View source for Projects:2018s1-140 Energy Storage Requirements for the SA Grid

== Team Members ==
*Julius Bullas

*Paul Citti

== Supervisors ==
*Derek Abbott

*David Vowles

== Advisors ==
*Holger Maier

*Angus Simpson

*Wen Soong

== Abstract ==
SA obtains approximately 45% of its electrical energy from renewable sources – large scale wind and small scale solar PV. 
To reliably integrate these intermittent sources into the grid will increasingly require energy storage in a variety of forms such as pumped-hydro and batteries together with virtual energy storage in the form of demand side management. The storage requirement will progressively increase as controllable fossil fuel generation sources are withdrawn from the system.
The objective of this project is to develop tools to assess the energy storage requirements to ensure reliable supply with high levels of intermittent generation.
This project commenced in 2017 and the two honours students involved did a very fine job in achieving several early objectives. Much work remains to be done, including the following:
*(1) Serve the time-series data via the web so that it can be analysed and graphically displayed in various ways online.

*(2) Extend the data analysis and graphical display toolbox to allow the user to rapidly identify and display interesting features and periods within the data.

*(3) Extend the storage optimization approaches to consider alternative optimization objectives and future scenarios.

*(4) To estimate storage requirements for Australia as a whole assuming a hypothetical 100% renewable scenario and hypothetical interconnected Australia.
(There are a number of similarities but also critical differences with the water storage problem. In this aspect of the project we will draw on the considerable experience in Civil Engineering in the area of water network optimization.)


'''The 2018 Project'''

Today’s world faces challenges with the way energy is being made and produced. This stems from the increasing use of renewable energy sources, instigated by global action to climate change and decreasing manufacturing and installation costs. However, these benefits incur the issue of the intermittent nature of renewables which presents reliability and unserved energy concerns in the South Australian grid.

This thesis documents the energy storage requirements for the South Australian grid. The Initial Problem involves the development of a supply-demand balance model of the state, given a set amount of energy storage. It aims to minimise a cost function that matches to the maximum controllable generation, whilst ensuring 100% reliability for a set period. The tool developed could potentially be utilised to form retirement strategies for fossil fuelled generation as more renewables come online. 

Genetic Algorithms will be used to optimise the Initial Problem. Previously been applied by the 2017 iteration of the project and to water distribution problems in the civil space, this technique incorporates the Darwinian concepts of natural selection and survival to the fittest. A simple genetic algorithm program has been developed on MATLAB which is to be later expanded and integrated into the initial problem.


== 1. Introduction ==


'''1.1. Motivation'''

RESs have become evidently popular but this rapid development incurs several challenges. A major issue is intermittent electricity, energy that is not readily available. Wind and solar are intermittent in nature, they only produce energy when the wind is blowing, or the sun is shining i.e. generation is largely dependent on weather conditions. This in contrast with controlled and synchronous generation that utilise fossil fuels. [5]

Although RES can be predicted, it cannot be dispatched to meet the electricity demand. [7] Therefore, during times of high demand (e.g. during the summer period), RES presents challenges to the power system in terms of reliability, power quality and system security. Security in the context of a power system is operating within the defined technical limits and parameters such as voltage, frequency and network loading. [4] To illustrate, renewable sources do not provide inertia as a by-product of generation due to the being connected to power electronics. This leads to issues related to the rate of change of frequency (RoCoF) and hence maintaining the balance between the demand and supply. [4] 

Another set of dilemmas are associated during periods of low demand and high generation. For instance, large energy flow from rooftop PV systems back to the distribution network can result in damage to transformers. This is due to the inherent design of the power system network that has catered to one-way flow. To avoid this, generation from RES are curtailed – restricting excess energy into the system. Overall, the intermittent characteristic of renewables poses a threat to both the power system security and reliability which only increases as more come online.

The issue of intermittency in apparent South Australia. A key evidence of intermittency is the demand variation of RES. Figure 5 indicates that for recent years, the variation distribution has become wider and flatter. [7] This implies more frequent and larger changes in output between five-minute intervals.  The curve also emphasises the greater variability in residual demand changes as more renewables come online. Therefore, this must be managed to maintain power system standards. The residual demand is met by other sources of generation or from power flow of the Heywood and Murraylink interconnectors. [7]

''Figure 5: South Australian residual demand (variation distribution)''

The potential effects of intermittency are becoming more recognised. It has made managing the power system challenging by introducing a heightened risk that involves unserved energy (USE), and the National Electricity Market (NEM) standard not being met. [8] AEMO forecasts that for during the 2017-18 period, the South Australian and Victorian power system will be at most risk to USE. Figure 6 conveys that during this period, USE is 0.0025% of the total demand. This exceeds the 0.0020% NEM reliability standard. Thus, if USE occurs it will be for two to four hours. [4] USE and reliability issues due to intermittency is therefore a clear threat to the reliability of the South Australian power system. 

To alleviate the intermittent nature of RES and its potential threats to the power system, energy storage is considered as a viable option. Utilising energy storage allows capacity firming which is maintaining the power output at a committed level for a period. [9] Power can be immediately dispatched to the grid to meet demand while also be charged at times of low demand. Additionally, energy storage eliminates the need to have generation capacity for the predicted highest demand and stabilises the effects of withdrawn controlled generation. 

Grid Energy Storage, a document produced by the United States Department of Energy (DOE) exemplifies the challenges associated with the U.S electric system and how the integration of energy storage can meet those challenges. [10] The main issue presented is the need for an electricity grid that is robust and reliable to meet the inevitable increase in electricity demand. The DOE summarises that modernising the electric system along with employing energy storage improves the operating capabilities of the grid, lower costs and ensure high reliability along with deferring and reducing infrastructure investments. [10] 

Energy storage can also provide immediate response to emergencies by supplying backup power and stabilising the electric grid. [10] In conjunction with these functions, different types of energy storage technologies provide multiple applications including energy management, load levelling, frequency regulation and voltage support. Hence, this offers flexibility in terms of applying a specific technology to meet requirements. [10] For example, large flywheel installations combined with power monitoring software ensures intermittent sources and variable load demands are maintained at nominal frequency. They can provide spinning reserve or curtailment which could reduce greenhouse gas emissions and improve the efficiency of infrastructure with industrial plant processes. [10]

The report further details the challenges involved in the deployment of energy storage. These are costs in manufacturing and installation, validated reliability and safety, equitable regulatory environment and industry acceptance. [10] Understanding these challenges and the strategies to tackle them is vital in supporting the commercial viability of energy storage. Table 2 summarises how the DOE will address these challenges. [10]

''Table 2: Strategy summary to address the challenges/goals of energy storage''
 
Effectively, the report establishes potential options to improve energy storage as well as actions to further encourage and maintain both scientific advancements and a pipeline of project deployments. [10]


'''1.2. Objectives and Project Significance'''

With the increasing integration of intermittent electricity sources in the South Australian grid and the clear requirement of energy storage application, it is appropriate to quantify the amount of storage. It is necessary to first formulate a problem and then identify an objective to accomplish. This can include ensuring reliability standard is met, reducing carbon emissions to a set level or even to reduce the cost of electricity. 

The main challenge of implementing energy storage in South Australia is the uniqueness of the grid - both its distribution and transmission networks. The distribution network covers a large area of 178,000 km squared while the transmission network operates the interconnectors that allow for energy flow from/to New South Wales and Victoria. [14] There is also the concern of different forecasted levels of solar and wind penetration in different regions, further emphasising the issue of intermittency. This presents the problem of quantifying renewable energy generation and hence the storage required. In addition, it also involves the assessment of costs of extending or upgrading the existing infrastructure to ensure safe and reliable operation. 

AEMO retains useful data that can be used to determine the energy storage requirements for a predetermined objective. These data include the interconnector flow, electricity demand and dispatchable generation from both non-renewable and renewable sources. However, the limitation is that they are only forecasts and therefore does not reflect the actual situation of the electricity grid. This presents a challenge to the problem, particularly in emergencies, where it a priority to restore the system to a stable and secure state. Nevertheless, the data is beneficial in providing a starting point to be modified in the project investigations. 

This project endeavours to parametrise the amount of energy storage required for several objectives and deliver a solution that combats the challenges previously mentioned.  The initial investigation aims to optimally utilise a given amount of storage to minimise the reserve generation capacity that is required to supply the system load, whilst ensuring 100% reliability for a set period. The results of this study provide information on the impact of using a certain amount of storage capacity and highlights the issues involved in this transition phase of the Australian energy system. It will formulate a relationship between storage capacity and installed reserve generation capacity, building a foundation for undertaking more complex problems. 

