Improving the accuracy of short-term offshore wind power prediction is extremely important for the integration of wind power generation into the electricity grid. Therefore, a short-term forecasting method of offshore wind power is proposed combining Gaussian mixture model (GMM) with conditional generative adversarial network (CGAN) in this paper. The datasets of an offshore wind farm with 15 wind turbines collected by a supervisory control and data acquisition (SCADA) system are first analyzed and preprocessed. A GMM clustering method based on data distribution is then used to cluster the wind turbines and determine the best grouping scheme, and the forecasting model CGAN designed with the convolutional neural network (CNN) and gated recurrent unit (GRU) is constructed for each group separately. Finally, taking the processed data as a practical example, forecasting results indicate that the proposed GMM-CGAN-based method can effectively predict the day-ahead hourly actual power output of the offshore wind farm, and provide higher accuracy compared with the other forecasting methods.