Large-scale advanced energy storage systems (ESS) have a short track record relative to other electric power technologies. This makes it difficult to understand and value the ways ESS can be used for financial benefit or reliability and to secure financing. A number of software tools have been developed to improve planning-level decision-making around ESS, including siting analysis, sizing analysis, technology selection, and cost-benefit analysis, all of which incorporate a calculation of the financial or societal value of ESS relative to its costs. The approaches taken by these tools range from simple spreadsheet calculations to physical simulations and optimizations. Many design decisions made when developing these tools impact their results and implicate when a tool is suitable and when another option is better. This document characterizes some of the common ESS planning tool types and locates popular modeling tools in this framework, which is meant to help the reader select the right tool for their job and to provide broader context to enable more robust storage value analysis.
Types of Tools
Each of the tools examined in this document was developed by a different team for a different reason using a different approach, so some variability in the results from these tools is to be expected. Even among tools that adopt similar approaches with similar objectives, some variability can be observed. In general, all the tools listed in this document were built to support some form of decision-making and support the decision with quantitative value estimates. The decisions made from tool results (e.g. where to locate an ESS, what type of ESS to select, the right size of ESS to build) are usually much more robust than the direct numeric estimate of value (e.g. the net dollars per year of savings expected from an ESS performing demand charge management). Different tools and different data sets used with the same tool might yield a similar decision result despite significant differences in the value estimate.
Energy storage is relatively unique among power sector technologies because of its strong energy limit. This energy limit couples the operation of the storage system in time, meaning that operation now could impact the ability of the system to operate in the future. Unlike generator technologies with near-unlimited energy generation potential, storage systems can run out of stored energy during their operation and lose value in the future based on how they were operated in the past. In most cases, this fundamental limit requires ESS modeling tools to internalize the chronological state of charge (SOC) of the ESS they are modeling and maintain the feasibility of their result by limiting the system to stay within SOC bounds (e.g. 0%-100%). This requires the modeling tool to make the complicated decisions about when to charge/discharge the system and when to participate in other services in such a way that the value estimate is useful for decision making. In some cases, the ability of a real-world controller to forecast market prices or a customer’s electric demand could be significantly sub-optimal and should be included in the model. In other cases, this is less important (usually when making a decision where the error is balanced on both sides).
Sizing vs Dispatch-Only
All tools in this document fit can be divided into two categories. Either the tool sizes a storage system automatically or it does not. Tools that do not automatically size the storage system could be considered ‘value calculators’ for storage and can still be used to inform sizing analysis by repeatedly calculating the value of systems with different sizes. Tools that automatically size the storage system may have included the power and energy capacity of the storage in their optimization algorithm or rule set directly or may execute repeated fixed-size value analysis behind the scenes. These tools will output the power and energy capacity that results in the highest project net value (or another financial metric).
The repeated fixed-size analysis has the advantage of giving the user visibility into the sensitivity of the results to storage power and energy capacity but may not arrive at a globally optimum solution if the range of power and energy capacities is not wide enough or if it performs too few value calculations to have a high likelihood of getting near the optimal solution. Still, it is good to have a wider understanding of the solution space around the optimum so a more robust solution can be found. If the optimum solution’s financial results are only slightly better than another solution and that other solution is less risky (e.g. indicates a smaller system size/cost or relies less on unpredictable events), then the less “optimal” solution might be the better solution. This is to say that all elements of a decision made by a human may not be included in every tool, so it is a good practice to interrogate the results and use several results to inform a robust decision.
Microgrid vs DER
A microgrid is a group of interconnected loads and distributed energy resources (DER) within clearly defined electrical boundaries that act as a single controllable entity with respect to the grid. A microgrid can connect and disconnect from the grid to enable it to operate in both grid-connected and islanded mode. Some tools are specifically designed to handle always-islanded microgrids, some are specifically meant to handle normal operation of grid-connected DERs, and some are meant to model grid-connected and islanded operation together. The objectives and types of decisions captured in these tools can differ, which may result in different results. For example, a tool designed for always-islanded microgrids that tries to minimize the cost of a microgrid to cover a given load may not be suitable for a grid-connected customer that is going to use a solar plus storage system to cover a fraction of their load during infrequent grid outages and to reduce their electricity bill.
