Abstract: With the increasing complexity of the power grid structure, the accuracy of automatic generation control (AGC) instructions executed by units is gradually increasing the impact on the online control of the power grid. In view of the fact that the effect of AGC command tracking in some stations is not accurate, this paper introduced "deep learning" technology to accurately perceive and evaluate the effect of AGC control command execution. Firstly, an accurate identification framework of AGC command execution effect based on deep learning was proposed, which used deep independent recurrent neural network (DIndRNN) to accurately perceive the effect of unit execution command. Secondly, a preprocessing strategy to speed up model training was proposed, which realized model input identification based on historical data of unit operation by analyzing the correlation characteristics between different input attributes. Finally, an uncertainty evaluation method for the effect of instruction execution was proposed, and the DIndRNN-RVM deep fusion technology was used to give the credibility of the prediction results within the given output deviation range, which enhanced the usability of the prediction results from the perspective of probability. The simulation results showed that the proposed identification framework and model optimization method could accurately perceive the accuracy of unit execution instructions and the reliability of execution results. The method proposed in this paper improves the power grid's accurate perception and prediction of AGC command execution effect, and can provide support for AGC online decision-making.