Research On Fuzzy Neural Network Based On Improved PSO

With the incessant development of science and technology, the neuralnetwork with its strong learning ability generalization and parallel processingcapabilities is followed with interest by academia. Fuzzy Logic has been widelyused in nonlinear systems, such as pattern recognition, fault diagnosis, predictionof stock with its strong reasoning ability of the human brain. Many scientistspredict that the fuzzy neural network (FNN) which is combined by the fuzzylogic and neural network will become the core technology in the field ofintelligent. Genetic algorithms and BP algorithm based on gradient descentmethod is currently being used by the height, but the genetic algorithm needs toset more parameters; convergence rate of BP algorithm based on gradient descentis too slow, and the algorithm is easy to fall into local minima. Particle swarmoptimization (PSO) that is used in this article can overcome the shortcomings ofthe above two algorithms, and has widely used in the fields of functionoptimization, neural networks, fuzzy control systems and so on.Linear PSO algorithm based on decreasing inertia weight cannot reflect thesearch process of nonlinear optimization. While dynamic particle swarmalgorithm can achieve nonlinear search, but is easy to fall into local optimum.According to the characteristics of nonlinear and complex process, while fuzzyneural training its weights. The above-mentioned particle swarm optimizationcannot do well, it is necessary to improve the conventional PSO.Improved PSO algorithm includes two aspects in this paper: First,improving the speed formula of PSO. Using the average of all the particles in theformula to replace individual optimal solution to make sure the particles learnfrom the experience of other particles in the search process to decide their ownconduct. Second, improving the method for calculating the inertia factor canimprove the adaptability of particle swarm optimization. Experimental results show that convergence rate of improved particle swarm is faster than standardparticle swarm optimization in the process of nonlinear search, and the error issmaller. Then, fuzzy neural network is optimized by improved PSO algorithm.Comparing the performance of fuzzy neural network based on improved PSO andfuzzy neural network by method of function fitting to prove that the former hasbetter learning ability and generic ability. Finally, fuzzy neural network based onthe improved PSO which is applied to the evaluation of water quality, andcompared with the prediction results of fuzzy neural network. The results showthat the error of fuzzy neural network based on the improved PSO is smaller, andthe method has better results for solving this kind of nonlinear problems, forexample water quality assessment.