Hierarchical Forecast Reconciliation via Summation Matrix
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current demand forecasting in supply chains faces significant errors due to inconsistencies when forecasting at different levels of a hierarchy, leading to cumulative errors and inefficiencies in manufacturing and inventory management.
Innovation Solution
A computer-implemented method that generates a multi-level hierarchy and a weight matrix to reconcile forecasts across various levels, using a non-linear Machine Learning model and non-negative least squares optimization to ensure consistency and accuracy, thereby reducing overall forecasting error.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If forecasts are made separately at different levels of the hierarchy, then the aggregation constraint is violated and forecasts do not add up correctly, but making forecasts at only one level (usually the most disaggregated level) avoids this inconsistency
Solution Approach 1:
The patent segments the forecasting problem into multiple hierarchical levels (e.g., product-family level and individual product level), allowing separate forecasts to be made at each level. These segmented forecasts are then reconciled through optimization to ensure they add up correctly across the hierarchy, thus maintaining consistency while preserving the benefits of multi-level forecasting.
Solution Approach 2:
The patent introduces an intermediary reconciliation process that acts as a mediator between forecasts made at different hierarchical levels. This intermediary optimization step adjusts the forecasts to satisfy aggregation constraints while minimizing the impact on individual forecast accuracy, effectively bridging the gap between consistency requirements and accuracy preservation.
2Measurement precision
If forecasts are made at multiple levels of the hierarchy, then more information is utilized for better overall accuracy, but cumulative errors increase when individual forecasts are added together
Solution Approach 1:
The patent implements a feedback mechanism where forecasts from multiple hierarchical levels are aggregated and the resulting cumulative errors are fed back into an optimization process. This feedback loop allows the system to adjust individual level forecasts to minimize overall error accumulation while maintaining the information benefits of multi-level forecasting.
Solution Approach 2:
The patent changes the parameters of the forecasting system by introducing optimization variables and constraints that control how forecasts are combined across hierarchical levels. By adjusting these parameters through optimization, the system can minimize cumulative errors while preserving the accuracy benefits of utilizing multiple levels of forecasting information.
3Measurement precision
If separate forecasts are made at different hierarchical levels, then the ability to capture local patterns is improved, but the complexity of reconciling these forecasts increases
Solution Approach 1:
The patent applies preliminary action by pre-defining the hierarchical structure and aggregation constraints before the forecasting and reconciliation process. This preliminary setup allows the optimization algorithm to efficiently reconcile forecasts without having to dynamically determine the complex relationships between levels, thus reducing reconciliation complexity while preserving local pattern capture capabilities.
Data Source
AI summary
Systems and methods for reconciling a forecast within a multi-level hierarchy, comprising a pre-processing module and a forecast reconciliation module. The pre-processing module reconstructs the structure of the hierarchy and captures the relationship between the nodes of the multi-level hierarchy in a summation matrix S. The forecast reconciliation matrix uses S, a weight matrix W (that reflects a weighting scheme between the nodes) and a base forecast to optimize the overall forecast error using a least squares procedure. The reconciled forecast has a zero consistency error.