The invention discloses a microgrid energy management method based on a NARMA-L2 model which comprises the following step: a step 10) utilizing the NARMA-L2 model to establish the relationship betweenthe output and the control input of the distributed power system. 20) generating a control signal input and a voltage output data set for the distributed power supply in the system based on the ideaof pinning control and taking the reactive power as an equal target; Step 30) training a neural network used in the expression in the step 10) to characterize a complex non-linear function with the data set obtained in step 20), and fitting the non-linear output dynamic characteristics of the distributed power supply under the condition of reactive power equalization; 40) generating a reference value by utilizing a gradient descent search algorithm to complete reactive power cooperative control of the microgrid. In this method, the neural network is used to fit the complex nonlinear function of the input and output characteristics, and the de-modeling control of the microgrid is realized. At the same time, based on the concept of pinning control, the collaborative control of micro-grid canbe realized, and the response speed and accuracy of control can be improved.