Interpretable Stochasticity for Adjustable Chance-Constrained Programming.

Abstract

Uncertainty poses a significant challenge for optimization algorithms in real-world applications. As these algorithms become more sophisticated, there is an increasing need for them to produce reliable and explainable decisions under uncertain conditions. This article introduces knowledge-enabled explainable chance-constrained programming (KECCP), a novel approach that considers the impact of uncertainty and human factors on chance-constrained programming. KECCP employs the differential evolution (DE) algorithm to construct and represent knowledge through the average treatment effect (ATE) and fuzzy cognitive map (FCM). This is based on analyzing how random variables and confidence levels influence the outcomes of chance-constrained programming. Subsequently, KECCP integrates knowledge-enabled chance-constrained programming with prediction and parameter feedback adjustments to achieve controllable, estimable, and adjustable programming solutions. Applied to stochastic resource allocation under uncertainty, this method demonstrates enhanced explainability and controllability in realistic scenarios.