Fixing the Climate, page 18
TWO WAVES OF BATTERY DEPLOYMENT
The big push to utilize batteries across the grid began in the 2010s. The first wave of efforts focused on inducing the procurement of the emerging technology by utilities—encouraging them to experiment with the installation of many different kinds of devices in many different settings, thereby driving down the costs and learning the value of various uses. That uptake of devices has been faster than expected, and efforts, still in the early stages, have shifted in a second wave to creating a regulatory framework that encourages a more efficient deployment of batteries yet safeguards the integrity of the network.
The instrument for the first wave had its origins in a 2010 law mandating CPUC to create an energy storage program to support the integration of renewables.41 The law, only a few pages long, set no targets. Instead it outlined a process by which CPUC would develop statewide goals through local planning. Each utility was required to procure storage systems and document how these storage devices functioned when connected to the grid.42 This kind of technology forcing under uncertainty recalls CARB’s transformation of the California vehicle market in the 1990s, discussed in chapter 4. The big difference is that in the case of CARB, vehicles did not interact with each other or the road network (apart from the need for charging stations), and thus CARB could concentrate more exclusively on the technological frontier and maximizing the deployment of devices on the road. In the CPUC battery case, by contrast, the storage devices and grids very much interact with their context. More devices, alone, would not cause a technological revolution.
Like CARB, CPUC began with an exploratory evaluation of possible uses, focusing on storage systems generally, not just the dominant lithium chemistry. Then (and now) solid-state lithium batteries were thought to be the technological winner, but better rivals might also thrive if given a chance, such as flow batteries that pumped liquid electrolytes through the device. From there, again like CARB, it set a provisional goal (a total procurement of 1,325 megawatts by 2024). That top-level goal included planning targets for each of the state’s three utilities, and subtargets for the portion of energy storage systems that each utility would deploy on the medium-voltage transmission system, lower-voltage distribution system, and at customer sites.43 This approach was designed to ensure maximum efforts at contextualization: each utility would be required to identify novel projects with many different use cases, technologies, and voltages.44 To underscore the centrality of novelty to the program, CPUC also expressly prohibited the inclusion of any projects that used the one electricity storage technology that was already technologically mature: large pumped hydro.45 In addition, CPUC required that the utilities contract at least half of their battery projects from independent suppliers. CPUC was creating an industry of the future, and feared that industries of the past (regulated utilities) would not be creative enough to understand and demonstrate creative new battery applications—especially where success might eat away at the traditional regulated utility model.46 A challenge in creating these industries of the future, of course, was that they would be deploying devices on grids controlled by the incumbents.
CPUC planned procurement rounds every two years, starting in 2014, so that the rules and choices in each round could be updated with local information learned through early procurement. In parallel, CPUC required a comprehensive evaluation of its whole battery storage program every three years, starting in 2016, so that “as more experience is gained … lessons can be applied.”47 Just as with the Montreal Protocol’s TOCs that evaluated exemptions for critical uses of ODS, discussed in chapter 2, CPUC created a compliance safety valve: a utility could delay meeting up to 80 percent of its planning target by showing that under its current circumstances, the target was for the moment infeasible.48
As CPUC turned from planning to implementation, information from the local levels often flowed in much faster than CPUC’s carefully planned biennial cycles of updates. For example, in just a year—from 2013, when CPUC set the first quotas, to 2014, when it approved the utilities’ first deployments—it became clear from the slate of proposed projects that a new procedure was needed for compensating utilities for the loss of electricity sales from power that was shunted into battery projects.49 After 2014, as actual procurement advanced, CPUC also learned that the expectations about the likely sites for battery deployment were wrong. At the start of the program, most experts thought that storage capacity would chiefly be deployed on the parts of the grid under direct utility control, including the higher-voltage systems, because that is where electricity flows are greatest and thus the gains from control over these grid-connected system are plausibly the largest too. Instead, many batteries were deployed at customer sites, in part because customers and the equipment vendors operating storage on their behalf quickly learned how to use these battery systems to cut costs as well as improve power reliability. In parallel with CPUC’s efforts to advance grid-connected batteries, large power users on their own were buying “behind-the-meter” battery systems that would help them cut electricity costs and make their electric supplies more reliable.
Because the behind-the-meter market was growing so much faster than expected, before it made any decisions about the 2016 procurement, CPUC adopted new accounting methods to allow utilities a 200 percent “ceiling” for overcompliance in deploying projects at customer sites, with overcompliance usable to offset procurement quotas elsewhere on their systems.50 (As in the sulfur case in chapter 4, what looks on the surface like a market—the trading of compliance quotas—is in fact active, conjoint learning by the regulator and regulated about how to set the rules, and the “trading” that follows is relatively inconsequential optimization within those rules.) At the same time, CPUC adjusted a state subsidy scheme that helped pay the cost of battery installations so that in providing electricity to users that were competing with utilities, known as Community Choice Aggregators, new actors could be engaged more fully in the battery revolution.51 Once again, the industrial policy of CPUC put it at odds with the utilities, for Community Choice Aggregators were designed to allow communities to take more control over their own procurement of energy services and shrink the revenues that might flow through utilities.
