Ali Meghdadi
IMPROVING THE SECURITY OF THE ELECTRIC POWER SYSTEM: USING MACHINE LEARNING TO PREDICT POWER SYSTEM STABILITY AND SECURITY
By Ms Nancy Van Nieuwenhove | March 2020 (edited April 2021)
Ali Meghdadi
3rd PhD candidate, Department of Data Science and AI, Faculty of Information Technology
Research interests: machine learning applied to stability and operation of power systems.
Ali received his MEng. and BSc. degrees in power system Engineering and had over three years work experience with the industry, and two years of teaching assistantship before joining Monash University. “The energy sector has been growing very fast during the last decade. It was the beginning of the integration of large solar and wind farms, as well as other new ideas to transition away from fossil fuels when I was an undergraduate student. The energy sector is now changing even faster as we proceed to 100% renewable grids which makes it even more interesting to researchers all around the world to create new ways for handling the challenges that these changes bring”.
Ali started a PhD with the Optimisation Lab, Department of Data Science and AI, Faculty of Information Technology at Monash University in February 2018, supervised by Professor Ariel Liebman and Dr Guido Tack from Monash University and Dr Nicolas Langrene from CSIRO (data61).. “I wanted to discover more and create a fast and reliable tool that improves the power systems stability. The classical method of stability assessment is very time consuming and you cannot rely only on them to deal with the new challenges, such as inertia reduction. Having access to top-edge research facilities and academics has enabled me to develop new ideas and create new tools. What I like the most as a PhD student at Monash is the support I receive from my supervisors and the excellent collaborations with industry”.
Ali shifted from Electrical Engineering to IT to innovate; he is using IT approaches, not only focusing on the data, but also the understanding of power systems stability. Ali is particularly interested in how we can integrate more renewables effectively into the Grid for a smarter, cleaner and more sustainable future. To do so, we need to deal with stability issues as we are going away from fossil fuels. “With the current pace of integrating renewable energy sources to the grid and the retiring of synchronous generators (and the consequent inertia reduction), there is a dire need to assess the stability of the power system operational scenarios. Since classical time-domain stability analysis is computationally expensive, there is a growing need for fast and reliable stability assessment tools. With the advent of advanced information technology, new methods with faster response time and higher accuracy were developed. As an alternative, data-driven approaches such as machine learning-based frameworks are exploited to detect stability status. My research exploits the understanding of power system dynamics to increase the performance of these data-driven approaches. My focus is in developing a power system dynamics based transfer learning scheme for security assessment of electrical grid operational scenarios.It drastically reduces the training effort of the machine learning tools to predict the stability status. This makes this approach ready to be used for real systems and be implemented rapidly”.
* RMS/EMT Simulations of schedules generations are computationally expensive. Data-driven solutions, like deep learning algorithms, have shown promising results with high accuracy levels but still a low generalisation, convergence, and interpretability. In their latest publication accepted at the IEEE Power & Energy Society's 2021 General Meeting titled "Versatile and Robust Transient Stability Assessment via Instance Transfer Learning", Ali and the team introduced a new data collection strategy, called "transfer learning". "The new data collection strategy is based on the understanding of power systems dynamics and how the disturbance effect propagates in the power network. This allows us to use conventional machine learning models (with better interpretability and convergence) in an instant transfer learning framework. This approach outperforms the existing data-driven methods in terms of accuracy, generalisation, and robustness".
Ali is currently working on a novel deep transfer learning algorithm to improve the training effort as well as robustness of the trained models against topology change scenarios, i.e. adding or removing transmission lines.
Is Australia going to meet the 2030 targets in terms of integrating renewables into the grid with enough confidence in the stability of the generators? Ali is positive about the energy future in Australia! For him, that energy future will be more multidisciplinary than ever.“I believe solar energy has great potential in Australia, (especially rooftop PVs) which will most probably be supported by energy storage systems. We soon will observe grids reliably and securely operating with 100% renewable generators. I assume there will be tighter collaborations of electrical engineers and data science experts in the close future to address new challenges as the classical operation of the grid will not necessarily be able to guarantee the optimum operation of the power system, unless IT methods are incorporated”.
After his PhD, he is looking to apply the results of his PhD to the real world and collaborate with industry to create specific tools based on their needs.
For further information:
