of the evolutionary process, the PDF of p is a linearly increasing function. Hence, there is a high probability of p having a high value. With successive generations, the PDF of p changes to a linearly decreasing function, and there is a high probability of p having a smaller value. Furthermore, a method for allowing a variable archive size is also proposed. After each generation, the archive size for the next generation is computed. If the archive size has to be reduced, the worst-ranking individuals are deleted from the archive. The accuracy of the proposed method is computationally compared with those of eight other advanced DE variants using benchmark functions from the IEEE Congress on Evolutionary Computation 2017. The results show that the proposed algorithm achieves the best performance among all the variants for most of the benchmark functions in the case of 30-dimensional problems.