| As an important part of municipal sewage treatment,total nitrogen detection is a necessary step to ensure the removal of total nitrogen and to monitor the total nitrogen content in effluent.In present,the detection of total nitrogen in most sewage treatment plants in Korea and China mainly depends on manual tests.It is difficult to adopt real-time control strategy of total nitrogen based on measurement,because the frequency of measurement and analysis is 1-2 times a week.As a result,the monitoring of total nitrogen in effluent is seriously inadequate and the total removal rate is not high,which affects the operation efficiency of sewage treatment plants and increases the risk of environmental hazards.Because total nitrogen is greatly influenced by environment and other sewage process elements,the content of total nitrogen in effluent also shows obvious wave.Hence,it is very difficult to implement real-time total nitrogen detection.Moreover,the current analytical methods and instruments extremely limited to detect total nitrogen in real-time.Therefore,this paper proposes a soft-computing method of total nitrogen based on fuzzy neural network(FNN)to predict total nitrogen,the detail as follows:First,the FNN model based on radial base function neural network(RBFNN)is designed.Second,partial least square(PLS)method is used to select variables correlated with total nitrogen.Then,the FNN model is trained using the selected variables as input values.Finally,the model is evaluated on the basis of actual experiment.At the same time,RBFNN has limitation to express the output of nonlinear systems because the connection value exists only between hidden layers and output layers.To solve this problem,the proposed FNN structure give an extension of the weight type to polynomial types.Also,the fuzzy C-mean(FCM)algorithm is also used to design RBFNN kernel functions.Experimental results show that the proposed soft-computing method can predict the total nitrogen rapidly. |
PDF Full Text Request