Tensorized LSTM Shared Memory for Multivariate Trend Forecasting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods struggle to effectively learn and forecast trends in multivariate time series data due to complex temporal patterns and dependencies among time series, especially in applications like traffic management and algorithmic trading, where detecting group behavior is challenging.
Innovation Solution
A deep learning model called DeepTrends, utilizing a tensorized LSTM with adaptive shared memory (TLASM) and a multi-task one-dimensional convolutional neural network (1dCNN), is employed to learn both long-term and short-term dependencies in multivariate time series, adapting to task relatedness and extracting salient features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If conventional prediction methodologies are used, then the analysis is simple, but the information about semantics and dynamics is insufficient
Solution Approach 1:
The patent segments the time series data into multiple segments and uses piecewise linear trends to represent different phases of the data. This allows the system to capture complex temporal patterns and dynamics by analyzing distinct segments separately while maintaining overall context, thereby reducing information loss without excessive computational complexity
Solution Approach 2:
The patent transforms the problem by adding a temporal segmentation dimension and using multi-resolution analysis. By converting the time series into piecewise linear representations with different resolutions and incorporating historical dependency modeling, the system extracts richer semantic and dynamic information while managing computational resources through structured dimensionality
2Measurement precision
If deep learning models are employed to learn temporal patterns, then trend forecasting accuracy is improved, but computational costs increase
Solution Approach 1:
The patent performs preliminary processing by segmenting the time series data and extracting piecewise linear trends before feeding into the deep learning model. This pre-processing reduces the complexity of the input data, allowing the neural network to focus on learning temporal dependencies rather than processing raw complex data, thereby reducing computational costs while maintaining accuracy
Solution Approach 2:
The patent uses a selective approach where the deep learning model processes only the extracted trend features and historical dependencies rather than the entire raw time series data. This partial action on the most informative features reduces computational requirements while preserving the essential patterns needed for accurate forecasting
3Loss of information
If multivariate time series data is analyzed, then comprehensive group behavior insight is achieved, but the complexity of detecting temporal patterns increases
Solution Approach 1:
The patent applies segmentation to multivariate time series by dividing each series into piecewise linear segments and then analyzing the segments across multiple variables simultaneously. This approach simplifies the detection of temporal patterns by reducing each variable to its dominant trends while maintaining the ability to detect group behavior through the relationship between segmented patterns across variables
Solution Approach 2:
The patent creates a universal processing framework that handles multiple time series variables through the same piecewise linear transformation and historical dependency modeling mechanism. This multi-functional approach allows comprehensive group behavior analysis while maintaining consistent methodology across all variables, reducing the complexity of detecting temporal patterns in multivariate data
Data Source
AI summary
A method for executing a multi-task deep learning model for learning trends in multivariate time series is presented. The method includes collecting multi-variate time series data from a plurality of sensors, jointly learning both local and global contextual features for predicting a trend of the multivariate time series by employing a tensorized long short-term memory (LSTM) with adaptive shared memory (TLASM) to learn historical dependency of historical trends, and employing a multi-task one-dimensional convolutional neural network (1dCNN) to extract salient features from local raw time series data to model a short-term dependency between local time series data and subsequent trends.