As electricity networks look to transform, grow and adapt to the ever-changing energy landscape, one of the biggest challenges is planning for an uncertain future. Energy Queensland’s solution was the development of a sophisticated automation system – one that maintains efficiency, safety, and cost-effectiveness. Energy magazine spoke to Dr Andrew Thomas, Principal Engineer Strategic Planning, Energy Queensland, about the company’s new technology and what it means for their electricity networks.
Energy Queensland, which is 100 per cent owned by the state of Queensland, comprises of electricity distribution, retail and energy services. Its electricity network businesses, Energex and Ergon Energy, manage a large area, meaning efficiency and accuracy are crucial.
Dr Thomas said the goal of automated system work is two-fold: to enable the networks to autonomously make accurate strategic planning decisions over the thousands of constraints that the automated network modelling can produce, and to build a system that can aid planners in breaking down the complexity in building the future network.
“Given the multitude of new factors which increase planning complexity – various photovoltaic (PV), electric vehicle (EV) and battery uptake scenarios, more dynamic control devices, dynamic operating envelopes, aggregator/retailer interventions in the distribution network – it is hoped that an artificial intelligence (AI) system can learn to make least-regret decisions in such an environment,” Dr Thomas said.
Developing a forecasting system
“The automation system that we have developed has a number of distinct components,” Dr Thomas said.
“The first is a new load forecasting system, developed by the Energy Queensland Strategic Forecasting team. It is based on standardised load profiles for normal residential and business customers, but also for new technologies such as PV systems, EVs and batteries.
“The forecasting team is involved with creating various uptake and technology use scenarios (e.g. various EV charging profiles), and these flow through to new load forecasts at various levels of the system – for instance, at our Bulk Supply Points (e.g. 132/66kV substations), our Zone Substations (e.g. 66/11kV substations) and distribution feeders (e.g. 11kV distribution network).
“This system generates maximum and minimum demand forecasts, to understand both peak load impacts on the network and also minimum demand/reverse power flow impacts.”
Programming and planning
For the next step of the automation system, Energy Queensland uses two key types of software: Python and PowerFactory. Python, a high-level programming language, allows for the effective integration of other systems. PowerFactory, on the other hand, is a “ leading power system analysis software application for use in analysing generation, transmission, distribution and industrial systems”.
Once the forecasts have been generated, they are then applied through Python programming language scripting to the PowerFactory models of the network. Automated load flows, through the PowerFactory-Python application programming interface (API), are then run utilising parallel processing over several virtual machines, under all the various modelled scenarios, maximum and minimum demand, out to 2040.
“This process produces lots of data – normal and contingency capacity of all the various transformers, lines and cables in the network, and also voltage information,” Dr Thomas said.
Following this, the next step is to apply the planning criteria to this information, again using automated analytics and scripts, to generate a forecast of expected constraints on the network under the various scenarios.
“This information is useful on its own terms, but even more useful when converting this forecast of constraints into expected future capital expenditure,” Dr Thomas said.
“Literally thousands of future network constraints could be forecast from this system, far too many for manual analysis – so the challenge is to determine expected strategic expenditure arising from these constraints in an automated way.”
The current system vs a future system
“The current version of the system uses a rule-based system to determine estimates of possible solutions to each constraint. For instance, if the voltage on a distribution feeder is projected to be outside of planning limits in 2030, we could assume that a new voltage regulator will solve this problem,” Dr Thomas said.
“The final stage of this system is to extract relevant information, summarise findings, and produce visualisations to aid and inform the rest of the business in decision making and planning.
“However, we are also working on a new system that uses machine learning/AI to determine the best solution in a more accurate way, using a machine-learning methodology called ‘reinforcement learning’. Early proof-of-concept work looks promising in this endeavour.”
Automated planning in the transition to net zero
When it comes to transition to net zero, Dr Thomas said automated planning has an important role to “enable us to make least-regret decisions”.
As major network augmentation means developing assets with long usable lives (which now stretch well into and through energy transition timeframes), the risk of regret increases, meaning these investments need to be carefully considered with a view to the possible future scenarios.
“Automated planning allows us to quickly and efficiently explore many possible future scenarios and analyse their impact on the network and future network expenditure,” Dr Thomas said.
“Armed with this information, we can produce better business cases that minimise regret in investment decisions now and into the future.
“Using these tools that we are developing, we hope to be able to help Energy Queensland enable, manage, and optimise the transition to the new energy future for our customers.”
The future of AI in energy applications
Dr Thomas said there are still many applications for AI in the energy sector that are yet to be discovered and applied.
“In the strategic planning area, Energy Queensland has the view that AI can be used to train intelligent ‘agents’ that can make good planning decisions, which is what we have been trying to prove or demonstrate in the proof-of-concept work,” Dr Thomas said.
“Using the reinforcement learning methodology, we are training an autonomous agent, whose ‘brain’ is a deep neural network, to interact with load forecasts and our PowerFactory network models to make good network augmentation decisions in order to lower the risk of constraints and also to minimise capital expenditure.”
Dr Thomas said that AI systems may never supersede a human planner, but that their applications definitely have promising benefits for the industry.
“It will be a long time – perhaps never – before an AI system can do the same, detailed planning work of a human planner, but it is hoped that this system will find possible new solutions or approaches, in the presence of network complexity, that improve the outcomes for planners, the network and customers.”