Projects:2019s1-140 Energy Storage Requirements for the SA Grid
Contents
Team Members
- Isaiah Turner
- Sean Fernandes
Supervisors
- Derek Abbott
- David Vowles
Advisors
- Angus Simpson
Abstract
South Australia is changing from the way energy is generated and used to satisfy the environment and economic development. However the problem with renewable energy is it's intermittent. The renewable energy generated is dependent on nature while coal energy isn't dependent and can be controlled to be supplied 24/7 without any issue. This raises issues on unreserved energy and renewable energy reliability to the SA grid.
The demand-supply model from 2018 is used to continue aiming at the original goal which is to reduce the maximum reserve generation capacity and increase the renewable energy intake. This model could be expanded upon.
Genetic Algorithms will be used to solve the energy usage by finding optimal solutions. This algorithm incorporates natural selection where it chooses the survival of the fittest. It is used to reduce the controllable generation in favor of more renewable usage. This was applied in the 2018 iteration of the project where they did several benchmark cases of various scenarios related to the supply and demand model using MATLAB. However the results had problems. This year will be used to improve their 2018 genetic algorithm code to satisfy the model and fitness result. MATLAB is used once again for this year.
Other objectives for this year project include measuring performance, forecasting and using AEMO real data that wasn't used 2018.
Introduction
Background
Australia's energy production is currently changing from using non-renewable energy sources to renewable energy sources. Non-renewable energy sources such as natural gas and coal fired powered generator are being retired to move towards generating energy from renewable energy sources such as the wind and solar. This is because support of the Australian government. Their supporting reasons are the reduction of environmental problems such as greenhouse gas emissions (GHG) and health problems such as breathing. Another reason is to meet the Australian Government scheme, Renewable Energy Target (RET) . This scheme’s purpose is to reduce GHG in the electricity sector by using sustainable and renewable sources for additional electricity generation. It aims to reduce the carbon dioxide emissions to achieve the Federal Government’s RET goal of annual renewable energy production of 33,000 gigawatt-hours (GWh) by the year 2020.
Motivation
Renewable energy sources have many of benefits such as unlimited energy supply and not generating GHG but intermittency is a major downside. Intermittency for renewable energy sources refers to power sources that change in intensity leading to a point where they are not continuously available to provide energy for electricity. Common issues include wind not blowing and sun not shining in certain times of the day. This makes them unreliable for energy generation. So, intermittency leads to a problem called dispatchability where we are unable to decrease or increase the power supply to meet the demand. This problem doesn’t occur with non-renewable fossil-fuel sources, for it can easily adjust to changes in demand by removal or addition of coal or fossil-fuels. That is why non-renewable energy sources are often referred to as ‘controllable generation’.
An additional problem with renewable energy solutions is the lack of inertia present within the individual systems. In Australia, the frequency of the system has to stay stable at 50 Hz. A way to control the rate of change of that frequency within the grid is to have non-renewable controllable sources make use of large turbines with sizable masses. As these masses rotate, they inherently produce inertia. This inertia slows down the rate that the frequency of the system changes, providing an opportunity for additional power to be deployed. Renewable systems do not have such a synchronously connected rotating mass, and as a result do not have the same inertial response.
A technique to combat these limitations is to implement Energy Storage Systems (ESS) within the grid. These energy storage systems have the ability to dispatch power during periods of high-demand and retain power in periods of low-demand, ultimately allowing a balance to be reached within the grid. With appropriate control systems, ESS can also contribute to frequency control within the grid.
