An Artificial Intelligence approach for the Identification of the normative frequency behavior of Decentralized Generators in an islanded Network
05 HYB24-77
Presented by: Claudia Bernecker-Castro
Isolated grid operation has been proposed as a sustainable emergency solution in case of long-lasting power blackouts by using the available generation units inside a local area. During scheduled maintenance, grid operators usually employ mobile diesel generators to continue providing electricity to low-voltage networks. Thus forming islanded grids intentionally and temporarily. Usually, these island-forming units are operated in such a way that power injection from decentralized generators (DGs), like rooftop PVs, is not allowed. This ensures that only loads are being fed in since conventional generators cannot absorb power.
The research project LINDA proposed an islanding concept that allows power injection from DG alongside traditional generators in isolated grids. The general idea is based on the frequency behavior of grid-connected DGs, which in grid-connected mode dictates a power reduction for higher grid frequencies, i.e., when the generation exceeds the consumption. In islanded mode, the grid-forming unit should be able to provide power to the net load, resulting from loads and power injection from DGs. Considering a battery inverter as a grid-forming unit, the frequency should be controlled in such a way that the battery's state of charge is not jeopardized.
Within the last twenty years in Germany, several grid codes have governed the behavior of grid-connected DGs, especially differing in their frequency characteristic. Two dominant behaviors can be identified as the grid frequency increases: the plant either reduces its injected power according to a defined gradient or disconnects itself from the network after exceeding a certain threshold. The same behavior is observed in an islanded mode of operation.
The field test application of the islanding concept has, therefore, revealed an important challenge for the stable operation of isolated German networks since, for times when the frequency is increased, the amount of power reduction from DGs is unknown, and thus, the grid-forming unit is subjected to sudden net load changes. An initial estimation of the number of DGs installed with a certain frequency behavior will help estimate the configuration of the controller parameters for the islanded mode of operation.
One approach is to estimate the configured grid connection code using the plant commissioning date. However, discrepancies between the estimated and implemented grid code configured in the plant controller or individual inverters will likely be identified during field tests. This is due to plant operators' lack of updated documentation, which could worsen in the upcoming years, considering that DG installation is expected to continue growing.
This paper proposes an artificial intelligence (AI) approach to identify the frequency behavior of the DGs connected to an isolated grid. The proposed concept is built upon a neural network architecture and leverages a power versus frequency profile recorded during field tests at the point of common coupling (PCC). The algorithm’s output is a prediction of the installed grid code associated with the DGs connected to the isolated network. The algorithm’s prediction is still compared against the commissioning date estimation method, showing some specific discrepancies. Initial tests hint at a promising ability to predict the connection grid code using field measurement data.
The research project LINDA proposed an islanding concept that allows power injection from DG alongside traditional generators in isolated grids. The general idea is based on the frequency behavior of grid-connected DGs, which in grid-connected mode dictates a power reduction for higher grid frequencies, i.e., when the generation exceeds the consumption. In islanded mode, the grid-forming unit should be able to provide power to the net load, resulting from loads and power injection from DGs. Considering a battery inverter as a grid-forming unit, the frequency should be controlled in such a way that the battery's state of charge is not jeopardized.
Within the last twenty years in Germany, several grid codes have governed the behavior of grid-connected DGs, especially differing in their frequency characteristic. Two dominant behaviors can be identified as the grid frequency increases: the plant either reduces its injected power according to a defined gradient or disconnects itself from the network after exceeding a certain threshold. The same behavior is observed in an islanded mode of operation.
The field test application of the islanding concept has, therefore, revealed an important challenge for the stable operation of isolated German networks since, for times when the frequency is increased, the amount of power reduction from DGs is unknown, and thus, the grid-forming unit is subjected to sudden net load changes. An initial estimation of the number of DGs installed with a certain frequency behavior will help estimate the configuration of the controller parameters for the islanded mode of operation.
One approach is to estimate the configured grid connection code using the plant commissioning date. However, discrepancies between the estimated and implemented grid code configured in the plant controller or individual inverters will likely be identified during field tests. This is due to plant operators' lack of updated documentation, which could worsen in the upcoming years, considering that DG installation is expected to continue growing.
This paper proposes an artificial intelligence (AI) approach to identify the frequency behavior of the DGs connected to an isolated grid. The proposed concept is built upon a neural network architecture and leverages a power versus frequency profile recorded during field tests at the point of common coupling (PCC). The algorithm’s output is a prediction of the installed grid code associated with the DGs connected to the isolated network. The algorithm’s prediction is still compared against the commissioning date estimation method, showing some specific discrepancies. Initial tests hint at a promising ability to predict the connection grid code using field measurement data.