The cerebral cortex plays an important role in human and other animal adaptation to unpredictable terrain changes, but little was known about the functional network among the cortical areas during this process. To address the question, we trained 6 rats with blocked vision to walk bipedally on a treadmill with a random uneven area. Whole-brain electroencephalography signals were recorded by 32-channel implanted electrodes. Afterward, we scan the signals from all rats using time windows and quantify the functional connectivity within each window using the phase-lag index. Finally, machine learning algorithms were used to verify the possibility of dynamic network analysis in detecting the locomotion state of rats. We found that the functional connectivity level was higher in the preparation phase compared to the walking phase. In addition, the cortex pays more attention to the control of hind limbs with higher requirements for muscle activity. The level of functional connectivity was lower where the terrain ahead can be predicted. Functional connectivity bursts after the rat accidentally made contact with uneven terrain, while in subsequent movement, it was significantly lower than normal walking. In addition, the classification results show that using the phase-lag index of multiple gait phases as a feature can effectively detect the locomotion states of rat during walking. These results highlight the role of the cortex in the adaptation of animals to unexpected terrain and may help advance motor control studies and the design of neuroprostheses.