# A reproducible method to generate multi-building, multi-climate HVAC operation datasets with a stochastic exploratory controller

## TL;DR

This paper introduces a reproducible method to generate HVAC operation datasets across different buildings and climates using a stochastic controller for diverse control scenarios.

## Contribution

A novel stochastic supervisory controller and a reproducible pipeline combining EnergyPlus and Modelica simulations for HVAC data generation.

## Key findings

- The method generates multi-year HVAC time series with diverse setpoint trajectories using a stochastic policy.
- Standardized datasets and code are released to support reproducible comparisons and transfer learning studies.

## Abstract

Building control research increasingly requires datasets that are reproducible, controllable, and rich in action–state coverage. We present a method to generate multi-year HVAC operation time series across heterogeneous buildings and climates using open-source building simulation frameworks, which expands control diversity using a deliberately stochastic supervisory controller. The workflow combines EnergyPlus-based simulation via Sinergym for multi-building/multi-climate “source” domains and Modelica-based simulation via BOPTEST for a distinct “target” domain to support transfer-learning evaluation and reproducible comparisons. Alongside the default rule-based controller (RBC), we implement a stochastic exploratory policy that interleaves stochastic drift, ramps, oscillations, jumps, and noisy holds to produce non-routine heating/cooling setpoint trajectories under operational bounds. The method produces standardized 15-minute multivariate time series including indoor temperature, outdoor weather, setpoints, and HVAC power, and releases both the datasets and the full code needed to reproduce or extend them.•Reproducible pipeline combining Sinergym (EnergyPlus) and BOPTEST (Modelica) under a common interface.•Stochastic HVAC supervisor that broadens setpoint distributions beyond standard schedules.•FAIR release of code + datasets to enable evaluation and reproducible comparisons, transfer learning, and robustness studies.

Reproducible pipeline combining Sinergym (EnergyPlus) and BOPTEST (Modelica) under a common interface.

Stochastic HVAC supervisor that broadens setpoint distributions beyond standard schedules.

FAIR release of code + datasets to enable evaluation and reproducible comparisons, transfer learning, and robustness studies.

Image, graphical abstract