Hybrid FCM Learning Algorithm Based on Particle Swarm Optimization and Gradient Descent Algorithm
Jun Chen,Yue Zhang,Xudong Gao
DOI: https://doi.org/10.1109/icarcv50220.2020.9305521
2020-01-01
Abstract:The learning algorithm of fuzzy cognitive map(FCM) is mainly based on expert intervention or effective historical data to learn the weight value in weight matrix W. With the increase of the number of nodes, the number of weights need to be learned will also increase square, the performance of the learning algorithm will decline sharply. Therefore, it is an important task to design a high-performance learning algorithm which depends on expert knowledge as little as possible. A single learning algorithm has some limitations, such as easy local optimization, slow learning rate and so on. Among them, particle swarm optimization algorithm has strong global optimization ability, which can quickly optimize the weight matrix to a certain accuracy, but the follow-up optimization time is too long. For the gradient descent algorithm, when the initial value is chosen properly, its can approach the global optimal solution quickly. In this paper, we design particle swarm optimization (PSO) and gradient descent algorithm (GDA) to learn FCM respectively. According to the characteristics of the two algorithms, a hybrid algorithm(HA) is proposed: first, the PSO algorithm is used to optimize weight matrix to a certain accuracy, then the gradient descent method is added, which makes up for the disadvantage that the error of single PSO algorithm decreases slowly in the later stage. Experimental results show that the hybrid algorithm proposed in this paper has a significant improvement in both accuracy and speed.