Computer models with collective intelligence can coordinate entire wind farms and gas turbines. This increases their output and slows down the rate at which they age.
In wind farms, depending on wind direction, different turbines experience different loads. With the help of simulations, experts can determine the direction of air flows and the strength of turbulence. Siemens researchers have now developed software that makes use of this data to optimize a park as a whole.
Turbine Fleet Intelligence:
Tapping a Data Gold Mine
open
- Text Size
- Share
Most of us have probably been at a concert only to find ourselves in the annoying situation of standing behind a giant person who blocks our view. If wind turbines had feelings, they would surely be equally frustrated to find themselves placed in the last row of a wind farm. That’s because the front rotors situated in the undamped wind will supply more power than the ones in the rear. On top of that, the rear turbines have to cope with kilometer-long trails of turbulence produced by rotors in the “front row,” which cause fluctuations in power output. It would thus be much better if the front turbines were to forgo some power in favor of their fellows in the rear. As a result, the wind farm would supply more energy.
With this in mind, Dr. Dragan Obradovic at Siemens Corporate Technology (CT) in Munich has been working closely with engineers from Siemens Wind Power to translate this insight into software that simulates the wind and the behavior of a whole wind farm within seconds and immediately transmits control commands to wind turbines.
Measurements are made of the output, rotor speed, temperature, and other factors. Each turbine is connected via fiber optic lines to a central controller that coordinates the whole system and, for instance, gives commands to change the angle of the rotor blades. “As a result, the entire wind farm functions as if it were a single power generator with collective intelligence,” says Obradovic.
For two years Siemens Wind Power in Brande, Denmark, has been testing the software at the Lillgrund wind farm off the Swedish coast. The results are expected in the summer of 2011. “We’re sure the energy yield will increase by several percent,” says Henrik Stiesdal, Chief Technology Officer at Siemens Wind Power. “It’s as if someone who buys 20 wind turbines gets an extra one for free,” adds Professor Thomas Runkler, head of the “Intelligent Systems and Control” Global Technology Field, which developed the algorithm.
Obradovic is already working on an updated version of his computer models that takes turbine aging into account. Turbulence that arises in wind streams sets off vibrations in components such as the rotor blades and towers in the rear rows of a wind farm and makes them age faster. “Maximizing power output and minimizing aging are actually contradictory goals,” says Obradovic, whose software will make it possible to optimize both of these factors in the future. The data gained in this way partially underlies the mathematical models used to calculate the interactions between turbines. If rotors age too fast, their output is reduced, or the output of the turbines in front of them is reduced in order to weaken turbulence. The idea is to prevent the “live fast, die young” fate of many rock stars.
Neural Networks for Turbines. The operation of gas or steam turbines for power generation is even more complicated. They have to provide a constant alternating current frequency through continuous rotation. If loads are switched on and off, or if wind farms provide less power on calm days, gas turbines must ramp up their output or current frequency will fluctuate. Sensors monitor such factors as air pressure, exhaust gas temperatures, emissions, and network behavior.
Volkmar Sterzing and his colleagues, who are also on Runkler’s team, have developed neural networks that use these parameters to generate turbine emission forecasts in seconds. Their software controls the fuel supply and ensures that the turbines are always running at ideal operational levels such that they generate the least emissions. Their neural networks are constantly learning from the resulting data, thus automatically optimizing turbine operations over time.
In a few years the software is expected to be mature enough for normal operation. At the world’s largest gas turbine in Irsching, Bavaria, 1,000 neural networks monitor the system’s components with data from approximately 5,000 measuring points. “This will lead to a measurable reduction in emissions, including during the load changes caused by solar parks and wind farms,” says Sterzing. In the future such learning processes will also be used to achieve a more uniform combustion process and thereby extend the service life of turbine parts.