The industrial significance of this investigation is that the tool developed can be used to assist in developing retirement strategies for fossil fuelled generators in the National Electricity System (NES) while meeting the NEM reliability standard. For instance, it can potentially avoid a state-wide catastrophe such as the SA State Blackout in September 2016. [15] A controversial opinion is that this could have been avoided if the Port Augusta coal-powered power station was kept online. [16] Furthermore, it can help the NES to transition to increasing amounts of renewable energy and become less reliant on fossil-fuelled generation capacity. Ultimately, this promotes reducing CO2 emissions and assists in attaining the Federal Government’s RET of 33,000 GWh by 2020. 


== 2. Technical Background ==

'''2.1. Problem Formulation'''

The aim of this section is to outline the problem formulation process of the project. It is important to establish the foundation in terms of the direction and objectives, especially in the early stages. This is to avoid over-scoping the problem and instead narrow it down. It also serves as a preliminary guide for the reader to form some understanding of the topic. This section is effectively a prelude to the formulation of the Initial Problem explained in Section 2.2. 

The electricity generation mix in South Australia consists of gas-powered generation '''P_C''', wind power from turbines '''P_W''', solar PV inverters '''P_P''' – both large and small-scale, energy storage '''P_S''' (solar thermal, pumped-hydro, battery), dispatchable load and interconnectors '''P_I'''. Figure 8 is a simple depiction of this model.  

From the scenario described above, the demand-supply balance equation is constructed as follows:  
'''P_G+P_S+P_I=P_L'''
Where '''P_G=P_C+P_W+P_P''' is the total generation from all sources (renewable and non-renewable) and '''P_L=P_LU+P_LC''' is the total demand from uncontrollable and controllable load respectively. 

The controllable load can be considered the base load, the state’s minimum level of demand for power at any given time while uncontrollable load can signify unpredicted fluctuations in the load e.g. during the summer period.

There are supply constraints associated with this model:
* '''〖P_(C(min))≤P〗_C≤P_(C(max))'''
* '''0≤P_W≤P_(W(max))'''  
* '''0≤P_P≤P_(P(max))'''
* '''〖P_(S(min))≤P〗_S≤P_(S(max))'''
* '''〖P_(i(min))≤P〗_i≤P_(i(max))'''

The power from renewables '''P_P''' and '''P_W''' have a lower limit of zero to represent intermittency. To supplement this and to supply the load, '''P_C''' and '''P_I''' will always be online and available up to a maximum value.

As described in the Motivation, as controllable generation is retired, and more renewables come online it becomes increasingly difficult to maintain supply-demand balance and deliver the base load. The problem then focuses on quantifying '''P_S''' to satisfy this condition. 

However, it is helpful to first consider some concepts:
'''Stored energy= E_S=∫_0^t▒〖P_S (τ)〗 dτ'''
Subject to the stored limits:
'''〖E_(S(min))≤E〗_S≤E_(S(max))'''
Where '''E_(S(min))≅0'''.
That is, the rate of change of stored energy is power from storage and stored energy has a lower limit of approximately zero. This means that stored energy cannot be negative. 

Moreover, there are characteristics in the simplified problem that are ignored or neglected:
* Some generation/storage classes
* Transmission network and associated congestion issues
* Power system security constraints e.g. '''P_(C(min))''' is constrained to ensure minimum synchronous generation
* Some supply constraints are overly simplified e.g. controllable generation is applied in ‘lumps’ that have long start and shutdown times. 

The question then to ask is ''What is the most “cost effective” way to satisfy demand-supply balance?'' This cost can be put in terms of the price of electricity, CO2 emissions or spilled resources from renewables. For instance, consider the issue of spilled resources from renewables. When there is excess power generation, supply is curtailed. This can lead into over capitalisation and thus signals the need for increased interconnection and/or storage. 

Once an objective is decided, the question now is ''How do we formulate the optimisation problem?'' With focus on minimising costs such that it accords with supply-demand balance equation and the supply constraints. The final question then is What optimisation technique are we going to use?


