Hierarchical Forecast Reconciliation via Time-Point Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing reconciliation processes for hierarchical forecasts are computationally intensive, inefficient, and slow, particularly when handling multiple levels of hierarchies, and often require pairwise reconciliation, which is difficult to implement and resource-intensive.
Innovation Solution
A distributed computing environment with parallel processing techniques that divide forecasts by time point and distribute them across multiple computing nodes, allowing for parallel processing using multiple threads to implement a reconciliation process that adjusts forecasts to meet aggregation constraints, while also validating S-matrices for consistency and handling missing values dynamically.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If pairwise reconciliation is used for hierarchical forecasts, then the reconciliation process can be implemented, but it becomes computationally intensive and slow
Solution Approach 1:
The patent segments the reconciliation process by dividing forecasts into separate time-point groups and distributing them across multiple computing nodes. Each computing node processes a specific subset of time points independently, transforming the monolithic pairwise reconciliation into parallelized discrete operations that reduce overall computational complexity and increase throughput.
Solution Approach 2:
The patent introduces a temporal dimension to the reconciliation process by organizing forecasts into time-point-based groups. Instead of processing all forecasts simultaneously in a single dimensional space, the system creates multiple discrete time slices that can be processed in parallel, effectively adding a temporal dimension to the computation space.
2Reliability
If forecasts are processed individually without regard for hierarchy constraints, then the forecasting process is simple, but the forecasts do not respect aggregation constraints
Solution Approach 1:
The patent performs preliminary actions by pre-organizing forecasts into time-point-based groups before the actual reconciliation computation. This pre-grouping structure is established in advance and used to guide the parallel processing, ensuring that forecasts are processed in a structured manner that naturally respects hierarchy constraints without adding complexity during the computation itself.
Solution Approach 2:
The patent introduces time-point-based groups as intermediary structures between the individual forecasts and the final reconciled values. These groups act as mediators that organize the forecast data into manageable units, enabling the system to handle hierarchy constraints systematically while maintaining processing efficiency through parallelization.
3Loss of time
If a single computing node processes all forecasts sequentially, then the system is simple to implement, but the processing time is long
Solution Approach 1:
The patent segments the forecasting data into time-point-based groups and distributes them across multiple computing nodes. Each node processes a specific subset of time points independently, dividing the total computational workload into parallel tasks that can execute simultaneously, thereby reducing overall reconciliation time while managing complexity through structured distribution.
Solution Approach 2:
The patent transforms the single-node sequential processing model into a multi-node parallel model by introducing a distributed computing dimension. The system organizes forecasts into time slices that can be processed across multiple spatial dimensions (computing nodes), enabling concurrent execution and significantly reducing the time required to complete the reconciliation process.
Data Source
AI summary
A parallel processing technique can be used to expedite reconciliation of a hierarchy of forecasts on a computer system. As one example, the computer system can receive forecasts that have a hierarchical relationship with respect to one another. The computer system can distribute the forecasts among a group of computing nodes by time point, so that all data points corresponding to the same time point in the forecasts are assigned to the same computing node. The computing nodes can receive the datasets corresponding to the time points, organize the data points in each of the datasets by forecast to generate ordered datasets, and assign the ordered datasets to processing threads. The processing threads (across the computing nodes) can then execute a reconciliation process in parallel to one another to generate reconciled values, which can be output by the computing nodes.