Selected work

Representative Papers

TIDE: Trustworthy and Interpretable Battery Degradation Estimation with Contextual Learning and Symbolic Distillation

Jul 16, 2026

Accurate estimation of battery state of health (SOH) is critical for effective control, maintenance, and longevity in battery management systems, particularly within intelligent interconnected systems where estimation errors can propagate widely. To address this challenge, this work proposes TIDE—a novel framework that uniquely integrates knowledge-guided degradation priors, monotonic residual modeling, and context-aware learning, while introducing symbolic distillation to achieve model-level interpretability. The synergistic three-component architecture jointly optimizes estimation accuracy, reliability, and interpretability, preserving component-wise transparency without compromising performance. Experimental results demonstrate that TIDE improves average SOH estimation accuracy by 19.7% over baseline methods, substantially reduces violations of aging consistency constraints, and thereby significantly enhances the practicality and robustness of SOH estimation in real-world applications.

0 citationsRead paper

Generative-Evaluative Agreement: A Necessary Validity Criterion for LLM-Enabled Adaptive Assessment

May 19, 2026

This study addresses the validity threat posed by self-referential bias in large language models (LLMs) when deployed in self-generated, self-scored adaptive assessments. To mitigate this issue, the authors propose “Generation-Evaluation Agreement” (GEA) as a novel validity criterion, which evaluates whether model-assigned scores accurately recover the intended proficiency levels embedded during item generation. The research implements a two-stage adaptive testing system grounded in skill decomposition, integrating item generation, response simulation, and automated scoring, and introduces the first quantitative metric for GEA. Empirical results reveal an overall GEA correlation of 0.698, with high consistency for grammar-related skills (r > 0.7) but near-zero agreement for design-oriented skills. Additionally, low-proficiency examinees near routing thresholds were systematically overestimated, underscoring the critical role of fine-grained scoring rubrics in enhancing assessment validity.

0 citationsRead paper

LLM-Upgraded Graph Reinforcement Learning for Carbon-Aware Job Scheduling in Smart Manufacturing

Dec 06, 2025

This work addresses the dynamic, sustainable, carbon-aware flexible job shop scheduling problem in smart manufacturing. We propose a graph reinforcement learning framework that integrates Graph Neural Networks (GNNs) with Large Language Models (LLMs). Through customized prompt engineering, the topological structure and semantic information of scheduling states are jointly encoded into high-quality embeddings; a multi-objective reward function guides deep reinforcement learning to jointly optimize makespan and carbon emissions. To our knowledge, this is the first study to incorporate LLMs into graph-based reinforcement learning for carbon-aware job shop scheduling, endowing the agent with intrinsic semantic understanding. Evaluated on both synthetic and public benchmark datasets, our method achieves an average 4.1% reduction in makespan (up to 12.2%) while significantly lowering carbon emissions, outperforming state-of-the-art algorithms in overall performance.

0 citationsRead paper

Machine Learning-Assisted Surrogate Modeling with Multi-Objective Optimization and Decision-Making of a Steam Methane Reforming Reactor

Jul 10, 2025

Steam methane reforming (SMR) reactor multi-objective optimization faces high computational cost and inherent trade-offs among conflicting objectives—e.g., methane conversion, hydrogen production rate, and CO₂ emissions. Method: This paper proposes an integrated framework comprising mechanistic modeling, data-driven surrogate modeling, multi-objective optimization, and decision-making support. An artificial neural network (ANN)-hybrid surrogate model replaces computationally expensive high-fidelity 1D fixed-bed simulations; non-dominated sorting genetic algorithm II (NSGA-II) efficiently computes the Pareto-optimal front; and a dual-criteria decision strategy—combining Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) and stochastic PROBID (sPROBID)—supports optimal operating condition selection. Contribution/Results: The surrogate model reduces average simulation time by 93.8%. The identified Pareto-optimal solution achieves methane conversion of 0.988, H₂ production of 3.335 mol/s, and CO₂ emissions of 0.781 mol/s—demonstrating substantial gains in optimization efficiency and engineering applicability.

