Abstract: With the large-scale and highly integrated development of wind power, the impact of output ramp events caused by regional weather processes on the safe operation of the power grid is becoming more and more serious, especially for the large-scale offshore wind power cluster with more changeable weather systems and more significant power aggregation effect. However, the existing research on the ramp-up events of large-scale offshore wind power clusters is less, and the time-space correlation characteristics of station output ramp-up are not fully considered, so it is difficult to improve the early warning ability of power supply shortage risk. Therefore, a power ramp events prediction method for offshore wind power clusters based on a spatiotemporal convolutional neural network is proposed. Firstly, a power ramp events description index system including power change rate, duration, and other indicators is proposed, and the fluctuation amplitude and time-lapse power change characteristics of different types of power ramp events under different wind speeds are analyzed, to realize the classification of power ramp events for the risk characterization of insufficient power supply in large-scale offshore wind power clusters. Then, a multi-objective prediction model based on a spatio-temporal convolutional neural network for slope power ramp events of offshore wind power base is established. The spatiotemporal correlation characteristics of meteorological changes in offshore wind power clusters under complex weather processes are deeply mined by graph convolution neural networks and temporal attention mechanisms. Through the parallel multi-task learning mode of multiple convolutional decoders, the coder feature extraction effect is improved by the joint feedback of multiple associated tasks, which realizes the simultaneous prediction of event types and event attributes and improves the prediction accuracy. The example analysis shows that the proposed method has higher prediction accuracy than the traditional methods.