Category Archives: Uncategorized

New paper in Applied Energy

I am pleased that our paper “Projecting battery adoption in the prosumer era” has been published in Applied Energy!
The website Science Trends also featured an article highlighting our work.

The paper investigates two aspects of battery adoption for residential consumers with PV generation. We use a data driven methodology based on 15-minute electricity consumption and PV generation data from the Pecan Street Project, from which electricity consumption data for several hundred households is available. The households are primarily located in Austin, TX, however other locations in the US are also included.

Firstly, the paper investigates what size of batteries are required for consumers to reach certain thresholds of self-sufficiency, and secondly, it investigates how the economic case for residential batteries looks under a range of electricity tariffs, including both current tariffs and future projections.

Figure 1: Illustrating self-sufficiency with PV and the effect of a battery. (A) The demand and generation from an example household – we see a significant amount must be exported. (B) The mean increase in household self-sufficiency for a given battery capacity. (C) The mean marginal increase in self-sufficiency as a function of battery size.

In the first part, we find that for the studied 369 consumers, the mean self-sufficiency from PV-only (without storage) is 36%, meaning that when a household installs a PV system that produces the equivalent of their yearly consumption, we expect them to export 64% of the total PV generation to the grid as it does not align with their consumption (see Fig 1A). Using batteries, we find that modest sizes can increase consumer self-sufficiency significantly, for example, a 10 kWh battery leads to a mean self-sufficiency increase of 25% across all households (see Fig. 1B). However, we find that (in terms of self-sufficiency), there is very little benefit of installing a battery with more than 3 kWh of storage per MWh of yearly PV generation (Fig. 1C).

In the second part of the analysis, we utilise consumer batteries explicitly for electricity bill minimisation. We find that while the economics of PV installations were very favourable in several regions (i.e. we find that PV has a Net Present Value (NPV) greater than $5000/kW for the average consumer under various CA tariffs), no current tariffs were favourable for battery storage and batteries produced negative NPVs (we studied both NEM and FIT based tariffs in our household regions – Austin, Boulder and San Diego). Therefore, projecting into the future, we considered a range of retail electricity price levels and FIT remuneration rates. Our results suggest that combined PV-battery systems generally become profitable with electricity prices above $0.40/kWh and solar rewards as low as $0.05/kWh or below. Interestingly, we find that decreasing PV installation costs has a strong negative effect on battery NPV, while decreasing battery costs and increasing efficiency had small positive effects (see Fig 2).

Fig 2: Comparing the difference in payback between PV-only and PV-battery systems for the median consumer. Payback difference = (Payback PV-only)-(Payback PV-battery). Average retail electricity prices and subsidy levels in different regions for the year 2015 are marked for context.

The main contributions of the work are to further understanding regarding the size of energy storage for households and to define the electricity prices and subsidy levels at which a large deployment of household batteries may be expected, which is important for grid operators and utilities to understand.

For anyone interested the battery scheduling optimisation code is available on my github and is described in this post.

Scheduling battery operation dependent on prices, demand and PV

Tags: Mixed Integer Linear Programming (MILP), Python, Pyomo, CPLEX, optimizing battery operation, residential batteries

In this blog post I will look at how to optimize a battery schedule using a Mixed Integer Linear Programming (MILP) formulation of the problem. I use data from a recent data-driven competition, which investigated using batteries to minimize a site’s electricity bill. I used this in a paper recently (Projecting battery adoption in the prosumer era in Applied Energy) and this post documents the problem formulation.

All the code and data is available in this repository containing a jupyter notebook which goes along with this post. This post is written to go along with the code in the repo.

The repository also contains an alternative formulation (also used in the paper) as well as notebook which applies the formulation to all test data set of the data driven competition to generate a score.

Getting set up

First of all, one needs to get set up with pyomo and CPLEX, which are the modelling language and the solver we will use. I wrote myself a rough set of notes for what I did to get set up and running with CPLEX/pyomo in jupyter notebooks on a linux server, and they might help… but I also imagine you can find some better resources for this – here is a link to the actual pyomo documentation.

Scheduling a battery using pyomo and MILP

Secondly, we need some data about electricity prices and battery properties. Additionally, since the problem is a little bit boring with only a single set of prices and more often than not people are interested in batteries deployed at a specific consumer site, we will consider batteries located at consumer sites with demand and generation, where there is a buy price and a sell price for electricity. Conveniently, the data can then come from the competition!

Our aim is to use the battery to reduce or increase the site load at different times, in order to minimize the site electricity bill. Since there are buy prices and sell prices for electricity, there is a difference between reducing the load to zero (i.e. reducing the amount the site must buy) and reducing the load below zero (i.e. increasing the amount that the site is selling). Each site has a PV installation, so there is both demand and generation, but the price at each period only depends on the net load. The demand (d(t)), PV generation (PV(t)), buy price (buyPrice(t)) and sell price (sellPrice(t)) are timeseries with a fixed value at each time period.

