This paper proposes a multi-level coordination of spatial load forecasting method based on prediction reliability. Firstly, this approach determines the cellular evaluation index of historical load data and calculates the prediction reliability for each cell using gray correlation theory. Secondly, it establishes a multi-level coordination model of spatial power load and applies cellular load prediction reliability to the multi-level coordination model. At last, use this model to adjust the forecasting results of cells which have different prediction reliability. The multi-level coordination model fully considers the association of the power load between different level. To a certain extent, it can eliminate the imbalance of the predicted results between different power level. It also improves the accuracy of spatial load forecasting results and lays a solid foundation for further network planning. This paper choses exponential smoothing forecasting method to predict and uses the multi-level coordination to adjust the forecasting results. In conclusion, numerical example is used to validate the effectiveness of this model.