Energy demand & price forecasting are an integral part of the decision making processes at energy suppliers. The costs associated with under-or over contracting and then trading energy on the wholesale market are generally very high.
In order to stay competitive in the market, energy suppliers have come to rely on Machine Learning models to make more accurate forecasts and to avoid imbalance costs. During disruptive times these new models, which are build using historical data, are challenged by structural changes. How can the energy sector use Machine Learning to stay competitive in times of structural change caused by COVID-19?
One of our partners for forecasting solutions, Tangent Works, offers their take on how their TIM (Tangent Information Modeler) tackles these challenges.
Forecasting in the Energy Sector
There are several types of forecasts on which energy suppliers rely to maintain their business and improve results. From a supplier’s point of view, the energy market is very competitive. In order to stay ahead, they need to accurately forecast energy prices and estimate their client portfolio’s future consumption.
The first main forecast, or Price forecasting, is needed to optimize trading on the different energy markets. These markets are characterized by considerable fluctuations in prices. By correctly anticipating these fluctuations, an energy supplier can directly improve its margins on client contracts.
Secondly, an energy supplier is required to maintain a balance on the grid for its client portfolio. Over-or underestimation of the demand will result in the supplier being charged imbalance costs by the grid operator. To purchase the right amount of energy and minimize these imbalance costs, they rely on Load Forecasting models which take historical energy consumption data from clients and estimate their future demand.
Machine Learning & structural changes
Energy consumption data and price data are prime examples of time-series data with recurring seasonalities and visible patterns. Typical Machine Learning models are excellent at recognizing these patterns and generating accurate forecasts. Energy suppliers are now relying more and more on these models to improve their forecasting capabilities, however they come with certain drawbacks.
In the energy sector, it is not uncommon for structural changes to take place. New energy production capabilities, new large volume grid users or unexpected changes in demand often create different patterns in the data. In these situations the performance of Machine Learning models is challenged since they are build using historical values.
In order to continue to make accurate forecasts and mitigate the risks accompanied with structural changes, energy suppliers will continuously need to adapt their models. However in practice, a high frequency of model building is not viable and can disrupt the business. Another solution is needed to ensure accurate forecasts.
Companies relying on ‘hand-crafted’ forecasting in Excel or other solutions are even worse off than the people relying on Machine Learning models. The differences between different modeling approaches are explored in detail in our previous blog:
Adaptive forecasting to respond to Covid-19
Tangent Works’ model building engine, TIM (Tangent Information Modeller), claims to generate and apply high-quality adaptive forecasting models for time series data in just a few seconds, which means TIM overcomes the challenges posed by structural changes (Tangent Works, 2018). How does TIM accomplish this?
Multiple innovative aspects come together to reach the solution. Firstly, in contrast with handcrafted modelling techniques and AutoML, TIM automates feature engineering. TIM goes even further, unifying this automated feature engineering, model building and model deployment (i.e. application of the model in production) into one single step.
The technology that accomplishes this is called RTInstantML (real-time instant Machine Learning). This extensive automation enables users to train a new model for each desired forecast, eliminating the need of designing and building models beforehand while ensuring optimal models are used during times of variable feature availability. RTInstantML thus helps to overcome the challenges of structural changes in forecasting through both ad hoc feature engineering and on-demand model building.
A recent example: In figure 1 we see the impact of COVID-19 measures (such as extensive lockdowns) on the grid load. With the use of adaptive forecasting models such as the RTInstantML solution of TIM, energy suppliers can avoid overestimation of the energy demand of their client portfolio and reduce imbalance costs.
In figure 2 we can see the Belpex spot prices and the distinction before (black) and after (red) the implementation of COVID-19 measures in Belgium. We clearly see a different pattern emerging, where steep price drops are more likely. This volatility increases pressure on energy suppliers who are trying to maintain their margins. With adaptive forecasting, the new data will quickly add information to the model which can be used to anticipate these price swings.
In situations where there is a significant difference between new and historical values, typical machine learning models are often rendered useless. RTInstantML allows energy suppliers to instantly take into account the new market conditions and avoid costly imbalance costs and unexpected price swings. The proven adaptive forecasting capabilities of TIM can help energy suppliers to stay competitive during disruptive times.
As mentioned in our previous forecasting blog we have neatly integrated this solution into our own SaaS platform for gas & power operations. We’re happy to discuss the benefits if you’re interested!