Time Series Forecasting via Self-Similarity Vector Pattern Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional time series forecasting methods, such as statistical and regression-based approaches, face challenges in accurately predicting future values due to their reliance on fixed time windows, which fail to account for the current state of the time series and often result in inaccurate predictions, especially when dealing with multivariate data sets that have different scales and lengths.
Innovation Solution
A system and method for time series forecasting using a pattern matching-based machine learning model that generates self-similarity vectors by comparing current time windows with historical subsequences, identifies key timestamps, and uses a read-out function to generate future projections and confidence scores, integrating multiple reference time series to improve prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If fixed time windows are used for forecasting, then the method is simple to implement, but the prediction accuracy deteriorates because it fails to account for the current state of the time series
Solution Approach 1:
The patent transforms the static fixed time window approach into a dynamic pattern matching approach. Instead of using predetermined fixed windows, the system dynamically identifies relevant historical patterns by comparing the current time series state with historical subsequences using similarity metrics. This allows the forecasting method to adapt to the current state of the time series while maintaining computational feasibility through efficient pattern matching algorithms.
Solution Approach 2:
The patent changes the parameter representation from fixed time window boundaries to variable pattern similarity thresholds. By using self-similarity vectors and configurable similarity thresholds, the system can flexibly adjust which historical patterns are considered relevant, thereby improving prediction accuracy without significantly increasing implementation complexity.
2Measurement precision
If multiple reference time series with different scales and lengths are integrated, then the forecasting accuracy improves, but the system complexity increases
Solution Approach 1:
The patent segments multiple reference time series into comparable units by normalizing them to a common time window structure. Each reference time series is divided into subsequences of uniform length, allowing diverse time series with different original scales and lengths to be processed uniformly. This segmentation approach enables the integration of multiple reference series while maintaining system manageability through standardized processing pipelines.
Solution Approach 2:
The patent creates a universal pattern matching framework that can handle multiple reference time series with different characteristics. The self-similarity vector computation and pattern matching mechanisms are designed to be universally applicable across different time series types, scales, and lengths, allowing the system to integrate diverse data sources without requiring separate processing systems for each type.
3Measurement precision
If pattern matching with self-similarity vectors is used, then the prediction accuracy improves by focusing on similar historic timestamps, but the computational complexity increases
Solution Approach 1:
The patent extracts only the essential similarity information from historical time series data by computing self-similarity vectors that capture the key characteristics relevant to pattern matching. Instead of processing entire historical datasets, the system extracts and stores compact similarity representations that can be efficiently queried and compared, significantly reducing computational complexity while preserving prediction accuracy.
Solution Approach 2:
The patent performs preliminary computation of self-similarity vectors and pattern matching indices before actual forecasting is needed. By pre-processing the time series data to identify and store similarity relationships in advance, the system reduces the computational burden during real-time forecasting operations, allowing accurate predictions to be generated quickly when needed.
Data Source
AI summary
Provided is a system for time series forecasting, including a computer hardware processor; and at least one non-transitory computer-readable storage medium storing instructions that, when executed by the computer hardware processor, causes the processor to perform: obtaining a target time series; obtaining at least one reference time series associated with the target time series; generating a self-similarity vector for the reference time series by determining a similarity between a current time window of the reference time series and one or more subsequences of the reference time series, different from the current time window; based on the self-similarity vector, identifying one or more historic timestamps of the reference time series with similarity values above a threshold similarity value; generating one or more future projections of the target time series based on the identified timestamps; and generating a forecasting of the target time series using a read-out function.