To calculate the net load, we sum the demand, the generation and the battery action at each period, as shown.

netLoad(t) = d(t)+PV(t)+ES(t) (1)

The price at each period is given either by the buy price, if the site is importing electricity, or the sell price, if the site is exporting electricity. This is summarised conditionally as follows:

IF  Net(t) > 0, price(t)=buyPrice(t) , ELSEIF  Net(t) < 0, price(t)=sellPrice(t) (2)

We can see from the first equation that the unknown is the input/output of the battery, ES(t) (if there is no battery then this term is equal to zero at all times)

The battery is then defined by its characteristics. We model a battery with the following parameters:

1. Maximum and Minimum states of charge, SOC_{max} and SOC_{min}
2. Charging and discharging power limits (which could be different, but in the example we assume they are equal)
3. Charging and discharging efficiency (\eta_{Char} and \eta_{Dis}, which again can be different, but we assume that they are equal and so the roundtrip efficiency is the product of the charging and discharging efficiency).

It must be noted here that this is a very simple battery model that does not take into account potentially important factors such as temperature, depth of discharge, applied current and voltage.

Now that we are done with the definitions we can go into the modelling procedure. As I mentioned earlier, we will use data from the recent data driven competition. The competition provided two datasets, a training set and a testing set. We will just use the testing set, since we are developing a data-independent model. In the competition, the demand was unknown and a forecast was given, and additionally the prices were only specified 1 day into the future. Hence, the training set was introduced so that competitors could learn specific features about each site and improve their algorithms.

Since we are interested in obtaining the optimum battery schedule, we will just take all the information we need from the testing dataset (in a folder called submit – you may need to sign up to get access – it is also all available in my repository). Therefore, for our chosen site (siteId=1) we will take values for the actual demand, actual PV generation and electricity prices for both buying and selling from the data provided (we won’t use the price, demand and generation forecasts at each period).

We will get the battery parameters from the metadata file from the competition repo:

The dataframe header showing battery properties

So we will initially model a battery with a capacity of 300 kWh, a charging power of 75 kW and a roundtrip efficiency of \eta_{Char} \eta_{Dis} = (0.95)^2.
Now let’s look at the data from site 1.

The data is in a csv file and it is easiest to access using pandas. It is split into different 10 day periods so we will just look at the first one.

Here is a plot of the demand and generation as well as the buy and sell prices for electricity.

Demand and generation from site 1 as well as the electricity prices

We can see that at certain times PV(t)>demand(t) and therefore the site will be exporting electricity at the sell price. Looking the price profile, we can also see that the export price is always a constant level which is less than the highest buy price. Therefore, we get the intuition that the optimum battery operation is likely to store electricity that would be sold and use it to replace electricity that would have to be bought at expensive times.

To set up the optimisation problem, we first define the variables and an objective function, which corresponds to the total price of the consumer electricity. Pyomo requires us to set up the problem algebraically. What this means is that we define a set of variables, the objective function and constraints on the variables. The majority of the difficulty of setting the problem up in this way is writing down the variables, their relationships to each other and constraints such that the problem makes sense and does exactly what we intend.

First off, we need to initialize the model, for which we will use the concrete model framework in pyomo (concrete model means that we will define all the parameters at the time of the model definition, as opposed to an abstract model where we formulate the problem algebraically and parameter values are specified at a later stage). We also set up the model index which will be our times.

# now set up the pyomo model
m = en.ConcreteModel()

# we use rangeset to make a sequence of integers
# time is what we will use as the index
m.Time = en.RangeSet(0, len(net)-1)

Now let’s write down the objective function taking in mind the Equations 1 and 2 above:

ObjFn = SUM [buyPrice(t)*posNetLoad(t) + sellPrice(t)*negNetLoad(t)] (3)

posNetLoad is the net load at every period if the net load is greater than zero and negNetLoad is minus the net load at every period if the net load is less than zero. posNetLoad and negNetLoad are the decision variables.

Let’s also write down rules governing the state of charge of the battery. This consists of two components; the rule governing the minimum and maximum states of charge and the rule governing how much the state of charge changes at each period. The rule governing the max and min states of charge is a simple constraint, and can be included using bounds on the variable SOC. In pyomo we can do this when initializing the variable:

 0 \leq SOC(t) \leq batteryCapacity (4)

# variables (all indexed by time)
m.SOC = en.Var(m.Time, bounds=(0,batt.capacity), initialize=0)

Integer constraints

Here is where the “mixed integer” part of the formulation comes in. We specify constraints ensuring that the battery can only either charge or discharge at a particular period. To do this we specify Boolean variables for charging and discharging, boolChar(t) and boolDis(t). We then bound the problem using the bigM method. Therefore, we formulate the following constraints:

posDeltaSOC(t) \leq M(1-boolDis(t)) (5)
posDeltaSOC(t) \geq -M*boolChar(t) (6)
negDeltaSOC(t) \geq -M*boolDis(t) (7)
negDeltaSOC(t) \leq M(1-boolChar(t)) (8)

The first constraint means that if the battery is discharging, the charging must be less than or equal to zero and the second forces it equal to zero. If the battery is charging then the first constraint imposes an upper limit that is at least two orders of magnitude greater than any feasible value and the second constraint imposes a redundant lower limit.

If discharging, the third constraint means gives a lower limit to the discharging (remember discharging is –ve) and the fourth gives a redundant upper bound. If charging, the third and fourth constraints force negDeltaSOC equal to zero.

To ensure that the battery cannot simultaneously charge and discharge at the same time we constrain that boolChar(t)+boolDis(t) = 1.

