Alsland – Detecting microgrids using artificial intelligence
Microgrids in the interconnected grid represent a real challenge, because the entire interconnected grid works as a single unit. However, grid faults can cause individual sections to be cut off. If there is additionally a local power balance, unintentional microgrids can develop. To prevent personal injury and plant damage, unintentional microgrids need to be locally detected and then specifically destabilized.
Unintentional microgrids in the interconnected grid can occur more frequently with an increasing number of generation plants and converters. This is where AI can make a contribution. The Alsland project investigates how converters can reliably detect such faults with the help of AI.
Creating a simulation model (Matlab/Simulink) of an exemplary low-voltage grid that is coupled with a simplified medium-voltage grid model. The low-voltage grid model is particularly characterized by a relatively high converter penetration.
Generating a dataset in Matlab/Simulink for the training and validation process of the machine learning method for microgrid detection. The dataset contains cases of dynamic microgrid development (“true cases”) as well as other dynamic grid events (“false cases”) during grid operation.
Selecting suitable machine learning methods for microgrid detection
Training and validation of the machine learning method
+49 (0) 561 7294-141
The machine learning method uses the measurement values that are recorded anyway, such as current, voltage, and frequency within the converter, to predict events of dynamic microgrid development.
To prevent faulty predictions, the microgrid detection method was not trained solely to detect cases of microgrid development, but also sensitized to detect dynamics that result from typical grid events.
To develop the machine learning microgrid detection method, LSTMs (long short-term memory) and XGBoost methods were compared with each other.