Intermediate data of the accuracy curves on possibility risk, consequence risk and final risk.

The data were calculated by the GSS-RiskAseer, we compare our GSS-RiskAsser with the following popular deep-learning models: (1) CNN-based methods. We only apply CNN architecture to process visual information and predict gas supply system risk. The models we compare are ResNet50 and ResNet101. (2) RNN-based methods. In these baselines, we only exploit score matrix by DNN, GRU or LSTM to predict risk value. (3) CNN-RNN fusion methods. We extract multi-modal features by CNN and RNN separately and concatenate them to predict risk.