Figure 7
The graph is titled “Training Convergence of Physics and Data Loss”. The horizontal axis is labeled “Epoch”, ranging from 0 to 200 with an interval of 25, and the vertical axis is labeled “Loss Magnitude”, ranging from 0.00 to 0.08 with an interval of 0.01. Two curves are plotted and identified in a legend at the upper right: a blue line labeled “Physics Loss” and an orange line labeled “Data Loss”. The blue curve starts at (0, 0.01) and decreases rapidly, approaching near 0.001 by around 75 epochs and remaining flat thereafter. The orange curve starts higher at about (0, 0.078), decreases steadily, and gradually levels off around 0.03 by epoch 200. Note: All numerical values are approximated.Loss decomposition averaged over five cross-validation folds, demonstrating the dominance of physics-based regularization over data fitting. The relative magnitude of the loss components confirms physics-governed learning and physically consistent state evolution