Research On GA-BP Algorithm Optimization And It's Application Of Soft-Sensing Of Sewage Parameters

Sewage treatment process, which is affected by water quality, water quantity, operational control and many other factors, has the characteristics of non-linear, time-varying, lag and so on, so it is difficult to establish accurate mathematical models. It is a kind of typical of the complex industrial processes. And a part of key water quality can't be on-line monitored, which make sewage water quality monitoring technology become a urgent issue to be resolved in the sewage treatment industry. As artificial neural networks have the excellent characteristics of non-linear approximation, parallel-distribution structure, better fault-tolerant and adaptive learning and the ability to summarize, especially applicable to the need to consider many factors and conditions at the same time and imprecise and ambiguous information processing. So the thesis researches sewage parameters soft-sensing methods based on the neural calculation.The main contents and methods of the thesis include:①first it analysis of their own characteristics of neural networks and genetic algorithms;②a new soft-sensing method—sewage parameters soft-sensing model based on GA-BP neural networks was raised on the basis of analysis several common soft-sensing methods;③The thesis established Model based on improved AGA BP neural networks. Reaching a soft-sensing modeling method based on artificial neural networks of improved adaptive genetic algorithms for the Shortcomings of the slow convergent speed, the choices of initial weights and thresholds are lack of basis, having great arbitrariness and being difficult to select the initial point with the overall situation. Genetic algorithms determines of BP neural network and a better search space instead of the randomly choices of general initial weights and thresholds, then trains and learns the network to convergent in this space, search out the optimal solutions or the approximate optimal solutions;④With the measured data of a sewage treatment plant for the training samples, for analyzing the sewage treatment process control parameters and water quality parameters, key parameters of aeration tanks SVI which can't be on-line measured being measured in the sewage treatment process, then comparing the measurement results with the results of BP networks and GABP networks.The results showed that the Forecast accuracy and generalization of the improved GA-BP model have a certain increase. The sewage parameters soft-sensing method based on the neural calculation not only contributes to achieving real-time control in the sewage treatment process, but also have a positive impact for other complex process control systems.