Demand response (DR) is one of the fastest growing and most enthusiastically received programs in the smart-grid space, with Barclays Capital predicting it could grow to a $20 billion market by 2020. But to reach that level, both the market for participating in DR and the systems for executing it need to become more efficient.
DR is a series of programs typically sponsored by the local utility, independent system operator (ISO) or retail energy provider. The most common DR structure pays electric customers to be on call to reduce electricity usage when the grid is being pushed to its capacity limits. In one form of DR, independent (i.e., non-utility owned) curtailment-service providers (CSPs) work with electric customers to curtail energy usage during times of peak demand, providing a valuable service to grid operators—one that can prevent blackouts and reduce the need for building new power plants. In return for this service, operators pay a set price per megawatt to CSPs, who in turn pass on a portion of this payment to their customers.
In the early evolution of the DR market, CSPs drove the process, selling a new and opaque service at a low financial reward to a relatively inexperienced customer base. Customers complied, treating DR payments as found money. Yet, they had little visibility into what a fair share of their payout should be or even knowing there might be more revenue available.
Those days should be over. Companies ready to participate in DR programs now have access to a market with transparency, price discovery and liquidity, all of which are keys to both helping the market scale and enabling businesses to convert their curtailable loads into strategic assets. Auctions provide the path into the DR market. No longer the exclusive tool of ISOs to allot load-relief capacity to DR providers, competitive DR auctions are moving downstream (i.e., closer to the customer) to dramatically re-shape how individual participants in DR programs can find the right service provider and maximize their share of DR revenue (see Figure 1).
Such markets are cultivating DR’s growth as a tradeable resource that benefits both utilities and customers. But realizing the true potential of DR requires utilities to apply today’s technology solutions and program structures—and to base their strategies on actual customer behavior and comfort, rather than on yesterday’s outdated assumptions about centralized load control.
While improving market efficiency is key to scaling the DR opportunity, advancing the systems and controls for enacting load curtailment is crucial to the program’s successful execution. Simply put, reaping the financial benefits of DR depends on being able to respond effectively to DR events and deliver the controlled load when required. Fortunately, breakthroughs in the technology of load reduction are meeting this challenge.
• Load Reduction: One technical issue faced by potential participants and program sponsors is that the cost of implementing load reduction sometimes exceeds its market value. Another issue is that from the customer’s perspective, the actual load reduction delivered by program participants varies greatly from customer to customer. These issues cause a significant inequity among program participants both from their value to the program and from the program’s impact on individual comfort, convenience, or business impact.
Consider a scenario where the DR provider is the electric utility and the program is residentially focused. A utility picks a DR technology vendor and deploys a pilot to test the performance (i.e., load reduction) of the load-control system. The results show 0.3-kW load reduction for each customer unit vs. the 1 kW needed for a cost-effective solution, so the utility then abandons the program on lack of financial merit.
Even though the data shows the load reduction was inadequate, no one questions the methodology or strategy being used to control the load. Often the control strategy being deployed isn’t examined in the context of the local appliance load conditions. Somehow, it became ingrained in the industry that only one kind of control schedule can be implemented with residential heating, ventilating, and air conditioning (HVAC) control—in this case the magic 50-percent duty cycle. The reality is that that the 50-percent control strategy being tested is only applicable to the utility that developed it, and it probably was developed more than 20 years ago.
When this real-life scenario occurred at a large West Coast electric utility, industry experts suggested that another control strategy could have delivered the necessary load reduction. That concept was rejected, however, because the tested strategy was the one commonly used throughout the industry. The 50-percent duty cycle approach didn’t work, however, because the typical residential HVAC system was hitting the grid with 2.3 kW of demand, not 4 kW as assumed in the test (see “HVAC Load Control—A Flawed Test”). It just happened that the local HVAC industry had a habit of over-sizing their installations to the point where the natural duty cycle of those systems was relatively low even during hot summer afternoons. A fixed duty cycle, based on program assumptions, rather than on local appliance run-time statistics, was doomed to failure. A correct control strategy that accounts for true demand, easily could deliver at least 1 kW of load reduction, while still allowing the HVAC system to deliver significant cooling.
• Adaptive Control: One of the methods employed today to improve load reductions is sometimes known as adaptive control, which has been tested and revised many times over the last 20 years. The first version was called the smart duty cycler, and was tested in the early 1980s at a large southern utility. The basic concept never has changed—install intelligence in the device connected to the load and program that device to forecast the next hour’s natural duty cycle. Once that forecast is made, it’s easy to calculate what the new run time of the appliance needs to be in order to deliver the desired load reduction.
