Gaussian embedded neural network model for time sequence prediction and modeling method

Abstract

The invention discloses a Gaussian embedded neural network model for time sequence prediction. The model comprises a long-short-term memory neural network layer, a Gaussian embedded module and a feedback path module. Wherein the long-short-term memory neural network layer is used for completing sequential feature modeling of a single sample in each iteration of the system, the Gaussian embedded module is used for sample uncertainty information, and the feedback path module is used for achieving iterative optimization of the network and improving the training effect of the model. The inventionfurther discloses a modeling method of the Gaussian embedded neural network model. According to the method, trainable probability distribution is inserted into the LSTM in an end-to-end mode to serveas feature representation, self-adaptive statistical feature learning is achieved on the basis, compared with the prior art, a better prediction effect is achieved, and a modeling result with better globality and better generalization performance can be obtained.

Description

technical field [0001] The invention belongs to the technical field of neural network model design, in particular to a Gaussian embedded neural network model and a modeling method thereof. Background technique [0002] The combination of uncertainty in time series and deep learning, especially LSTM, is becoming a research hotspot. Uncertainty is divided into model uncertainty and data uncertainty. Bayesian recurrent neural networks and Monte Carlo pruning have yielded good results in model uncertainty modeling, enhancing model interpretation. As for data uncertainty, noise uncertainty and quantile regression (QR) raise some concerns. With regard to noise uncertainty, related studies assume a Gaussian distribution with zero mean and homoscedastic or heteroscedastic variance. Regarding QR, the model combines the advantages of LSTM with parametric QR, which not only captures the volatility in the time series, but also captures the temporal dependence. However, the combinati...

AI Technical Summary

Problems solved by technology

[0003] The present invention aims at the problem of the lack of probability information considerations in time series prediction in the above-mentioned prior art. The present invention proposes a Gaussian embedded neural network model and its modeling method for time series prediction, which can be trained in an end-to-end manner. The probability distribution of the LSTM is inserted into the LSTM as a feature representation, and on this basis, adaptive statistical feature learning is realized, so that better prediction results can be obtained, and modeling results with better globalization and better generalization performance can be obtained.

Images