What’s going on in there? Interpretable Machine Learning for Deeper Energy Savings
Efficiency Vermont engineers provide technical assistance to building energy managers by creating energy models of customer facilities. These models use historical consumption data to predict future energy use from equipment and processes, and the drivers of that energy consumption. Predicting, however, is not the same as explaining. Achieving these outcomes requires demonstrating that these models provide consistently superior performance than current approaches, and that Efficiency Vermont can communicate the model outputs to energy modelers, evaluators, and customers, in simple but more insightful ways than the current visualizations. Efficiency Vermont researchers piloted a hybrid model that supports the use of cutting-edge algorithms and the design of insightful, informative visualizations. The researchers produced predictions that are highly accurate, intuitive, and easy to understand.