Study On Soft Sensor Of Atmospheric Fine Particulate Matter Based On Artificial Neural Network

In recent years as China’s atmospheric fine particulate matter pollution is gettingworse, the real-time online monitoring of fine particulate matter concentrations becomesone imperative thing. As atmospheric fine particulate matter concentrations show strongnonlinear characteristics by the ambient effects, the traditional soft sensor technology isdifficult to make accurate monitoring. And the soft sensor technology based on neuralnetwork has become a hot topic in various fields because of its superiority. Therefore,this paper studies the establishments of soft sensor models based on BP and RBF neuralnetworks to do the soft sensor experiments for atmospheric fine particulate matterconcentrations, and provides strong data supports for the air pollution control.Firstly, this paper describes the pollution status, hazards and national monitoringtechnologies of the fine particulate matter. This accumulates a certain amount of prioriknowledge for the establishment of soft sensor models. Then make systemicpresentations for the structures and learning processes of BP and RBF neural networks,and determine the experimental feasibility for soft sensor models based on BP and RBFneural networks.Then this paper uses the improved correlation analysis to screen8preselectedimpact factors of fine particulate matter, and selects6greater impact factors as inputvectors of soft sensor models. On this basis uses experiences and heuristics comparingthe results to determine the parameters of BP and RBF network models, and constructsthe optimal soft sensor models.After that this paper selects another set of data to inspect the well-establishedmodels. The results show that BP network model and RBF network model can accuratelymeasure the fine particulate matter concentrations, and soft measurement accuracies canmeet the needs of practical works. Compared to both, RBF network model is simple andstable, and BP network model has higher accuracy.Finally, this paper use MATLAB GUI to write soft sensor system interface, makingthe whole system more clear, avoiding the inconvenience of reading programs, andfacilitating the users.