Tensor Time Series Modeling for Multi-Mode Forecasting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning models struggle to effectively capture relationships among co-evolving time series with multi-mode tensor structures, leading to inefficiencies in computer memory usage and suboptimal analysis and forecasting in applications like smart building monitoring, financial analysis, and traffic monitoring.
Innovation Solution
A network of tensor time series (NeT) model that incorporates Tensor Graph Convolutional Networks (TGCN) and Tensor Recurrent Neural Networks (TRNN) to structure and analyze co-evolving time series, generating node embeddings and temporal dynamics, and forecasting future values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If separate machine learning models are applied to each time series, then model simplicity is maintained, but the ability to capture relationships among co-evolving time series deteriorates
Solution Approach 1:
The patent combines multiple separate time series models into a unified tensor-based model that processes co-evolving time series together. The tensor representation merges multiple modes (different time series) into a single multi-dimensional structure, allowing the model to capture relationships among them while maintaining computational efficiency through shared parameters and operations.
2Ease of manufacture
If separate machine learning models are applied to each time series, then model training is simpler, but computer memory usage efficiency deteriorates
Solution Approach 1:
The patent creates a universal tensor-based model structure that can handle multiple time series simultaneously with shared parameters. The same tensor decomposition and neural network operations are applied across all modes, making the model multi-functional and reducing the total number of parameters needed compared to separate models for each time series.
3Ease of operation
If traditional models are used for co-evolving time series, then implementation is straightforward, but forecasting accuracy deteriorates
Solution Approach 1:
The patent transforms the traditional flat or hierarchical data structure into a multi-dimensional tensor structure that explicitly represents the co-evolving nature of multiple time series. This dimensional transformation allows the model to capture temporal correlations and cross-series relationships that are invisible to traditional models, improving forecasting accuracy while maintaining a systematic implementation approach.
Data Source
AI summary
One or more machine learning models for a network of tensor time series can be provided. Co-evolving time series having multiple modes can be received. A tensor graph convolutional network can be trained, using the co-evolving time series and adjacency matrices associated with the multiple modes in the co-evolving time series, to generate node embeddings associated with a snapshot of the co-evolving time series at time t. A tensor recurrent neural network can be trained to generate temporal dynamics associated with the co-evolving time series based on the generated node embeddings. A neural network model can be trained to forecast a prediction for the co-evolving time series based on the generated node embeddings and the generated temporal dynamics. The tensor graph convolutional network, the tensor recurrent neural network and the neural network model can be trained jointly.