Water Quality Assessment Based On Improved PSO - BP Neural Network Algorithm

For the governance and protection of water environment, scientific assessment method is needed to classify water environment. In our country, the single-factor assessment is adopted, which carries the principle of "One-vote veto ". The single-factor evaluation seems simple and easy to operate, while it cannot fully utilize the data and its result of evaluation tends to be pessimistic.In this paper, based on the monitoring data of water quality in Yong’an River of the area Erhai, the water quality has been assessed comprehensively by using Principal Component Analysis, BP Neural Network and PSO-BP Algorithm. The evaluation function built by PCA method does not have an evidently physical significance. And PCA method cannot focus on the index which has a greater impact on contaminants. Then BP method is adopted with a better nonlinear mapping and self-learning ability. Its result is more targeted and has a clearer physical meaning when assessing none-linear and complex water environment. However, BP method has its weakness such as slow convergence, weak generalization ability, being easy to fall into local extremum and sensitive to network initialization parameters. This paper uses the Particle Swarm Optimization (PSO) Algorithm to optimize the network parameters of BP Neural Network. PSO-BP Algorithm has the advantages of strong global search ability, it helped to optimize the connection parameters of BP Method and solved the problems that BP method is sensitive to network parameters initialization and easy to fall into local extremum. At the same time, PSO algorithm is easy to implement with a simple structure, and easy to combine with other algorithms. This algorithm uses parallel computing and has not only fast computing speed but also high resource utilization. With the combination of the two algorithms, the convergence precision and generalization ability of the BP Neural Network Algorithm is improved. Meanwhile, in the process of optimizing the BP neural network, new variables and the iterative process is involved, and the running time of the algorithm increases. In conclusion, the improvement of the inertia weight attenuation function in PSO algorithm has increased the convergence rate of the algorithm and saved the running time, under the condition of guaranteeing the convergence accuracy of the algorithm.The result is validated by experiment simulation we made.