Thermal process abnormal data detection method based on DBSCAN-SVC

Abstract

The invention discloses a thermal process abnormal data detection method based on DBSCAN-SVC. The method comprises the following steps: A, inputting an EXCEL file containing original data, and carrying out the preprocessing of the original data; B, normalizing the data, segmenting a data set, and storing the segmented data as a training data set and a test data set; C, inputting a training data set, determining an optimal clustering number i, and performing clustering analysis by adopting the optimal clustering number i; D, performing binary classification on each category of data, and adjusting hyper-parameters to enable the training model to achieve the optimal performance; E, inputting a test set into the training model, and returning an accuracy rate and a thermal process abnormal data value position; the abnormal data value of the thermal process can be accurately and efficiently detected.

Description

technical field [0001] The invention relates to the technical field of anomaly detection in a thermal system of a power plant, in particular to a DBSCAN-SVC-based method for detecting anomaly data in a thermal process. Background technique [0002] In recent years, with the rapid development and progress of the power industry, thermal power units have continued to develop towards ultra-supercritical. Since the thermal system of the power plant is quite complex, in order to understand and control the operation status of each link in real time, many settings have been set in the thermal system. The measuring points are distributed in each link. The measured physical quantities include interrelated physical quantities such as temperature, pressure, flow rate, and valve opening. These measurement points provide massive data for the thermal system. However, due to sensor failure and other reasons, the massive data generated during the operation inevitably contains abnormal data....

AI Technical Summary

Problems solved by technology

This method does not pay attention to the detection of specific outliers, and the filtering will also have a certain impact on the original data

Statistical methods can find measurements with significantly larger deviations, but it is easy to misjudge dynamic processes

For the actual thermal process, the operating state of the thermal power unit is constantly switching between the steady state and the dynamic state, and the data distribution cannot be simply assumed to be a certain standard distribution, which makes the method applied to the abnormal data detection of the thermal process, and the detection results may be There will be a large deviation

The machine learning method can better fit the data, but the disadvantage is that the algorithm complexity is high and the physical meaning is not clear

Once the fitting is inaccurate, the detection results may have a large deviation

Images