The engine that drives storage: Software

The engine that drives storage: Software

April 10, 2018 | By Deanne Barrow in San Francisco

Software is the brain that makes a storage system work. It optimizes use of a battery that is supposed to serve multiple functions, weighing the potential revenue from competing uses against the potential wear on the battery. If it does not work, the battery performs sub-optimally or does not work at all. Bankers are starting to focus on the software as part of their diligence. 

A group of software experts talked about warranties, bankability and other issues around software at an Infocast storage conference in San Francisco in early March. The panelists are Jennifer Worrall, CEO and co-founder of software company Iteros, Ryan Wartena, president and co-founder of software company Geli, Pedro Elizondo, manager of energy storage business development at NEXTracker, and Michael Atkinson, vice president of sales and business development at Doosan Gridtech. The moderator is Deanne Barrow with Norton Rose Fulbright in Washington.

Market data

MS. BARROW: Let’s begin by setting the scene. Jennifer Worrall, you have some figures on the current size of the market for energy storage software and how the market size is expected to explode in the near term.

MS. WORRALL: Based on research done by Navigant and our own analysis, between 7% and 8% of project costs today, and maybe 5% and 6% in years to come, will be the software costs that go with a storage project. That translates into a $2.9 billion market today for energy management software in general, growing to about $8 billion in upcoming years. The energy software component of storage looks like a $400 to 500 million market this year, growing to $3 billion in the next five or six years.

MS. BARROW: Does anyone have a different view?

MR. WARTENA: The other side of the equation is battery prices are falling. So guess what? Our software costs — the 7% to 8% of system cost — need to be falling as well. Software is a big market, but it is also a competitive one.

MR. ATKINSON: Utility-scale software is a growing market. More utilities are putting in storage that they own. They need more flexibility in how they can operate their systems. Storage software also starts blending into how they manage distributed energy resources on their systems as well, which takes the potential market for storage software from hundreds of millions to many hundreds of millions to possibly a $1 billion market, purely from a utility perspective.

MR. WARTENA: When smart grid was big, Cisco suggested that 10% of the smart grid was going to be software. I remember buying my first Gateway computer. It was a $3,000 computer, and I had to buy a $300 operating system for it. That was 10% of the cost.

To do what we need to do over the next 20 to 30 years is equivalent to about an $80 trillion investment if we want to see global electricity supply move to 100% renewables. If software is 10% of an $80 trillion investment over the next 20 or 30 years, that is still a really good business.

MS. BARROW: Not only is it a really good business, but also software is often called the secret sauce of successful storage projects. I have a feeling the sauce has a slightly different flavor for each of you. So let’s go across the panel and hear why you think software is vital to the storage industry and what you perceive to be the problems that software is solving today.

MR. ATKINSON: Storage is the secret sauce. Without it, you have a lot of bricks, a bunch of lithium, now ions, and they do not really do anything.

Storage itself is the most flexible asset that the electrical industry has ever seen, and the ability to use it productively depends on the software being able to pull together lots of different inputs.

It is reading in weather data to do forecasting. It is reading in the deregulated markets data to do pricing. It is optimizing batteries from both a physics and an economics perspective. The operating case and use cases are getting more complicated.

You have to make sure you model the operating and use cases properly ahead of time with an eye to an optimal solution and then, after the fact, you must be able to track what actually happened. Are they within the parameters you set?

Software is managing all of the disparate pieces of the system and sending operating data to other software programs and into utility control rooms. It is everywhere within a storage system.

MR. WARTENA: Now that we have systems that pencil out financially, it is starting gates open.

The flavor of our Geli sauce is called Geli Rapid Energy. It covers the full life cycle of a process from the front-end design to helping with financials to make sure they are accurate, to automating contracts. It is a holistic system.

Our goal is to reduce the design, construction and development time. We are starting at the beginning. This is a hard industry. It only takes a couple weeks to build, but development time was 18 months.

MS. BARROW: Pedro Elizondo, how does NEXTracker fit in?

MR. ELIZONDO: We build hardware. We build containers for batteries, lithium ion mainly. In order to be able to comply with the performance warranties, we need data. Data allows us to take decisions relating to the infrastructure we provide for the batteries. Data allows us to optimize battery performance and do preventive maintenance on the most significant equipment for the battery life, which is the H-pack. Batteries are great, but no H-pack, no batteries. Data is needed to prevent issues, prolong battery life and comply with the battery performance warranties.

MS. WORRALL: In order to fit into this rapidly changing landscape, you must have a software layer. Otherwise, your equipment is just a pile of cells over there in a corner.

