学术成果与研究发表

Dynamic Uncertain Causality Graph Theory and Applications

Author: 冯宇

Abstract: Proposed Dynamic Uncertain Causality Graph (DUCG) model for handling uncertainty causal relationship modeling and inference in complex systems. The model uses virtual random events to express uncertain causal mechanisms and employs weighted logic reasoning algorithms for efficient dynamic inference.

Keywords: Probabilistic Graphical Models, Causal Inference, Uncertainty, Dynamic Systems

Applications: Legal reasoning, Financial risk control, Software reliability, Cybersecurity

Status: Under research

Game AI Algorithm Optimization Based on Deep Reinforcement Learning

Author: 冯宇

Abstract: Researched AlphaZero algorithm applications in board games. Through improved neural network architecture and training strategies, significantly enhanced AI performance in Gomoku. Experimental results show the improved algorithm reaches professional level in shorter time.

Keywords: Deep Reinforcement Learning, AlphaZero, Monte Carlo Tree Search, Game AI

Contributions: Network architecture optimization, Training efficiency improvement, Strategy evaluation enhancement

Status: Preparing for submission

DURCA: A Dynamic Causal Reasoning-based Large-scale Microservices Root Cause Analysis Framework

Authors: Chunling Dong, Yu Feng

Affiliation: School of Computer and Cyberspace Security, Communication University of China, Beijing 100024

Status: Under Submission

Abstract: DURCA addresses the dual challenges of second-level localization and black-box fault propagation in microservices. Built upon Dynamic Uncertain Causality Graph (DUCG), it constructs a temporal-evolving Cubic-DUCFG to unify multi-modal data (logs, metrics, traces) and explicitly model temporal dependencies, causal delays and dynamic negative feedback. A weighted logic root cause analysis (WLRCA) decouples domain knowledge-based causal strength and data-driven uncertainty for efficient and interpretable posterior reasoning; a dynamic inference pruning strategy adapts the graph with real-time alerts to significantly reduce computation with linear-time complexity. In simulated temporal propagation scenarios, DURCA achieves 99.9% Top-5 root cause recall (+4.7% over SOTA) with 10 ms average diagnosis latency (30x faster than lightweight baselines like NSigma).

Keywords: Dynamic Uncertain Causality Graph, Root Cause Analysis, Fault Prediction, Causal Inference

The Method of Cubic DUCG for Dynamic Causal Modeling and Fuzzy Temporal Causality Reasoning in Industrial Fault Diagnosis

Authors: Chunling Dong (Corresponding Author), Yu Feng

Journal: Reliability Engineering & System Safety (JRESS), Research Paper

Manuscript No.: JRESS-D-24-01378

Status: Under Review

Keywords: Dynamics and uncertainties; fault diagnosis; dynamic reasoning; fuzzy temporal causality reasoning; fuzzy evidence

Abstract: We propose Cubic DUCG for industrial fault diagnosis under time-varying operating conditions and fuzzy uncertainty in data and knowledge. A generative cubic causal modeling scheme and a fuzzy temporal causality reasoning algorithm are introduced, together with weighted logic inference, causality simplification and EELA strategy to reduce modeling and reasoning complexities. Comparative and simulator-based experiments (nuclear power plant) validate accuracy, interpretability and efficiency in large-scale dynamic systems.

Survey and Practice of Machine Learning Explainability Methods

Type: Technical Report

Content: Comprehensive survey of mainstream ML explainability methods including SHAP, LIME, Integrated Gradients, with practical case studies demonstrating applications across different domains.

Case Studies: Financial risk model explanation, Medical diagnosis decision support, Recommendation system transparency

Depth: Theoretical analysis + Code implementation + Case studies