Adaptive updating method applied to time sequence LSTM parameter prediction model

Abstract

In practical application, parameters of a research object deviate from an initial data operation range along with time, and even an evolution rule of data changes, so that an average prediction error of an original prediction model exceeds an expected value, the prediction precision of the model is reduced, and at the moment, a current prediction model needs to be updated. In order to solve the problem, the invention designs a self-adaptive updating method applied to a time sequence LSTM parameter prediction model based on an evolution rule of time sequence parameters. According to the method, when the prediction error of the original model exceeds the preset expected value, the current prediction model can be updated in time by using the standby prediction model, and the prediction value of the prediction model on the parameters can be corrected by using the dynamic adjustment coefficient when the model needs to be updated but the standby model is not trained, so that the prediction precision of the model is improved.

Description

technical field [0001] The present invention relates to an adaptive updating method applied to a time series LSTM parameter prediction model, more particularly, the present invention relates to an automatic prediction model that can be applied to the time series parameter prediction model when the evolution law of the parameters changes The updated method finally improves the prediction accuracy of the parametric prediction model. Background technique [0002] Time series is ubiquitous in all walks of life. It is a collection of data collected at a series of equally spaced time points, such as the sales of a product in commercial activities; in industrial production, the trend of monitoring parameters; meteorology Precipitation in a certain city, etc. The order of data and the size of data in time series contain the inherent information of a thing or phenomenon in the objective world, and represent a dynamic process that changes with time. The significance of time series a...

AI Technical Summary

Problems solved by technology

However, in practical applications, the parameters of the research object will deviate from the initial data operating range over time, and even the evolution law of the data will change, which will cause the average prediction error of the original prediction model to exceed the expected value, and the prediction accuracy of the model will decline. At this time, the current forecasting model needs to be updated

Images