Artificial intelligence (AI) is emerging as a disruptive technology in the energy sector and raising the attention of different stakeholders, such as electric system operators, energy retailers, energy services companies, consumers, etc. The main motivations are:
(a) The electrical grid operation is becoming more complex due to the higher number of actors and services, that range from new technologies and decentralized power generation and electrical storage to new market frameworks, players and services. The current electric system was not designed to accommodate diversified and distributed energy sources, particularly renewable ones characterized by their variable behavior. The more restrictive regulation requirements together with the increased complexity of the energy network system operation requires a more resilient power system, requiring automated procedures supported by new software and hardware technologies, with a strong emphasis in digital technologies.
(b) Investment in internet-of-things technology driven to foster new business opportunities that extract value from collected data and improve energy efficiency (e.g., smart homes and buildings, industry 4.0), asset management and maintenance policies.
(c) Revolution in the buy & sell of energy at the wholesale and retail markets. New market schemes, e.g. peer-to-peer trading, require distributed autonomous trading at the community level.
Recent Research and Technical Breakthroughs
The application of AI techniques to the energy sector is not new and the first state-of-the-art review was published in 1989 under the name “expert systems”  and several AI applications were reviewed in 1997 . In fact, neural networks and expert systems have been used for more than 30 years in the energy sector and for different applications, such as energy optimization and forecasting. However, recent breakthroughs in AI research showed a potential for the revival of this technology in the energy sector, in particular the following breakthroughs:
- AI hardware: Semiconductor performance is a key driver behind progress in AI applications. Graphics processing units (GPU) mean fast training and model iteration.
- Deep learning: Breakthrough in computer vision (detecting objects, understanding actions) and in learning goal-oriented behavior (reinforcement learning).
- Neuroscience: Better understanding of biological brains inspired new concepts for AI: episodic and working memory, attention, continual learning, imagination, etc.
- Transfer learning: Knowledge acquired by a trained model can be re-applied during the training process for a new task. Reduces the amount of data needs.
- Automated AI: Automatically discovers the best model architecture for a specific task (reduces human craft).
For example, very recently several potential applications from Google AlphaGo, namely deep reinforcement learning and Monte Carlo tree search, in electric power systems were discussed in . If fundamental requirements related with scalability and replicability are met in different energy use cases, it will represent a game changer for the sector.
Use Cases for the Energy Sector
The key challenge is on how to identify interesting use cases in the energy sector for AI technologies. The following figure, inspired by the Cognitive Bias Codex, summarizes the main drivers of AI impact and is being applied by INESC TEC to find AI use cases in the energy domain.
When applied to the energy sector, and for a non-exhaustive list of examples, we identify the following use cases with high potential for AI technologies.
In this figure it is important to underline the transversal application of AI in the energy sector, covering use cases for consumers/citizens, market players, energy services companies, distribution and transmission system operators.
Use Case Example at INESC TEC: Energy Optimization in Wastewater Stations
Urban wastewater sector is being pushed to optimize processes in order to reduce energy consumption without compromising its quality standards. Energy costs can represent a significant share of the global operational costs (between 50% and 60%) in an intensive energy consumer. Pumping is the largest consumer of electrical energy in a wastewater treatment plant. Thus, the optimal control of pump units can help the utilities to decrease operational costs.
In the European Project InteGrid, INESC TEC developed an innovative predictive control policy for wastewater variable-frequency pumps that minimize electrical energy consumption, considering uncertainty forecasts for wastewater intake rate and information collected by sensors accessible through the SCADA (Supervisory Control and Data Acquisition) system  (and patent pending). The proposed control method combines statistical learning (regression and predictive models) and deep reinforcement learning (Proximal Policy Optimization — PPO) for addressing the following set of objectives:
- Predictive control of variable-frequency pumps
- Anticipate periods of high wastewater intake
- Include different levels of wear and tear of the pumps
- Be easily implemented and it is scalable to other systems
The following main original outcomes were produced:
- Model-free (avoids the physical modelling of the wastewater system) and data-driven predictive control
- Control philosophy focused on operating the tank with a variable wastewater set-point level
- Use of supervised learning to generate synthetic data for pre-training the reinforcement learning policy, without the need to physically interact with the system
The results for a real case-study during 90 days show a 16.7% decrease in electrical energy consumption while still achieving a 97% reduction in the number of alarms (tank level above 7.2 meters) when compared with the current operating scenario (operating with a fixed set-point level) — see more results in . The numerical analysis showed that the proposed data-driven method is able to explore the trade-off between number of alarms and consumption minimization, offering different options to decision-makers.
Some Predictions and Impacts
AI technology will be a building block of the future energy sector, but without waiving domain knowledge and “classical” analytical method. Therefore, we leave here our five predictions and impacts for the next 5 years.
 Zhang, Z. Z., Hope, G. S., Malik, O. P. (1989). Expert systems in electric power systems-a bibliographical survey. IEEE Transactions on Power Systems, 4(4), 1355–1362.
 Madan, S., Bollinger, K. E. (1997). Applications of artificial intelligence in power systems. Electric Power Systems Research, 41(2), 117–131.
 Li, F., Du, Y. (2018). From AlphaGo to power system AI: What engineers can learn from solving the most complex board game. IEEE Power and Energy Magazine, 16(2), 76–84.
 Filipe, J., Bessa, R.J., Reis, M., Alves, R., Póvoa, P. (2019). Data-driven predictive energy optimization in a wastewater pumping station. Applied Energy, 252, 113423. https://arxiv.org/abs/1902.03417
Ricardo Bessa, Senior Member IEEE, received his Licenciado (five-year) degree from the Faculty of Engineering of the University of Porto, Portugal (FEUP) in 2006 in Electrical and Computer Engineering. In 2008, he received the M.Sc. degree in Data Analysis and Decision Support Systems on the Faculty of Economics of the University of Porto (FEP). He obtained his PhD degree in the Doctoral Program in Sustainable Energy Systems (MIT Portugal) at FEUP in 2013. Currently, he is Assistant Coordinator and Senior Researcher at INESC TEC in its Center for Power and Energy Systems and co-founder of a start-up company Prewind that sells renewable energy forecasting services.
His research interests include renewable energy forecasting, electric vehicles, data mining and decision-making under risk. He worked in several international projects such as the European Projects FP6 ANEMOS.plus, FP7 SuSTAINABLE, FP7 EvolvDSO, Horizon 2020 UPGRID, Horizon 2020 InteGrid and an international collaboration with Argonne National Laboratory for the U.S. Department of Energy. At the national level, he participated in the development of renewable energy forecasting systems and consultant services about energy storage. He is co-authors of more than 38 journal papers and 88 conference papers and member of the editorial board in IEEE Transactions on Sustainable Energy and MDPI Energies.