Method for predicting time sequence of multiple measuring points on gas turbine based on steady-state characteristic composition

Abstract

The invention discloses a steady-state feature composition-based time sequence prediction method for multiple measuring points on a gas turbine, which comprises the following steps of: extracting steady-state features of an input sample in an end-to-end manner, constructing an incidence relation network, guiding incidence relation composition through steady-state loss, and establishing a space-time neural network sequence prediction model based on the steady-state feature composition; and extracting steady-state features of the system under a fixed working condition by adopting the space-time neural network sequence prediction model, performing sequence prediction by using space-time convolution, realizing adaptive steady-state feature learning, and finally obtaining a gas turbine multi-sensor time sequence prediction curve by utilizing the prediction model. In time sequence prediction, instantaneous features of a sequence are introduced into a composition module, and dynamic construction of an association network is carried out by inputting the sequence. Compared with the prior art, the method has a better prediction effect, can extract the steady-state characteristics of the system under a fixed working condition, improves the time sequence prediction effect, and is used for analyzing the abnormality of the associated network to check the working condition of the system.

Description

technical field [0001] The invention belongs to the design and application of a neural network model, in particular to a space-time convolutional network model of a steady-state feature composition and the use of the model to realize time series prediction of various measuring points on a gas turbine. Background technique [0002] The problem of multivariate time series forecasting has been a hot topic in statistics and deep learning research. Uncertainty is divided into model uncertainty and data uncertainty. The traditional ARIMA (Auto-regressive Integrated Moving Average model) model can handle non-stationary sequences. The main idea is to first differentiate the non-stationary sequence to make it a stationary sequence, and then use the ARMA model to fit the sequence after the difference. Linear models applied to univariate, homoscedastic cases. ARIMA requires that the time series have univariate, homoscedasticity, and the series follow linear regression, but these assu...

AI Technical Summary

Problems solved by technology

Although this part of the composition method gets rid of the limitations of the traditional composition method, it still does not take into account that in a stable system, there will be some relationships between nodes that can remain stable at any time

Images