machine learning in energy - part two

10 minute read

This is the second of a two part series on the intersection of machine learning and the energy industry. Read the first part here

Data sceptic ai in industry

Key is to figure out where to focus / spend energy

”..too early to do everything with AI… but it’s too late to do nothing” Steve Guggenheimer, Microsoft’s corporate vice president of AI Business

Data is still a mess in most places (maybe put this in part one - but rehash the challenges here in context of the applications) Included tomorrow and the Swedish company in the energy in ml post, also include the UK demand side company, also include Tempus, Siemens ???

This post will detail three applications of machine learning in energy

  • forecasting of electricity generation, consumption and price
  • energy disaggregation
  • reinforcement learning

We will also take a look at one of the most famous applications of machine learning in an energy system - optimization of Google’s data centers.

forecasting of electricity generation, consumption and price

what’s the problem

The time of energy generation and consumption has a major economic and environmental impact. Large scale storage of electricity is only just becoming economic - without storage generation must match consumption. When demand is high the grid can be forced to use expensive and inefficient peaking plants. In periods of low demand electricity can be so abundant that the price becomes negative.

The more accurately a system operator understands what generation and demand will be in the future, the better they can manage the grid. Historically there wasn’t major uncertanity on either side.

The major uncertantity with dispatchable fossil fuel generators is unscheduled maintenance - usually expected to cause around 5-10% downtime annually. On the demand side the aggregation of an entire grid’s demand made the forecast a challenging but acceptable problem.

Our current energy transition is making the forecasting problem more difficult. Our transition towards intermittent and distributed generation is introducing uncertantity on both the generation and demand side.

Our current energy transition is moving us away from dispatchable, centralized and large-scale generation towards intermittent, distributed and small scale generation.

Like the weather, intermittent generation is hard to forecast. Wind generation depends on forecasting wind speeds over vast areas. Solar power is more predictable but can still see variation as cloud cover changes.

As grid scale wind & solar penetration increase balancing the grid becomes difficult. Higher levels of renewables can lead to more fossil fuel backup kept in reserve in case forecasts are wrong.

The distributed and small scale of renewables (solar in particular) also makes demand forecasting more difficult.

A solar panel sitting on a residential home is not directly metered - the system operator has no idea it is there. As this solar panel generates throughout the day it appears to the grid as reduced consumption. The most famous example of this the ‘duck curve’. In California the loss of distributed solar power as the sunsets appears to the system operator as a massive increase in demand, requiring dispatchable generation to quickly ramp up to meet the loss of solar.

Our current energy transition is a double whammy for grid balancing. Forecasting of both generation and consumption is becoming more challenging.

This has a big impact on electricity prices. In a wholesale electricity market price is set by the intersection of generation and consumption. Volatility and uncertainty on both sides spill over into more volatile electricity prices. South Australia is a prime example of this - the combination of high penetrations of wind and solar with little out of market subsidies makes the South Australian wholesale electricity market one of the most volatile commodities markets in the world.

how machine learning will help

Classical time series forecasting models such as SARIMA are well developed for forecasting energy time series. Machine learning adds additional tools to the time series forecasting toolbox.

Both regression and classification supervised machine learning models can be used for time series forecasting. Regression models can directly forecast electricity generation, consumption and price. Classification models can forecast the probability of a spike in electricity prices.

Well trained random forests, support vector machines and neural networks can all be used to solve these problems. Of particular interest are recurrent neural networks - networks that model the temporal structure of features and targets implicitly by feeding in and generating sequences.

A key challenge is data. As renewables are weather driven forecasts of weather can be useful exogenous variables. It’s key that we only train models on data that will be available at the time of the forecast. This means that historical information about weather forecasts can be more useful than the actual weather data.

Other useful variables for forecasting electricity generation, demand or prices include

  • lagged values of the target (i.e. what the price was last half hour
  • prices in neighbouring markets
  • prices of primary fuels such as natural gas or coal
  • interconnector flows or flows through constraints in the grid
  • externally supplied forecasts of the target variable (and the errors of those forecasts)

what’s the value to the world

Improving forecasts allows us to better balance the grid, reduce reliance on fossil fuel peaking or backup plants and reduce curtailment of renewables.

It’s not only the economic & environmental cost of keeping backup plant spinning. Incorrect forecasts can lead to fossil fuel generators paid to reduce output. This increases the cost to supply electricity to customers.

There are benefits for consumers of electricity as well. Improved prediction can also allow flexible electricity consumption to respond to market signals. More accurate forecasts that can look further ahead will allow more electricity consumers to be flexible. Using flexible assets to manage the grid will reduce our reliance on fossil fuels for grid balancing.

sources and further reading

Forecasting UK Imbalance Price using a Multilayer Perceptron Neural Network

energy disaggregation

what’s the problem

Imagine if every time you went to the restaurant you only got the total bill. Understanding the line by line breakdown of where your money went is valuable. Energy disaggregation can help give customers this level of infomation about their utility bill.

