Multiple Time-Series Forecast Reconciliation Across Granularities
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing forecasting techniques struggle to produce self-consistent forecasts for quantities with different granularities and groupings, leading to inconsistencies in forecasts across items and time scales.
Innovation Solution
A method involving determining individual forecasts for different items and time sequences, reconciling them using relationship information to form second forecasts that satisfy data series requirements and approximate the first forecasts, employing data processing procedures like Quadratic Programming to enforce consistency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If separate forecasts are made for different items or categories, then forecast detail and specificity are improved, but consistency and self-accuracy across forecasts deteriorates
Solution Approach 1:
The patent combines multiple separate forecasts into a unified reconciled forecast that satisfies consistency constraints. The reconciliation process merges forecasts for different items/categories while enforcing relationships between them, ensuring that the sum of category forecasts equals the total forecast and that time series forecasts are consistent across different granularities.
Solution Approach 2:
The patent implements an iterative reconciliation process that uses feedback from consistency checks to adjust forecasts. The system repeatedly compares forecasts against consistency constraints and modifies them until self-accuracy is achieved, where the reconciled forecast satisfies both the detail requirements and the consistency relationships.
2Adaptability or versatility
If forecasts are made at different time granularities (daily, weekly, monthly), then flexibility and applicability are improved, but consistency between different time scale forecasts deteriorates
Solution Approach 1:
The patent introduces a new dimension of reconciliation that operates across multiple time granularities. The system reconciles forecasts at daily, weekly, and monthly levels simultaneously, ensuring that aggregations across time scales are consistent. This multi-dimensional approach maintains flexibility for different time scale analyses while enforcing internal consistency.
3Device complexity
If direct forecasts are made for aggregated quantities, then forecast simplicity is improved, but ability to capture detailed patterns and relationships deteriorates
Solution Approach 1:
The patent segments the forecasting process into two stages: first generating detailed forecasts for individual items or categories, then reconciling them to produce aggregated forecasts. This segmentation allows the system to capture detailed patterns at the item level while producing simple, consistent aggregated forecasts that satisfy relationships between different levels of aggregation.
Data Source
AI summary
An approach for forecasting multiple related data series includes first determining individual forecasts for different items or groups of items and/or over different time sequences or granularities. These first forecasts are then reconciled to form second forecasts that satisfy relationships between the data series while best approximating the first forecasts.