The second wave of contextualization began in 2018. With the conclusion of the second biennial procurement, the state was exceeding its deployment goals for batteries, yet still had not learned much about how batteries would transform grid operations. Under the procurement program, the utilities were required to operate the new equipment in a program-specific experimental zone. Neither the regulator nor the investing utilities knew how procured devices could be put to efficient economic use while also meeting the requirements of the ongoing reliability of the grid. A combination of regulatory mandates and guarantees of fair return on investments allowed for the necessary experimentation while limiting the scale of deployments so that adverse effects on the network would be safely contained. To scale deployment beyond this sheltered zone would require new rules to allow for the efficient use of the technology while continuing to assure network reliability.
The central issue in devising these rules is a shared language of evaluation; as explored in chapter 3, experimentalism requires the ability of those running the experiments and reviewers to understand what is being learned. The utilities that manage their local and regional grids under the supervision of the CPUC, the manager of the statewide high-voltage grid (CAISO), and the private companies that deployed at least half the battery systems all needed the capacity to understand how battery systems might create value and affect the reliability of the grid, and they needed a common language to communicate that understanding.
This language took two forms. One, focused on the benefits from deployment, is “value stacking.” Batteries can do a lot more than just store energy. If coupled to advanced software that can communicate with the grid and power users, a battery storage system can perform many other functions or services, from arbitraging time-sensitive price differentials to stabilizing grid voltage and frequency. Some of these services can be supplied simultaneously, while others can only be provided in sequence. In practice, it is as if no single service requires a substantial fraction of a storage system’s available resource, and no single service is profitable enough to amortize the cost of investment. The economic use of storage therefore requires rules that allow the combining or stacking of different uses, with the proviso that the combination of committed services not put the stability of the network at risk.
What makes value stacking especially difficult for regulators and storage users alike is that the performance of storage devices is highly context specific, and the exact value from combinations of services was unknowable without experience. As a leading research institute put it in 2015, “The range in value that energy storage and other distributed energy resources can deliver to all stakeholders varies dramatically depending on hundreds of variables. These variables are specific to the location where resources are deployed, making generic approximations of value difficult.”52 It is possible to distinguish adverse uses conceptually (where a commitment to one service jeopardizes serving another) from compatible ones. But the actual performance of particular devices, and hence their value (and risks) to the grid, can be known only through intense and ongoing observation in local contexts.53
The other shared language concerns the reliability of the grid and has already been discussed above: grid modeling. Grid operators use a suite of power flow models that simulate how every component (node) of a grid reacts to changes in the behavior of any of the others. Each utility or other local grid operator must maintain its own power flow models that simulate all the nodes under its control, customized to its local setting (with the help of the model software provider) and periodically calibrated to the actual behavior of each grid. These local models are regularly compared to one another under the supervision of the regulators to produce a single, contemporary, system-wide set of best practices for grid modeling even as each grid operator maintains its own grid-specific configuration.
The practice is dynamic and interactive, and begins before any large new devices are connected to a utility’s grid, including battery systems. The utility uses simulations of the performance of a proposed device to assess the values of all the services that the device might provide. These simulations of power flow on the grid also identify places where the device might create conflicts or congestion that might require grid upgrades or adjustments to the design of the device itself. Through an iterative process of project proposal, simulation of power flow on the grid, adjustment, and then resimulation, a viable project emerges. With value stacking, the economic value of the project can be assessed. With power flow models, the risks to the reliability of the grid can be assessed. The convergence is formalized in an interconnect agreement between the operator of the battery project and the grid operator, stating in effect what the grid in its various layers and nodes can expect from the project under varying conditions, and constraining how the device behaves on the grid.54 The power flow calculations made by the operator of the grid—in California, that’s the utility for medium- and low-voltage grids, and CAISO for the high-voltage grid—are the ultimate arbiter in this process. But the methods and assumptions are transparent so that other third-party contractors can run their own simulators and developers can design projects without always being under the utility’s thumb. Along the way, the regulator could check these calculations as it makes decisions such as whether to approve the inclusion of a utility storage device in a utility’s rate base or the cost of storage services that a utility purchases from an independent supplier.55 This is why the concept of value stacking was so important to accelerating investment in battery systems: most of the services provided by batteries have never been valued properly in markets and thus regulators needed a way to determine whether the full range of benefits from deploying batteries could be aligned with the prudent bearing of costs. That interconnect agreement is in turn updated as needed, if only to alert the regulator and grid operator to new power service commitments, or variations in the actual performance of a device, that might eventually interfere with (or complement) other devices or the grid’s reliability.56