Other constraints and variable relationships

We then define posDeltaSOC and negDeltaSOC in terms of energy coming from the grid to the battery and energy coming from the PV, at this point ensuring that the charging efficiency decreases the amount of energy that actually makes it into the battery with the opposite for the discharging energy.

posEInGrid(t)+posEInPV(t) = posDeltaSOC(t)/ \eta_{Char} (9)

Similarly,

negEOutLocal(t)+negEOutExport(t) = negDeltaSOC(t)*\eta_{Dis} (10)

For the charging and discharging power limits, we must also use constraints rather than bounds, since we have split charging into the component which comes from the PV and that which comes from the grid. Therefore we have:

posEInGrid(t)+posEInPV(t) \leq ChargingLimit (11)
negEOutLocal(t)+negEOutExport(t) \geq DischargingLimit (12)

Then, to make sure that energy available for charging from PV is always less than or equal to the excess PV and the energy used locally must be less than or equal to the demand that is not met by PV, we specify the following constraints:

posEInPV(t) \leq \mbox{ Initial negative }netLoad(t) (13)
negEOutLocal(t) \geq \mbox{ -Initial positive }netLoad(t) (14)

Finally, we then define the decision variables in terms of the other variables:

posNetLoad(t) = \mbox{ Initial positive } netLoad(t)+posEInGrid(t)+negEOutLocal(t) (15)
negNetLoad(t) = \mbox{ Initial negative } netLoad(t)+posEInPV(t)+negEOutExport(t) (16)

Optimisation

Now that the problem is defined we can run the optimisation. We need to specify what tool we will use for the optimisation, which in our case will be cplex. Therefore we set:

opt = SolverFactory("cplex", executable="**insert you path here**")

And finally we can run!

# time it for good measure
t = time.time()
results = opt.solve(m)
elapsed = time.time() - t

Outputs

To get the model outputs, we cycle through the variables that we defined in the model component objects, and I prefer assigning each to a numpy array. Therefore, we loop through the model variable objects and assign the variables we want to arrays.

# now let's read in the value for each of the variables 
outputVars = np.zeros((9,len(sellPrice)))
j = 0
for v in m.component_objects(Var, active=True):
    print v.getname()
    #print varobject.get_values()
    varobject = getattr(m, str(v))
    for index in varobject:
        outputVars[j,index] = varobject[index].value
    j+=1
    if j>=9:
        break

Finally, we can then compare the consumer’s electricity bill with and without the battery, and look at what the optimal battery schedule and the consumer’s net demand are! We can also calculate the consumer’s electricity cost with the battery and compare this to their cost without the battery.

# get the total cost
cost_without_batt = np.sum([(buyPrice[i]*posLoad[i]/1000 + sellPrice[i]*negLoad[i]/1000) for i in range(len(buyPrice))])
cost_with_batt = np.sum([(buyPrice[i]*outputVars[7,i]/1000 + sellPrice[i]*outputVars[8,i]/1000) for i in range(len(buyPrice))])

The optimum battery operation for the site

Looking at the results we can see that the battery is doing the things we would expect – charging at the times when the price is low and discharging when the price is high. Finally, it is worth noting that in this example, there are a range of different optimum solutions – there are lots of different time options when the battery could charge and discharge at the same price (and the cost to the battery of PV-that-would-otherwise-be-exported is the same as the grid import price).

Concluding notes

The nice thing about using this setup is you can just keep adding more and more constraints and different forms of the cost function. For example, one could stipulate that a portion of the consumer’s electricity bill was related to their peak power usage. This allows one to model more and more complex electricity bill scenarios. However, as the problem gets more complicated, you need to be careful about what methods you are using to solve the problem, which is beyond the scope of this post (see the CPLEX manual for details regarding the type of problems it can solve). Additionally, as the number of constraints increases the computational time required will increase.

I wrote this post primarily because when I was first looking into MILP formulations, I didn’t find many easily accessible examples of code that had been used for this type of exercise in academic papers. Therefore, I wanted to document my process.

Finally, thanks to my colleague, Alejandro Pena Bello at the University of Geneva first suggested that I try pyomo for MILP formulations in python and shared some code with me, which was super helpful. Additionally, I found these notes from MIT opencourseware useful when trying to understand how to set the problem up. The tutorials were also helpful.

OSES 2018

I recently heard that the OSES (Offshore Energy and Storage Summit) conference series may still be considering abstracts for their conference on 4-6 July 2018 in Ningbo, China.

This is a great conference and a great place for ideas and discussion surrounding all aspects of offshore energy. Thermo-mechanical energy storage technologies in particular are a major theme due to their potential for use with large scale offshore renewable technologies, and the conference committee certainly contains some of leaders in this field; Seamus Garvey, Rupp Carriveau, Tonio Saint, Daniel Friedrich, Jonathan Shek and Andrew Pimm.

I was lucky enough to go to the inaugural OSES conference in 2014, where I presented a concept for isobaric CAES using carbon dioxide to maintain high pressure storage.  It was thoroughly enjoyable as well as introducing me to a lot of interesting people and ideas.

Unfortunately I won’t be able to attend this year, but I plan to return to the thermo-mechanical scene soon so perhaps in 2019…

Also – for anyone wondering – I never got around to publishing the “isobaric CAES using CO2” concept in a journal article, but last year another team from China had the same idea. I’m definitely annoyed that I didn’t get around to submitting it as I have had a very similar article ready to submit (the extension of my conference paper), but it will teach me to be more proactive and I guess it is good that other people are thinking hard about CAES too!

