Research Of Outlier Detection For Process Control System

With the rapid development of the process control theoretical research, the modeling methods for the controlled object of this process are more and more, but most of the modeling methods often rely on data collected at the scene, requiring high accuracy of the modeling data, so the accurate data is a prerequisite for an establishment of accurate model. But as we all know, there are many uncertainties in the industrial field, such as noise, instrument failure and on-site accidents etc, which will result in the inaccurate data collected at the scene, so outlier detection for the process control system is particularly important.In this paper, based on lots of related literatures, some common methods of outlier detection at home and abroad summed up, two accurate outlier detection methods is brought forward. The first one is based on the BP network. Compared with the traditional detection methods, the method can not rely on statistical assumptions of the underlying data, and deal with noise data better, and automatically adjust the output of the weight of the measure, then establish a non-linear of mapping exactly reflect the inherent law of the control object. The second one is based on the wavelet transform. The wavelet transform is the localization analysis of time and frequency, and it can multi-scale refine the signal by calculating of flex and transition. It can extract feature information of the signal effectively, and it can find the location and the amplitude of outliers. By experiment, compared with that based on the BP network, the method based on wavelet transform has simple and fast operation, and higher checkout precision, and can be used to detect on line.In this paper, outlier detection based on wavelet-transform is applied to electric arc furnace control system, combined with the feather of output data in process control system, contacting the output with the input, and considering the coupling of three-phase AC electric arc furnace. This outlier detection provides reference System Parameter Identification (SPI), Control Strategy Development, etc.Data processing program is mainly written in the MATLAB environment. First the output data is processed preliminarily, and then using the neural network toolbox, BP network is designed and trained. Besides, wavelet-transform program is also written in the environment. Finally the simulation is made. The experimental results show that the method based on wavelet transform surpass that based on BP network, so it has broad application prospects in the field of process control.