The invention discloses a large-scale energy storage power station operation unit fault early warning method and system. The method comprises the following steps of: collecting discharge data of an energy storage battery as reference historical data; collecting operation state parameter data of the energy storage main system, the energy storage subsystems and energy storage units contained in theenergy storage subsystems; constructing an importance analysis model, analyzing the correlation degree of the energy storage subsystems and the energy storage total system, and determining the importance factor of each energy storage subsystem; analyzing the correlation degree of the energy storage units and the energy storage subsystems, and determining the importance factor of each energy storage unit; selecting an energy storage subsystem and an energy storage unit of which importance factors are greater than a preset threshold value, and analyzing the operation state parameter data of theenergy storage subsystem and the energy storage unit and the reference historical data by adopting a long-short term prediction neural network to obtain a prediction result of a next time unit; and when the deviation between the prediction results of the energy storage subsystem and the energy storage unit and the reference historical data is greater than a preset threshold, fault early warning issent out.