Software is necessary to connect to different devices and to be able to take in the different inputs. For example, what happens if a rate plan changes for a behind-the-meter customer? What if there is a new market program that will allow you to continue to work within your warranty specifications and is a better use case for that particular battery?

Software enables equipment manufacturers to focus on what they do best, while leaving the core competency to a software provider to understand policies and what is the best economic use case at any given time. It provides the data for engineers who need to be able to understand how their products are working. It makes adjustments in real time when outside parameters, like rates, change.

Optimizing usage

MS. BARROW: Let’s delve into how software optimizes performance when there are different parameters and sometimes competing priorities for the use of a storage system. If you have a behind-the-meter system that is installed on a customer’s site partly to reduce demand charges, and excess capacity is being sold to the utility, how does software optimize dispatch in that situation?

MS. WORRALL: We organize each of those use cases, whether it is demand charge, energy arbitrage or market participation, into small pieces that we call orchestrations. We then create a customizable value stack that prioritizes those particular use cases based on how we think the economics should look.

We have an intelligent engine that does a simulation at the front end of a project to assist with financials and help our equipment manufacturers decide which projects are worth pursuing. We use the same engine on a daily basis to determine the right way to operate that device. We are always looking for the most economic way to operate.

Some of it will be rule based. In a must-offer system, capacity must be released at particular times. If the software is handling demand-charge management versus energy arbitrage, it must weigh the economics of both. It must evaluate both, which one will make the customer money and also whether the amount of money is worth the wear and tear on the battery.

If you are stacking energy arbitrage on top of peak reduction, but you only have a little sliver of battery to use for energy arbitrage, then the software must evaluate whether the arbitrage is worth the wear on the battery.

MR. ATKINSON: We build operating modes in much the same way. From a utility standpoint, storage systems must be able to run either autonomously or be able in real time to be shifted to whatever rapidly-developing situation the utility is seeing.

This involves reading in load-flow data and prioritizing at all points between economics and physics. It is done to decide the most productive way to use the system as a whole, or the circuit that it is on, by working through the potential use cases.

The use cases are prioritized. VAR support and frequency regulation are priorities one and two. Voltage support and demand-charge management are further down the line. The software cycles through those cases, and the use can be changed in real time. The software looks at weather forecasting and what is happening with the solar on the system. It also looks at market pricing and what are the wholesale market prices or market signals in non-market areas. It adapts in real time to these variables.

MS. BARROW: Ryan Wartena, Geli backs up its ability to optimize with a performance guarantee on demand-charge management. Can you tell us more about that?

MR. WARTENA: It is an industry first. It is an analytical insurance around demand-charge management. We model out demand-charge management performance for a specific system at a specific site, and we provide a guaranty around that, and we put our software maintenance fees up as collateral.

There is shared upside because we know we are constantly improving our algorithms.

We are going through a bankability study right now with Wells Fargo and DNV GL on how demand-charge management affects battery degradation. How the battery degrades affects how much demand-charge management you can do.

We have an online design tool called Geli ESyst and, within five or 10 minutes, you could have a solar storage system designed. We give you a performance guarantee.

MS. BARROW: Do you guarantee a specific dollar amount of savings in demand-charge reductions?

MR. WARTENA: It is in kilowatts. Tariff risk and load-change risk is often put on the customer. The contracts give us the right to recalculate if load or tariffs change. Everyone walks into this knowing that tariffs may change.

MR. ELIZONDO: Demand management is a good example of a software application. From the hardware perspective, the batteries are capable of injecting the necessary power to reduce the maximum demand, but the key question is the time when this is done. Maximum demand is being measured every 15 minutes.

The key issue is how to inject the power at a precise time to avoid the charge. That is the part that software adds to the hardware.

MR. WARTENA: It is a hard question. I tell my team once we get to X%, we are getting out of the energy storage game and we are going to the stock market.

Predicting valleys and peaks is forecasting load.

Putting solar into the mix adds another level of difficulty because solar drops in and out, so you have to forecast both solar and load. We have a model predictive control loop which we can stack. Each site can optimize itself, and they can work like gears with an upper-level optimizer also. Each is solving its own equations, each is modeling, each is optimizing, and we move between multiple optimization algorithms, from convex algorithms to random-forest approaches, to standard last-day approaches.

Going through that loop constantly is how we manage multiple value streams. That is how we focus on the value at that site, and we can also decouple value at the site versus value for the grid. We can co-optimize where the site and grid have different owners.

MS. BARROW: This is big data analytics and machine learning.

If a solar project is supposed to qualify for an SGIP payment or the investment tax credit, then you have to control the source of charging. I bet that adds another layer of complexity. Does software control all of that?

MR. WARTENA: Absolutely. When you have a good model predictive control loop, it can solve for multiple constraints. We weigh all the constraints by assigning dollar values. If you want something to happen, make it really valuable. That is how we do it.


