Soft-sensing Research For Water Ammonia Nitrogen Based On Dynamic RBF Neural Network

With the rapid development of social economy,industrialization and urbanization,the consumption of the water is increased sharply.The increase of sewage discharge exacerbates the shortage of fresh water resources and water pollution levels.The ammonia nitrogen is one of the main pollutants in water,which can be set as an important indicator of the water quality.The exceed ammonia nitrogen will led to water eutrophication and environmental pollution.Therefore,in order to reduce the harm done by excess ammonia nitrogen concentration,the measurement and study of ammonia nitrogen in the wastewater becomes crucial.In the actual sewage treatment plant,the existing measurement method of ammonia nitrogen is very cumbersome and the measurement accuracy is low.In addition,because the lag time of the measurement is long and the instrument maintenance cost is high,which makes it difficult to realize the on-line detection.Therefore,the ammonia-nitrogen soft-sensing model of self-organizing RBF neural network is proposed,and the effluent ammonia nitrogen intelligent soft measurement system development is completed,the water ammonia nitrogen soft measurement system interface visualization and accurately prediction are realized.The research content of this paper consists of following parts:?1?Design of RBF neural network based on the relative contribution index?RC-RBF?;In this paper,a design method to the structure and parameters of RBF neural network is proposed.Firstly,an improved adaptive Levenberg-Marquardt algorithm is used to train all the parameters for RBF neural network with fixed structure.Secondly,the relative contribution index is defined as the important level between the hidden layer nodes and the output layer neurons.The dynamic adjustment of the RBF network structure is realized and the convergence of the algorithm is proved by the neuron's parameter compensation.Finally,compared with other self-organization methods,the simulation results show that RC-RBF neural network ensures to achieve high prediction accuracy with a compact structure,and also improves the generalization ability and nonlinear modeling performance of the RBF neural network,which lays a solid foundation to the soft measurement model of water ammonia nitrogen.?2?The soft-sensing model of ammonia nitrogen based on RC-RBF neural network;The input variables of effluent NH₄⁺-N soft sensor model are chosen through analyzing the mechanism of ammonia nitrogen in water and the characteristics of the principal component analysis method.Then,water ammonia nitrogen soft measurement model is established base on dynamic RBF neural network based on relative contribution index and combined with the improved adaptive LM algorithm.And which ensures the rapid convergence of the network and the prediction performance of the algorithm,realizing the on-line prediction of ammonia nitrogen.Compared with other methods,the results show that the prediction model based on RC-RBF neural network avoids the structural identification of complex models and requires less prior knowledge,which can realize the on-line prediction of effluent ammonia nitrogen effectively.?3?Development of intelligent system for ammonia nitrogen soft measurement.The water ammonia nitrogen intelligent soft measurement system is designed and developed in this paper.It mainly includes user registration,landing module,sample data management module,the neural network model selection,neural network model online training module,real time prediction module and other modules.In the system design process,the SQL Server 2008 database is used to store user information and secondary variables data.The mixed programming technology is used by c#and Matlab.Dynamic RBF neural network in the interface is called,and the training and prediction results of water ammonia nitrogen are displayed and saved.The water ammonia nitrogen predicted output and display are achieved through the user management module,data processing,neural network model training and forecasting information transmission between modules,achieving the goal of soft measurement system interface visualization.