基于贝叶斯岭回归的联合作战油料需求预测决策研究

Research on Joint Operational POL Support Demand Forecasting and Decision-Making Based on Bayesian Ridge Regression

摘要: 针对联合作战油料需求预测中不确定性量化不足、决策支持薄弱的问题，本文提出一种基于贝叶斯岭回归的概率化预测与决策框架。该框架通过输出后验预测分布，实现了对预测风险的科学量化；进而将军事损失函数与后验分布相结合，构建了可依据风险偏好调整的贝叶斯最优分位数决策模型。实例验证表明，相较于岭回归，所提框架在保持略优的点预测精度与更高稳定性的基础上，提供了校准更佳的预测区间，从而显著降低了决策的综合代价。研究为联合作战油料保障提供了集预测可靠性、不确定性透明化与决策适应性于一体的理论方法，有助于推动保障模式从依赖点估计的经验决策，向基于风险量化的主动优化转型。

Abstract: Aiming at the problems of insufficient uncertainty quantification and weak decision support in joint operational POL support demand forecasting, this paper proposes a probabilistic forecasting and decision‑making framework based on Bayesian Ridge Regression. The framework outputs a posterior predictive distribution to achieve scientific quantification of forecasting risk. Furthermore, by integrating a military loss function with the posterior distribution, it constructs a Bayesian optimal quantile decision model that can be adjusted according to risk preference. Case studies demonstrate that, compared with ridge regression, the proposed framework not only maintains slightly better point‑forecast accuracy and higher stability, but also provides better‑calibrated prediction intervals, thereby significantly reducing the integrated cost of decisions. This research offers an integrated theoretical and methodological support for joint operational POL support, combining forecasting reliability, uncertainty transparency, and decision adaptability, and helps to promote the shift of support mode from experience‑based decision‑making reliant on point estimates to proactive optimization based on risk quantification.