 

Is community energy storage a smart choice for a smart grid?

Our paper entitled Community Energy Storage: A smart choice for the smart grid? has recently been accepted in Applied Energy. The paper looks at the economic drivers for energy storage when considered from the perspective of an individual household compared with a community perspective.

The main motivation behind the paper was the engineering intuition that a battery in every household does not seem likely to be an optimal solution for the future smart grid, given high battery costs and the rare and expensive nature of the constituent raw materials for battery manufacture. In the paper, we study the economic arguments for both individual households and communities, which illustrated a number of key points as outlined below:

  • Most importantly, aggregating household demand profiles reduces the required battery capacity, with each kWh (kilo-Watt hour) of installed storage being significantly more effective at integrating PV generation when deployed at the community level.
  • It is possible to realize significant cost benefits at a community level from the economy of scale of both batteries and inverters.
  • However, the rate of economic return for individual household batteries is more robust to the changes in the solar resource, despite being less effective for PV integration. This is due to the fact that it doesn’t make economic sense for an individual household to purchase a battery which can always store all the surplus generation, rather an economically optimum size is likely to be a size which can be fully utilized most days.

Due to the points above, our work highlights the concern that residential storage may gain a significant market foothold despite community-level batteries being much more effective from both an economic and a costs perspective. Accordingly, it is crucial for energy policy to consider new market mechanisms to encourage community storage projects in areas where they are more economic and environmental.

Forming communities of neighbours by joining households along the road network in Cambridge, MA

Important modelling assumptions

The paper considers a future scenario wherein the electricity price ($0.35/kWh) is high enough that batteries do provide a return for some users, and we model demand, PV generation and network structure based on empirical data. We also assume a FIT rate which is much lower than current rates ($0.05/kWh), which is based on average wholesale prices for electricity production. Hence in a regime wherein solar PV is widespread, we may expect PV to be rewarded at near-wholesale price levels.

The location data comes from an electric utility in Cambridge and the raw data cannot be shared, however to model the demand and PV generation data we use the Pecan Street project, which supports free academic access to data. Using the location data, we generate hypothetical communities of households, which are grown outwards along the road network, using a multi-source breadth-first search method based on the Dijkstra shortest path algorithm. This could also be useful for estimating the topology of the real distribution network, however, validation in this regard is difficult as the exact topology is deemed sensitive for security reasons.

For each community, and for each individual household, we find the level of storage that maximizes the Net Present Value (NPV) of the battery system. We simulate approximately 4500 consumption patterns for a month using 484 actual demand patterns in the Pecan Street data. Furthermore, we generalize our results to a high degree by considering many different PV sizes, based on empirical distributions of the sizes that households tend to install.

All the code available for running the simulations is available on github.

 

Lightsail – another setback for thermo-mechanical energy storage

Recently there has been another blow for thermo-mechanical energy storage, and in particular adiabatic CAES, as Lightsail – the much hyped company bankrolled by Bill Gates – has entered “hibernation”. This appears to be due to running out of cash, with the company ceasing operations. This is can be added to list of failures in CAES, including SustainX and ApexCAES.

Lightsail’s concept – the $70 million idea

Many have watched the company with interest given its hype, larger than life founder and large amounts of investor funding – sources indicate this exceeded $70 million. Their various claims of high thermodynamic efficiencies which had already been achieved were also interesting – although it is hard to tell from the little available information whether this was simply a result of including the compression heat in their calculations (if you are interested in compressor performance you typically only consider the difference in air enthalpy between compressor inlet and outlet). As such, a little skepticism regarding this is probably healthy, which is emphasized by the company’s switch from CAES system to high pressure storage container manufacturer. Their water-spray injected compressor was an interesting idea, but was probably difficult to manufacture in practice. Moisture in compressed air can cause many problems including corrosion, aiding the build up of sludge deposits by mixing with oil/dust and as a result increasing pipe pressure losses, and causing faults in sensor and control equipment to name just a few issues off the top of my head. Crucially, it must also have been very difficult to avoid freezing during their expansion process, and water droplets/icicles in fast moving machinery can also be incredibly damaging. I should also note that their patent does also refer to oil spray injection.

Lightsail’s move to high pressure container storage manufacturer smacks of desperation and suggests to me that the management realised that they had completely underestimated the scale of the ACAES challenge – indeed a review from a former employee on glassdoor quotes the Cons of working at Lightsail as “Large scale energy storage is NOT an easy problem to solve”! The high pressure storage tank is probably the only part of the ACAES system that actually exists off-the-shelf, which throws up questions of why this would be the component of choice to manufacture. I have discussed in previous posts my opinion that it is most certainly a misconception to think that a ACAES system can be constructed with “off-the-shelf” components. The primary reason would seem to be that they thought they could implement a marked reduction in container costs compared to leading manufacturers, and a recognition that building high pressure vessels was certainly possible, as these type of containers do exist and have a market. This would have potentially allowed them to make incremental scientific progress and make a better or cheaper version of a product that already exists rather than the more ambitious aim of a game-changing energy storage technology. The problem with this turned out to be that they couldn’t make a product that was enough of an improvement to ship many orders.

