The POWER Interview: The Importance of AI and Machine Learning

Artificial intelligence (AI) and machine learning (ML) are becoming synonymous with the operation of power generation facilities. The increased digitization of power plants, from equipment to software, involves both thermal generation and renewable energy installations.

Both AI and ML will be key elements for the design of future energy systems, supporting the growth of smart grids and improving the efficiency of power generation, along with the interaction among electricity customers and utilities.

The technology group Wärtsilä is a global leader in using data to improve operations in the power generation sector. The company helps generators make better asset management decisions, which supports predictive maintenance. The company uses AI, along with advanced diagnostics, and its deep equipment expertise greatly to enhance the safety, reliability, and efficiency of power equipment and systems.

Luke Witmer

Luke Witmer, general manager, Data Science, Energy Storage & Optimization at Wärtsilä, talked with POWER about the importance of AI and ML to the future of power generation and electricity markets.

POWER: How can artificial intelligence (AI) be used in power trading, and with regard to forecasts and other issues?

Witmer: Artificial intelligence is a very wide field. Even a simple if/else statement is technically AI (a computer making a decision). Forecasts for price and power are generated by AI (some algorithm with some historic data set), and represent the expected trajectory or probability distribution of that value.

Power trading is also a wide field. There are many different markets that span different time periods and different electricity (power) services that power plants provide. It’s more than just buying low and selling high, though that is a large piece of it. Forecasts are generally not very good at predicting exactly when electricity price spikes will happen. There is always a tradeoff between saving some power capacity for the biggest price spikes versus allocating more of your power for marginal prices. In the end, as a power trader, it is important to remember that the historical data is not a picture of the future, but rather a statistical distribution that can be leveraged to inform the most probable outcome of the unknown future. AI is more capable at leveraging statistics than people will ever be. 

POWER: Machine learning and AI in power generation rely on digitalization. As the use of data becomes more important, what steps need to be taken to support AI and machine learning while still accounting for cybersecurity?

Witmer: A lot of steps. Sorry for the lame duck answer here. Regular whitehat penetration testing by ethical hackers is probably the best first step. The second step should be to diligently and quickly address each critical issue that is discovered through that process. This can be done by partnering with technology providers who have the right solution (cyber security practices, certifications, and technology) to enable the data flow that is required.

POWER: How can the power generation industry benefit from machine learning?

Witmer: The benefit is higher utilization of the existing infrastructure. There is a lot of under-utilized intrastructure in the power generation industry. This can be accomplished with greater intelligence on the edges of the network (out at each substation and at each independent generation facility) coupled with greater intelligence at the points of central dispatch.

POWER: Can machines used in power generation learn from their experiences; would an example be that a machine could perform more efficiently over time based on past experience?

Witmer: Yes and no. It depends what you mean by machines. A machine itself is simply pieces of metal. An analogy would be that your air conditioner at home can’t learn anything, but your smart thermostat can. Your air conditioner needs to just operate as efficiently as possible when it’s told to operate, constrained by physics. Power generation equipment is the same. The controls however, whether at some point of aggregation, or transmission intersection, or at a central dispatch center, can certainly apply machine learning to operate differently as time goes on, adapting in real time to changing trends and conditions in the electricity grids and markets of the world.

Artificial intelligence and machine learning will be major aspects of future power generation, helping make various generation types more efficient, reliable, and safe, and helping customers better interact with electricity providers. Courtesy: Wartsila

POWER: What are some of the uses of artificial intelligence in the power industry?

Witmer: As mentioned in the response to question 1, I think it appropriate to point you at some definitions and descriptions of AI. I find wikipedia to be the best organized and moderated by experts.

In the end, it’s a question of intelligent control. There are many uses of AI in the power industry. To start listing some of them is insufficient, but, to give some idea, I would say that we use AI in the form of rules that automatically ramp power plants up/down by speeding up or slowing down their speed governors, in the form of neural networks that perform load forecasting based on historic data and the present state data (time of day, metering values, etc.), in the form of economic dispatch systems that leverage these forecasts, and in the form of reinforcement learning for statistically based automated bid generation in open markets. Our electricity grids combined with their associated controls and markets are arguably the most complex machines that humans have built.

POWER: How can AI benefit centralized generation, and can it provide cost savings for power customers?

Witmer: Centralized power systems continue to thrive from significant economies of scale. Centralized power systems enable equal access to clean power at the lowest cost, reducing economic inequality. I view large renewable power plants that are owned by independent power producers as centralized power generation, dispatched by centralized grid operators. Regardless of whether the path forward is more or less centralized, AI brings value to all parties. Not only does it maximize revenue for any specific asset (thus the asset owner), it also reduces overall electricity prices for all consumers.

POWER: How important is AI to smart grids? How important is AI to the integration of e-mobility (electric vehicles, etc.) to the grid?

Witmer: AI is very important to smart grids. AI is extremely important to the integration of smart charging of electric vehicles, and leveraging of those mobile batteries for grid services when they are plugged into the grid (vehicles to grid, or V2G). However, the more important piece is for the right market forces to be created (economics), so that people can realize the value (actually get paid) for allowing their vehicles to participate in these kinds of services.

The mobile batteries of EVs will be under-utilized if we do not integrate the controls for charging/discharging this equipment in a way that gives both the consumers the ability to opt in/out of any service but also for the centralized dispatch to leverage this equipment as well. It’s less a question of AI, and more a question of economics and human behavioral science. Once the economics are leveraged and the right tools are in place, then AI will be able to forecast the availability and subsequent utility that the grid will be able to extract from the variable infrastructure of plugged in EVs.

POWER: How important is AI to the design and “construction” of virtual power plants?

Witmer: Interesting question. On one level, this is a question that raises an existential threat to aspects of my own job (but that’s a good thing because if a computer can do it, I don’t want to do it!). It’s a bit of a chicken-and-egg scenario. Today, any power plant (virtual or actual), is designed through a process that involves a lot of modeling, or simulations of what-if scenarios. That model must be as accurate as possible, including the controls behavior of not only the new plant in question, but also the rest of the grid and/or markets nearby.

As more AI is used in the actual context of this new potential power plant, the model must also contain a reflection of that same AI. No model is perfect, but as more AI gets used in the actual dispatch of power plants, more AI will be needed in the design and creation process for new power plants or aggregations of power generation equipment.

POWER: What do you see as the future of AI and machine learning for power generation / utilities?

Witmer: The short-term future is simply an extension of what we see today. As more renewables come onto the grids, we will see more negative price events and more price volatility. AI will be able to thrive in that environment. I suspect that as time goes on, the existing market structures will cease to be the most efficient for society. In fact, AI is likely going to be able to take advantage of some of those ‘legacy’ features (think Enron).

Hopefully the independent system operators of the world can adapt quickly enough to the changing conditions, but I remain skeptical of that in all scenarios. With growing renewables that have free ‘fuel,’ the model of vertically integrated utilities with an integrated resource planning (IRP) process will likely yield the most economically efficient structure. I think that we will see growing inefficiencies in regions that have too many manufactured rules and structure imposed by legacy markets, designed around marginal costs of operating fossil fuel-burning plants.

Darrell Proctor is associate editor for POWER (@POWERmagazine).