An automatic filling method for missing values ​​in time series based on long short-term memory network

Abstract

The invention discloses a method for automatically filling missing values ​​in time series based on long-term and short-term memory networks, which comprises the following steps: obtaining an original data set without missing values, and performing preprocessing; randomly deleting certain values ​​according to a given missing rate to form Missing data sets with different missing rates; according to the idea of ​​deep residual network and graph structure dependence, the residuals based on graph dependence are introduced to connect to the LSTM model to form RSU, and at the same time, the hidden state of the LSTM model at each moment and the RSU's The historical state information is fused and transmitted; the new model constructed by training is filled with the value of RSU according to whether the data is missing or not during the forward propagation process of network training. RSU integrates the previous residual and hidden state information, so that the built model can associate enough historical information about missing data, and at the same time realize the end-to-end training process for time series data containing missing values.

Description

technical field [0001] The invention relates to the technical field of automatic filling of missing values ​​in time series, in particular to an automatic filling method for missing values ​​in time series that can learn end-to-end data containing missing values ​​based on a long-short-term memory network. Background technique [0002] Time series is one of the most common forms of data in practical applications, including weather, health and medical, motion capture, financial markets and urban traffic control, etc. However, these actual time series data inevitably contain missing values ​​due to sensor failures. Generally, the methods for dealing with missing values ​​include zero filling, mean filling, polynomial fitting method or EM algorithm, etc. to fill missing values. [0003] However, time series imputation is a very challenging task compared to static data imputation. Because missing values ​​often have nonlinear and dynamic correlations with their previous values...

AI Technical Summary

Problems solved by technology

Similarly, the Temporal Regularized Matrix Factorization (TRMF) model proposed by Dhillon et al. uses autoregression to simulate the time dependence between the corresponding latent variables, and they generalize the structure of this autoregression as a graph , used to establish dependencies between missing values ​​and their previous non-missing values ​​at different stages, but this relies on human prior knowledge and requires manual design of the graph structure

In addition to the above-mentioned graph-based methods, Long Short-Term Memory (LSTM) can also be used to establish the transition relationship of time. However, the traditional LSTM model cannot handle time series data with missing values, so it cannot Conduct end-to-end network training

Images