It is very difficult to work out exactly what went on at Lightsail – I would love to talk to someone who worked there as I’m sure they were full of talented employees. There were also various reports about questionable spending habits on behalf of the senior management – again a look on glassdoor also suggests this. At least they were probably good to their employees, even offering pet insurance as a benefit…

In any case, it’s a huge shame for thermo-mechanical energy storage, and what’s most frustrating is that there is almost no ability for anyone to learn from the technical failures that undoubtedly sank the company. Picking through the Lightsail patents (now owned by the Silicon Valley Bank) is difficult, and they just seem to be full of quite generic ideas, with no indication of any ingenuity in the designs. This reinforces why these early stage companies are so frustrating, and can ultimately be bad for the technology development, as they simply put-off would be investors while no new scientific knowledge can be gained. Maybe I will email Bill Gates and see if he can push for any knowledge to be made public…

CAES: A simple idea but a difficult practice

Download available here.

In the mainstream there are two main branches of Compressed Air Energy Storage (CAES) – conventional and adiabatic.

  1. Conventional CAES

Conventional (also known as diabatic) CAES plants are essentially gas turbines in which air is pre-compressed using off-peak electricity, rather than running a turbine and compressor simultaneously. In these plants, off-peak grid electricity is used to compress air which is stored, and then mixed with natural gas and combusted during expansion. Compression is staged and the majority of the compression heat wasted (although some may be stored in a recuperator to pre-heat the air before combustion). Currently there are two commercial CAES plants worldwide; the Huntorf plant in Germany and the McIntosh plant in Alabama.

  • Huntorf CAES plant: Data from [1]. 310,000m3 cavern at a depth of 600m, pressure tolerance between 50 – 70 bar, converted from a solution mined salt dome. Daily charging cycle of 8h, output of 290MW for 2 hours. 0.8kWh of electricity and 1.6kWh of gas required to produce 1kWh of electricity. Notably, built when the price of gas turbines was historically high.
  • McIntosh CAES plant: Data from [2]. 538,000m3 salt cavern at a depth of 450m, pressure tolerance between 45-76 bar. Originally it provided an output of 110MW for 26 hours but in 1998 two extra generators were added and its total output capacity is now 226MW. 0.69kWh of electricity and 1.17kWh of gas to produce 1kWh of electricity.

Both plants are commercially viable and still running in their respective markets!

CAES

Figure 1: Schematic of diabatic CAES system.

As with Pumped Hydro Storage (PHS), CAES also requires favourable geography to provide the underground air storage caverns. However there are many more suitable sites worldwide than for PHS, although the costs are highly site specific. The costs of mining a suitable underground cavern where suitable geology doesn’t exist or creating an above-ground equivalent storage container are potentially prohibitive, whereas alternatively a naturally occurring cavern or somewhere easily minable may offer a very attractive price of storage in terms of $/kWh (or dollars per metre cubed of air storage).

Caverns can be created in salt geology (typically using salt solution mining techniques) or existing caverns can be exploited provided that they are capable of housing the desired pressure. Geological formations such as aquifers and salt formations (bedded salt and domal salt) offer potential locations. Costs can also be reduced if existing well infrastructure is in place from previous underground drilling operations. While specific geology is required, this geology is relatively widespread. For example, the EPRI suggests that up to 80% of the US could have favourable geology [3] (see Figure 2).

US CAES map with wind resources marked

Figure 2: US geology for compressed air caverns. Regions with high wind resources are also indicated with the idea that CAES sites and wind turbines could be co-located [4].

Estimates for the costs of cavern mining can be as low as $1/kWh of storage capacity if solution mining techniques can be used [5]. In solution mining, fresh water is pumped in a salt deposit, becomes saturated with salt and is then removed. One problem however is that disposal of this brine can cause environmental issues.

1.1 CAES Performance Characteristics and Applications

CAES systems have traditionally been designed as centralised storage facilities which are intended to cycle on a daily basis and to operate efficiently during partial load conditions. This design approach allows CAES units to swing quickly from generation to compression modes and means that they are well suited to ancillary services markets, providing frequency regulation. Their ability to operate on a (intra) daily cycles means that they are also useful for load-following/peak shaving. The air storage caverns can also be very large, allowing for multiple days worth of electricity storage.

It should be noted that the inlet pressure (45-76 bar) for the CAES high pressure turbine is much higher than the equivalent for a typical gas turbine (about 11 bar) so a typical gas turbine can only be used as the low pressure expander. The high pressure turbine at Huntorf is based on a small-intermediate steam turbine design.

1.2 Table of Cost Estimates

Typical Capacity Typical Power Efficiency Storage Duration $/kWh $/kW Lifespan Cycling capacity
500MWh – 2.5GWh 50 – 300MW n/a Hours – days 4-7 [6], 2-50 [7], 60  – 120 [8] 300-600 [6], 400-800 [7], 1000-1250 [8] 20-40 years High

Table 1: CAES cost characteristics

 

  1. Adiabatic CAES

Adiabatic CAES is an energy storage concept that removes the natural gas combustion from conventional diabatic CAES. In adiabatic CAES the heat generated by the compression of air (the charging process) is stored in a Thermal Energy Store (TES) which is separate from the ambient temperature high pressure air store. When the system is discharged the high pressure air is reheated using this stored heat and then expanded. Without the stored heat, the process has an unacceptably low efficiency – this is because significant exergy is stored in the heat as well as the cool high pressure air. When the heat is recovered, the expected practical efficiency of these systems is debated – though the second law of thermodynamics does not pose a ceiling on the efficiency as for  heat engine – it just means that the real process has to be less than 100% efficient. Pragmatic estimates of the real efficiencies of this type of system are debated; most of the academic literature estimates practical efficiencies in the range of 60-75% [9,10]. If a plant could be constructed with no inefficiencies in any process – the theoretical efficiency would approach 100%.

