This is a brief article on how AI can help solve problems in local energy markets and distribution grids, focusing on state-of-art AI technologies and most recent literature.

Materials used in this article are from the trial lecture for my PhD defense under the same title. There is also a video version of this article:

• Youtube | The role of AI in distribution grid: future prospects and interplay with local energy markets

The slides can be found here.

Problems for local energy market in the distribution grid

In the context of this lecture, which is to discuss the role of artificial intelligence in distribution grid control and local energy markets, we can try to map out the relevant problems a local energy market in the distribution grid might face. These are:

1. The need for accurate forecast of demand and variable renewable energy generation.
2. The need for time-efficient and reliable decision-making either in the market or for dispatching flexible capital. Given that most market participants in an local energy market might not be energy experts dedicated in the market all day long, they need time-efficient and reliable decision-making process when participating in the market and dispatching their flexible capital.
3. The need for a well-functioning distribution grid. Obviously a local energy market cannot function if the distribution grid it lies in malfunctions.
4. The need for secure and private ways of utilizing sensitive data. In the scale of local energy market, highly disaggregated demand and supply data are needed, which requires extra care in this regard.

Now we will go through each problem one by one, discussing why artificial intelligence is beneficial in addressing the problem and how it can be implemented to that end.

Accurate forecast of demand and variable renewable energy generation

The stochastic nature of electricity demand and variable renewable energy generation means that sufficiently accurate forecast of these variables in the near future is essential for economically efficient scheduling of flexible capital, either for the owners of the assets or the distribution grid and local energy market in general.

So we want accurate forecast of demand and variable renewable energy generation. But why exactly does artificial intelligence help here?

The main reason is that demand and variable renewable energy supply are related to various factors. So for example, demand profiles can be affected by household sizes and types, type and size of residents, and outdoor temperature. Solar generation can be affected by solar radiation and temperature. Wind generation can be affected by wind speed, terrain, and large scale weather systems.

Many of these factors exhibit non-linear relationships with broad spatial and temporal interdependencies. These characteristics make forecasting the profiles suitable for machine learning applications.

Time-efficient and reliable decision-making

Compared with human, artificial intelligence can perform routine tasks seamlessly with automated and more efficient workflow. This gives it potential contribution in helping owners of flexible assets to make time efficient and reliable decisions regarding the dispatch of their assets and participation in local energy market.

Why? That is because for professional traders and market operators, conventional optimization methods can be time-consuming, especially if the problem is non-convex.

In addition, local energy markets consist of many ordinary households. Households and small businesses might not acquire the necessary domain knowledge to participate in local energy markets on their own. Their dispatch and bidding strategies may therefore rely on simple heuristics.

Artificial intelligence can therefore strike a balance between the two extremes.

Artificial intelligence can be trained with supervised or reinforcement learning for decision-making. When trained with supervised learning, the “correct” decisions are labeled on the data using either conventional optimization methods or existing datasets. When trained with reinforcement learning, the model adjusts its policy iteratively to improve a reward function.

Well-functioning distribution grid

A well-functioning grid is crucial to the normal operation of a local energy market. Artificial intelligence can help in this regard because of these reasons.

1. The failure of critical components is affected by various factors such as weather, age, and utilization rate. Similar to forecasting demand and variable renewable energy generation, non-linear relationships of these factors make predicting it difficult.
2. The power flow equations of the distribution grid make the optimal control problem non-convex when voltage control is considered. Conventional optimization methods can be time-extensive.
3. During an outage, deciding whether to turn on or turn off switches and controlling critical components is also computationally challenging for conventional optimization methods
4. During a contingency, obtaining a robust and stable control policy is more important than obtaining the most economic efficient policy.

Data security and privacy

Finally, digitalization of the distribution grid creates huge volume of data storage, data transmission, and exploitable vulnerabilities for the distribution grid. The usage of artificial intelligence in distribution grid and local energy market applications can be designed using federated learning to mitigate the security and privacy risks accompany by digitalization.

However, artificial intelligence serves as a double edge sword for cyber-security in this regard, since both the system operator and the attacker can leverage computational intelligence at their favor.

Conclusion

1. There are many existing and potential applications of artificial intelligence in the control of distribution grids, including accurate forecast, time-efficient and reliable decision-making, maintaining distribution grid’s function, and data security and privacy.
2. Artificial neural network and classical machine learning once dominated the research, but in very recent years deep learning and generative artificial intelligence become hot topics. Blockchain and other encryption techniques are commonly applied together with artificial intelligence, especially in a federated learning environment.
3. For local electricity markets specifically, artificial intelligence technologies provide market participants, market operators, and distribution system operators improvement of supply and demand forecast. They can also offer time-efficient and privacy preserving alternatives to traditional market clearing algorithms based on centralized and conventional optimization methods. However, loss of optimality, as a trade-off, can often be overlooked.
4. Most literature around artificial intelligence and distribution grid discusses positive synergies between artificial intelligence and the grid; still, negative impacts should be considered more holistically alongside, including:

• Higher energy demand
• More sophisticated cyberattacks
• Social inequality due to differences in computational resources

Questions?

If no…