Uploaded March 2024 | Updated September 2026, 2 weeks ago
The race towards net-zero requires intensive collaboration across a myriad of industry, research, and governmental organisations. Often, that collaboration hinges on access to raw data, both to develop solutions for pressing Energy concerns, and to enable the sharing of expertise. This need is amplified given the push to utilize neural networks and deep learning techniques within various arms of Energy research, especially given the urgency of the transition to clean energy.
However, the sharing of said raw data represents a major stumbling block. Organisations within Australia are hesitant to share possibly sensitive information, due to privacy, legal, and/or competitive concerns. For example, government agencies who are trusted operators of the power grid cannot expose data from industry participants within the electricity market.
This project investigates the use of AI techniques to securely generate synthetic data for use within the Energy domain. It goes beyond traditional techniques such as anonymisation and is similar to the use of synthetic data within bio-medical research. This also may represent a more achievable pathway compared to other techniques on the horizon such as homomorphic encryption.
CSIRO Energy is the trusted custodian of several significant datasets. This includes data from building energy management systems, power grid operations, electricity market prices, and electric vehicle charging infrastructure. The overarching aim is to apply various AI techniques to learn from existing datasets, and then synthesize statistically similar data to act as stand-ins.
To evaluate our approach, we can leverage the wealth of internal models, algorithms, and expertise within CSIRO Energy. We have at our disposal a unique combination of traditional physics-based computations, classical machine-learning approaches, and newer data-driven models that serve as an ideal testbed. It is noted that these algorithms require either (and more likely both) the HPC and GPU infrastructure within Pawsey.
The actual input data to be used will be determined by CSIRO Energy staff later. This is to allow consultations with our various internal teams that are each specialised in their respective sub-domains (e.g., transport electrification, grid control, commercial buildings). This also allows us to prioritise datasets according to potential impacts, initial pre-processing required, and in-line with the number and quality of students selected.
The race towards net-zero requires intensive collaboration across a myriad of industry, research, and governmental organisations. Often, that collaboration hinges on access to raw data, both to develop solutions for pressing Energy concerns, and to enable the sharing of expertise. This need is amplified given the push to utilize neural networks and deep learning techniques within various arms of Energy research, especially given the urgency of the transition to clean energy.
However, the sharing of said raw data represents a major stumbling block. Organisations within Australia are hesitant to share possibly sensitive information, due to privacy, legal, and/or competitive concerns. For example, government agencies who are trusted operators of the power grid cannot expose data from industry participants within the electricity market.
This project investigates the use of AI techniques to securely generate synthetic data for use within the Energy domain. It goes beyond traditional techniques such as anonymisation and is similar to the use of synthetic data within bio-medical research. This also may represent a more achievable pathway compared to other techniques on the horizon such as homomorphic encryption.
CSIRO Energy is the trusted custodian of several significant datasets. This includes data from building energy management systems, power grid operations, electricity market prices, and electric vehicle charging infrastructure. The overarching aim is to apply various AI techniques to learn from existing datasets, and then synthesize statistically similar data to act as stand-ins.
To evaluate our approach, we can leverage the wealth of internal models, algorithms, and expertise within CSIRO Energy. We have at our disposal a unique combination of traditional physics-based computations, classical machine-learning approaches, and newer data-driven models that serve as an ideal testbed. It is noted that these algorithms require either (and more likely both) the HPC and GPU infrastructure within Pawsey.
The actual input data to be used will be determined by CSIRO Energy staff later. This is to allow consultations with our various internal teams that are each specialised in their respective sub-domains (e.g., transport electrification, grid control, commercial buildings). This also allows us to prioritise datasets according to potential impacts, initial pre-processing required, and in-line with the number and quality of students selected.