2.1 Status

As no demonstration plant has ever been successfully constructed, Adiabatic CAES must be considered as an unproven technology. It does however have significant promise for use with renewables integration, energy management, peak shaving and grid reserves. The largest planned demonstration ACAES facility is a 290 MW adiabatic CAES project based in Germany called project ADELE [11]. It is a consortium between German utilities RWE and GE, the German Aerospace Center DLR, construction company Zublin, the Fraunhofer IOSB and the Unversity of Magdeburg.

Adiabatic CAES

Figure 3: A simple schematic of an ACAES configuration. There is a thermal store for each compression stage.

A schematic diagram of an ACAES system is shown Figure 3. In this configuration, air is compressed and then cooled using counter-current heat exchangers that transfer the heat from the air into a thermal fluid. This thermal fluid could then be stored in an insulated tank and used to reheat the air prior to each expansion stage. Several people have also suggested the use of Packed Bed regenerators to store the compression heat in the air.

2.2 Underwater CAES

Underwater CAES is a sub-type of ACAES which exploits an underwater Compressed Air Store at a depth of typically around 400m. The ambient pressure at this depth is approximately 40 times the atmospheric pressure, and the air store is either a flexible bag or a dome structure open at the bottom. As air is pumped into the storage container it displaces water and thus the store can operate at a constant pressure. This idea was pioneered by Prof Seamus Garvey and Dr Andrew Pimm at the University of Nottingham, as well as by researchers at the University of Windsor Ontario and Canadian startup Hydrostor (whose work is ongoing at the time of writing).

2.3 Fuelless CAES

The usage of the term “adiabatic CAES” is also somewhat ambiguous, as the term “adiabatic” is sometimes used to refer to the compressions and sometimes to refer to the overall process – i.e. the energy storage process aims to be adiabatic in the sense that ideally, it would exchange negligible heat with the surroundings. Therefore some authors therefore prefer the use of the umbrella term Fuelless CAES. This then clearly encompasses all compressed air processes which aim to store and return energy without the use of fossil fuels. This includes systems which have typically been labelled as isothermal CAES.

2.4 Isothermal CAES

In isothermal CAES the compressions aim to be isothermal and reversible. This is theoretically achieved by minimising the temperature differences which drive heat flow from the compressors to the environment (which is at a lower temperature). A huge challenge here is to make an isothermal compression process which operates sufficiently quickly to be of practical industrial importance but which is still slow enough to maintain the small temperature differences required for high reversibility. One idea for near-isothermal compression which has been suggested by LightSail (a start-up company in California) involves a water spray into the compression chamber of a specially designed reciprocating compressor/expander unit (see Figure 4). The water droplets absorb the heat of compression and their high specific heat capacity causes the temperature increase in the compression chamber to be much smaller. This warm water is then stored and on discharge is re-injected as a mist into the reciprocating machine which now acts as an expander.

Figure 4: Illustrating a near-isothermal CAES concept [12]

Isothermal CAES was also being pioneered by SustainX, however this company has ceased operations citing spiralling system costs. Lightsail Energy and SustainX had a similar goal of an efficiency above 60% for their first generation of machines and believe that 75% is achievable in the long term. The SustainX prototype was a 1.5 MW machine.

2.5 ACAES Challenges

There are several challenges which must be overcome before adiabatic CAES can become a viable energy storage technology option.

  • Specialised compressor equipment must be developed, in which the heat generated during the compression procedure is stored in a highly reversible manner. This process seems most likely to consist of a series of adiabatic compressions in which heat losses from the compressor to the surroundings are minimised. The compressors must also operate with much higher compression ratios than current compressors which do not involve cooling during the compression. Each of the compressions is then followed by a cooling stage which aims to reversibly extract the compression heat. Possible options for heat extraction include packed bed regenerators or counter-current indirect contact air-to-fluid heat exchangers. This type of compression equipment is fundamentally different to industrial many industrial compressors. Why? Because the vast majority of compressors are designed to minimise the work required to achieve air at a given pressure. Most industrial compressions then typically involve trying to shed as much heat as possible from the compression process – as hot air takes more work to compress. The ACAES process is fundamentally different as reversibility should be maximised rather than work minimised. In fact, the greater the reversible work is per cubic metre of compressed air the higher the energy density of the storage system.
  • Specialised expansion equipment must also be developed. Air turbines which provide highly isentropic expansions and operate within the desired pressure ratios are required. The expansion process of an Adiabatic CAES system should aim to mirror as closely as possible the reverse compression process. Therefore it should include the same number of expansion stages and heating stages, and expansion stages must aim to minimise heat gain and return all heat reversibly during the heating stages. While these turbines do not currently exist on the industrial market, it is anticipated that their design can learn much from the current generation of gas turbines for power generation. The pressure ratios will likely be smaller than most current gas turbines. One specific advantage is that the material demands will be much less (in terms of temperature tolerance) than current gas turbines which operate with inlet temperatures up to 2200K.
  • Sliding pressures. Unless the system can be operated between constant operational pressures, both the compression and expansion machinery must operate at maximum efficiency over a range of pressure ratios. A single constant high pressure air storage is a primary advantage of UnderWater CAES.
  • High pressure air storage. Depending on the chosen method of storage high pressure, air storage tanks must be developed which have minimum cost. This has apparently been a problem area both for SustainX and LightSail, however LightSail have released statements which hint that they may have found a method of lowering the costs.
  • highly reversible heat exchangers will also be required which can minimise the temperature difference between the working fluid and the thermal storage medium while introducing minimal pressure drops.

2.6 Notable experimental ACAES development

Lightsail (California) – startup. http://www.lightsail.com/

Hydrostor (Ontario) – startup. https://hydrostor.ca/

SustainX (Massachusetts) – startup (liquidated)

Project Adele (Ongoing utility/academic collaboration – big unexplained delays??)

University of Windsor – Prof. Rupp Carriveau and Dr. David Ting

University of Nottingham – Prof Seamus Garvey and Dr Andrew Pimm

 

 

 

 

References

[1] BBC Brown Boveri. Huntorf Air Storage Gas turbine Power Plant. https://www.eon.com/content/dam/eon-content-pool/eon/company-asset-finder/asset-profiles/shared-ekk/BBC_Huntorf_engl.pdf

[2] M. Nakhamkin, L. Andersson, E. Swensen, J. Howard, R. Meyer, R. Schainker, R. Pollak, and B. Mehta, J. Eng. Gas Turbines Power 114, 695 (1992). https://doi.org/10.1115/1.2906644

[3] Compressed Air Energy Storage: Renewable Energy (2010, March 17) retrieved 22 April 2017 from https://phys.org/news/2010-03-compressed-air-energy-storage-renewable.html

[4] Succar, S & Williams, R.H.. Compressed Air Energy Storage: Theory, Resources, and Applications for Wind Power, Princeton University (published April 8, 2008)

[5] De Samaniego Steta, F. Modeling of an Advanced Adiabatic Compressed Air Energy Storage (AA-CAES) Unit and an Optimal Model-based Operation Strategy for its Integration into Power Markets. EEH – Power Systems Laboratory. Swiss Federal Institute of Technology (ETH) Zurich

[6] Kaldellis, J. K. & Zafirakis, D., 2007. Optimum energy storage techniques for the improvement of renewable energy sources-based electricity generation economic efficiency.. Energy, Volume 32, p. 2295–2305.

[7] Chen, H. et al., 2009. Progress in electrical energy storage system: A critical review. Progress in Natural Science, Volume 19, pp. 291-312.

[8] EPRI, 2010. Electricity Energy Storage Technology Options. http://large.stanford.edu/courses/2012/ph240/doshay1/docs/EPRI.pdf

[9] G. Grazzini, A. Milazzo. A Thermodynamic Analysis of Multistage Adiabatic CAES. Proc IEEE, 100 (2) (2012), pp. 461–472

[10] Barbour, E, Mignard, D, Ding, Y,  Li, Y. Adiabatic Compressed Air Energy Storage with packed bed thermal energy storage, Applied Energy, Volume 155, 1 October 2015

[11] RWE Power. ADELE – Adiabatic Compressed Air Energy Storage for Electricity Supply. https://www.rwe.com/web/cms/mediablob/en/391748/data/364260/1/rwe-power-ag/innovations/Brochure-ADELE.pdf

[12] Fong, D. Insights by Danielle Fong. https://daniellefong.com/

 

 

 

Storage and the duck

The California duck curve is now infamous and is very often features in discussions around storage. The duck phenomenon is a result of several factors coming together at once to create a scenario in which there is significant strain on the electricity generation system.

The infamous California duck

The now infamous California duck.

Typically the output from solar panels is well-aligned with times of high electrical demand, especially in systems which have large cooling dominated loads. This is because it often gets hot when the sun is shining and people tend to be most active during the daylight hours.

Solar generation typically occurs when demand for electricity is high - during the middle of the day.

Solar generation typically occurs when demand for electricity is high – during the middle of the day.

However when there is a cool sunny day in systems which have a lot of solar panels that are typically used to meet cooling-driven loads, then the situation can arise in which the net demand for electricity which must be generated by conventional powerplants (i.e. coal, nuclear, gas) becomes depressed, as most of the demand can be met by the solar. This is a problem for utilities in itself as turning down the output on some of these plants (especially nuclear, to a lesser extent coal) is difficult and costly, so instead they sometimes opt to sell their electricity very cheaply (or even pay for it to be used when prices go negative). For utility-scale renewables this is also a problem, as they can end up in the situation where they simply have to stop producing electricity. On top of this, the power output from all the solar panels in a local region is very well correlated. Therefore they all start and stop producing power at close to the same time (there is some spread due to orientation and location). This leads to a sharp increase in the net demand leading up to the evening peak which typically occurs after the sun goes down. There are only certain types of plant which can react to changes in demand quickly (they have high ramp rates), for example gas and hydro and only hydro can do it cheaply, as conventional gas plants must already be running for some time at their Minimum Stable Generation levels before ‘ramping up’, which is often less economic and more polluting per unit of electrical output.

