Considering that the existing prediction models of satellite observation task schedulability are difficult to model the potential dependencies between observation tasks with long time interval, a novel predictive model for satellite earth observation task schedulability based on Bidirectional Gated Recursive Unit (Bi-GRU) network is proposed. The model can learn from the historical satellite observation task scheduling results and forecast the observation task scheduling result accurately without a time-consuming scheduling computation. Firstly，the model adopts a multi-layer fully connected forward neural network to extract the relationship between the features of observation tasks. Then，a multi-group and multi-layer Bi-GRU network is designed to formulate the temporal features between the current task and its precursors and successors in task sequence bi-directionally. Lastly，the outputs of Bi-GRU groups are fused in order to enhance the accuracy of the prediction result. The experimental results show that，compared with the state-of-the-art approaches，the accuracy, precision, recall and F1 score of the proposed method are improved by 2.27%, 2.36%, 3.45% and 2.37%, respectively.