Utilities have long sought to gain greater control over customer loads for the simple reason that the cheapest, most environmentally friendly megawatt is the one you don’t have to generate. Industrial customers have taken advantage of favorable rates associated with load control programs for decades, but despite the technology being available, programs for residential customers have been slow to follow.
Beyond overcoming consumer anxiety over relinquishing control of the thermostat, utilities have lacked the means to make participation in load control programs transparent to the user. ABB undertook a research project to address this issue and the result is EnergySolver, a plug-and-play, price-responsive residential load control solution.
EnergySolver works by creating an automated virtual energy market within the home. For each device, whether generation (via grid power or on-site distributed energy resources) or load (appliances, lighting, HVAC, etc.) an “agent” is created to represent the goals, constraints, and behavior of the device. Agents communicate with a “central solver,” which settles the market to achieve some global goal (e.g., reduce cost, maximize comfort, maximize energy self-consumption).
EnergySolver can model a variety of loads found in residential locations including:
- Solar PV
- Energy storage
- Continuously controllable (e.g., lights)
- Scheduled (e.g., heat)
- Isolated thermal (e.g., water heater)
- Sheddable plug (e.g., dishwasher)
- HVAC (with single zone)
Different central solver algorithms can be used for different contexts (e.g., real time pricing, power export limiting, peak power limiting, etc.). Even if the context is not related to overall cost savings, the central solver operates in terms of energy price, which can be used to influence the agents to change their device schedules. Price, then, acts as a common language that allows agents and market to communicate under a variety of scenarios.
The devices, automation system, and interoperability layer for such a system are nearing commercialization. ABB’s work focused specifically on developing the algorithms and software implementation to allow these components to function in an “energy market” within the home. Essential to this was making the system as close to plug-and-play as possible to minimize effort from the homeowner and in turn maximize the likelihood they would use it.
How it works
EnergySolver uses an incentive-feedback optimization (IFO) approach to arrive at a solution using one of two processes (see fig. 1). The central solver sends transactive incentive signals (TIS), essentially a price forecast, to resource agents that in turn self-optimize—with no knowledge of other resources—to generate transactive feedback signals that are sent back to the central solver. The central solver sums all forecasts from the agents, checks the total against global constraints (e.g., avoiding export to the grid), and then adjusts its TIS in the intervals (the system works in 5min increments) where those constraints are violated. For example, the central solver might raise the power price when net power exceeds the threshold.
This process repeats until all constraints are met or a pre-determined maximum number of iterations is reached. The central solver then sends the results to resource agents as final dispatch instructions.
It’s important to note that agents are only responsible for their own resources; there are no dependencies between resources represented in the market model. New agents come from a library of agents for different common resources and can be added automatically. The agents also capture user preferences such as desired temperature, one of the parameters described by the HVAC agent. In addition, there is a single global “comfort” parameter that determines the weighting of user preferences versus cost savings in the EnergySolver algorithm.
Assumptions and processes
EnergySolver operates on a variety of assumptions. Loads must be controllable, individually metered and able to transmit data. With machine learning techniques, it is possible to build predictive models of individual resource behavior that can be updated periodically.
Inside the central solver, an energy management algorithm uses the resource models, user preferences, and external data (e.g., weather forecasts, electricity prices, DR commands) to develop optimal resource schedules that minimize energy usage/costs and fulfill user preferences. The communications abstraction layer enables interoperability between the algorithms and automated devices from different vendors, allowing for plug-and-play functionality. An automation system controls the devices themselves.
Creating a virtual energy market as a mechanism for intelligent load control has tremendous potential, with benefits accruing not only to the user but to the utility as well. From the utility’s perspective, a system like EnergySolver would enable high penetrations of DERs and renewables, defer investments in new generating capacity, reduce reliance on expensive wholesale spot markets and even generate revenue by potentially offering services up to the transmission market. It likely would encourage stronger participation rates among residential customers, too, with simple and low-cost setup.
For residential utility customers, participating in demand response or other load control programs would be completely transparent once user preferences were set. They would be able to take advantage of time-of-use pricing (which is becoming mandatory in some jurisdictions) and could avoid demand charges with no additional effort.
EnergySolver is a framework for interoperable load control algorithms that operate via a simple, common market interface that enables interoperability among all resources and reduces commissioning effort. ABB has prototyped the algorithm in Simulink and demonstrated moderate to very large benefits depending on the comfort preference. To wrap up the project, the ABB team is building a hardware demonstration that includes REACT, ABB free@home components, and a Mikado algorithm implementation.
Further investigation of the system’s potential would allow the development of new use cases that require immediate action (e.g., islanding from the grid). More study of the execution interval is also warranted to explain how the pace of price fluctuations improve participation of some resources over others (i.e., rapid fluctuations improve energy storage participation while gradual ones improve participation for thermal resources). With more work, EnergySolver could prove to be a useful tool for utilities to realize the potential of demand side resources while delivering additional value to residential customers.