GRID: Graph-based Reasoning for Intervention and Discovery in Built Environments

2025-09-19

AI Summary

Manual fault diagnosis in commercial building HVAC systems is time-consuming and error-prone, while existing data-driven methods identify only statistical correlations and fail to model underlying causal mechanisms. To address this, we propose the first three-stage causal discovery framework tailored for built environments. It integrates constraint-based causal search, neural structural equation modeling (Neural SEM), and domain language model–guided prior injection to automatically and interpretably recover causal structures from heterogeneous sensor data. Furthermore, we incorporate intervention scheduling optimization to enhance diagnostic accuracy and risk mitigation efficiency. Evaluated on six benchmarks, our method achieves F1 scores of 0.65–1.00; it perfectly recovers ground-truth causal graphs (F1 = 1.00) in three controlled experiments; and in a real-world ASHRAE case study, it attains an F1 score of 0.89 with low intervention cost and significant system risk reduction.

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

Manual HVAC fault diagnosis in commercial buildings takes 8-12 hours per incident and achieves only 60 percent diagnostic accuracy, reflecting analytics that stop at correlation instead of causation. To close this gap, we present GRID (Graph-based Reasoning for Intervention and Discovery), a three-stage causal discovery pipeline that combines constraint-based search, neural structural equation modeling, and language model priors to recover directed acyclic graphs from building sensor data. Across six benchmarks: synthetic rooms, EnergyPlus simulation, the ASHRAE Great Energy Predictor III dataset, and a live office testbed, GRID achieves F1 scores ranging from 0.65 to 1.00, with exact recovery (F1 = 1.00) in three controlled environments (Base, Hidden, Physical) and strong performance on real-world data (F1 = 0.89 on ASHRAE, 0.86 in noisy conditions). The method outperforms ten baseline approaches across all evaluation scenarios. Intervention scheduling achieves low operational impact in most scenarios (cost <= 0.026) while reducing risk metrics compared to baseline approaches. The framework integrates constraint-based methods, neural architectures, and domain-specific language model prompts to address the observational-causal gap in building analytics.

Problem

• Manual HVAC fault diagnosis is slow and inaccurate in commercial buildings
• Existing analytics stop at correlation rather than discovering causal relationships
• There is an observational-causal gap in building sensor data analysis

Innovation

• Combines constraint-based search with neural modeling
• Integrates language model priors for causal discovery
• Uses graph-based reasoning for intervention scheduling