This repo collects the training sets used in paper "A learning-based power scheduling method for thermostatically controlled load providing flexibility with guaranteed feasibility and optimality."
All these samples are generated by PandaPower based on the IEEE 123 Bus system. For more details, please refer to: http://www.pandapower.org/
We have trained four MLPs. The first three MLPs are regression MLPs and used in the proposed learning-based methods. The last MLP is a binary-classification MLP and used in the Benchmark B3.
feature_123Bus.csv:
This csv file contains the features (inputs) for MLP training. It is a matrix with 258 columns (Active load injections: column 0 to 121; reactive load injections: column 122 to 243; actual PV output: column 244 to 257).
I_vio_123Bus.csv:
Labels (outputs) of the regression MLP for predicting current safe distances.
V_vio_123Bus.csv:
Labels (outputs) of the regression MLP for predicting voltage safe distances.
loss_123Bus.csv:
Labels (outputs) of the regression MLP for predicting power loss.
label_123Bus.csv:
Labels (outputs) of the binary classification MLP for judging the feasibility of the given sample.