Method for predicting fault of electromechanical device based on combined prediction model

Abstract

The invention relates to a method for predicting fault of an electromechanical device based on a combined prediction model. The method comprises the steps of: (1) acquiring monitoring data of an operating state of an industrial filed from an industrial filed monitoring system, and extracting fault sensitive characterization factors to be taken as a prediction time sequence; (2) primarily selecting uniterm predicting models, and predicting various signal data in the prediction time sequence respectively at one prediction interval by utilizing the primarily selected uniterm predicting models; (3) determining a proper prediction accuracy evaluation index in the conventional evaluation index according to experiment, so as to carry out containment detection on the primarily selected uniterm predicting models for determining that whether the models are selected in a combined prediction model bank; (4) calculating comentropy value of the uniterm prediction models at the ith moment for the prediction time sequences, and determining weight coefficient of the uniterm prediction models at the ith moment; and (5) predicting the (i+1)th moment according to the weight coefficient omega j(i) by utilizing the combined prediction value fj(i+1) of the jth prediction method at the (i+1)th moment, so as to obtain the prediction value at the (i+1)th moment. The method is widely applied in various large electromechanical devices.

Description

technical field [0001] The invention relates to a method for predicting failure of electromechanical equipment, in particular to a method for predicting failure of electromechanical equipment based on a combined prediction model. Background technique [0002] Most of the equipment for fault diagnosis and prediction on industrial sites are large-scale units with complex structures and harsh operating environments, and the factors affecting their operation are intricate. For a complex fault predictability problem, it is first necessary to establish a suitable prediction model and prediction method. Since different forecasting models and methods have their own advantages, disadvantages and scope of application, the data used by each forecasting method is also different, and different data provide a wealth of information. How to choose different single-item forecasting models and effectively combine them It has become a difficult problem to be solved urgently. [0003] Since m...

AI Technical Summary

Problems solved by technology

[0003] Since most of the existing combined forecasting models are models with fixed weight coefficients, once the weight coefficients are determined, the weights in the combined forecasting model will no longer change. Timeliness

There are three main reasons for the change of the weight coefficient: 1. Timeliness problem: Since the sample data used for prediction is a time series, as time goes by, the prediction effect of each single prediction method will change, so the corresponding weight coefficient should also change over time

3. The relationship between individual forecasting methods will change: as the forecasting environment changes, the relative performance between individual forecasting models may also change, changing their importance in the combined forecasting model, resulting in changes in weight coefficients

Images