TLDR: InsightBuild is a new two-stage framework that combines causal inference with a fine-tuned large language model (LLM) to provide human-readable, causal explanations for anomalous energy usage in smart buildings. It first identifies cause-and-effect relationships from building data and then uses an LLM to generate clear, actionable explanations, helping facility managers diagnose and mitigate energy inefficiencies. Evaluated on real-world datasets, InsightBuild significantly improves explanation accuracy and user satisfaction.
Smart buildings are equipped with countless sensors that collect vast amounts of data on everything from temperature and CO2 levels to occupancy and HVAC settings. While this data is crucial for optimizing energy use, facility managers often struggle to understand why sudden spikes or anomalies in energy consumption occur. Traditional dashboards might show the data, but they don’t explain the ‘why’ behind an event, leading to generic alerts that aren’t very helpful.
This is where InsightBuild comes in. Researchers have developed a new two-stage framework designed to provide clear, human-readable explanations for anomalous energy usage in smart buildings. The goal is to help facility managers quickly diagnose and fix energy inefficiencies.
How InsightBuild Works
InsightBuild operates in two main stages. First, it uses a lightweight causal inference module. This module applies statistical tests, specifically Granger causality tests, and structural causal discovery techniques to building data. It looks at various factors like temperature, HVAC settings, and occupancy to identify direct cause-and-effect relationships. For example, it can determine if a sudden increase in occupancy directly led to a rise in energy consumption.
The second stage involves a fine-tuned large language model (LLM). Once the causal inference module identifies the likely causes of an energy anomaly, these detected causal relationships are fed into the LLM. This LLM has been specially trained on a dataset of sensor-level causes paired with expert-written textual explanations. This training enables the LLM to translate complex causal relationships into concise, actionable explanations that facility managers can easily understand.
For instance, if the system detects that a spike in energy was caused by increased occupancy and a rise in zone temperature, the LLM might generate an explanation like: “The surge in energy usage was chiefly due to a sudden influx of occupants in Zone 3, leading to elevated zone temperature. As a result, the chiller ran at higher capacity. Adjusting the setpoint or redistributing occupants could prevent such spikes.”
Also Read:
- Unpacking AI Decisions: A New Logic for Contrastive Explanations
- ChainEdit: Enhancing LLM Knowledge Consistency Through Logical Rule Propagation
Real-World Validation
InsightBuild was evaluated using two real-world datasets: the Google Smart Buildings dataset (2017–2022) and the Berkeley Office Building dataset (2018–2020). The system’s performance was measured based on explanation accuracy, precision and recall of the top causes, and expert satisfaction. The results showed that InsightBuild significantly outperformed traditional rule-based systems and even vanilla LLMs in providing accurate and helpful explanations. Facility managers rated InsightBuild’s explanations as clearer and more practical, demonstrating its superior ability to assist in diagnosing and mitigating energy inefficiencies.
The research highlights that combining explicit causal discovery with LLM-based natural language generation is key to producing precise and trustworthy explanations, avoiding the ‘hallucinations’ that purely LLM-based systems might produce without proper causal grounding. This innovative approach ensures that the explanations are rooted in the actual physical system of the building.
For more in-depth information, you can read the full research paper: InsightBuild: LLM-Powered Causal Reasoning in Smart Building Systems.
