Research On Anomaly Detection And Trend Forecasting Technology Of Small Supercharged Boiler Based On Deep Learning

With mature technology and high reliability,supercharged boiler is the first choice for large steam power plants.However,the boiler conditions change frequently and the parameters fluctuate greatly when the ship on mission,resulting in frequent accidents.In addition,due to the complex structure of the supercharged boiler and the huge system,it is difficult to conduct the boiler with health management with traditional methods,and it is urgent to develop a more efficient technology in this field.With the vigorous rise of big data analysis and artificial intelligence,and the improvement of industrial digitalization level,intelligent and digital fault prediction and health management technology provide solutions for improving the safety of supercharged boilers.Whereas,the current domestic research on the health management technology of power equipment and related deep learning theories is not deep enough and the space for development need to be expanded.In this paper,supercharged boiler is taking as the object to carry out research on abnormal detection and trend prediction of safety performance parameters while targeting two key tasks under the health management system – anomaly detection and trend prediction.Firstly,the characteristics of the operation data of the small supercharged boiler are analyzed and researched through the correlation analysis method.Secondly,feature engineering techniques for data noise reduction and feature selection are carried out according to the data characteristics,so as to prepare for improving the performance of the trend prediction model.Thirdly,denoising convolutional autoencoders are designed and trained to monitor and detect abnormal device states based on the increase load operation data of boiler.Finally,based on the feature engineering results,a deep learning model is designed and trained to achieve real-time trend prediction of superheated steam temperature,drum pressure and drum water level.The research results denote that: The anomaly detection model for small supercharged boilers established based on denoising convolutional autoencoder can output anomaly scores reflecting the degree of anomaly in real time,which is of great significance for timely discovery of equipment safety hazards;data difference and noise reduction processing technology.Data difference and noise reduction technology decrease the maximum error of the trend prediction models to less than 0.8% and expand the prediction range to 60 seconds,indicating that they are effective feature engineering technologies to improve the performance of the model.In terms of deep learning algorithm selection,convolutional neural network and long short-term memory network both can achieve higher prediction performance,but the debugging cost of the former in the training process is 1/5 of that of the latter,so it is a better alternative.In addition,for the trend prediction model,the naive prediction ratio proposed in this paper can effectively make up for the problem that the traditional evaluation system cannot judge the validity of the models.The research results have certain reference value for relevant researchers in the fields of data-driven modeling and intelligent power equipment health management.