Time-Series, Instantaneous, or Design Day
Time-series modeling is generally required for storage value estimation because the state of charge (SOC) of an ESS couples all moments in time, meaning the model must track the SOC over time as the storage system operates and restrict the SOC from going out of bounds. Performing a service that charges or discharges the storage system at one point in time can impact the storage system’s ability to provide another service in the future. This means that, unlike in many power system models, each time step cannot be considered independent.
Tools that incorporate time-series modeling arrive at different answers to the question, “how many hours/days/weeks/months/years should be included in this time series analysis?” depending on the objectives of the tool. Design days are an approach adopted by tools like DER-CAM that divide a year into a few representative days of time-series data per month and use those days to perform the analysis. This approach can be well-validated for some of the decisions DER-CAM focuses on but would break down when trying to model other cases, like those that include long duration energy storage or uncommon loads. Other tools, like EPRI’s DER-VET, incorporate at least one year of time series data into every analysis. This comes at the expense of computational intensity but allows for more flexibility.
There are a series of challenges for storage system valuation models around the interdependence of each time step. For example, how likely is it that a bad solar day will align with a high load day for a solar plus storage microgrid? How likely is it that a market service will deplete the state of charge of the storage system and make it unable to meet other objectives in the future? Tools with different focuses will adopt different answers to these questions.
Some older tools do not model the storage system’s operation in time, relying instead on assumptions about how each service constrains every other service. This can be a good approach for a small set of very well-defined use cases but breaks down when trying to solve a more general problem.
Optimization vs Non-Optimization Control
Some storage valuation models determine how to operate the storage system based on a set of rules, some attempt to simulate a real controller, and others use an optimization algorithm. Rules-based dispatch is easily understood by the tool user and is simple but tends to limit the usefulness of the storage, especially in cases when flexibility is required from the storage system. (See Realistic Control) Simulating an ESS controller is another option. This approach implements logic like what exists in real ESS controllers (or hypothetical controllers) and is only given information that a real controller would have. This simulation could have an imperfect forecast of future market prices or loads but would not have perfect knowledge of these and would need to make decisions that maximize benefit and minimize risk based on this imperfect information.
This approach can be useful when a lot is known about the storage system before modeling begins but may introduce an additional unwanted source of variance in planning studies. Optimization-based dispatch attempts to find the best schedule for the storage system that maximizes its value and minimizes costs. This could be “perfect foresight” optimization, where the storage system is given perfect knowledge of future data or the optimization could be given imperfect information.Optimization-based dispatch tends to have a major drawback in that the optimizer needs access to information about the future to make dispatch decisions for one point in time. Given that valuation tools need to be generally applicable, this usually means that the optimization algorithm is given perfect access to all future data in some time window. This is called ‘perfect foresight’. Perfect foresight dispatch optimization yields the best possible results for the storage. This is a good approximation of real-world results where perfect knowledge of the future exists. For example, applications like shifting energy based on a customer’s retail tariff, where future energy prices are well-known, should yield very similar results between perfect foresight and non-perfect foresight approaches. However, in cases like reducing the monthly demand charge of a customer, where the customer’s future load is not well-known, the perfect foresight approach can dramatically overestimate the value of storage.11826632
Each of the tools makes assumptions or takes user inputs about how the storage system being modeled generates value. This section talks about some of the major avenues for generating value that are modeled by these tools, what they mean, and contains some discussion. The tools discussed below each contain some of these services and a full list of which tools model which services is available in Section 2.
Energy Time Shift
Energy time shift is the process of charging a storage system from low-cost electricity and then discharging it during times with a high electricity cost/price. In the process, some energy is converted to heat through roundtrip efficiency losses, the storage system may degrade somewhat, and the operation may incur a variable operation and maintenance cost. If the difference between the charging price and the discharging price is high enough to overcome these costs and there are no other opportunity costs, then it is beneficial to shift energy between the low-price time and the high-price time.
Wholesale energy market prices, a power purchase agreement, a tariff, or some other contract may define what the energy price is for an entity at any point in time. When the energy price is known perfectly in advance, a storage system may be scheduled in advance for this service with very little uncertainty. However, in the case of an energy market, the uncertainty in future prices may impact the amount of value a storage system can achieve through this service. Despite this uncertainty, a good operator should be able to achieve something close to the perfect foresight result, even using a simple strategy. One such strategy is called the ‘persistence method’ whereby the storage system is scheduled for a day using known energy prices from the previous day. Small variations exist with this method, including adjusting for weekends vs weekdays (or day of the week), using intervals other than one day, etc.
Frequency regulation involves following a signal provided by the grid operator to maintain the balance between generation and load in an area usually on an interval of several seconds. Battery energy storage tends to provide this service well due to the fast ramping requirements, low response time, and low energy throughput required by this service. However, valuation planning models cannot typically model the operation of a storage system at a time resolution of seconds. So, valuation models parameterize the effect of providing frequency regulation on the storage system. This involves characterizing the net change in state of charge over a large time step and calculating how the storage system was cycled during the time step. Additionally, the power and energy capacity reserved for the regulation service cannot be used by any other service during normal operation.
To provide regulation, there are some region-specific requirements that ESS must follow. Importantly, there are limits on the state of charge of the storage system when providing regulation. If the storage system is providing regulation, it must be able to realize the worst-case scenario defined by the market rules. This could be defined as the power capacity as was selected for the regulation service for the entire duration of the interval or it could be lower, depending on the location. If the system were providing regulation in the up direction as part of an hourly day-ahead market, then the worst-case scenario might be to discharge at its full power for an hour, meaning that the system must have at least one hour of charge stored at the beginning of the hour. Similarly, a storage system providing regulation down must be able to charge without exceeding its maximum state of charge.
Spinning and Non-Spinning Reserves
These reserves are power and energy capability that is not called on in normal operation but is instead held in reserve for a contingency event, such as the unexpected loss of a generator or transmission line. For storage systems, this represents an amount of power capacity and stored energy that are held in reserve and not used during normal operation. Valuation models typically assume that a storage system will never be called to provide power for this service, but always reserve enough power and energy capability to do so if required. Because these contingency events are relatively rare, this assumption yields good value estimates.
These services can be differentiated by response time, ramp rate, and duration.
Load following can be provided by resources that can adjust their power output in response to changing load and uncontrollable generation. While the exact power output of resources providing load following cannot be known far in advance, their ability to change can be known, so can be modelled similarly to regulation or other services that require reserved power capacity that will not be used by other services.
This service produces value by reducing the peak power drawn by a customer’s site and correspondingly reduces the customer’s demand charge. Demand charges are typically proportional to peak load, but there can be complexity on how demand charges are assessed. Most demand charges are calculated monthly, but other options exist, including daily. Some demand charges only apply to the peak load achieved during certain times of day. Many demand charges are calculated based on 15-minute or 30-minute interval data – the longer the interval, the lower the peak power that will be used to calculate demand charges.
A storage system providing demand charge reduction discharges when the load it is servicing is high and charges when it is low. This reduces the overall peak power drawn by the customer’s site and can reduce demand charges for a customer, utility, etc. Performing this service well requires a good idea of what the load is going to do in the future and can be a case where a perfect foresight dispatch optimization will dramatically overestimate the value of a storage system without adjustment. If a storage system underestimates the load in future hours, it can end up fully discharged and unable to reduce the load anymore. Because monthly demand charges are typically based on 15-minute or 30-minute average peak load in the month or in certain hours of the month, even failing to offset a small amount of energy for a short time can completely eliminate the benefit from the storage for the month. This risk is lower for demand charges that are assessed more frequently, such as daily, because only one day’s worth of demand reduction is at stake at one time and day-ahead forecasts are much better than month-ahead forecasts.
Grid Investment Deferral
Grid investment deferral involves offsetting or delaying the need for investment in the power system by using a storage system to mitigate the need. The deferral could be driven by a growing load exceeding a thermal power limit on a piece of grid equipment (N-0 deferral), a growing load 11826632 1-6 approaching the planned load limit during an N-1 contingency (N-1 deferral), or there could be some other problem, like the voltage on a distribution system going out of bounds. Wherever there is a point in the grid where a power limit is exceeded for a short period of time, a storage system could be installed and programmed to mitigate the overload by discharging during the peak load times. The overloaded asset could be a line, transformer, duct bank, or any other piece of transmission or distribution equipment. By alleviating the overload, a storage system can offset the need for a traditional upgrade to the overloaded equipment or the need for additional equipment.
In the case where the overloaded equipment is serving a growing load, the storage system may only be able to defer the need for investment instead of offsetting it entirely. This is still valuable, but care needs to be taken in the planning phase to understand how the storage system will be used after it is no longer capable of deferring the investment.
