Research On Fuzzy Reasoning In MBR Membrane Flux Simulation

In today’s society, water resources are becoming increasingly scarce.Compared with the traditional wastewater treatment methods, MBR (membrane bioreactor) has many advantages, such as small size, high volume load, good effluent quality, and less residual sludge, so it is widely used in the treatment of domestic wastewater, organic wastewater and industrial wastewater, and become the research hotspot in the field of sewage treatment in twenty-first Century.And the biggest obstacle is the membrane fouling problem in the process of promoting the large-scale promotion and application of MBR. Therefore, it is very important to find out the main factors affecting membrane fouling and the technology of controlling membrane fouling.In this paper, the working principle of membrane bioreactor is introduced, and the main factor that hinders the development of MBR technology is membrane fouling, and the mechanism of membrane fouling is studied.The main factors affecting the membrane flux are found by using the fuzzy model of MBR, so that the input parameters can be found to improve the input parameters, so as to achieve better water quality.In view of the problem that the self-learning ability is poor, this paper introduces an improved gradient based real-time learning algorithm to construct a dynamic error transfer factor, which can not only improve the accuracy of the fuzzy model, improve the convergence speed and precision of the system, but also can solve the conflict between the convergence speed and the convergence process.And the average relative error of the predicted results is reduced, and the prediction of the membrane flux is more accurate.Because the structure identification and parameter optimization of the system are the two interactions in the process of data mining, and the phase of each other, so we introduce the adaptive method based on the normalized variance in the last chapter. This method can be used to optimize the two stages of the data mining process simultaneously and adaptively. Using this method, a simple and accurate model can be obtained, and it only takes a very little computational cost.