Aiming at the issue of low fault identification rate in existing power load management terminals, an operating fault diagnosis method is proposed for the power load management terminals based on auto-encoder multilayer perception. Firstly, operational status indicators and its related faults in the power load management terminals are analyzed. Secondly, the fault diagnosis method’s related model is depicted including its implementation method. Furthermore, the model obtains a candidate set of features representing the terminal operating status through equidistant sliding window operations on the timing data of terminal operating technical indicators. The auto-encoder feature selection sub-model is used to extract the optimal feature subset, and the multi-layer perceptron fault diagnosis identification submodel is used to achieve terminal operating fault classification and diagnosis. Finally, the effectiveness of proposed method is verified through experiments. Experimental results illustrate that the proposed method can improve accuracy, recall, and identification efficiency of power load management terminal operating fault identification, which can achieve refined diagnosis performance for power load management terminal operation faults.