Demand Forecasting System Using Time Series Clustering and Trend Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing demand forecasting systems fail to accurately forecast demands for commodities and services due to missing short-term variations and inability to handle time series data with different scales, leading to imprecise predictions.
Innovation Solution
A demand forecasting system that performs time series clustering, extracts trend component data, calculates pattern data, and uses regression learning to generate a learning model that considers variations in time series scales, enabling precise demand forecasting by combining trend and pattern data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Stability of the object's composition
If clustering is performed on smoothed actual results, then forecasting stability is improved, but short-term variations are lost leading to reduced forecasting precision
Solution Approach 1:
The patent segments the forecasting process into two distinct components: (1) clustering on smoothed actual results to capture long-term trends and stable patterns, and (2) separate processing of short-term variations to preserve transient information. This segmentation allows each component to be optimized independently, resolving the contradiction between stability and precision.
Solution Approach 2:
The patent introduces smoothed actual results as an intermediary between raw data and clustering. This intermediary preserves long-term trends while filtering out noise, enabling stable clustering without complete loss of variation information. The smoothing acts as a mediator that balances stability requirements with information preservation.
2Adaptability or versatility
If clustering is performed on time series data with different scales, then versatility is improved, but measurement accuracy deteriorates due to scale differences
Solution Approach 1:
The patent applies parameter transformation by standardizing time series data with different scales before clustering. This parameter change (standardization) converts data with varying units and magnitudes into a common scale, enabling accurate clustering across diverse time series while maintaining the ability to handle different scales.
3Device complexity
If a single forecasting model is used for all provision targets, then device complexity is reduced, but forecasting precision worsens due to unique demand characteristics
Solution Approach 1:
The patent segments the forecasting approach by creating separate forecasting models for each cluster of provision targets with similar demand characteristics. This segmentation balances model complexity (fewer models than individual forecasting) with precision (tailored models for each cluster), resolving the contradiction between simplicity and accuracy.
Solution Approach 2:
The patent applies local quality by assigning different forecasting models to different clusters based on their specific demand characteristics. Each cluster receives a forecasting model optimized for its local patterns and variations, rather than a uniform global model, improving precision while maintaining reasonable complexity through clustering.
Data Source
AI summary
A demand forecasting system obtains first forecasting data indicating a representative demand forecasting value of a provision target for each cluster by executing a time series clustering process, extracting trend component data for each cluster, and calculating pattern data. The demand forecasting system obtains second forecasting data that is data indicating a forecasting value of a difference for each provision target and obtains third forecasting data indicating a demand forecasting value for each provision target.