Particle swarm optimization performance improvement using deep learning techniques

Deep learning is widely used to automate processes, improve performance, detect patterns, and solve problems. Thus, applications of deep learning are limitless. Particle swarm optimization is a computational method that optimizes a problem by trying to improve a candidate solution. Although many researchers proposed particle swarm optimization variants, each variant is unique and superior to the existing ones. Among them, inertia weight-based particle swarm optimization has its own identity. By adjusting the inertia weight, the performance of the swarm can be improved. This paper proposes two new particle swarm optimization models using the convolutional neural network and long short-term memory to predict the inertia weight in moving the swarm for improving the swarm performance. The performance of the two new inertia weight models is compared in terms of mean absolute error and standard deviation, with the existing inertia weight based particle swarm optimizations like constant inertia weight, random inertia weight, and linearly decreasing inertia weight particle swarm optimizations. Experiments are conducted with swarm sizes 50, 75, and 100 with dimensions 10, 15, and 25 using the five most commonly used benchmark functions. The results show that the new models have significant performance gain over existing constant, random and linearly decreasing inertia weight particle swarm optimization models.