Hydraulic supports serve as crucial equipment in a fully mechanized coal mining face, and the performance degradation assessment is a core topic for the prevention of roof disasters. Existing support degradation assessment is empirical, and transferable approaches from other machinery suffer from inadequate representation of fault-induced degradation behaviors and overlook multi-parameter correlations. This paper proposes a novel predictive performance degradation assessment approach using multivariate statistical fusion. An attention-enhanced TCN-BiLSTM model is developed to predict the future evolution of multiple operating parameters of a hydraulic support. Multi-domain features are then extracted and selected to characterize the support behavior under various fault-induced degradation conditions. Two statistics are constructed to quantify the deviation from the healthy state by evaluating both amplitude deviations and correlation changes in multiple features. A novel overall degradation indicator DI is then proposed, and the baselines at different degradation levels are determined by evaluating the probability density function using adaptive kernel density estimation. Multiple fault-induced degradation experiments are carried out on a support test rig at different severity levels. Results show that the proposed indicator achieves a degradation assessment accuracy above 85% for all fault types and above 95% for most fault types.