Introducing radically new technologies raises huge challenges for modeling because it is hard to identify the right functional form and assumptions to govern how novel devices are represented. While analogous cases can be identified, the problem is emblematic of the uncertainty discussed in chapter 3—uncertainty that is irreducible in the absence of real-world experience in context.57 One place to observe this readily is at CAISO, the high-voltage grid operator, because it makes its deliberations around modeling and the procedures for setting interconnect agreements so transparent. As the first battery projects were advancing, each required the modeling of how the battery would affect the surrounding grid and an interconnect agreement. For the high-voltage grid, this meant that the regulator (CAISO) needed to resolve novel questions such as whether batteries were generators or sources of demand, the traditional categories for grid-connected resources.58 The answer was both and neither, so CAISO created a new category of grid-connected resources, published the code it would use in modeling these resources, and outlined an approach to modeling and interconnect agreements that would be good enough until practical experience might reveal better approaches.59 As each cluster of projects in California advanced—each needing modeling and interconnect agreements—grid operators also watched as a few other jurisdictions around the world ran experiments in allowing battery interconnection. Until the California market surged in size, making it today the biggest deployer of grid-connected batteries, the largest battery experiment was in Australia, where a large system was installed at the interconnection of two grids. Comparing model-based studies that tried to predict how that Australian battery would operate with what was learned when the battery actually functioned helped reveal how little was predictable reliably in the absence of real-world experiments in system context.60
A fundamental task for regulation under uncertainty in this setting is to strike a balance. On the one hand, the regulator is advancing a form of industrial policy; it is pushing for the emergence of a nascent industry and thus wants as much experimentation as possible.61 On the other hand, it must ensure that the experimentation needed to encourage innovation and deployment does not degrade network reliability or impose unacceptable costs on ratepayers. Thus CPUC’s 2018 provisional rules for the deployment of storage authorized utilities to provide stacked services in any combination they think useful, subject to the requirement (in rule 6 of CPUC’s “Decision on Multiple Use Application Issues”) that “a single storage resource must not enter into two or more reliability service obligation(s) such that the performance of one obligation renders the resource from being unable to perform the other obligation(s).”62 In effect, rule 6 is a kind of guardrail. Deployers of battery systems are authorized to experiment with value stacking and have strong incentives to find ways to maximize the total value, so long as their experimentation does not involve multiple claims on services that could be needed when the grid is under stress. Neither CPUC nor the battery operator can predict the exact grid configurations or conditions that might affect how batteries perform in real time, but they can control how reliability-related services are offered to the marketplace and contracted. After that, real-world observation is essential to looking, in context, at whether the regulator has struck a balance that is too conservative; if so, the guardrail would need changing.63
Periodically, the real-world offers extreme experiences that can test experimental deployments at limits that would be too dangerous to create under normal circumstances and too unusual to simulate reliably. Large-scale grid outages offer a unique opportunity for that kind of testing; at this writing, the most important of these were two days of statewide power shortages in August 2020, a time when hundreds of megawatts of battery projects had been deployed—each operating under interconnection agreements with grid operators that were designed with protective guardrails yet their actual operations would help reveal whether batteries in the context of grid duress could do a lot more for grid reliability than had been allowed. CAISO, an operator of the state high-voltage grid, ran an interagency process after those outages that mapped power flow models to the actual performance of all devices on the grid—looking in particular at whether batteries could be operated in different ways when the context on the grid required more resources. What CAISO found was that the guardrails were too conservative and led to the underutilization of battery resources in practice.64 Put differently, the rules had run out. Prior to such real-world experiences it was impossible to set an effective rule that could anticipate all circumstances. Supervision using common languages of assessment, rather than rules, was required—and when the opportunity of extreme conditions arose, that supervision made it possible to write new rules and set the context for a new wave of experimentation in real-world contexts.65
In sum, the problem for the regulator and investor in regulated storage is to arrive, initially, at an approximate estimate of the performance of the device that is detailed and reliable enough to warrant a decision on the soundness of the investment, and then correct an understanding of the network with regard to that type of device as local experience accumulates. The common languages of evaluation—value stacking and power flow models that simulate reliability—made this possible. Guardrails such as rule 6 made it safe. This is, again, a paradigmatic case of experimentalist contextualization: local adaptation is necessary (because performance in place can’t be adequately estimated without place-based data), but the stability and adaptability of the overarching system depends on continuing the central adjustment to these ground-level changes. Through this process, regulators oversaw a massive deployment of batteries on the California grid—deployments that in just a few years, rose from essentially zero to the largest in the world.