MR. ATKINSON: It is important to track usage both in real time and for history. In eight years, when suddenly an issue comes up with the batteries, or you have a disagreement about whether there is enough battery life left to meet your performance guarantees through the full 10 years, you must be able to look back at how the battery was used. Did the usage fall within the parameters of the contract? Is the battery still under warranty, or did you do things that made it go out of warranty?

That kind of real-time tracking of data is critically important. We are putting together a live warranty tracker for a project on which we are working now. It allows the customer to model what it must do in the future. Ten years from now, when there are wildly different potential use cases and you want to do something different, they would take you out of warranty, and that worries you.

We are looking at real-time warranty tracking. We are modeling how certain usage could affect the warranty. For example, if you are now going to use the battery in a certain manner, and you originally had a 10-year performance guaranty, we will have to reduce it to a 9 1/2-year performance guaranty if you use the battery in that manner.

The maintenance of the data and then the backwards analysis of that data is going to become more and more important as we move forward.

MR. WARTENA: There are two parts to this question. One is warranty tracking. You are tracking the warranty that came with the equipment.

What is actually happening on the battery is the delta. If you did the best optimization, then you have more capacity than the equipment manufacturer’s warranty says you should have. Now you have some extra play room — basically a buffer on your warranty.

MR. ATKINSON: Yes and no. What we are talking about is either extending or reducing the length of the warranty in real time. We provide a warranty to the system for the owner, and we are taking it on ourselves to be able to predict whether that warranty is going to extend or be cut short. It is a little bit of a risk for us, but we have worked enough with the battery and inverter companies to understand what will happen.

MS. BARROW: So these are back-to-back warranties? A warranty from the equipment supplier to GridTech and then GridTech provides a warranty to the project?

MR. ATKINSON: Yes, we provide the overall performance guarantee to the customer.

MS. BARROW: And your warranty tracker tracks whether a warranty will be scaled back based on usage patterns?

MR. ATKINSON: Yes, because we wrapped the entire project. The customer does not care if the problem is the battery, the inverter or the software. He does not care because he has a wrapped performance guarantee from us that the system will be able to operate within certain parameters for a certain number of years.

We are working on implementing the first stages of this on a project now. We are giving people the flexibility to operate in more of a freewheeling environment as opposed to being locked in today to operating this many cycles, this depth of discharge and that’s it. We are not there yet, but we are working on it.

Predictive maintenance

MS. BARROW: Pedro, you mentioned the role of software in predictive and preventative maintenance. Can you explain how that works?

MR. ELIZONDO: Predictive maintenance is implementing maintenance routines before the equipment fails, which is by far more cost-effective than corrective maintenance.

Maintenance-free batteries are common, but there is no such thing as inspection-free. You need to inspect the batteries. Things can happen. If you get data about duty cycles, meaning depth of discharge, state of charge, state of health and the charge rate, you can predict what maintenance will be required. You are able to say that after 1,000 cycles, a maintenance routine should be done.

Batteries are like circuit breakers. After 1,000 operations, they require maintenance. The key part is getting the data to know when maintenance must be done. This prolongs the life of the battery. Batteries are the most expensive asset in the energy storage system. Without batteries, nothing happens.

MR. WARTENA: Pedro, you did not mention the type of temperature control.

MR. ELIZONDO: You have to keep the temperature inside the container steady. We do that with software. We do not do it with a thermostat like in the home.

MS. BARROW: So temperature control is important. Ryan, you also said high state of charge is the new smoking.

MR. WARTENA: I think they say sitting is the new smoking, and high state of charge is the new sitting. We found that keeping a battery at a high state of charge is just as bad as having it hot. You are starting to see that in warranties. If you do it right, you could end up with 1% to 2% degradation a year. If you do not do it right and keep a high state of charge and you let the temperature drift a little bit, you can have 9% to 10% degradation a year.

MR. ELIZONDO: If you keep the temperature at 28 degrees Celsius, for example, you get certain warranty terms. If you design for 21 degrees Celsius, terms are better.

MR. ATKINSON: Not only is high state of charge bad for the battery, it is also bad for the business cases. If you stay at a high state of charge all the time, you are limiting what that battery can do. Batteries are not there just to discharge. Batteries are there to add flexibility to the system. They are there to charge when it is beneficial and to discharge when it is beneficial.

MS. WORRALL: I want to tie together this concept of preventative maintenance and planning for operations. Machine learning is important for forecasting load and solar production. But machine learning can really be a great guide in terms of predictive maintenance, too.

MR. WARTENA: This is a complicated thing. I remember the first time I walked up to an internet browser. I thought, “I see the internet, but I have no idea how this computer really works.” We are building all of that now. The level of complexity is high.

