Thermal power plant primary fan fault early warning method based on auto-encoder

Abstract

The invention discloses a thermal power plant primary fan fault early warning method based on an auto-encoder, and the method comprises the following steps: S1, selecting data of a fan during normal operation in a whole data set, and dividing the data into a training set, a verification set and a test set; S2, performing normalization processing on the measurement value of each measurement point according to the maximum values and the minimum values of the training set, the verification set and the test set; S3, establishing an auto-encoder model and carrying out model training; S4, calculating a test residual error between an output value and an input value of the auto-encoder model of the test set; S5, performing rolling calculation of a mean value and a standard deviation on the test set data residual error, and determining a threshold value; and S6, calculating a residual error of the fan operation real-time data by utilizing the auto-encoder model, comparing the residual error with a threshold value to determine an equipment state, and carrying out early warning according to the equipment state.

Description

technical field [0001] The invention relates to an autoencoder-based early warning method for a primary fan failure in a thermal power plant, which belongs to the technical field of electric power equipment. Background technique [0002] The primary fan is an important auxiliary equipment of the thermal power plant, which provides guarantee for the stable operation of the boiler combustion system by outputting the primary air with a certain flow rate and pressure. Due to the complex environment in which the primary fan is located, this will seriously affect the safe and reliable operation of the equipment. Once the equipment fails, the host production line will be shut down, resulting in serious economic losses and even personal safety. [0003] At present, due to the complexity of the power plant system, the early warning of primary wind turbine faults is mainly implemented using data-driven methods, including multivariate statistical methods, neural network methods, and s...

AI Technical Summary

Problems solved by technology

However, the linear correlation between the vectors in the memory matrix will cause the model to fail to obtain the optimal solution. Therefore, it is necessary to use principal component analysis (PCA) to eliminate the linear relationship, and to design a nonlinear operator to ensure the reversibility of the matrix.

Here, because PCA cannot identify the nonlinear relationship between variables, and there are certain difficulties in the design of nonlinear operators, this makes it difficult for MSET to obtain robust results.

[0004] The indicators that affect the working conditions of primary wind turbines in thermal power plants are complex and diverse. In the face of high-dimensional data, the traditional multivariate state estimation technique (MSET) method needs to solve the problem of feature compression and matrix reversibility

In addition, as the amount of data increases, the calculation speed of traditional algorithms will also decrease significantly, which will affect the actual working effect of the model.

Images