Monitoring quantity uncertainty prediction method and system based on variation auto-encoder

Abstract

The invention discloses a monitoring quantity uncertainty prediction method and system based on a variation auto-encoder, and the method comprises the steps: taking environment sample data as an independent variable for changing the physical quantity of a monitoring object, and taking the physical quantity of the monitoring object as a dependent variable; establishing a time index corresponding tothe physical quantity of the monitored object according to the acquisition time of the environmental sample data, and taking the acquisition time corresponding to the time index as a time index pointof the data set; constructing a probability distribution model of an implicit variable causing the change of the physical quantity of the monitored object; obtaining a prior probability of an implicit variable by using a prior distribution function; constructing a deep neural network model for the generation process of the prior probability, constructing an identification model, predicting the missing value of the missing physical quantity according to the identification model, obtaining the probability distribution of the physical quantity of the monitoring object, and sampling the probability distribution of the physical quantity of the monitoring object to obtain missing value uncertainty estimation. According to the method, the prediction result of the monitored object has good uncertainty estimation, and the performance of the monitored object can be better mastered.

Description

technical field [0001] The invention relates to the technical field of object monitoring, in particular to a method and system for predicting uncertainty of monitoring quantities based on variational autoencoders. Background technique [0002] At present, many objects in the real world need to be monitored continuously, so as to ensure that the monitored objects have the accuracy to meet specific performance. During the continuous monitoring of the monitoring object, a series of monitoring data with time attributes will be obtained, and people can judge whether the monitoring object meets the ideal state through the monitoring data. Since there is uncertainty in the monitoring data of the monitoring object, predicting this uncertainty will make the monitoring of the monitoring object more accurate and reliable. [0003] Time series is the numerical sequence formed by arranging the values ​​of certain statistical indicators in chronological order. By compiling and analyzing...

AI Technical Summary

Problems solved by technology

However, these models have high model assumptions, and the model's ability to model data is not strong

[0006] On the whole, although some methods in the existing technical schemes have high accuracy, they lack the estimation of uncertainty, and some technical schemes can obtain the uncertainty of monitoring data, but the accuracy is low

Images