Early tests were performed with a very basic level of intelligence built in to the units—microprocessors were in their infancy, at least in these applications—and as a result forecasts were very crude. The end result was that those original units tended to occasionally over control because they couldn’t adapt in real time to intermittent variations in appliance usage. In one case, the customer had turned the unit off, and, when control started, the device was trying to reduce the already 100-percent off-duty cycle even further. When the customer set the unit to run, the control device kept the unit off completely.
Newer versions of this solution are becoming better at predicting the next hour’s natural run time and, as a result, now are being deployed and operated at many utilities. In every case, the utilities are reporting improved load reduction because they are capturing the oversized HVAC systems, as in the example above, and forcing them into duty cycles that actually deliver load reductions. There’s still the risk of over control, and each vendor’s solution offers its own method of mitigating that risk. Some of these smart-control algorithms are moving to the point where the utility simply can send a command to the control device and direct it to determine exactly what the control strategy should be for a specified load-reduction request. From a control-optimization perspective, that is the Holy Grail of direct load control. It maximizes the average load reduction per participant, while moving closer to equalizing each customer’s load-reduction contribution.
Generally speaking, each solution attempts to predict how much the customer’s HVAC would have run during the control period if it were uncontrolled. The controller then implements a control strategy (i.e., cycle strategy) in a manner that delivers the desired demand reduction. Without this technique, many over-sized central air conditioning systems would not yield significant or appropriate load reductions.
Some solutions make this control strategy adjustment based on compressor run-time data collected the hour prior to the start of the control period. The previous hour’s duty cycle is used to forecast the next hour’s duty cycle. From that forecast, a control-shed percentage based on that predicted duty cycle is implemented. For example, if the previous hour’s natural duty cycle was 50 percent and the new shed cycle was 50 percent, the resulting effective control strategy would be off 75 percent of the time (i.e., 50-percent reduction of the 50-percent natural duty cycle). If the unit’s connected load was rated at 4 kW, the load reduction would be 1 kW—assuming the duty cycle forecast for the next hour (e.g., 50 percent) was accurate in the first place.
It’s obvious that there are a lot more variables than just the previous hour’s duty cycle that can be used to predict the next hour’s natural duty cycle. Site conditions (e.g., sun angle, shading, time of day, humidity, outside and inside temperature, etc.) as well as a longer observation period all can be influence. As these algorithms become more sophisticated, they’ll become more accurate and more efficient. In today’s world, it’s still best to have some default constraints programmed into the controller in case the algorithm calculates what would appear to be an extreme control strategy, like the 100-percent off strategy that was implemented in that original test.
• Customer Comfort Impacts: Although the results of adaptive control strategies do deliver increased load reductions, there’s no mechanism in them to help levelize comfort impacts. Ensuring comfort is an important part of the overall opportunity to increase equity, fairness and customer acceptance across program participants.
Obviously, it’s much easier for a customer who has a 10-kW HVAC system to give up 1 kW of load than it is for a customer who has a 3-kW HVAC system. If they’re both driven to deliver 1 kW, there would be a significant discrepancy in customer comfort impacts between them. Also, customers who have multiple zones of HVAC might be more willing to allow some zones to be curtailed more aggressively than others—as long as the zones are physically isolated.
Considering that fact, newer versions of the so-called adaptive algorithms now are being designed with the capability of being customized by downloading customer-specific parameters that will go even further in minimizing comfort impact variations, while simultaneously managing the load reductions being delivered from various size installations. By providing the algorithm with parameters such as the HVAC connected load (kW), the control strategy can be optimized to not only guarantee that all participants contribute load reductions, but that customers’ relative BTU reductions per hour also are considered. This optimization example is one more method of increasing equity amongst program participants, which in the long run will help insure the program’s overall success.
Most discussions are on the assumption that control devices directly manage the running characteristics and duty cycle of HVAC systems. But another variant that’s now common is the so-called smart thermostat, which allows the option to implement either duty-cycle control or temperature setbacks.
Both approaches accomplish load reductions, but aren’t tied to the same control methodology. For a temperature setback, the utility effectively is reducing the natural run time of the unit, except now it’s hardwiring that duty cycle to a temperature. This can be attractive to the customer because it offers a clearly defined level of comfort impact and guarantees it across all participants (i.e., everybody’s setpoint is increased by 3 degrees). This will occur at the expense of levelized load reduction, but in aggregate the program still will deliver an average load reduction per participant that meets the cost-effectiveness test. Most utilities, having a choice, will let load reductions vary from customer to customer, while opting for a levelized comfort impact. The control strategy—temperature setback in this case—still must be modulated by the load-control system to meet the required aggregate load-reduction values, but at least comfort impacts are levelized across all program participants.
