Rolling bearings are widely employed in a variety of industries. These include petrochemical, energy, manufacturing, and transportation. On account of the harsh working environment, it is easy to produce varying degrees of damage. However, the traditional signal processing algorithms are not easy to recognize the early failure of rolling bearings, because the defect impulse pattern is often obscured by background noise and other interference. To settle the question of accurately extracting and separating the characteristics of rolling bearings early fault signals, the paper proposes the compound faults feature extraction algorithm according to parameter optimized maximum second-order cyclostationarity blind deconvolution (CYCBD). For different types of failures, the reciprocal of envelope spectrum fault characteristic energy ratio is taken for fitness function. The Horned Lizard Optimization Algorithm (HLOA) is applied to automatically obtain the optimum filter length, cyclic frequency, and sample index. The optimized parameters set CYCBD to deal with compound faults signal. The each individual fault feature associated with the individual failed part can be extracted. The simulated and measured data results demonstrate that this means can effectively and accurately separate the fundamental frequency and harmonic multiples of the inner and outer race faults of rolling bearings, even under conditions of strong background noise.