The Study Based On Neural Network Of Water Quality Prediction In Sewage Disposal

Sewage disposal is a complex process, which is often affected by many factors such as the quality and volume of water supply, equipments and processes. Water quality changes have the non-linear feature and nondeterministic features. The traditional water-quality prediction, ASM Model, has not produced good results in application because it involved many reaction process and parameters, which cannot be determined by sophisticated methods for measuring . Artificial Neural Network has self-adaptive, self-learning and fault-tolerant capacity as well as large-scale concurrent operation, which is particularly applicable for inaccurate and fuzzy information processing with many factors and conditions involved. Therefore, the application of Artificial Neural Network provides a new approach to the prediction of yielding water quality in sewage disposal with unique advantages.The paper adopts MATLAB 6.5 as computational platform and BP neural network configuration which consists of input level, implied level and output level. Six data of water supply quality is determined after process analysis, which are water volume, PH, temperature, COD, sulfide, MLSS. MLSS refers to input nerve cell whereas COD refers to output nerve cell. Based on monitoring data of sewage disposal station, BP network is analyzed in terms of the factors affecting learning efficiency and prediction accuracy. Furthermore, after its application to COD prediction of the station is optimized with regard to the amount of neural cell in implied level, training frequency, excitation function in implied level, learning sample numbers, BP Artificial Neural Network Model is introduced to the prediction of yielding water quality in sewage disposal. After removal of four points with discrete tendency , monitoring data is divided into six types, among which 2/3 is selected as learning sample and the rest as check sample.The result indicates that relative errors vary from 9.2% to 0.9% between the predication results of BP model and the monitoring data of COD. Mean value of relative errors is 3.4% and discrimination is up to 92.3%. Such a high accuracy can meet the requirements of the predication of water quality predication..