Thermal Process Alarm Data Filtering Based On Hybrid Model

With the continuous expansion of the share of new energy power generation,thermal power units are participating in grid peak shaving more and more frequently,and the operating status of the units often fluctuates,resulting in the generation of a large number of thermal alarm data.If these data cannot be accurately identified and processed,it will directly affect the stable operation of subsequent units.The large amount of thermal process historical data generated during the operation of a thermal power unit contains a wealth of information on the operating status of the unit,which provides an important information basis for the early warning and fault analysis and diagnosis of the unit.At present,most of the traditional thermal process data analysis is simple operations such as data summary and unit operation trend display.There has never been a filter analysis method that is more accurate for the thermal process alarm data.Most of the staff can only rough the alarm data based on experience.Screening and analysis cannot deeply understand the various laws behind the data.Therefore,using the hybrid model to accurately and efficiently filter the alarm data of thermal power plants and identify abnormal alarm data is of great significance to the safe and economic operation of thermal power plants.This paper conducts research based on the above status quo,the main research work is as follows:(1)Aiming at the problem that the data of thermal power plants are mixed with each other under different working conditions,which affects the accuracy of data filtering,this paper proposes a thermal process alarm data filtering algorithm based on the AHC-GP hybrid model.The class algorithm clusters the thermal process data,distinguishes different working conditions,and then selects the Gaussian process model to filter the alarm data.The validity of the proposed algorithm is verified by using 5000 sets of data including the actual main steam temperature and main steam pressure of a1000 MW unit in a power plant to verify the effectiveness of the proposed algorithm.The experimental results show that the algorithm has good data filtering accuracy.At the same time,the algorithm is compared with traditional algorithms such as Gaussian process model and support vector machine,K-nearest neighbor,and it is concluded that the AHC-GP hybrid model has better data effects.(2)Thermal process data has the characteristics of non-linearity,strong coupling,high dimensionality,and time-varying characteristics.Simple models are difficult to fit.At the same time,labeling thermal alarm data is very time-consuming and cumbersome.Combining the above problems,this paper proposes an unsupervised thermal process alarm data filtering algorithm based on Gaussian mixture model.At the same time,it also proposes the concept of mixed probability density,which combines the mixed probability density with the Gaussian mixture model for accurate and efficient Alarm data filtering.The algorithm first uses the Gaussian mixture model to cluster the preprocessed data,and then combines the clustering results to calculate the mixed probability density of each point in the data set,and finally filters according to the set threshold to determine the location of the alarm data.It is verified by using 10,000 sets of data such as the actual load,main steam temperature and main steam pressure of a1000 MW unit in a power plant as the experimental data set.The results show that the algorithm proposed in this paper can accurately locate the alarm data position under various error conditions,and has a high detection accuracy.At the same time,compared with algorithms such as hierarchical clustering and K-Means,the algorithm proposed in this paper has better performance in accuracy,false detection rate and missed detection rate.