PART 4: Why You Can't Predict the Class A Peaks

In this five-part series called "Tame the Beast," the goal is to help you understand:

  1. The different players in the Ontario electricity market - setting the field;
  2. What happened to Ontario's electricity prices - the beginning signs of trouble;
  3. The impact of electricity peaks and the Global Adjustment - the serious situation;
  4. The story of how many companies are trying to reduce their electricity costs - the struggles of demand response; and,
  5. How companies are successfully reducing their Global Adjustment costs - how to tame the beast.What is the Global Adjustment?

As covered in the previous section, there are two major groups of customers in Ontario – the Industrial Conservation Initiative (ICI) Class A participants who are the largest energy consumers in the Province, and the Class B composed of everyone else. This former group is the one that is trying to predict the energy demand in the Province and try to reduce their energy loads during peak periods. At the same time, the number of businesses that are eligible for Class A has grown rapidly and this expansion of Global Adjustment Class A eligibility in the ICI Program has led to an increasing number of interested parties trying to take advantage of the Program. This is mostly because, by properly managing the Program, a business can save up to 70% of their annual electricity costs as a Class A customer.

In order to do so, a business would need to predict when each of the top five peak hours will occur each year, and then lower their electricity usage as much as possible during these hours. The top five peak hours are the five hours each year where the electricity usage in the Province of Ontario is at its highest. With an increased number of companies designating as Class A and consequently attempting to predict the peak hours, however, it is becoming increasingly difficult to accurately predict when these peaks will occur.

The reason for this is a concept known as chaos theory, which simply means that the more participants involved in a certain event, the more difficult that thing becomes to predict. With the expansion of the ICI Program, and the addition of many new Global Adjustment Class A customers, there are now more players than ever trying to predict and respond to the top five peak hours. Most of these Class A customers are attempting to use the publicly available peak predictions offered by the IESO through their peak tracker, however, it is very difficult to beat the market when you are using the same data as everyone else.

The IESO publishes their predictions of the Province’s electricity usage publicly each day, by updating the information on their peak tracker portal, and this data is used by many companies to predict when a peak hour will occur. These publicly available predictions are often used by Global Adjustment Class A customers and other market actors to decide when, or if, they should reduce their electricity usage. In fact, the majority of energy consultants serving the Province of Ontario do not go beyond this publicly available peak tracker for their analyses of the energy market. As a result, most of the decisions made by Class A customers to respond to peak hours are based on the same publicly available information. This was not a problem when there were only a few dozen Class A companies responding to peak days; however, with thousands of companies now responding, this publicly available peak tracker is becoming increasingly unreliable.

This is not because the IESO peak tracker is itself particularly inaccurate. In fact it has a high accuracy of around 95%. Rather, it is the very fact that it is publicly available that makes the data from the peak tracker unreliable. This is because as many market players use the same publicly available peak tracker predictions as the basis for their demand response efforts, Ontario has begun to experience a large increase in what are now known as “False Peaks” and “Double Peak” days.


A False Peak occurs when the publicly available predictions provided by the IESO estimate that a peak day will occur, causing many businesses to reduce their electricity usage, sometimes causing over 2,000 MW of electricity usage to suddenly shutdown. As a result of this demand response, the usage becomes much less than predicted and no peak day actually occurs. This becomes an even more severe problem during periods of prolonged cold, such as the cold snap that occurred in late December 2017 and early January 2018. During this period, the publicly available IESO peak tracker repeatedly predicted that a peak day was imminent, but this model was repeatedly overestimating the actual usage in the Province. In fact, out of the 12 days between December 28th and January 8th, the IESO predicted that 9 days had the potential to be a peak day. In actuality, only 2 of these days became peak days. For many businesses operating during the holidays, this caused major disruptions to their operations. Due to the inaccuracy of the publicly available IESO data, many businesses were forced to repeatedly shut down their operations in an attempt to reduce usage during peak days, most of which never actually happened.