0 citationsRead paper

Recent publications

Latest Papers

TIDE: Trustworthy and Interpretable Battery Degradation Estimation with Contextual Learning and Symbolic Distillation

Jul 16, 2026

Accurate estimation of battery state of health (SOH) is critical for effective control, maintenance, and longevity in battery management systems, particularly within intelligent interconnected systems where estimation errors can propagate widely. To address this challenge, this work proposes TIDE—a novel framework that uniquely integrates knowledge-guided degradation priors, monotonic residual modeling, and context-aware learning, while introducing symbolic distillation to achieve model-level interpretability. The synergistic three-component architecture jointly optimizes estimation accuracy, reliability, and interpretability, preserving component-wise transparency without compromising performance. Experimental results demonstrate that TIDE improves average SOH estimation accuracy by 19.7% over baseline methods, substantially reduces violations of aging consistency constraints, and thereby significantly enhances the practicality and robustness of SOH estimation in real-world applications.

0 citationsRead paper

Generative-Evaluative Agreement: A Necessary Validity Criterion for LLM-Enabled Adaptive Assessment

May 19, 2026

This study addresses the validity threat posed by self-referential bias in large language models (LLMs) when deployed in self-generated, self-scored adaptive assessments. To mitigate this issue, the authors propose “Generation-Evaluation Agreement” (GEA) as a novel validity criterion, which evaluates whether model-assigned scores accurately recover the intended proficiency levels embedded during item generation. The research implements a two-stage adaptive testing system grounded in skill decomposition, integrating item generation, response simulation, and automated scoring, and introduces the first quantitative metric for GEA. Empirical results reveal an overall GEA correlation of 0.698, with high consistency for grammar-related skills (r > 0.7) but near-zero agreement for design-oriented skills. Additionally, low-proficiency examinees near routing thresholds were systematically overestimated, underscoring the critical role of fine-grained scoring rubrics in enhancing assessment validity.

0 citationsRead paper

LLM-Upgraded Graph Reinforcement Learning for Carbon-Aware Job Scheduling in Smart Manufacturing

Dec 06, 2025

This work addresses the dynamic, sustainable, carbon-aware flexible job shop scheduling problem in smart manufacturing. We propose a graph reinforcement learning framework that integrates Graph Neural Networks (GNNs) with Large Language Models (LLMs). Through customized prompt engineering, the topological structure and semantic information of scheduling states are jointly encoded into high-quality embeddings; a multi-objective reward function guides deep reinforcement learning to jointly optimize makespan and carbon emissions. To our knowledge, this is the first study to incorporate LLMs into graph-based reinforcement learning for carbon-aware job shop scheduling, endowing the agent with intrinsic semantic understanding. Evaluated on both synthetic and public benchmark datasets, our method achieves an average 4.1% reduction in makespan (up to 12.2%) while significantly lowering carbon emissions, outperforming state-of-the-art algorithms in overall performance.

0 citationsRead paper

Machine Learning-Assisted Surrogate Modeling with Multi-Objective Optimization and Decision-Making of a Steam Methane Reforming Reactor

Jul 10, 2025

Steam methane reforming (SMR) reactor multi-objective optimization faces high computational cost and inherent trade-offs among conflicting objectives—e.g., methane conversion, hydrogen production rate, and CO₂ emissions. Method: This paper proposes an integrated framework comprising mechanistic modeling, data-driven surrogate modeling, multi-objective optimization, and decision-making support. An artificial neural network (ANN)-hybrid surrogate model replaces computationally expensive high-fidelity 1D fixed-bed simulations; non-dominated sorting genetic algorithm II (NSGA-II) efficiently computes the Pareto-optimal front; and a dual-criteria decision strategy—combining Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) and stochastic PROBID (sPROBID)—supports optimal operating condition selection. Contribution/Results: The surrogate model reduces average simulation time by 93.8%. The identified Pareto-optimal solution achieves methane conversion of 0.988, H₂ production of 3.335 mol/s, and CO₂ emissions of 0.781 mol/s—demonstrating substantial gains in optimization efficiency and engineering applicability.

0 citationsRead paper