In view of the problem that the fault diagnosis of the existing tank track, only considered the balance state of the hoist, but the unbalanced load of hoist caused by the failure of suspended cylinder in ultra-deep mine is ignored. A fault pattern recognition method between sliding tank ear and tank for normal, bulges and dislocation of cage guid is proposed under the unbalanced load of hoist, whose original feature set is consisted of energy entropy, singular value, Standard deviation and waveform index of each frequency band after the wavelet packet decomposition of the lateral vibration signal of the hoist, and the irrelevant and redundant features are removed by neighborhood rough set, to obtain the sensitive feature set for pattern recognition based on support vector machine optimized by cuckoo algorithm. Compared with genetic algorithm, particle swarm optimization algorithm and firefly algorithm optimization, the support vector machine optimized by cuckoo algorithm has higher classification accuracy (91.7%) and shorter running time which is of great significance to ensure the safe operation of the lifting system.