This goes all the way down to cell manufacturers and battery manufacturers. They may buffer capacity. They may put a battery management system in that will limit the capacity by limiting the voltage range.

Cell makers like to sell cells. They like to sell all the kilowatt hours in them. A manufacturer may sell a 100 kilowatt-hour battery, but install 120 kilowatt-hours and eat that cost. This way the customer gets a 10-year battery with 100 kilowatt-hours and can use 100% of it. In reality, the manufacturer never wants someone to use 100% of the battery.


MS. BARROW: You are saying suppliers are oversizing their batteries.

MR. ATKINSON: No battery system is installed with the exact amount of nameplate capacity. We put in a project within the past year that is more than 50% oversized because of the end-of-life requirements and because of the heat and the environmental conditions.

However, from a warranty standpoint, the customer is not allowed to draw out more than nameplate capacity — more than was signed up for in the beginning. That ensures at the end of 20 years it can still draw nameplate capacity.

MR. ELIZONDO: One of the main reasons batteries are oversized is because of lack of data. There are no batteries with 25 years of operating history at this point. Better to oversize to be on the safe side.

MS. WORRALL: Which increases the cost of the project in general.

MR. ELIZONDO: There is always a reason why prices are going down.

MR. WARTENA: All the ones in our favor.

MR. ELIZONDO: When we talk about warranties, project life and performance, that is one thing. When we put that in a contract, then we are not friends.

MS. BARROW: Why is that?

MR. ELIZONDO: Because you have to comply with the contract terms.

MS. BARROW: And the penalty for not doing so is liquidated damages?

MR. ELIZONDO: Money. If something affects your pocket, then you think differently.

MR. WARTENA: Good fences make good neighbors. Good contracts make good friends. They have to be good contracts.

Where to probe

MS. BARROW: They have to be good contracts, so you need good lawyers.

Let’s move this in a different direction. Ryan, you mentioned bankability in a study you did with Wells Fargo. So, probing on bankability, what questions are you being asked by lenders and investors, and on what issues should they be focused?

MR. WARTENA: People are very concerned about whether the battery will work. It took a long time for investors in solar to believe that the sun will rise every day. That is kind of a universal fear. The sun is definitely rising. Whether the software will work is not like that. That is a much different risk factor. If the software does not operate, the system is a brick. That is the biggest concern.

It is really not about the volume of money. It is the predictability of the revenue stream. That is why half of our software development team is in analytics, working on this exact problem. It is hard to predict.

Lenders and investors are looking for predictability. They want a warranty. A performance guarantee will be necessary for the next two to five years until the rest of the industry gets comfortable that the software works.

Geli has an interesting approach on how we do battery modeling. A number of other universities and national labs, including Sandia, have battery models, too. So we are using all each other’s battery models, which is kind of cool.

We are headed toward an algorithm war, but it will produce some really great algorithms that the whole industry can stand behind. Some banks have looked into degradation. We pair algorithms with models from universities and national labs to get a better handle on degradation. It is important to give financiers visibility into the algorithms.

There is something else that financiers like, but do not always request. In solar, you do monthly wrap ups on how your projects are performing. It gives financiers 12 data points a year on how their billions of dollars of investment are doing. In our case, we can tell them every single moment how well the systems are performing. That is an entirely different level of data and visibility, and it allows people to aggregate and hedge in different ways. They like that.

MR. ATKINSON: Bankability also turns on whether the software company will be around in 10 to 20 years.

I am sure people are seeing required escrows. It is an advantage to be a company that looks stable. These are assets with long lifetimes. They have to be able to operate as promised at the back end. A 20-year performance guaranty is really worth nothing if the manufacturer is not around after three years. Bankability relies on a number of things, and one of them is clearly the perceived longevity of the firm.

MS. BARROW: So software companies are being asked to escrow source code. Jen Worrall, how do you license your software?

MS. WORRALL: We have a couple of different models.

Something that we get asked about by everyone is cybersecurity. That is a major risk that can affect the operation of a project. An external factor could completely interrupt the operation of a project, cause irreversible damage and result in lost profits to the customer.

MS. BARROW: Pedro Elizondo, any thoughts on bankability?

MR. ELIZONDO: Since we are talking about 20 years of project life, one of the problems is the lack of data. As suppliers move to more efficient racks that provide more energy, there is less data. For example, one of the suppliers in less than two years has moved from 100 kilowatt-hours per rack to almost 200 kilowatt-hours, double per rack. Data is going to be a really key part to model the new battery racks, but there is obviously no operating history. Advances can sometimes be less bankable for this reason, even though they are improvements in what existed before.