Research On Abnormal Point Detection Of Complex Industrial Process Modeling Data Based On Improved DBSCAN

The industry which contained complex industrial processes occupies an important position in the national economy and people’s lives. Some variables are difficult to be measured directly or online measured quickly under the existing technical conditions in the complex industrial processes. These variables can only be measured by other variables.Its quality requirements are guaranteed indirectly. The soft sensor technique solves such problems. The soft sensor in the industry is the modeling technique which is based on the data of the industrial process. The effectiveness of modeling data is related to the accuracy of the soft sensor model directly. However, it can cause abnormal results which called as the outlier in the actual measurement of the complex industrial processes. The reason is the mistakes when the measurement data being read or recorded, or the random interference of the testing instruments. It is very important for soft sensor modeling complex industrial processes to determine the outlier and remove it.In this context basis, in-depth study of the characteristics of the modeling data in the complex industrial process, a new method which combines the clustering algorithm with soft sensor modeling is proposed in this paper. A method which combination with soft sensor modeling is proposed for the specificity of the factors modeling data outlier detection in the complex industrial processes. This method uses K-means clustering algorithm to improve DBSCAN algorithm. Firstly, the method which improved algorithm with K-means algorithm makes up for the shortcomings of the traditional DBSCAN algorithms.Again, this method combines outlier detection with soft sensor modeling. It uses modeling error guide the outlier detection process. This will ensure the quality of outlier detection, while the soft sensor model will be completed. This method is used in the modeling data of Baosteel Group 300t LF temperature. The experimental results show good results.Aiming at the particularity of data outliers of soft sensor modeling in complex industrial processes, a new outlier detection method is proposed.Firstly, the method segments the original dataset. Then each piece of data will be detected for outlier with DBSCAN algorithm. Then it consolidates and fills the data set after excluding the outlier. Modeling errors will return the new data set to guide the selection of DBSCAN algorithm parameters. The soft sensor modeling errors are used as the guidance of outlier detection process and replace the traditional manual intervention in the clustering process. Meanwhile the outlier detection is completed as well as the more accurate model is established. Use this method to detect the outlier of Shanghai Baosteel Group power load modeling data. The simulation results show that the new method can guarantee modeling accuracy while detect the outlier successfully.