Method
Commenting
One of the flaws within previous iteration of the GA code was the lack of commenting. Less comments made it harder to understand what each line of code represents. So one of the first stages done was to write comments next to each line of code to clearly understand how it is processing. This made it easier to modify the code in the later stage. An example of this is demonstrated in Appendix A with the original code having lack of comments followed by code revised for commenting
Vectorization
The next step was to vectorize various sections of the code in order to optimize its speed. It was done so the code could be easier to read, less prone to errors and have better performance. For loops were vectorized as illustrated in Appendix A. The old revision of the code had for loops with arithmetic vector calculations. So these were vectorized to increase its performance speed. In the StorageControl_GA.m file, the crossover section of the code was vectorized. The original crossover code had an entire population going through a loop with each pair of members getting crossover based on the probability of crossover. So, the code was revised to now do matrix calculations so it can crossover the entire population in one process. This saves time and avoids processing multiple times in a loop. Appendix A shows the original code and the revised vectorized code.
Fitness MEX File
Another method that was devised to improve the speed of the GA was the use of an external MEX file that would do all the fitness computations. An MEX file is a type of MATLAB compatible file that allows the use of C++ programming. Using C++ to complete all the computations allows a quicker runtime due to the power behind the language. In order to integrate this process into the GA, the MATLAB code would have to call the MEX file while also passing in all the relevant information across. This would occur after the initialisation of all the values that are required but before any calculations have taken place. The MEX file would then take the values passed in by the MATLAB function and complete all the calculations needed. When finished, the MEX function would then have to pass all the final values back to the MATLAB function in order for plots of the solution to be created. It should be noted that the MEX function does not have complete functionality and can therefore be considered a work in progress. The source code for this MEX file can be found in the appendix.
Results
Figure 1 compares the quickness benchmark 1 after using the previous (red) and current iteration (blue)of the GA code. For a maximum population size of 1000, it takes about 8 seconds for the previous iteration to complete. For the current iteration, it takes about 1.2 seconds. The gap between gets larger with each population size. This plot shows that the current iteration is 8 times faster than the previous iteration. The reason for this difference in the speed is because the previous iteration of the code was not vectorized whereas the current iteration was vectorized. This plot proves that we have successfully achieved one of our initial goals which was to increase the speed at which the program completes
Conclusion
From the results, it can be stated that the Genetic Algorithm produced, in conjunction with the supply-demand model, provide a solid foundation for future work. The results indicate that the GA can quickly and accurately produce a solution that minimises the controllable generation capacity while also satisfying the demand. In the isolated system that was used throughout this project, the GA provides near ideal solutions on a regular basis.
Future Work
There are many tasks that need to be done for future work. The following objectives for the future include:
- Convergence Monitoring Implementation – The GA Storage Control module doesn’t have the correct implementation of convergence monitoring. Implementation of this will help the GA program to stop once it converges at the optimum solution.
- Additional benchmarks cases – GA reability would be better if more benchmark needs to be created and tested.
- Real data – The use of real data within this GA will allow it to be used in a wide array of applications.
- Increase GA accuracy – Increasing the accuracy of the GA would allow it to be used to precisely determine the amount of power may be needed at a particular time.
References
[1] AEMO, “South Australian Electricity Report 2018”. 07 November 2018. [Online] AEMO, “http://www.aemo.com.au/-/media/Files/Electricity/NEM/Planning_and_Forecasting/SA_Advisory/2018/South-Australian-Electricity-Report-2018.pdf,” [Accessed 29 Mar. 2019]
[2] Clean energy and the electricity market | energy.gov.au", Energy.gov.au, 2019. [Online]. Available: https://www.energy.gov.au/government-priorities/energy-supply/renewable-energy-and-technology. [Accessed: 05- Apr- 2019].
[3] Goldberg, D. E. (1989). Genetic Algorithms in Search, Optimisation & Machine Learning. Alabama: Addison-Wesley Publishing Company, Inc.
[4] A. Savic, G. Walters, R. Atkinson and M. Smith, "Genetic Algorithm Optimization of Large Water Distribution System Expansion", Measurement and Control, vol. 32, no. 4, pp. 104-109, 1999.
[5] F. Zheng, A. Zecchin and A. Simpson, "Self-Adaptive Differential Evolution Algorithm Applied to Water Distribution System Optimization", Journal of Computing in Civil Engineering, vol. 27, no. 2, pp. 148-158, 2013.