Machine learning prognosis for system imbalance volumes (IMPALA)

Electric power is a perishable commodity - it must be produced the moment it is consumed. For us to have a stable and reliable supply of power, production and consumption must always be balanced with great precision. Statnett has the balancing responsibility in Norway. Their starting point is a power system that is planned (1-day to 1-hour ahead) to be in balance, but since no one can know exactly when the wind blows or the power is consumed, imbalances occurs. Their means of dealing with imbalances is reserves. These reserves can be divided into automatic reserves - which are activated instantaneously in the event of imbalances, and manual reserves - which Statnett can manually be activated. As the volume of automatic reserve is limited, the quality and security of supply of power depends on how well Statnett can counteract imbalances using manual reserve activations. This requires good predictions for the upcoming imbalance as it typically takes 5-20 minutes to active these reserves. The aim of this project is thus to develop intelligent machine learning systems that can predict the upcoming imbalances. Artificial intelligence and machine learning are disciplines that have received a big boost recently and described by many as the next industrial revolution. These fields and methods differ from today's IT-systems mainly because the systems they produce can be trained to specialized tasks itself without being directly instructed by a human. Thus, a system based on machine learning methods and enough data can develop an expertise far beyond human capability. An analogy to a machine learning system in this specific context is a person who develops an empirical intuition, becomes and expert, and has an infinite appetite for processing information. The R&D project - Impala - have developed an Aritificial Intelligence capable of accuratly predicting upcoming system imbalances in the Nordic power system - in real time. So far, the projected has succeeded in developing this in an offline environment, and the emperical results displays significantly better preformance than the existing systems. This has convinced the pilot customers - Statnett and SvK - that Impala will enable them to reduce frequency deviations in the Nordics, and they have expressed so explicitly in their publications. The project is now in its final phase and the artificial intelligence is being implemented in a real-time prototype system, enabling the system operators to use Impala in real time operations.

The outcome of the project is a prototype software proving that artificial intelligence can make a significant improvement to both the quality and security of supply in the Nordics. This is of great value now, but of even greater value in the future as more intermittent renewable productions is entering the power system and thereby increasing the operational challenge of power system balancing. The project has been carried out in the context of the Nordic power systems, but the results are applicable to other large power systems. The potential social economic benefits are large as the AI-based decision support tool developed is able to support the operational management of today's systems and thereby improve security significantly and reduce cost. The benefits are even larger in power systems with less flexible power supply, such as continental Europe and the power systems of tomorrow, because it enables operators to counteract upcoming disturbance proactively.

The costs that TSOs have related to mitigating imbalances and frequency deviations have increased steadily the last 10 years, especially in the Nordics but also in many other regions around the world. The drivers behind the increase in the Nordics (higher intermittent production from renewables and tighter integreation with surrounding countries) in costs are expected to become stronger going forward. In general, the shorter the time frame that a system operator has to respond to a system imbalance, the fewer resources are available for mitigating actions and correspondingly the more expensive it is for the system operator to take action. In addition, in extreme periods unexpected large imbalances can lead to system blackouts, which can have a very high societal cost. Being able to forecast system imbalance volumes therefore has a great value to system operators. Currently, there are no effective tools available in the market for providing such forecasts. Additionally, published literature on forecasting system imbalance volumes is very limited and of a quality that leaves great room for improvement. The question that this project is designed to address is thus: How can we design an algorithm that is fast enough to be used in a real-time environment while at the same time providing a highly reliable forecast of system imbalance that can easily be updated as new knowledge becomes known?