Multivariate time series multilayer space-time dependence modeling method based on deep learning

A multivariate time series and deep learning technology, applied in the field of deep learning, can solve problems such as the influence of different layers of features, achieve advanced performance, enhance interpretability and robustness, and improve accuracy

CN111651504AActive Publication Date: 2020-09-11HUNAN UNIV
3 Cites 21 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Publication Date
2020-09-11

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
  • Figure 3
    Figure 3
Patent Text Reader

Abstract

The invention belongs to the field of deep learning, and discloses a multivariate time series multilayer space-time dependence modeling method based on deep learning. According to the method, a novelattention mechanism is introduced to carry out finer-grained processing on space-time dependence features extracted from different layers in the neural network; the model provided by the invention iscomposed of a stacked long-short-term neural network-convolutional neural network (LSTM-CNN). The network is composed of a spatial attention mechanism based on a CNN, a CNN-based channel attention mechanism, a time attention mechanism and an autoregressive assembly. The concept of multi-layer space-time dependence is introduced; a CNN-based channel attention mechanism and a CNN-based spatial attention mechanism are used to pay attention to space-time dependence characteristics of different layers respectively. According to the method, filtering of redundant information and effective extractionof features having greater influence on a prediction result are realized, the purpose of improving the prediction result is achieved, and the method is excellent in performance on multivariate time series data in different fields and can be extended to a task of unit time series prediction.
Need to check novelty before this filing date? Find Prior Art

Description

Technical field:

[0001] The invention belongs to the field of deep learning, relates to a multivariate time series prediction method based on deep learning, and belongs to the application of deep learning on time series data. Background technique:

[0002] Multivariate time series commonly exist in various fields, and they have complex time and space characteristics. By using deep learning tools to represent the temporal and spatial characteristics of multivariate time series data, it can predict its future development trend, which has important practical application value in various fields such as energy consumption, traffic flow and air quality.

[0003] Various methods have been proposed to capture the inherent spatiotemporal properties of multivariate time series. In the traditional time series forecasting method, the autoregressive moving average model (ARIMA) can effectively extract the long-term dependence of the sequence itself, but often ignores the spatial correla...

Examples

Embodiment Construction

[0036] Following result accompanying drawing and specific embodiment are described in further detail to the present invention:

[0037] The purpose of the present invention is to solve the problem that the influence of different layer features on the prediction results is ignored in the process of current deep learning to capture spatiotemporal features of multivariate time series data to complete the prediction task.

[0038] Such as figure 1 , 2 , 3, 4, and 5, what the present invention provides is a multivariate time series multi-layer space-time dependent modeling method based on deep learning, comprising the following steps:

[0039] Step 1: Raw data is preprocessed, including the following steps:

[0040] (1) Perform z-score standardization on multivariate time series data;

[0041] (2) Introduce a sliding window to scroll multivariate time series data. The value in the sliding window is used as an attribute, and the value at the next moment is used as a label. The step...