3.3.2. One Anomaly with Fault

In this group of experiments, the velocity field is divided into three layers, with only one abnormal point. We made predictions for a single shot point and multiple shot points. The effect of predicting the velocity field and abnormal points is shown in Figure 11. There is no obvious difference in the prediction of both velocity field and anomaly. In order to better view the prediction effect of the velocity field, we have produced vertical velocity profiles, as shown in Figure 12. The prediction error of a single shot point at an anomaly is smaller than that of a multi-shot-point prediction.

**Figure 11.** Prediction for velocity field and anomaly with one anomaly and fault.

**Figure 12.** Vertical velocity profiles of velocity field with one anomaly and fault.

3.3.3. Multiple Anomaly without Fault

In this group of experiments, the velocity field is divided into two layers, with two anomalies. We made predictions for a single shot point and multiple shot points. The effect of predicting the velocity field and anomalies is shown in Figure 13. There is no obvious difference in the prediction of the velocity field, but the method of single-shot prediction is more accurate for the prediction of anomalies. In order to better view the prediction effect of the velocity field, we have produced vertical velocity profiles, as shown in Figure 14. The prediction error of a single shot point at an anomaly is smaller than that of a multi-shot-point prediction.

In summary, the neural network algorithm based on U-Net can accurately complete the process of predicting the velocity field and anomalies from the seismic waveform. Both single shot point and multiple shot points have good prediction effects, but the effect of multiple shot points is not necessarily better than single shot point.

**Figure 13.** Prediction for velocity field and anomaly with 2 anomalies and no fault.

**Figure 14.** Vertical velocity profiles of velocity field with 2 anomalies and no fault.
