Reaching our climate goals with forecasting models
Both the EU as well as Germany have set ambitious climate protection goals: By 2050, the annual greenhouse gas (GHG) emissions should be reduced by 80 to 95 percent compared to 1990.
Decentralised energy generation will play a crucial role in reaching these goals while simultaneously meeting the global demand for energy. Generation from locally available energy sources such as solar, wind, and other renewables must increase steeply in future. Decentralisation disrupts traditional, centralised and linear energy generation. This results in an extraordinarily complex system of many sources of energy and stakeholders. The situation is complicated further by the range of fluctuation of renewables, making the network even more complex.
A highly dynamic market is born out of this, both for providers as well as consumers. The latter can also simultaneously turn into providers, so-called “prosumers”. Every stakeholder on the market will have to make decisions by the minute about which behaviour is most economically efficient: consuming energy, selling energy, storing energy. Forecasting models and comprehensive networking of all stakeholders are key, as this will ensure optimisations in terms of costs as well as CO2 emissions.
The following stakeholders are classically represented on the energy market (simplified model):
- Energy consumers
- Grid infrastructure operators and grid agencies
- Energy producers and distributors
Artificial intelligence (AI) can play a major role in aligning the needs of these three stakeholders while offering significant support in the process. Subsequently, we analyse various solutions in the roles they can play to help reduce CO2 emissions through the optimisation of energy consumption, storage and generation, while choosing the most economically viable option.
Artificial intelligence on the energy consumption side
Private households are increasingly developing into “smart homes”. Some platforms assume all networking tasks within a home, capable of reacting to changes in weather conditions. The washing machine can be operated using the solar power generated by the home itself, and an electric vehicle can at the same time be charged if the weather is sunny and sunny weather is forecasted for the next day as well.
If, on the other hand, rain is predicted for the next day and no trip by car is scheduled either, the car will not charge, and the generated power is stored in the internal battery of the house instead. Solutions for connected homes include the Bosch Energy Manager and SMA Sunny Home Manager.
To date, energy managers are generally only used in houses with their own photovoltaic installation or heat pump. In the future, they will be a staple of every household, to disconnect consumers from the grid when there are high network loads (e.g. with electric vehicles simultaneously charging) and prevent grid collapse. On the other hand, they also help when the other extreme kicks in: whenever excess power is generated by wind or PV installations, appliances are automatically started and operate. Examples include automatically charging the battery of an e-bike or starting the washing machine.
There are still obstacles to overcome when it comes to electric cars. Only the Tesla models can be controlled remotely from a smart home. It is important to note that electric cars cannot simply be switched on or off externally using a relay. The car’s own computer usually handles this process. Before pulling a plug, the car must be instructed to halt the charging process. Only then can the plug be pulled. Otherwise, the batteries might experience spikes.
To facilitate this process, all electric vehicles should have the option of being remote-controlled through an open API. Then, the AI of the smart home as well as the AI of the grid operator can switch the vehicles on and off as required.
Companies also have a range of options available to them to improve their energy efficiency and reduce their CO2 footprint. There are many solutions available, which illustrate the energy consumption and allow for (partial) optimisations with forecasting models. Some examples are laid out in the following.
The energyControl software developed by Recogizer uses artificial intelligence to optimise air-conditioning technology in office buildings. The self-learning system utilises weather and occupancy data as well as information on building use and opening hours to make calculations and prognoses. Not only can these energy requirement forecasts be created for the entire building, but for individual climate zones as well. Depending on need, the climate can be controlled fully automatically in cycles of 15 minutes. If the sun is shining on the building in summer, the air-conditioning is turned on. On cloudy days, it might be possible to shut down the system and save energy.
Google follows a different path. The group from Mountain View has developed an algorithm which does not aim at reducing energy consumption, but rather strives to increase computing power in its data centres whenever renewable sources of energy are available. All non-critical services – such as adding new words to “Translate” or creating new photo filters – will be carried out whenever the sun is shining brighter or the wind is blowing harder. The forecasts of two different models are compared every day. Google has developed its own model to predict the hourly energy consumption of its data centres. The Tomorrow model by the eponymous Danish start-up is used for comparison. It predicts changes to the average hourly carbon intensity of local power grids over the course of a day. Both sources are compared using the “carbon-intelligent platform” by Google to only have computing carried out at times of low-carbon energy supply on an hourly basis.
Artificial intelligence in the grid infrastructure
Smart grids are essential to aligning both consumers and prosumers in the future. The intelligent electrical grids link the available information and data to balance load peaks.
The ifesca company offers such a solution. Its software ifesca.AIVA can provide information on grid demand on short notice. These data can be used to carry out simulation calculations on the feed and exit situation as well as overloads. Grid operators can gain insight into the effects of various load distributions relative to the prognoses and which intelligent actions can be taken to keep the grid in balance.
