Forecasting Cohort Half-Life Weighting for Resource Allocation
Solution Overview
Problem
Computer systems face challenges in accurately predicting resource usage over varying time horizons due to the differing accuracy of forecasting algorithms, leading to uncertainty and reliability issues in resource allocation.
Innovation Solution
A method that combines forecasting algorithms with different time horizons using half-life weights, where the half-life weight decreases the effect of each algorithm over time, ensuring that only algorithms with a confidence level of ½ or greater contribute to the forecasting cohort, and an adjustment factor is used to refine weights based on past accuracy, thereby maintaining reliability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple forecasting algorithms with different time horizons are combined, then the forecasting coverage is improved, but the reliability of individual algorithm predictions deteriorates due to varying accuracy over time
Solution Approach 1:
The patent applies dynamics by making the weighting of forecasting algorithms time-dependent through half-life weights. Each algorithm's contribution to the cohort forecast dynamically changes over time, with weights decreasing as time progresses. This allows the system to adaptively adjust the influence of each algorithm based on its temporal characteristics, resolving the contradiction between comprehensive forecasting coverage and reliable individual predictions.
Solution Approach 2:
The patent changes the parameter of algorithm weighting from static to dynamic by introducing half-life weights. The weight parameter w(t) = (1/2)^(t/h) changes continuously based on time t and half-life h, allowing the system to optimize the contribution of each algorithm at different time points. This parameter transformation enables the cohort to maintain reliability while covering diverse forecasting horizons.
2Ease of manufacture
If all forecasting algorithms in the cohort are given equal weight, then the system is simple to implement, but the confidence level of the forecast deteriorates due to inclusion of less accurate algorithms
Solution Approach 1:
The patent applies local quality by assigning different weights to different algorithms based on their individual characteristics (half-life parameters). Instead of uniform weighting, each algorithm receives a localized weight adjustment according to its temporal decay pattern. This allows the system to maintain simplicity in the overall cohort structure while improving confidence levels through differentiated algorithm contributions.
3Duration of action of moving object
If the effect of forecasting algorithms is maintained over long periods, then the forecast horizon is extended, but the reliability deteriorates as algorithm accuracy decreases over time
Solution Approach 1:
The patent implements periodic action through the exponential decay of half-life weights over time. The weight of each algorithm follows a periodic-like decay pattern w(t) = (1/2)^(t/h), where the influence of algorithms naturally diminishes as time progresses. This allows the forecast horizon to be extended while automatically reducing the reliability contribution of older predictions, maintaining overall system reliability.
Data Source
AI summary
A forecasting cohort includes a first set of forecasting algorithms and a second set of forecasting algorithms. A half-life of each of the first set of forecasting algorithms and the second set of forecasting algorithms are determined. A half-life weight for each of the first set of forecasting algorithms and the second set of forecasting algorithms at a subsequent time are determined, such that the half-life weights decrease an effect of a forecasting algorithm as time elapses. A confidence level of the forecasting cohort at the subsequent time is determined and used to adjust resource usage.