Meeting peak energy demand reliably and at less cost is a critical challenge for utilities around the world. Demand Response programs, which incentivize customers to reduce energy when called upon, have emerged as a strategic tool to help utilities offset the costs of capital upgrades. Yet managing participation in these voluntary programs – understanding who will reliably shed load when called upon – is difficult.

To manage its Demand Response portfolio and predict event-level reductions, this natural gas and electric utility internally developed an Excel-based forecasting tool. While helpful, this tool would provide it with program-level event impact estimates that were typically off by 50 – 70% and even less visibility into impacts at individual Sub-Load Aggregation Points (SUBLAPs).

The utility is required to report daily to the ISO on how many megawatts (MW) of power it has available in its demand response programs. Inaccurate forecasts can lead to fines, but more importantly undermine confidence in the programs themselves. At the utility, the ultimate goal is to manage demand response as a “virtual power plant,” a large, dependable and cost-effective energy store that can be deployed where and when needed to help ease demand on the grid and ensure 24x7x365 power for all customers.

Making Data Useful: A “Solver” for Demand Response

TROVE, the predictive data science company, proposed developing the operational Demand Response forecast differently, building a model that would not look at annual program participation averages, but rather at individual customer behavior.

With its Demand Response Solver, which combines customer behavioral models, customer-specific machine learning (bringing customer smart meter data into the modeling mix) and a proprietary store of 3rd party consumer data, TROVE developed an individual forecast for every customer enrolled in the demand response program. TROVE then aggregated those forecasts up into a new program-level view.

This new approach has proven so successful it has been adopted across several of the utility’s demand response programs, which currently enroll 350,000 customers. Leveraging the scalability of the TROVE Predictive Data Science Platform, the utility has moved from running one demand response forecast daily to running 500,000 in 30 minutes.

The Benefits of Granularity: Precision, Resolution and Operational Agility

By using the TROVE Platform to generate customer-level forecasts, the utility has been able to deliver precise daily program forecasts to the ISO. It is also using the Platform to aggregate and analyze this customer-level information in different ways beneficial to its operations.

For example, the TROVE Platform can aggregate the forecasts to specific load control areas or even to individual circuits in the distribution network. With this higher resolution view into where load reductions are coming from, the utility has gained more options when calling an event. Now, rather than calling an event for an entire service area, it can call one for a particular substation to meet the same reduction, or avoid calling one altogether, putting in place a switching plan instead to offload demand.

More accurate forecasting, enabled by predictive data science, has made the utility’s demand response programs more reliable and more nimble, bringing them one step closer to becoming the strategic asset – i.e. the virtual power plant – every utility dreams of owning.