'''2.2. The Initial Problem'''

The objective of the Initial Problem is to investigate a control strategy for energy storage focusing on minimising the amount of reserve generation capacity or reserve generation energy supply. In the model, a lumped representation of the South Australian power system is considered for simplification. The power magnitudes in MW at time '''k∆T''', where '''∆T''' is the sampling interval in hours and k is the sample number, are as follows: 
* '''P_L [k]''': Total load of South Australia
* '''P_R [k]''': Total generation from renewables – wind and solar
* '''P_G [k]''': Total reserve generation i.e. non-renewable sources from gas and power from the interconnectors
* '''P_S [k]''': Total output power from all storage devices

With consideration of the conservation of energy, the demand and supply balance equation is therefore:
'''P_R [k]+P_G [k]+P_S [k]=P_L [k]'''
 
''Figure 9 – Incomplete. Placeholder.''  

The known parameters in this model are the time-series forecasted data of renewable generation '''P_R [k]''' and load '''P_L [k]'''. These are obtained from the AEMO NEM data archive. The extraction process is achieved utilising the software tool developed in last year’s iteration of the project. Further information of the 2017 project is noted in Section 2.5.  AEMO data is represented in 5-minute intervals, this is transformed to indicate 30-minute intervals and hence '''∆T=0.5 hours'''.

Conversely, the parameters to be determined are the time-series reserve generation '''P_G [k]''' and energy in the storage '''E_S [k]'''. Values of '''P_G [k]''' are compared for all timesteps and the largest value will be obtained. 

It is observed that '''P_G [k]''' can be a negative or positive value. When it is a positive value '''P_G [k]>0'''  this means that at this time, power from reserved generation is used to supply the load. In contrast, when '''P_G [k]<0''', this indicates that there is a surplus of generation and the excess is exported to the interconnectors. 

Supply constraints and power limits associated with the model include:
'''P_(S(min))≤P_S [k]≤P_(S(min))''' (MW)
Similarly, '''P_S [k]''' can also be positive or negative. When '''P_s [k]<0''', storage is being charged or consuming power from the network, up to a maximum rate '''P_(S(min))'''.  Hence, this implies that when '''P_s [k]>0''', power from the storage is being discharged to supply the network, up to a maximum rate '''P_(S(max))'''. 

Adding complexity, the model will consider the storage efficiency η which is user defined to possibly model different storage technologies. Firstly, to distinguish between charging and discharging, the superscripts c and d are adopted respectively. To represent the '''P_S [k]''' before/after losses, the subscript i is applied. The power storage can now be charactersied as: 
'''P_S^d [k]=n^d P_Si^d [k]≥0'''
'''P_S^c [k]=(P_Si^c [k])/η^c <0'''
Specifically, '''P_Si^d [k]''' is the power discharged by the storage before at time '''kT''' losses while '''P_Si^c [k]''' is the power consumed by the battery after losses also at time '''kT'''. 

This then establishes the stored energy with consideration of losses using the trapezoidal method which estimates the energy accumulated by the storage between successive samples. 
'''E_S [k]=E_S [0]-(∆T/2) ∑_(i=1)^k▒{(P_Si^c [k-1]+P_Si^c [k])+(P_Si^d [k-1]+P_Si^d [k])}'''    (MWh)
In addition, the energy capacity of the storage is limited such that:
'''〖E_(S(min))≤E〗_S [k]≤E_(S(max))'''  (MWh)
The maximum amount of storage in the battery '''E_(S(min))''' is zero. While the minimum storage '''E_(S(max))''' is user specified. 

Forecast data of '''P_L [k]''' and '''P_R [k]''' from AEMO and the described equation constructs a model that aims to minimise '''P_G [k]''' for a given storage capacity. The optimisation problem is therefore determining the maximum value of '''P_G [k]''' such that it minimises “cost” '''C''' given a set amount of energy storage, whilst ensuring the demand is always met i.e. 100% reliability. In other words: 
'''min⁡(C) where C=max⁡〖(P_G [k])〗''' subject to the above equations and constraints. 

The optimisation time period '''T=N∆T''' is then set by the user. This will be the forecasting horizon: for forecasted values '''P_L [k]''' and '''P_R [k]''', the model will output the forecasted schedules of '''P_G [k]''' and '''E_S [k]'''. Note that forecasting errors ignored. 