In an ideal world we would have visibility of each individual consumer of energy. We would know when a TV is turned on in a home or a pump is running in an industrial process. One solution would be to install metering on every consumer - an expensive, complex and impractical process.

Energy disaggregation is a more elegant solution. A good energy disaggregation model can estimate appliance level consumption through a single aggregate meter.

how machine learning will help

Supervised machine learning is all about learning patterns in data. Many supervised machine learning algorithms can learn the patterns in the total consumption. Kelly & Knottenbelt (2015) used recurrent and convolutional neural networks to disaggregate residential energy consumptions.

A key challenge is data. Supervised learning requires labeled training data. Measurement and identification of sub-consumers forms training data for a supervised learner. Data is also required at a very high temporal frequency - ideally less than one second.

what’s the value to the world

Energy disaggregation has two benefits - it can identify & verify savings opportunities, and can increase customer engagement.

Imagine if you got an electricity bill that told you how much it cost you to run your dishwasher that month. The utility could help customers understand what they could have saved if they ran their dishwasher at different times. This kind of feedback can be very effective in increasing customer engagement - a key challenge for utilities around the world.

sources and further reading

Energy Disaggregation: The Holy Grail (Carrie Armel)

Putting Energy Disaggregation Tech to the Test

reinforcement learning

what’s the problem

Optimal control of energy systems is hard. Key variables such as price and energy consumption constantly change. Operators control systems with a large number of actions, with the optimal action changing throughout the day.

Our current energy transition makes this problem harder. Increased uncertantity on the generation and demand side leads to more volatility in key variables such as electricity prices. The need for smarter ways of managing energy systems, such as demand flexibility, introduce more actions that need to be considered.

Today deterministic sets of rules or abstract models are commonly used to dispatch plant. Deterministic rules for operating any non-stationary system can’t guarantee optimality. Changes in key variables can turn a profitable operation to one that loses money.

Abstract models (such as linear programming) can account for changes in key variables. But abstract models often force the use of unrealistic models of energy systems. More importantly the performance of the model is limited by the skill and experience of the modeler.

how machine learning will help

Reinforcement learning gives a machine the ability to learn to take actions. The machine takes actions in an environment to optimize a reward signal. In the context of an energy system that reward signal could be energy cost, carbon or safety - whatever behavior we want to incentivize.


What is exciting about reinforcement learning is that we don’t need to build any domain knowledge into the model. A reinforcement learner learns from its own experience of the environment. This allows a reinforcement learner to see patterns that we can’t see - leading to superhuman levels of performance. Another exciting thing about reinforcement learning is that you don’t need a data set. All you need is an environment (real or virtual) that the learner can interact with.

It’s important to clarify that modern reinforcement learning, as advanced as impressive as it is, is not currently able to take control of our decisions for us. Modern reinforcement learning is very sample inefficient, and learning requires simulation. But the promise of intelligent machines that can see patterns we can’t see and take optimal decisions is a future that could happen.

what’s the value to the world

Better control of our energy systems will allow us to reduce cost, reduce environmental impact and improve safety. Reinforcement learning allows us to do this at superhuman levels of performance.

Letting machines make decisions in energy systems allows operators to spend more time doing routine or preventative maintenance. It also allows more time to be spent on upgrading existing plant.

sources and further reading

energy_py – reinforcement learning in energy systems

  • [Minh et. al (2016) Human-level control through deep reinforcement learning

]( - Reinforcement learning course by David Silver (Google DeepMind)

Alphabet/Google data centre optimization

One of the most famous applications of machine learning is Google’s work in their own data centers. In 2014 Google used supervised machine learning to predict the Power Usage Effectiveness (PUE) of data centres.

This supervised model did no control of its own. Operators used the predictive model to create a target PUE for the plant. The predictive model also allowed operators to simulate the impact of changes in key parameters on PUE.

In 2016 DeepMind published details of a how they applied machine learning to optimizing data centre efficiency. The technical details of this implementation are not as clear as the 2014 work. It is pretty clear that both supervised and reinforcement learning techniques were used.

The focus on the project again was on improving PUE. Deep neural networks predicted future PUE as well as future temperatures & pressures. The predictions of future temperature & pressures simulated the effect of recommended actions.

DeepMind claim a ‘40 percent reduction in the amount of energy used for cooling’ which equates to a ‘15 percent reduction in overall PUE overhead after accounting for electrical losses and other non-cooling inefficiencies’. Without seeing actual data it’s hard to know exactly what this means.

What I am able to understand is that this produced the lowest PUE the site had ever seen’.

This is why as an energy engineer I’m so excited about machine learning. Google’s data centers were most likely well optimized before these projects. The fact that machine learning was able to improve PUE beyond what human operators had been able to achieve before is inspiring.

The potential level of savings across the rest of our energy systems is exciting to think about. The challenges & impact of our energy systems are massive - we need the intelligence of machine learning to help us solve these challenges.

Sources and further reading

  • [Jim Gao (Google) - Machine Learning Applications for Data Center Optimization

]( - DeepMind AI Reduces Google Data Centre Cooling Bill by 40%

Thanks for reading!