Increasing hosting capacity on the distribution system is becoming a more important use case for energy storage systems. When controlled explicitly for the purpose of mitigating problems on the grid, a storage system can allow for more intermittent generation. However, many systems, especially customer-sited systems, are not and can decrease the calculated hosting capacity because they are considered to potentially charge or discharge at any point in time.
Resource adequacy is a reliability requirement which ensures that there are enough generation and non-generation resources available to meet the forecasted next-year peak load along with reserve requirements, generally one to three years ahead. In California, to qualify for system or local area resource adequacy, a storage resource is rated at the maximum output which can be sustained for at least four consecutive hours and be available for at least three consecutive days. For flexible capacity, a “bi-directional” storage resource is rated at the output which can be charged for 1.5 hours and discharged for 1.5 hours.
Voltage/Reactive Power Support
Storage can be used to supply reactive power to keep the grid operating within bounds. This can be used to offset the need for other, traditional equipment, and maintain power quality in isolated cases, such as islanded microgrids. This can limit the operation of the storage system by consuming inverter capacity and leaving less for other services. This service does not require real power beyond what is required to keep the storage system operational.
Storage systems that are co-located with a load can be used to supply power to the load in the case of a grid outage. This impacts normal operation and valuation if some stored energy is always reserved in case of a grid outage, leaving less state of charge for other services to work with. If the storage system is used for other purposes, then its state of charge will follow some probability distribution which can be used to estimate the likelihood of “surviving” a random outage for which the storage system had no notice. This probabilistic approach allows the storage system to generate value in other ways while not neglecting its commitment to cover load during grid outages.
While resilience has several definitions, it is related to ensuring that load ends up being served even during extreme events and that normalcy can be established quickly following extreme events. Being ready for extreme events can impact how a storage system would otherwise operate, ensuring it has enough stored energy to contribute to resilience.
Demand response is when controllable loads are curtailed during periods of grid stress to reduce the overall system load. Energy storage can be used to modify a site’s load so is sometimes allowed to participate in demand response programs similarly to controllable loads. The details of demand response programs vary but a key distinction can be made between programs that notify participants far in advance (e.g. 24 hours) that an event is coming and those that give short notice. If a storage system controller has knowledge of a demand response event far in advance, it can cease participation in other services and charge before the event. If the event is not known far in advance, then the storage system must maintain a state of readiness. Standing ready with a high state of charge limits the availability of the storage system to other services, reducing its economic value, which must be balanced against the value of participating in the demand response program.
Gaps in Current Modeling Tools
Despite the breadth currently available tools, there is some functionality that is both lacking and useful. This section contains a discussion of some of the key elements that are missing from current energy storage valuation tools. EPRI endeavors to constantly improve its modeling tool capability to address the gaps below.
Connections Between Tools
One of the most useful expansions to the current suite of available tools is software connections with common power system and market modeling tools that can better capture the storage systems’ effects on its environment. Little modeling occurs in a vacuum and some of the highest-value contemporary use cases for ESS require grid simulation tools in addition to a storage valuation tool. Most currently available tools make assumptions about the storage system’s effect on its surroundings and rely on the user to transfer results manually between tools. These external tools include grid simulation tools, such as EPRI’s DRIVE and OpenDSS tools, long term procurement planning tools, effective load carrying capacity calculators, or market models that capture the effect a storage system will have on local energy and ancillary service market prices. As the amount of storage being deployed increases and storage gains market power, this functionality will become more and more valuable.
Most storage modeling tools available today take one of two approaches to dispatching the storage system. The first is to design a set of rules that govern the operation of the storage system. This could be as simple as, “Charge fully between midnight and 5am then discharge fully between 6pm and 10pm”. Approaches like this tend to leave a lot of value on the table, depending on the use case, but have the benefit of being very simple. More complicated rule sets include external stimuli, e.g. “keep the net load at a customer’s side below a threshold level”. The second common approach is to use an optimization algorithm to decide how to operate the storage system. A well-formulated optimization problem will be able to arrive at the best possible operational profile for the storage system given the physical limits of the system and the data it was given. If the optimization is given errorless data, it will produce the perfect foresight result, which is the true optimum. This removes any variance in the result due to the differences between controllers and between forecasts and provides a benchmark to test more realistic controllers against. Most tools do not simulate a controller directly, instead relying on one of the approaches above, but work is being done in this area. As energy storage controllers become more standardized and their performance better-known, the variability they introduce should go down, allowing for higher-fidelity valuation models.
