We are a data-driven company. We use energy audit data about a building to automatically create a physics-based model of the building. Then, this model is simulated to generate a low-fidelity dataset that augments the high-fidelity dataset of the actual energy consumption. In this way, we generate datasets with enough information to perform counterfactual predictions and estimate the expected impact of different energy efficiency measures.
This is a challenging task and we have found probabilistic programming to be the most useful part of our toolset. The fusion of multifidelity data relies on probabilistic modeling techniques, and probabilistic and do-algebra approaches are used for the counterfactual predictions.
We continuously learn and document the ability of our methods to accurately evaluate the uncertainty in the energy savings from an energy efficiency measure plan, as well as the impact of the type and granularity of the available data on both the predicted uncertainty and the actual error in the estimation of the expected energy savings.
To support the operational stage of our transaction model, we rely on our energy efficiency metering and analytics service that is able to:
Compare in real time the actual energy consumption with the initial expectations that govern the agreements with the ESCOs;
Make the data on the realized energy savings transparent and easily accessible to all actors involved in the transaction model;
Monitor the expected value of an agreement at the portfolio level, since the energy efficiency measures in the buildings of the portfolio will come in effect in a stage-wise fashion during the period between the signing of the agreements and the actual delivery of the energy savings;
Timely identify spatial or temporal trends of degradation of the agreement’s value.