Wind energy is a key component in the race to achieve the climate control targets. Wind energy will only be able to measure up to its crucial role, however, if the wind turbines put in an optimum performance. One way of unlocking their full potential is to have intelligent control systems, preferably based on artificial intelligence (AI). Wind turbines can harness the power of AI to adapt independently to changes in environmental conditions and variations in load.
The ZSW has been involved with the study and application of machine learning (ML) for many years, investigating methods which enable very precise wind power forecasting. The complex interaction of weather model forecasts, satellite data, meteorological measurements and wind output data history at any given location are taken into account, as is the future wind output. The ZSW is now significantly expanding the application of AI in wind energy research as part of the WINSENT brief. We are using new machine learning methods to improve feed-in projections and to optimise models for the integration of storage solutions in future energy systems (e.g. power-to-gas, energy storage batteries).
The ZSW is also using AI to develop a bird recorder, a name given to a device which can detect birds in the immediate vicinity of wind turbines and identify the various species. If a flock of birds is flying towards the rotor blades, the control unit can then react immediately and take appropriate action to prevent a collision, such as reducing the speed. This is one way in which advances in technology are contributing to effective nature conservation.
The integration of renewable sources of energy in the energy system is becoming more and more important as the proportion of wind power and solar energy increases. In order to feed the total amount of energy generated into the power grid if at all possible, there is a need to optimise feed-in projections, especially for sun and wind because these energy sources are particularly volatile. Forecasts for the quantities fed in from run-of-river power plants are also of great importance, however, both in terms of grid supply and commercial exploitation.
The ZSW has been researching and developing feed-in forecasts for wind farms, solar power stations and run-of-river power plants for more than 10 years. The development and marketing side of the work is done in tandem with the domestic industry. The main input comes from weather service providers and electricity distribution companies in the field of renewable sources of energies and public utilities. For this purpose, the ZSW is developing and running an innovative, operational forecast system which is also used for the forecasts by the GridSage tool for the Redispatch 2.0 grid management scheme.
The latest AI methods are used for the feed-in forecasts in order to plot the complex relationship between numerical weather model data, meteorological measurements, satellite data and ongoing feed-in data. The AI systems are trained in a validation process using the data history from previous years before going into live operation for ongoing feed-in forecasts for up to 180 hours or, in certain applications, for up to 384 hours.
Both deterministic and probabilistic forecasting techniques are used depending on the purpose for which the feed-in forecasts are used. Numerical ensemble weather model predictions are usually taken as input parameters for the AI models for probabilistic forecasts. The ensemble models are very data-intensive, however, and require extensive computational input. This is why the ZSW has been developing new AI methods in the last few years with the capability to take deterministic weather models as a basis for valuable probabilistic feed-in forecasts. These have major advantages, especially in the management of power grids and the marketing of renewable sources of energy.
Wind turbines are complex, large-scale systems with a minimum service life of 20 years during which they will ideally be permanently in operation and free of malfunctions. In addition, the systems are sometimes exposed to extreme environmental conditions bringing immense stress to bear on the components, such as the rotor, bearings, gearbox and generator. It is therefore of great importance to monitor all the main components, keeping a constant watch over the system so as to detect any irregularities as early as possible, and to predict the probability of loss of normal operation, especially the occurrence of faults in modules which will necessitate downtime.
This is another ideal field of application for innovative AI methods which are best suited to accomplishing the relevant tasks online and in real time with the help of so-called SCADA data from wind turbines. The ZSW works with wind turbine manufacturers and their suppliers to this end, seeking to develop innovative solutions based on AI methods and follow through with their application in practice.
The assessment of a potential site for the generation of wind power is one of the first steps involved in the development of wind farms in a new location. If there are one or more wind farms in the immediate vicinity with comparable orography, the data on the output achieved at these neighbouring sites are often used to calculate the expected yield. This is not possible in complex terrain, mainly because the inflow and the intensity of the turbulence in the complex terrain are heavily dependent on the orographic conditions. It is then necessary to take wind measurements with a wind measurement mast or remote sensing equipment over a relatively long period of time – usually over a year – at potential wind farm sites. The measurements must then be set in a context of long-term reference with the help of reanalysis data using measure-correlate-predict methods (MCP) in order to substantiate the reported yield figures.
Again, the ZSW has developed innovative AI-based methods for MCP processes which are superior to classic processes, especially in complex terrain with all the challenges it creates in this field of wind power for the calculation of the anticipated yield of wind farms and wind turbines. In a detailed study comparing AI-based MCP methods with previous MCP methods, the ZSW was able to show a significant reduction in the yield data errors encountered with conventional MCP methods, especially in complex terrain.
When developing the wind test site in Stötten, the ZSW fine-tuned the AI methods for the MCP algorithms in order to set not only the wind speed at hub height in a context of long-term reference, but also wind profiles over the entire height of the rotor. This is particularly important for modern systems with large rotor diameters and hub heights often exceeding 140 metres.
The current approach is to continue focusing on the turbulence intensity – if it was measured at the site – and placing it in a context of long-term reference as well with AI-based MCP methods and reanalysis data. This is a key factor when identifying site parameters and selecting suitable wind turbines for a specific site, because the stress and load on the systems are very heavily dependent on the turbulence intensity of the inflow.