To understand how the grid infrastructure must be designed to handle future energy needs, we have trained models for forecasting at Cloudflight. They use, among other resources, mobility data and patterns deducible from them, to make predictions on the expected energy demand of a certain region.
Of course, the grid infrastructure can be damaged and fail as a result of the load situation, weather conditions, or age of the grid infrastructure. To prevent outages before they occur, artificial intelligence is used for predictive maintenance.
This process has been in use since 2017 in the medium-voltage grid of Schleswig-Holstein Netz. The algorithm includes a range of data in its predictions of potential faults. These include fault reports, load data, weather data, and facility data. This data analysis not only helps prevent grid outages but is also used to better plan and prepare maintenance activities. The majority of maintenance work is now scheduled, with fewer “emergency interventions”.
We at Cloudflight, too, have been active in this field for approximately two years for an Austrian energy supplier. Through ongoing monitoring of various facility parameters, we can recognise emerging problems before they negatively affect operations.
Artificial intelligence in energy generation and distribution
Prognoses of certain parameters play a crucial role in the economic efficiency of energy generation.
In district heating, power is a “waste product”, which can, therefore, be sold separately. It is worth noting that heat can be stored more easily and more affordably than power. The challenge lies in creating the most cost-efficient solution, striking a balance between generation and storage. With a link between the two types of energy.
In a Cloudflight project for an energy supplier, we have developed a model which predicts the demand for district heating on the basis of changes to the outdoor temperature as well as influencing factors which vary with the time of day. Heat is produced whenever power can be sold at a higher price. As soon as power gets cheaper, heat and associated power production are halted, and district energy is taken from the energy stores.
Such a model uses, among other data, very fine-mesh input data, while taking temperature curves into consideration.
AI can not only help realise greater revenue through optimised sales, but also offers protection against expensive fines. In some markets, the operators of solar and wind plants risk getting fines whenever the power they actually generate deviates from the prognosis. Energy & meteo systems offer a solution. Their customer portal operates at the interface of meteorology and the energy economy, giving plant operators access to prognoses on their generation data from five minutes up to 15 days in the future. The model uses various weather forecasts and wind generation prognoses for its predictions.
Power supply company awattar has an offer for its customers which is still unique in Germany. Rates are based on the hour. With a link to the spot market price for energy, rates fluctuate over the course of a day, depending on the amount of (renewable) power available in the grid. If there is excess supply and lacking demand, the spot market price can even drop below zero, as is illustrated in the following diagram. A smart meter is installed at the customer’s site. The distributor can offer its customers a highly innovative product which helps cutting costs while contributing to the energy transition.
Overarching data exchange for better AI
The examples discussed before illustrate that artificial intelligence opens up a range of options to reduce the CO2 footprint by allowing easier integration of renewables into the existing grid infrastructure.
Nevertheless, AI can only unfold its true potential if the three previously addressed fields have completed the process of digitalisation and are connected to each other. Overarching data exchange is immensely important to make highly exact predictions and automate all parameters, facilities, and energy-consuming equipment. Only then can the algorithms learn from each other and continuously improve themselves.
Smart meters, which transmit data on an hourly basis, must be installed in households to deliver current data. Many models only transmit data once a week. This falls far short of what’s needed for active controlling.
How well-positioned is your company?
Have you already taken these issues into consideration? In addition to monetary incentives, being seen as a “green” company will give you an edge with both customers and potential new employees. Today, the purpose of companies is a crucial factor in people’s decision for a new supplier or employer. But, realistically, not every company or business model is suitable for having a profound purpose. Therefore, the economical and environmentally friendly design of daily processes offers an attractive option to position yourself.
You can focus on controlling loads to aid in balancing the grid using AI. Enormous price benefits are possible depending on how your power consumption is spread out over the day. Even if this “only” concerns charging e-cars at the employee parking lot. The example of awattar illustrates that a significant increase in the use of renewables is possible.
There is no better time than now to start studying the possibilities of energy optimisation using artificial intelligence.
Whether this concerns saving energy, cutting costs, or increasing the share of renewables to use “clean” energy, there have never been fewer obstacles to overcome. Technologies such as artificial intelligence, Digital Twin, and Cloud Native are ready to go, now is the time to start implementing.
The same applies to the field of energy producers and grid operators. The foundations for the future are laid now. Intelligently position yourself and your infrastructure to be optimally prepared for the decentralisation of the energy economy. On the one hand, you should start preparing for new business models, while complexity will increase on the other hand, as mentioned in the introduction. But your budget does not have to. You will not need to hire new staff; the only way to master this greater complexity is by using better technology. You don’t have to go “the full monty” right away. Start with some proof of concepts and explore this “new world” by gathering initial experiences.
As you can see, artificial intelligence has some options in store to help you contribute to climate goals. Do you take the future of the following generations seriously? Then you should act now.