3D Temporal Tensor Multivariate Time Series Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current multivariate time series forecasting methods require statistical assumptions and extensive preprocessing, failing to adequately capture non-linearities and requiring significant computational resources, especially when dealing with large feature and temporal dimensions.
Innovation Solution
Transforming historical 2D multivariate time series data into a 3D temporal tensor and training deep volumetric 3D convolutional neural networks (CNNs) to predict future values, which simplifies the training process and reduces computational requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If statistical assumptions and manual preprocessing are used for multivariate time series forecasting, then forecasting can be performed with current methods, but additional manual preprocessing and feature engineering are necessary and non-linearities are not adequately captured
Solution Approach 1:
The patent replaces manual preprocessing and feature engineering operations with a deep learning-based automated system. The neural network automatically performs feature extraction and transformation from raw multivariate time series data, eliminating the need for manual statistical preprocessing while capturing non-linear relationships that traditional methods miss.
Solution Approach 2:
The patent transforms the data representation from traditional 2D matrices to 3D temporal tensors, fundamentally changing the parameter structure. This transformation enables the model to simultaneously capture temporal dynamics and cross-variable relationships, improving forecasting accuracy while reducing the need for manual feature engineering.
2Reliability
If current forecasting methods are used, then forecasting can be performed, but large amounts of computational resources are required, especially when provided input data has very large feature and temporal dimensions
Solution Approach 1:
The patent segments the multivariate time series data into 3D temporal tensors with specific dimensions (time steps, variables, and features). This segmentation allows the model to process large datasets more efficiently by breaking down the computational task into manageable volumetric units that can be processed through convolutional operations.
Solution Approach 2:
The patent adds a temporal dimension to transform 2D data into 3D tensors, enabling more efficient computational processing. This dimensional transformation allows the model to exploit temporal locality and redundancy, reducing overall computational resource requirements while maintaining forecasting capability.
3Ease of manufacture
If 3D temporal tensor transformation and deep volumetric 3D CNNs are used, then non-linear feature-temporal and auto-correlational features are captured and manual preprocessing is reduced, but a transformation process and neural network training are required
Solution Approach 1:
The patent implements a self-service system where the deep volumetric 3D CNN automatically performs feature extraction and data transformation without human intervention. The model learns optimal feature representations during training and automatically applies these transformations to new data, eliminating the need for manual preprocessing while capturing complex non-linear relationships.
Data Source
AI summary
A computer-implemented method according to one embodiment includes receiving historical two-dimensional (2D) multivariate time series data; transforming the historical 2D multivariate time series data into a three-dimensional (3D) temporal tensor; training one or more deep volumetric 3D convolutional neural networks (CNNs), utilizing the 3D temporal tensor; and predicting future values for additional multivariate time series data, utilizing the one or more trained deep volumetric 3D CNNs.