Meval Methodology
An M&V method designed to estimate and verify energy savings from HVAC system upgrades, representing the evolution from research-stage code to an operational, practitioner-facing application.
Beyond Predictive Models
Unlike traditional approaches, the Meval tool does not rely exclusively on pure predictive models to estimate energy savings. Instead, it uses both pre- and post-retrofit data to infer the indoor activity levels that act as context for the HVAC system operation. Given the observed outdoor temperature and the inferred indoor activity levels, the model calculates energy savings by estimating the energy consumption that would have occurred under pre-retrofit efficiency and post-retrofit activity levels.
From an IPMVP perspective, this is equivalent to an IPMVP Option B and C model with an additional conditioning variable that captures operational intensity. However, while traditional approaches assume activity stability, Meval :
Explicitly verifies whether activity levels remain comparable before and after the upgrade
Offers a way to adapt to activity shifts so that energy savings attribution remains valid
Our Four-Step Modeling Process
Physics-Based Pre-Retrofit Model
A physics-based model is fitted on the pre-retrofit data. The model has parameters with physical meaning and captures the dynamics of energy consumption with respect to weather conditions along with seasonal variations of the baseload. The physics-based model is flexible enough to capture thermal inertia, free-cooling operation and capacity saturation effects.
Activity Level Estimation
Pre-retrofit activity levels are estimated as a hidden feature that explains variability around the physics-based model's predictions. Calendar features are used to regularize activity and remove noise by imposing a prior belief that similar calendar features map to similar activity levels more likely than not.
Post-Retrofit Model Transformation
A physics-based model is fitted on the post-retrofit data. Meval learns a transformation from pre-retrofit model parameters to post-retrofit ones that is regularized so different parameters require different evidence strength to get updated.
Aligned Activity Comparison
Post-retrofit activity levels are estimated using the pre-retrofit activity distribution as reference, ensuring "high activity" post-retrofit means the same operational intensity as "high activity" pre-retrofit.
Core Capabilities
Model Families
Meval blends predictive models with physics-based models. A novel method is applied to quantify the extent to which the assumptions behide the predictive models still hold after the upgrade.
Feature Engineering
Users define and type the data features before analysis, supporting sensor features for both indoor and outdoor conditions, and meter features linked to specific HVAC systems and thermal zones.
Uncertainty Quantification
Meval reports CV(RMSE) and NMBE metrics in an interactive dashboard. Uncertainty intervals are calculated using conformal prediction — a distribution-free statistical framework.
NRE Detection & Adjustment
Meval deals with non-routine events by supporting efficiency upgrades, activity shifts, and transient events with two causal paths that isolate retrofit-attributable savings from occupancy changes without full re-baselining.
Structured Web-Based Workflow
Execute M&V analyses following a standardized blueprint.
Model training and prediction histories are maintained, supporting auditability and longitudinal tracking. The tool operates at the building and HVAC-system level, with a structured hierarchy (buildings, zones, systems) that supports programmatic aggregation of savings across portfolios.
Try the Tool
The tool is freely accessible and was developed by HEBES Intelligence under the SRI2MARKET project (EU LIFE programme, Grant Agreement No 101077280).