Mean Field Clustering for SKU Demand Forecasting Scalability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional dynamic programming and optimal control methods are impractical for managing large-scale business systems with numerous stock keeping units (SKUs) due to the complexity of representing correlations between SKUs, leading to scalability issues.
Innovation Solution
The system divides SKUs into Mean Field clusters, with a tracker identified for each cluster to generate product decisions, and then deconstructs these clusters to obtain decisions for individual SKUs, using Mean Field clustering and cluster deconstruction components to spread risk and improve scalability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional dynamic programming and optimal control methods are used to model correlations between SKUs, then measurement precision of demand forecasting is improved, but device complexity and computational scalability deteriorate
Solution Approach 1:
The patent segments the large set of SKUs into multiple Mean Field clusters, where each cluster contains SKUs with similar demand characteristics. This segmentation reduces the complexity of modeling correlations by grouping SKUs into manageable clusters rather than treating all SKUs individually, thereby maintaining forecasting accuracy while improving scalability.
Solution Approach 2:
The patent introduces Mean Field trackers as intermediary representations for each cluster. These trackers aggregate the demand information of multiple SKUs within a cluster, serving as a mediator that simplifies the correlation modeling process. Instead of directly modeling correlations between all individual SKUs, the system models correlations at the cluster level using trackers, reducing computational complexity while preserving essential demand patterns.
2Measurement precision
If conventional methods model correlations between all SKUs, then measurement precision is improved, but productivity and computational efficiency deteriorate
Solution Approach 1:
By dividing SKUs into Mean Field clusters, the patent enables parallel processing of demand forecasting for different clusters. This segmentation allows the system to compute forecasts for multiple clusters simultaneously, significantly improving computational efficiency and productivity while maintaining the precision needed for accurate demand prediction.
Solution Approach 2:
The patent creates Mean Field trackers as simplified copies or representations of the actual SKU demand patterns within each cluster. These trackers capture the essential demand characteristics without requiring full detailed modeling of every SKU, enabling faster computation while preserving the accuracy needed for effective demand forecasting and inventory management.
3Productivity
If SKUs are grouped into clusters, then productivity and scalability are improved, but loss of information at individual SKU level may occur
Solution Approach 1:
The patent applies local quality by maintaining distinct Mean Field trackers for different clusters, allowing each cluster to have its own specialized representation tailored to its specific demand characteristics. This approach ensures that local information about individual SKUs and their cluster-specific patterns is preserved, while still benefiting from the scalability of clustered organization.
Solution Approach 2:
By segmenting SKUs into meaningful clusters based on demand similarity, the patent ensures that SKUs with comparable characteristics are grouped together. This segmentation strategy minimizes information loss because SKUs within the same cluster share similar demand patterns, making the aggregated cluster-level representation sufficiently accurate for forecasting while maintaining individual SKU traceability when needed.
Data Source
AI summary
A set of SKUs is divided into a plurality of different Mean Field clusters, and a tracker (or sensor) is identified for each cluster. Product decisions for each Mean Field cluster are generated based on the tracker (or sensor) and each Mean Field cluster is then deconstructed to obtain product decisions for individual SKUs in the Mean Field cluster.