Solar Panel outputs from the Pecan Street project (https://dataport.pecanstreet.org/)

Solar Panel outputs from the Pecan Street project (https://dataport.pecanstreet.org/) all producing electricity at the same time. Red line is the average

The concern about the duck is a prime driver for energy storage development. This storage can come in several forms – i.e. not just batteries coupled with the solar panels. Some of these are highlighted in this NPR discussion which includes fuelless Compressed Air Energy Storage, Concentrated Solar Power with thermal storage in Molten Salts and Ice Storage for cooling.

Ultimately it is all down to the economics. If the costs of storage are less than the increased costs of utilities as a result of having to provide the additional flexibility the duck requires, or if storage can increase the value of renewable energy sufficiently then it will become a viable option. At present the costs of curtailment are likely to be less than storage, but as the amount of curtailment increases and storage costs fall then this could rapidly change.

 

 

Energy Policy and the UK 2015 general election

This is just a very quick post to link to some useful resources and discussions regarding energy policy and the UK general election. Although it is very difficult to tell exactly what will arise from the tiny snippets of information given by the manifestos of the UK political parties, I think that energy and climate change are such important questions that these sections of the manifestos are worth looking at.

The Carbon Brief has a good blog post looking at the energy policies of the various parties that I would encourage anyone to read.

Between Labour and the Conservatives there isn’t too much to choose from, both support the climate change, support North Sea oil and see Nuclear as part of the future electricity mix. The Conservative plan to put a blanket halt to the development of onshore wind is slightly worrying, given the sizable onshore resource still available and the huge expense of installing turbines offshore. While onshore turbines aren’t suitable everywhere, current planning restrictions are pretty tight, and given the choice between the pollution and climate change threat associated with fossil fuels or “unsightly” wind turbines, it seems foolish to me to completely rule out the wind turbines! Their stance on fracking is another difference, with the Conservatives more strongly in favour.

The SNP and the lib dems are more supportive of renewables in general and of the development of CCS. The SNP also heavily supports the North Sea Oil and Gas industry.

Obviously the green party has an agenda heavily dominated by climate change, and they advocate ending tax breaks for fossil fuels and oppose nuclear power. They also have some fairly extreme goals for energy efficiency – i.e. a 50% reduction in energy demand by 2030. That’s not to say that with the right measures this isn’t achievable! Their position on nuclear power is difficult and I suspect that a softening on this view would see them gather more support, as many people – including Prof David MacKay – believe that some nuclear is probably a good idea given our current energy needs.

For anyone who believes that climate change poses a clear and present danger the UKIP manifesto is pretty scary – proposing to repeal the climate change act, remove renewable subsidies and use coal for cheap electricity.

 

 

 

China up to second for installed capacity of Pumped Hydro

While Pumped Hydroelectric Energy Storage (PHES) development has stalled in much of Europe and the USA, in the People’s Republic of China development is booming and the installed capacity had exceeded 22.5 GW by the end of 2014. This moves China into second place for installed pumped hydro globally above the USA which has approx. 21 GW; only Japan has more with 24.5 GW.

In addition, there is currently an additional 11.5 GW of pumped hydro under construction in China which is likely to see it take the lead by 2017. Japan is also currently constructing 3.3 GW of additional pumped storage. Figure 1 shows the development of PHES in Europe, Japan, China, USA and India.

Development of Pumped Hydroelectric Energy Storage in Europe, Japan, China, USA and India

Development of Pumped Hydroelectric Energy Storage in Europe, Japan, China, USA and India

Figure 1: PHES development in Europe, USA, China, Japan and India. Data from numerous sources including US DOE energy storage database. Available in text format here or from the downloads page.

.

While there are undoubtedly many reasons why investors in China and Japan are currently more willing to fund PHES schemes than those in Europe and the USA, the main difference seems to be due to the different regulatory and market structures that exist. In much of the US and Europe, PHES must be rewarded by the market and compete for services that are generally provided by power generation units – and it is treated in a very similar manner to these units. Treating electricity storage as generation makes little sense as storage makes pretty poor generation – the second law of thermodynamics forbids it from outputting more electricity than that which is inputted. Crucially, legislation normally forbids Transmission and Distribution (T&D) network operators from owning PHES (as well as generation). This means it is difficult to reward PHES for its use as a network asset and although the storage could provide benefits across the wider electrical network, the revenue available to them only reflects a small fraction of this value.

In China and Japan PHES plants are rewarded in a cost-of-service manner and can be used as network assets. The network operators can then dispatch these plants as they require for a variety of uses, including ancillary services (frequency response, voltage support, fast reserve etc), peak electric capacity and network congestion alleviation. If the plant can introduce an overall cost saving to the wider network then it is worth the investment.

Can negative electricity prices encourage inefficient electrical energy storage devices?

Following on from my earlier blog post I have written a journal article on this subject which has recently been published in the “International Journal of Environmental Studies”. The article is available to download here. It also gives a nice description of the mechanisms that can lead to negative electricity prices.

One of the main points is that we should only expect a less efficient storage device to be able to generate a higher revenue if the duration of the negative prices is long enough that a more efficient device with the same charging power would become fully charged before the end of the negative price period. Hence whether an inefficient device can indeed generate a higher revenue depends on the ratio between the charging power and the storage capacity. In the case of bulk storage (with the ability to charge and discharge for many hours) it seems unlikely that this situation will arise and the more efficient the device the more revenue it can generate.