跳转至

超越阈值：面向冷链物联网系统的质量感知决策智能框架

文章背景与核心概要

当前的冷链物流系统大多仅作为被动监控工具存在，即在温度超过预设阈值时触发警报，却无法评估产品的累积变质程度，也难以将这些信号转化为可执行的物流决策。这种局限性导致了资源浪费和产品质量控制的盲区。

为了解决这一问题，本文提出了“质量感知决策智能”（QADI）框架。该框架通过结构化的质量状态表示（\(S_q\)）、结合物理模型与数据驱动的混合建模层，以及基于 Microsoft Phi-4 和检索增强生成（RAG）的推理层，实现了从单纯监控到主动决策的跨越。在牛奶和西兰花冷链场景的基准测试中，QADI 显著降低了保质期预测误差，并在 99.5% 的场景中实现了最优决策，证明了其在复杂物流环境下的卓越性能。

📌 摘要 (Summary)

Current cold chain logistics systems largely function as reactive monitors—triggering alerts when temperature thresholds are breached without assessing cumulative product degradation or converting those signals into actionable logistics decisions.

To overcome this limitation, the authors introduce the Quality-Aware Decision Intelligence (QADI) framework. QADI combines three core components: 1. Structured Quality State Representation: Evaluated via \(S_q = [L, Q, U, R]\), which tracks remaining shelf life, degradation rate, estimation uncertainty, and operational risk. 2. Hybrid Quality Modeling Layer: Merges physics-based microbial kinetics with a data-driven correction term. 3. Advanced Reasoning Layer: Leverages Microsoft Phi-4 with retrieval-augmented generation (RAG) over a domain knowledge base.

Across rigorous benchmarking on milk and broccoli cold chain scenarios, the QADI framework achieves a significantly lower mean absolute shelf-life error (7.2 hours vs. 30.9 hours for physics-only models) and produces oracle-optimal decisions in 99.5% of scenarios.

📋 摘要 (Abstract)

冷链物流在技术上已取得长足进步，但大多数已部署的系统仍停留在被动监控层面，而非决策代理：阈值仅用于触发警报，却无法将违规情况与产品的累积变质联系起来，也无法将变质信号转化为物流决策。我们通过“质量感知决策智能”（QADI）框架填补了这一空白，该框架结合了三种能力：一种结构化的质量状态表示 \(S_q = [L, Q, U, R]\)（即剩余保质期、变质率、估计不确定性和运营风险，均可从框架方程中导出并计算）；一个结合了基于物理的微生物动力学与数据驱动修正项的混合质量建模层；以及一个基于 Microsoft Phi-4 并通过结构化领域知识库进行检索增强生成（RAG）的推理层。

Cold chain logistics has advanced technologically, yet most deployed systems remain reactive monitors, not decision-making agents: thresholds trigger alerts, but nothing relates violations to cumulative product degradation or converts degradation signals into logistics decisions. We address this gap with a Quality-Aware Decision Intelligence (QADI) framework combining three capabilities: a structured quality state representation, \(S_q = [L, Q, U, R]\) -- remaining shelf life, degradation rate, estimation uncertainty, and operational risk, all derived and computable from the framework equations; a hybrid quality modeling layer combining physics-based microbial kinetics with a data-driven correction term; and a reasoning layer built on Microsoft Phi-4 with retrieval-augmented generation over a structured domain knowledge base.

我们针对五种基准（阈值监控、纯物理模型、物理加噪声模型、基于优化的决策以及基于规则的专家系统）进行了对比，涵盖了八种冷链场景，并以巴氏杀菌奶作为主要案例，其保质期真值取自独立于我们模型的已发表乳制品研究。比较采用了带有 Holm 校正的 Wilcoxon 符号秩检验。在牛奶和西兰花场景中，该框架的平均绝对保质期误差达到 7.2 小时（相比之下，纯物理模型为 30.9 小时；\(p<0.001\)），变质率为 14.5%（相比之下，纯物理模型和基于规则的系统为 16.6%；\(p=0.08\)），并在 99.5% 的场景中实现了最优决策。移除大语言模型（LLM）推理组件后，最优性下降至 45.5%（\(p<0.001\)）。专家评估的解释质量达到 83%（\(\kappa = 0.71\)）。消融实验表明，混合建模和 LLM 推理分别贡献了独特的增益，而 RAG 检索主要提升了解释质量。

We benchmark against five baselines -- threshold monitoring, physics-only, physics-plus-noise, optimisation-based decisions, and a rule-based expert system -- across eight cold chain scenarios, using pasteurised milk as the primary case, with ground truth shelf-life drawn from published dairy studies independent of our model. Comparisons use Wilcoxon signed-rank tests with Holm correction. Across milk and broccoli scenarios, the framework attains mean absolute shelf-life error of 7.2 hours (versus 30.9 hours, physics-only; \(p<0.001\)), spoilage rate of 14.5% (versus 16.6%, physics-only and rule-based; \(p=0.08\)), and oracle-optimal decisions in 99.5% of scenarios. Removing the LLM reasoning component drops optimality to 45.5% (\(p<0.001\)). Expert-rated explanation quality reaches 83% (\(\kappa = 0.71\)). Ablations show hybrid modeling and LLM reasoning contribute distinct gains, while RAG retrieval mainly drives explanation quality.

📊 元数据与分类 (Metadata & Classification)

license icon View License