Clustering Time Series Data Using Forecast Distribution Divergence
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing time series clustering methods focus on historical data, failing to effectively cluster based on forecast data for future time periods, which limits their predictive capabilities in applications like inventory management and sales forecasting.
Innovation Solution
A system utilizing a forecasting model to generate forecasted values and confidence intervals, combined with a symmetric Kullback-Leibler divergence algorithm to create a distance matrix, allowing for clustering of forecasted values based on distribution information, enabling clustering operations on future data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional clustering methods are used on historical time series data, then clustering can be performed on observed data, but the method cannot effectively cluster forecast data for future time periods
Solution Approach 1:
The patent transforms the clustering approach by changing the parameters used for comparison from simple mean values to full distribution characteristics (variance, skewness, kurtosis). This allows the clustering method to effectively handle forecast data by considering the complete probability distribution rather than just central tendency, thereby improving adaptability to future time periods while maintaining reliability through comprehensive distributional analysis
2Measurement precision
If only mean values are used for clustering, then the clustering process is simple, but divergences in forecasted values cannot be identified when means are similar
Solution Approach 1:
The patent extends the clustering methodology from one-dimensional mean value comparison to multi-dimensional distributional comparison by incorporating variance, skewness, and kurtosis as additional dimensions. This dimensional expansion enables precise identification of divergences in forecasted values even when means are similar, achieving high measurement precision while managing complexity through systematic use of distributional moments
3Reliability
If distribution information is incorporated into clustering, then forecasted values can be clustered more accurately, but the computational complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-calculating and storing the distributional moments (mean, variance, skewness, kurtosis) of forecasted values before performing clustering operations. This preprocessing step organizes the distribution information in a structured format that facilitates efficient comparison during clustering, thereby maintaining high clustering reliability while reducing computational complexity during the actual clustering process
Data Source
AI summary
In accordance with the teachings described herein, systems and methods are provided for clustering time series based on forecast distributions. A method for clustering time series based on forecast distributions may include: receiving time series data relating to one or more aspects of a physical process; applying a forecasting model to the time series data to generate forecasted values and confidence intervals associated with the forecasted values, the confidence intervals being generated based on distribution information relating to the forecasted values; generating a distance matrix that identifies divergence in the forecasted values, the distance matrix being generated based the distribution information relating to the forecasted values; and performing a clustering operation on the plurality of forecasted values based on the distance matrix. The distance matrix may be generated using a symmetric Kullback-Leibler divergence algorithm.