Double Peaks have also become quite common during summer days, for example as on September 26th, 2017. This usually occurs during extreme weather events, when every Global Adjustment Class A customer can be confident that the day will become a peak day. The problem, however, becomes in predicting precisely which hour will be the peak hour. Most Class A businesses continue to rely on the publicly available IESO predictions to estimate when the peak hour will occur. As an example for the issues that can occur due to this, on September 26th the IESO predicted that the peak hour would occur between 6 and 7pm. This consequently caused many businesses to reduce their usage from 6 to 7pm to catch the peak hour. As a result, however, usage decreased so much that two so-called “shoulder peaks” occurred from 5 to 6pm, and from 7 to 8pm. As a result, all of the businesses that relied on the IESO data missed the peak hour, and therefore those businesses earned no savings in the ICI Program.

In response to this inaccurate data, some businesses have decided to shut-down their production for much longer periods of time. For example, they may decide to shut down from 4 to 9pm. However, this causes enormous disruptions to their production and can be extremely costly.

The new phenomenon that is beginning to occur is that the greater the projected peak usage, the more likely it is that Global Adjustment Class A customers will respond. As a result, with a greater response from the market, it becomes more likely that the peak will not actually occur as originally predicted. Clearly, a different solution is needed rather than just the publicly-available information from the IESO and basic spreadsheet models that are run off the same information. Given the phenomenon described above, a complex algorithm becomes necessary that will not only estimate what the Province’s electricity usage will be, but will also estimate how the rest of the market will respond. This model would then be able to avoid False Peak days, and accurately predict which hour will be the true peak hour. There is a solution that may be able to achieve this and is described in further detail in the next section.


Naturally, demand response is not the only factor affecting the pattern of energy demand in the Province of Ontario. There are also baseline factors that have always occurred in the market. With extreme weather conditions, either too hot or too cold, energy usage spikes as residents and businesses alike attempt to control the temperature in their buildings.

There are also fairly reliable and consistent energy usage patterns that occur between 4 and 8pm in the summer, and between 5 and 7pm in the winter, which are strongly correlated with when homeowners are returning home each day. Although a single individual coming home at a consistent time and turning on their heating, lights, and home appliances is obviously not enough to affect the Provincial energy demand; millions of individuals all living on roughly the same schedule can cause major changes in energy usage on the provincial scale.i

Most Class A customers have a system in place to reduce their usage during peak hours, with some businesses simply turning off their lights, while others will completely shutdown during peak hours. Each strategy comes with its own set of advantages and disadvantages, especially since strategies can affect production schedules, be labour-intensive, prohibitively expensive, or some combination of the above.

The majority of Class A customers hire external energy consultants to predict the energy market to help shift electricity usage during peak hours. Many of these so-called experts simply just use the publicly-available IESO data, which is an approach that has many issues as mentioned above. However, some experts will go further. These experts will usually attempt to tackle this issue by using a statistical data analysis approach called linear regression which analyzes and models a relationship between the scalar dependent and one or more independent variables. What this means is that given a number of points of data, these consultants will try to essentially “connect the dots,” and based on this effort, predict the placement of future points. Through the use of this model, linear regression analyses can be used to fit simple predictive models. However, these models are designed for simple predictions, while the prediction of energy usage in the Province of Ontario, which relies on hundreds of different variables, is most certainly not a simple prediction.

When considering simple sets of variables and only a few influencing factors, linear regression analyses are indeed highly effective at displaying trends. However, in order to forecast energy demand accurately, the model must, at a minimum, consider the following factors:

  • Air Temperature;
  • Wind Speed;
  • Dew Point;
  • Humidity;
  • Probability of Precipitation Forecasts;
  • Human Behavior;
  • Energy Response Fluctuations; and
  • Many other factors.

Exacerbating the problem further is that all of these factors interact with one another, and therefore the complexity of these relationships increases exponentially. Moreover, even if one were to account for every factor that affects the energy demand in the Province of Ontario, they would still need to consider how human behavior influences the energy response fluctuations as one example, or how air temperature changes human behavior.

The result is that models built using linear regression algorithms are simply unable to accurately predict Ontario’s electricity demand; especially as it becomes more important to not only predict the base electricity usage, but also predict how other market participants will respond during peak days.

Thankfully, a better option exists, which is described in more detail in the next section.


i “Ontario Energy Quarterly - Data Catalogue.” Ontario Energy Report, Government of Ontario, 1 Sept. 2017,