The opportunity to use DR for delivering some of the ancillary generation services required to operate a utility has been promoted for years. Efforts to deliver under-frequency response were tested in the early 1980s, but were shown to be too slow. Instead, DR was used to provide load reductions where real-time operating requirements were not so rigorous, such as supplemental or operating reserves, which typically require that the DR system respond in less than 10 minutes. Now things have changed.
In recent years, there has been a lot of press associated with development of smart appliances, where sensors are installed within the appliance in order to enable it to respond to certain grid events—like under-frequency. This application of DR continues to hold great promise, especially as home area networks become more prevalent in residences so that these appliances can become active participants in a utility’s under-frequency strategy. In the meantime, the utility can it retrofi its existing customer appliances with DR equipment that has under-frequency response built in.
The first example of using direct load-control equipment with built-in under-frequency sensing was in Indiana where similar DR systems were deployed by two large Midwest electric utilities. Those DR systems are still being deployed today, and the control devices contain the ability to implement autonomous under-frequency response. The key issue in the development of these systems was to verify that the under-frequency sensing and response (i.e., shedding of the connected appliance, in this case residential HVAC systems) was fast enough to win the race to shed that would exist between the DR equipment and the existing under-frequency relays already installed and operational in substations.
To ensure that the DR equipment shed fast enough to win the race, the DR shed logic included setting the appliance under-frequency trip point to operate at a significantly higher frequency than traditional under-frequency feeder control. A typical set point was 59.8 Hz. The other component was to verify that the under-frequency sensing and shedding within the DR equipment was fast enough to compete with traditional under-frequency shedding schemes. Testing with an industry standard event recorder verified that the DR equipment could sense the frequency shift and shed its connected load—including the A/C compressor contactor drop out time—within approximately 11 cycles or 183.3 milliseconds, thereby qualifying it as a true under-frequency resource. The other key specifications designed into this solution were that the under-frequency sensing would be able to be armed and disarmed remotely by the system dispatcher, as well as remotely setting the under-frequency trip point. This capability was tested and verified, and those devices now are being deployed as an accessory application of the DR system. The system also contains the ability to restore load automatically after the frequency has returned to normal, but that restoration time intentionally was delayed by a utility’s configurable time period, so that those affected loads wouldn’t reconnect to the grid at an inappropriate time. The normal operating procedure would be for the dispatcher to enable the load restoration when the system was stable enough to accept the load.
The reason for deploying this type of under-frequency capability was to clearly demonstrate the ability to use DR systems for the provision of true frequency responsive type of spinning reserves that can supplement or replace the governor response of generators and diminish the need for the under-frequency shed of feeders. DR systems often have been advertised as providing spinning reserves, but almost all of those systems don’t possess this under-frequency sensing capability. Without autonomous sensing and response, the DR systems can’t provide frequency responsive spinning reserves, because they otherwise would depend on the broadcast of a shed command from the DR head-end. The time it would take to sense under-frequency and send a shed command through the system would be too long to meet under-frequency response requirements.
The other type of reserves that are commonly supplied by DR often are referred to as supplemental reserves or daily operating reserves, which are typically dispatcher initiated and have response times in the order of 10 minutes, which easily can be met by existing DR systems. This type of DR application provides significantly more capacity response than regular DR events, because under these emergency conditions, the loads are shed and not duty cycled, so the available load reduction is the full value of the actual appliance diversified demand at the time of the event. This value easily could be two to three times the load reduction being delivered during normal load-reduction events.
When DR is used for under-frequency response, the customer loads that have been curtailed likely will have been off for an extended period of time (i.e., 15 to 30 minutes is typical), which means that when they are turned back on, the duty cycle will be very close to 100 percent, creating a load that might be as high as four times the normal appliance diversified demand. Turning all the devices on at once and at the wrong time very likely would create its own secondary grid event. This risk can be totally avoided by designing the system to prevent unrestrained automatic load restoration after an under-frequency event. Automatic load restoration is allowed only after a significant (i.e., configurable) time period and is only enabled as a precaution against operator failure to manually release these loads after the under-frequency event has passed.
Full deployment of these technological advances within the DR systems can add significantly to the value and reliability of DR loads both in the DR marketplace and with the utility-owned programs. They specifically will improve customer acceptance, participation and their persistence over time—a critical component to the long-term success of any DR program.
There are three critical success factors with any DR program: customer participation, technology effectiveness, and the delivery of the required load reduction at the time it’s needed. Systematic technological approaches will contribute significantly toward each of those success factors.