Incorporating the Cost of Degradation
A component of optimization-based control in storage valuation tools that should be captured better is the cost of degradation, especially for battery models and others with significant degradation. Most optimization-based methods either do not include the cost of degradation in the optimization objective function or apply a linear penalty that scales with energy throughput. But degradation is a complicated and nonlinear process that depends on factors not visible to most valuation tools, like the battery’s internal temperature. Given the cost of the energy storage medium, especially batteries, there is strong incentive to use them as efficiently as possible and including some mechanism to incorporate the cost of degradation into the dispatch optimization can improve project economics dramatically by extending the lifetime of the storage system. Increasing the accuracy of the degradation model internalized by the valuation tool’s dispatch algorithm will enable more robust degradation-based decision making.
The effect of degradation is most prominent when the storage system is performing a relatively low-value, high-energy-throughput service, such as shifting energy in time based on the price of energy. Including the cost of degradation increases the price spread (max price – min price) required before the storage system finds it economical to shift energy, meaning the storage system charges and discharges less, which reduces degradation due to cycling and extends its life at the cost of some energy shifting revenue.
A rigorous, well-formulated term calculating the cost of degradation in the optimization objective function is needed, which will penalize cycling based on the actual cost of degradationexpected from the specific system.
Degradation is usually calculated based on two categories of degradation – cycling degradation and calendar degradation. Calendar degradation occurs all the time simply due to the passage of time, but can be accelerated by extremes of temperature and SOC. Cycling degradation depends on how much the battery is cycled, the depth of discharge of those cycles, the SOC range through which it is cycled, and can also be accelerated by temperature. These two terms, which are both complicated and nonlinear by themselves, combine to erode the energy capacity of the battery. Through the process of degradation, other changes occur, such as to the battery’s internal resistance (and therefore roundtrip efficiency), power capacity, etc. The rate at which a battery experiences calendar and cycling degradation also depends on the battery’s chemistry. This, along with differences in active cooling systems, geometric layout, etc. between systems can 11826632 1-9 result in dramatically different degradation rates between two otherwise similarly capable batteries.
Uncertainty, Reliability, and Resilience
Energy storage systems have one more key constraint beyond typical power technologies, which is the amount of energy they can store. When using a storage system for a reliability purpose, this means that it is important to understand both the likelihood of meeting the power requirements and the likelihood of having enough energy stored at each point in time to meet the reliability requirement.
For example, when designing a storage system as part of a microgrid that is designed to increase the reliability and resilience of a site’s power supply by islanding during grid outages, it is important to have a very good understanding of the site’s load. Even modest load growth or an unusual coincidence of unfavorable conditions can push the energy required of a storage system beyond its capabilities.
Many storage valuation tools neglect to model the costs associated with a full storage system well, including the cost of electric upgrades to enable microgrid capabilities, interconnection costs, etc. This is very important in the planning and sizing phase of a project when the cost estimate needs to cover a range of sizes, durations, etc. The best option is to use real cost information, but this is not always available or applicable. The next best option is to develop a good cost estimate based on all the components in a system. This is a challenge for valuation tools, which do not have access to detailed design information. But estimates can be made based on the scale, form factor, and complexity of a system.
Equipment Reliability and Downtime
In most valuation-level tools, storage systems are modeled as having 100% availability and 0% chance of failure. Depending on the specifics of the system, this can overestimate its usefulness. There is a need to include more design specifics in valuation tools that can enable estimating the likelihood of a storage system successfully providing the power and energy that is asked of it.
Detailed Storage System Models
There are some details of the storage system models that, depending on the use case, can dramatically impact the system’s overall value. These may be the self-discharge rate, auxiliary power, and the difference between operating modes of a storage system when the cost of energy to supply these is significant. Other components that are sometimes included but not always and may have an impact on the results are inverter efficiency curves, cell efficiency curves, and the dependence of performance on cell temperature.
High-Fidelity Market Participation Models
Energy storage modeling tools vary in the complexity of their wholesale energy and ancillary service market participation models. Generally, it is very difficult to capture the full range in current market rules let alone prepare for changes that may or may not happen over the life of a storage system.