Implementation of this model is developed in MATLAB. Description of its operations and characteristics are discussed in Section 3.3.

The optimisation technique that applied to this investigation will be a process known as Genetic Algorithms (GAs). This approach is described in the following section as well as the motivation for its employment.

'''2.3. Genetic Algorithms'''

GAs are optimisation techniques (search procedures) that are built on the mechanics of natural selection and natural genetics. They integrate the concept of “survival of the fittest” with a unique information exchange process. [18] GAs centre on the theme of robustness and the balance between efficiency in a variety of different problem spaces.  

A simple GA consists of several heuristic or selection operators that are applied to a “population” of '''X''' solutions. Initially, this population is randomly generated but controlled by constraints and limitations to the problem space. Each individual member of the population (solutions) is analogous to a chromosome while their individual decision variables '''N''' are relative to genes. [18] The heuristic operators update the current population initially by selecting a pair of solutions (“parents”) based on their fitness function. This ''fitness function'' corresponds to the optimisation objective. Then through selection, identifies which parents will reproduce. 

The selection operator is inspired by the “survival of the fittest” concept and hence will choose the probabilistic fittest candidate solutions. [18] In this investigation, the selection operator used is ''Tournament Selection''. This compares the fitness values of the selected parents and in general, the larger value or “winner” proceeds to the ''mating pool''. All members of the population are selected at random for Tournament Selection and members already chosen cannot be selected again. This process is repeated, and the outcome from the two tournaments will have '''X''' winners in the mating pool. Tournament Selection essentially increases the average fitness value of the population. [17] [18]  

Two random parents are then selected from the mating pool. These become subject for “breeding” by ''crossover'' and ''mutation'' operators producing “children” or “offspring” to form the next-generation population. [18]

Crossover of the two parent chromosomes involves swapping their genes. This process occurs if the crossover probability '''P_Crossover''' criteria is met. If a randomly generated value between '''0''' and '''1 – P''' is less than '''P_Crossover''' i.e. '''P<P_Crossover''', the crossover operator will be applied. The crossover point '''Y''', an integer between '''1''' and '''N-1''', is then randomly determined. The crossover point controls the location along the chromosomes where their genes will swap. Genes positioned from '''Y+1''' to '''N''' of the chromosomes will swap, producing offspring. [18] This process is reiterated until all chromosomes from the mating pool are selected, yielding '''X''' children. Ultimately, crossover results in individual solutions exchanging decision variables. 

After crossover, the offspring are exposed to mutation. Mutation involves in one or more decision variables modified in the offspring. Like crossover, if a randomly generated factor '''P''' is less than the probability of mutation P_Mutation i.e. '''P<P_Mutation''', then the child will mutate. Genes that are mutated are replaced by a uniformly distributed random value in its feasible range. The purpose of mutation is to preserve and introduce diversity into the new population or next generation and assists in the EA in delivering an optimal solution. [17]

As this process is repeated, over successive generations, the population will head towards optimality. In most cases, the solutions and their decision variables will be the same. Figure 10 is a diagram to show the process of GAs. An exercise to better understand GAs, its functionality as well as the process is achieved in Section 3.1.

''Figure 10 – Incomplete.''

'''The 2017 Project'''

The 2017 iteration of the project by Daniel Bondarenko and Ryan Standing builds a foundation for the 2018 project. The work completed last year consisted of two main stages: producing a software package and applying it to perform investigations relevant to energy storage. They developed a software package that collected data from AEMO and created a database to store and maintain that data. Once this data was extracted, it was used to explore how energy storage technology can be used to change the behaviour of intermittent generators in the NEM. A case study was performed on the Hornsdale wind farm, which at that time the Tesla battery was still to be installed.  The objective of the investigation was to minimise output variance using GAs. Reducing output variance is vital to make the power output more stable and act more like high-inertia generators. Results show that the installed 129 MWhr battery would at very best reduce the output variance by 38.5%. Overall, the 2017 project examined how energy storage can impact the South Australian power system and NEM whilst forming a baseline for the 2018 project. 


== 3. MATLAB Programming and Implementation ==


== References ==


== Glossary and Symbols ==