<p>The independent metering valve-controlled hydraulic systems (IMVCHS) decouple the valve structure and independently control the inlet and outlet of cylinders, providing precise motion control, high flexibility, and reduced energy consumption compared with conventional valve-controlled hydraulic systems. Component-level fault diagnosis of IMVCHS under few-shot scenarios remains challenging due to data insufficiency and imbalance caused by the increased control degree of freedom and unmeasurable system parameters. To address this, we propose a novel model-reconstruction domain-mix few-shot learning (MR-DMFSL) strategy. At the pretraining stage, the strategy develops a reconstructor network using 1DLCNN and ResNet to reconstruct the unmeasurable internal pressures in the system from simulation data. For meta-training, a domain-mixer is constructed to extract domain invariant and variant features to reduce domain imbalance, and a few-shot graph neural network classifier is constructed to achieve component-level fault diagnosis. Experimental results validate that the MR-DMFSL strategy effectively identified 12 faults from specific system components and significantly enhanced the fault diagnosis accuracy from 55.23% to 84.10% under a 5-shot scenario. The proposed strategy can overcome the data insufficiency and domain imbalance in fault diagnosis of hydraulic systems, demonstrating a promising solution for the reliable, precise, and energy-efficient operation of fluid power machinery.</p>