Dynamic Hierarchical Forecast Reconciliation for Structural Change
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing forecasting systems fail to provide coherent and accurate forecasts across different levels of hierarchical datasets due to static reconciliation methods that do not adapt to structural changes in time series data, leading to sub-optimal decision-making.
Innovation Solution
Implement a dynamic reconciliation process that adapts the mapping matrix P in real-time based on structural changes in time series datasets using machine learning techniques, such as change point detection, clustering, stochastic processes, or deep learning, to ensure consistent and accurate forecasts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If static reconciliation methods are used, then the system is simple and stable, but the forecast accuracy and coherence deteriorate when structural changes occur in time series data
Solution Approach 1:
The patent implements dynamic reconciliation by training a machine learning model to learn the mapping matrix P from historical data, allowing the system to automatically adapt to structural changes in time series data. The model dynamically adjusts reconciliation parameters based on learned patterns rather than using static predefined methods.
Solution Approach 2:
The system changes the reconciliation parameters (mapping matrix P) from static to dynamic by using machine learning to learn optimal parameters from historical data. The learned mapping matrix P adapts to different structural conditions in the time series data, improving forecast accuracy under varying conditions.
2Reliability
If dynamic reconciliation using machine learning is implemented, then forecast accuracy and adaptability improve, but system complexity increases
Solution Approach 1:
The machine learning model performs self-training on historical reconciliation data to automatically learn the optimal mapping matrix P. The system serves itself by autonomously adapting to structural changes without requiring manual intervention or complex configuration, reducing operational complexity despite the advanced algorithms used.
Solution Approach 2:
The system performs preliminary training on historical data before actual forecasting operations. By pre-learning the mapping matrix P from historical reconciliation data, the system prepares optimal parameters in advance, which simplifies the actual forecasting process and reduces runtime complexity.
3Productivity
If traditional reconciliation methods are used, then the process is fast and simple, but decision-making quality deteriorates due to incoherent forecasts
Solution Approach 1:
The system performs reconciliation model training in advance using historical data before actual forecasting is needed. This preliminary action prepares the learned mapping matrix P so that during actual forecasting operations, the system can quickly apply the pre-learned model without time-consuming computations, maintaining both speed and accuracy.
Data Source
AI summary
Mechanisms are provided for performing automated and dynamic reconciliation of forecasts for hierarchical datasets. A machine learning training is executed on a dynamic reconciliation computer model engine to train the dynamic reconciliation computer model engine, based on historical data and forecast data, to learn an association of reconciliation computer models with structural changes in a hierarchical dataset. Runtime forecast data is generated based on a runtime hierarchical dataset, and the trained dynamic reconciliation computer model engine is executed on the runtime forecast data to reconcile the runtime forecast data across a hierarchy of the runtime forecast data. The trained dynamic reconciliation computer model applies different reconciliation computer models to the runtime forecast data based on structural changes in the runtime forecast data. Reconciled runtime forecast data is generated based on results of executing the trained dynamic reconciliation computer model engine on the runtime forecast data, which is then output.