Soft Measuring Technique Research Applied In Wastewater BOD Based On Neural Computing

Due to the bad operating conditions during wastewater treatment process which ischaracterized of severe random disturbance, strong nonlinearity, time-variantproperties, and severe lag, it is almost impossible to model the process throughconventional mechanism approaches. Moreover, in monitoring and controllingwastewater treatment processes, on-line information of some essential wastewaterparameters is accessible. In order to acquire the real time data and contribute to realtime closed loop control, this thesis researches soft-measuring approaches based onneural computing to predict wastewater parameters. This thesis consists of threemodels as follows:First, establish a soft-measuring model which is based on BP neural network. Throughcomparison among standard BP algorithm, momentum improved BP, self-adaptivelearning rate BP and both improved BP algorithm, the self-adaptive learning rate andmomentum grads descend algorithm is better than the other three algorithms.Second, put forward a new soft-measuring approach which is based on GABP neuralnetwork. The algorithm GABP applies genetic algorithm to optimize the weights ofneural network. In this algorithm, the self-adaptive learning rate and momentum gradsdescend algorithm is applied to train the neural network so that the fitness function iscalculated and the weights of the neural network are optimized. After the simulationcomparison among standard BP algorithm, improved BP algorithm and GABPalgorithm, it can be concluded that GABP algorithm has improved a lot inconvergence speed, convergence precision and stability.Third, put forward a soft-measuring approach to on-line predict BOD based on PCAtime-delay neural network. This approach consists of three aspects such as PCA,time-delay neural network and model updating. This thesis applies principlecomponents analysis (PCA) to reduce the dimension of process variables and realize IIAbstractthe choiceness of assistant variables and reduce the input number of neural network. Time-delay neural network is established to develop a dynamic neural networkmodel to map the changes during the retention time by means of importing delay cells.On-line updating of soft-measuring model applies short-term learning and long-termlearning approaches which improve the model's adaptive ability to dynamic processchanges, therefore, advance the forecast precision of soft-measuring model. Thismodel, which is of good real-time property, good stability, high precision and easyupdating, can be applied to on-line predict wastewater BOD.The soft-measuring approach based on neural computing not only is helpful to realizereal time closed loop control during wastewater treatment, but is also active tooptimize control during other complex processes.