Controlling strategies of UASBAF system based on BP neural network

Based on the prototype experiment of UASBAF as the methanogenic phase of two-phase anaerobic treating herb wastewater, adopting a fast convergence backpropagation algorithm with momentum and adaptive learning rate, an artificial neural network (ANN) model was established. Besides, comparison was conducted using method of partitioning connection weights to investigate influence of key factors (pH, COD, ALK, HRT) to the performance of the reactor. As a result, pH value is found to be the most important controlling parameter to maintain the system performance especially under high loading conditions. On the ground of model built, by fractionally fixing affecting factors tridimensional figures with two conjoint affecting factors were plotted to analyze ways of input parameters affecting the reactor visually and qualitatively. And ultimately a series of strategies were proposed to optimize the working condition of the system, which broke through the traditional predicting role of ANN and provided an effective way of exploiting advantage of ANN in controlling reactors.