Transmission expansion planning: A deep learning approach

Abstract

This paper proposes a transmission expansion planning (TEP) method based on deep learning (DL) to address the increasing complexity and excessive reliance on mathematical formulas in current TEP models. First, we utilize a traditional mathematical programming model to obtain unit outputs and line construction decisions by varying loads, thereby generating the dataset required for DL training. Next, we build a convolutional neural network (CNN) based DL model, which includes convolutional layers, pooling layers and fully connected layers, and whose inputs consist of load data and unit output data, while output is line construction data. We use Bayesian optimization (BO) to select the best hyperparameters for the model. We conducted both single and multiple training experiments on the Garver’s 6-bus, IEEE 24-bus and IEEE 118-bus systems. In the single training experiments, the R² values achieved by our proposed method on these three systems were 0.99471, 0.99594 and 0.99676, respectively, with K-fold cross-validation showing stable results. In the multiple training experiments, we repeated the CNN training 50 times and obtained confidence intervals for each metric to further validate the model’s effectiveness. Additionally, we performed significance testing on the BO results, showing that among the three comparative experiments, two had P-values less than 0.001, indicating a significant difference. The remaining one has a P-value is larger than 0.05 indicating a difference but not significant.