Large Language Model-Based Interpretable Machine Learning Control in Building Energy Systems

Aug 3, 2026

Upload by Hazel

Machine Learning Control (MLC) in HVAC systems is constrained by limited transparency, making it difficult for users and modelers to understand inference behavior and resulting decisions. This work studies Interpretable Machine Learning (IML) to enhance model transparency and improve the credibility of MLC in practical HVAC applications. It introduces an interpretable framework that merges Shapley values with large language model in-context learning to produce human-understandable narratives of rule and non-data-driven elements. A demand-response precooling case study in a virtual testbed validates feasibility and explains generated control signals.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages19

Reading Time48 min

AccessFree to download

Highlights

Abstract

Keywords

1 Introduction

Building energy use and the role of control

Machine learning control in building systems

Interpretable machine learning as a solution

Shapley values and SHAP

LIME and local explanations

Large Language Model-Based Interpretable Machine Learning Control in Building Energy Systems Liang Zhang1,2 *, Zhelun Chen3 1. University of Arizona, Tucson, Arizona, U.S. 2. National Renewable Energy Laboratory, Golden, Colorado, U.S. 3. Drexel University, Philadelphia, Pennsylvania, U.S. *Corresponding Author: [liangzhang1@arizona.edu](liangzhang1@arizona.edu) Highlights • Innovative interpretable machine learning framework for machine learning control • Shapley values and large language models are combined for improved interpretability • Case study demonstrates interpretable control processes in demand response events • Bridging trust gap in machine learning control usage for building energy management Abstract The potential of Machine Learning Control (MLC) in HVAC systems is hindered by its opaque nature and inference mechanisms, which is challenging for users and modelers to fully comprehend, ultimately leading to a lack of trust in MLC-based decision-making. To address this challenge, this paper investigates and explores Interpretable Machine Learning (IML), a branch of Machine Learning (ML) that enhances transparency and understanding of models and their inferences, to improve the credibility of MLC and its industrial application in HVAC systems. Specifically, we developed an innovative framework that combines the principles of Shapley values and the incontext learning feature of Large Language Models (LLMs) . While the Shapley values are instrumental in dissecting the contributions of various features in ML models, LLM provides an in-depth understanding of thenon-data-driven or rule-based elements in MLC; combining them, LLM further packages these insights into a coherent, human-understandable narrative. The paper presents a case study to demonstrate the feasibility of the developed IML framework for model predictive control-based precooling under demand response events in a virtual testbed. The results indicate that the developed framework generates and explains the control signals in accordance with the rule-based rationale. Keywords Building control, machine learning control, interpretable machine learning, Shapley value, large language model 1 Introduction Building operations account for approximately 30% of the world's total energy consumption and 26% of global greenhouse gas emissions (EIA 2023) . As predicted by the Department of Energy [1], the US needs advanced technologies to achieve substantial decarbonization and transition to clean energy by 2050, considering the projected increases in both population and business activities. In this endeavor, the control and optimization of building operations emerge as a pivotal aspect. The precise control and adjustment of various building systems based on real-time data and conditions helps to eliminate wasteful energy consumption, prevent system conflicts, and ensure that energy-consuming equipment operates at optimal levels [2] . Given the surge in building automation systems adoption and the influx of a large amount of data, the industry's pivot to machine learning control (MLC) [3, 4] has become imperative. MLC has the capabilities to handle complex nonlinear systems where traditional linear control methods fall short [5, 6] . Machine Learning Control (MLC) applications in building energy systems have demonstrated great potential and can be categorized into different categories based on their control strategies [4] . Machine Learning-based Model Predictive Control (MLbased MPC) utilizes ML models for system identification. For instance, Cole et al. [7] employed a neural network (NN) to predict building energy load, and Kim et al. [8] used NN to forecast indoor air temperature, energy consumption, and daylight illuminance. In contrast, ML-based model-free control, such as Reinforcement Learning (RL), directly learns control actions from environmental interactions. Examples include Wei et al. [9], who applied RL for zone airflow con

FAQs

Why is interpretability a challenge for machine learning control in HVAC systems?

MLC decisions are often driven by opaque inference mechanisms that are hard for users and modelers to fully comprehend, which reduces trust in the resulting control actions.

What is the proposed interpretable machine learning framework?

The framework combines Shapley values with large language model in-context learning so that feature contributions and non-data-driven or rule-based elements are transformed into a coherent, human-understandable explanation.

How is the framework validated in the paper?

A case study demonstrates interpretable control processes for model predictive control-based precooling under demand response events using a virtual testbed, showing that control signals can be generated and explained according to rule-based rationale.

Aug 3, 2026

Upload by Hazel