Design Of Real Time Comdition Monitoring System For Thermal Power Plant Based On Neural Network

Thermal power still occupies a leading position in China.It’s still a long way to improve the monitoring level for thermal power plant.The auxiliary equipment of fans is very important in thermal power plant.Its operating conditions are related to the safety and stability of the entire unit.With the development of intelligent algorithms and the improvement of the digital level of the power plant,it is possible to find more information from the operating data of the equipment.This thesis is mainly based on BP neural network.Firstly,the basic principle of this algorithm and the gravitational search algorithm(GSA)is introduced.By the combination of BP neural network and GSA,the performance is improved.The performance is intuitively demonstrated in the testing of standard datasets in the field of machine learning,which proves the effectiveness of the improved method of BP neural network.Secondly,the detail of air system is introduced,especially the induced draft fan.The research background of this subject is the actual operation status of a 660MW generating unit in Shanxi Province,data and information are provided by Vestore-SIS platform.Comparing the operating data of the fan under the same unit load in different periods,the abnormal situation of the induced draft fan was found.After studying the mechanism characteristics of the induced draft fan,supplemented by consulting the field operators and searching relevant paper,the possible causes are discussed.In industry,the performance of machinery is gradually declining.At present,most studies are based on fault diagnose by using fault data.However,in the actual scene,the fault data is hard to obtain and the amount of fault data cannot meet the requirements.These bring to a lot of hurdles.From another perspective,if we can monitor the performance index of the device in real time,when the index drops,it may attracts field operator’s attention,and timely maintenance can reduce the loss caused by the equipment failure.Therefore,a prediction model of air quantity based on GSA-BP neural network is proposed in this thesis.After training the neural network by normal operation data,this model can accurate predict the air quantity of the induced draft fan under normal conditions,the prediction error is not more than 3%,which meets the industrial error requirements.At the same time,when the abnormal condition of the induced draft fan occurs,there will be a deviation between the predicted air quantity and the real air quantity,which implies abundant equipment information.Drawing the error curve in the chart demonstrates the operation status of the equipment which provides more visual information to the field staff.