Motif-Based Forecasting Algorithm Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Choosing the most accurate forecasting algorithm for time series data is challenging due to differences in how each algorithm works, leading to varying predictions for the same future time interval, even when trained on the same historical data, and existing methods do not effectively account for specific patterns within the data.
Innovation Solution
The method involves identifying motifs in time series data, which are repeating patterns, and selecting the forecasting algorithm with the lowest forecast error for each motif to predict future demand, using a system that processes time series data to determine the best-suited algorithm for specific time intervals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple forecasting algorithms are used to predict demand for different time intervals, then forecasting accuracy can be improved by selecting the best algorithm for each pattern, but the complexity of selecting and managing multiple algorithms increases
Solution Approach 1:
The patent segments the time series data into different motifs (repeating patterns) and assigns different forecasting algorithms to different motifs. This segmentation allows each algorithm to be optimized for specific patterns rather than requiring a single algorithm to perform well across all patterns, thereby improving forecasting accuracy while managing complexity through structured organization.
Solution Approach 2:
The patent applies local quality by selecting different forecasting algorithms for different local patterns (motifs) within the time series data. Each motif receives a locally optimized algorithm based on its specific characteristics, rather than applying a uniform algorithm globally. This approach improves accuracy for each pattern type while the systematic motif-based framework manages the overall complexity.
2Device complexity
If a single forecasting algorithm is used for all time intervals, then the system is simpler to manage, but forecasting accuracy decreases because different algorithms perform better on different patterns
Solution Approach 1:
The patent creates a universal motif-based framework that can accommodate multiple forecasting algorithms through a common selection mechanism. The motif identification and selection system serves as a universal layer that manages multiple algorithms, providing simplicity in system management while enabling accuracy improvements through pattern-specific algorithm selection.
3Reliability
If forecasting algorithms are trained on historical demand data, then predictions can be made for future time intervals, but different algorithms produce different predictions for the same time interval making selection difficult
Solution Approach 1:
The patent replaces the manual or ad-hoc process of algorithm selection with an automated motif-based selection mechanism. The system automatically identifies motifs in the time series data and selects appropriate algorithms based on motif matching, eliminating the need for manual evaluation and comparison of different algorithm predictions. This substitution improves prediction reliability through consistent, data-driven selection while making the process easier to operate.
Data Source
AI summary
The present disclosure describes methods and systems for selecting the forecasting algorithm to use for a prediction based on motifs. A motif is a pattern of interval values that is found to repeat in time series data. Time series data that includes historical demand data (e.g., average communication volume) for an entity at various time intervals in the past is received. The time series data is processed to identify motifs. For each identified motif, the forecasting algorithm that best predicts the historical demand data for time intervals associated with the motif is determined. Later, when the entity desires to receive a forecast for a future time interval, the motif associated with the future time interval is determined. The forecasting algorithm determined to best predict demand for the determined motif is then used to predict the demand for the future time interval.