The operating process of complex equipment has strong non-linearity, and it is often affected by some unknown factors, bringing much non linear and non gaussion monitoring data, and the calculation time grows up like exponential form as calculated amount increases. If these data are used directly for equipment residual life prediction, it is hard to complete model parameters' estimation and realize equipment's online maintenance. Aiming at settling the above problems, especially for the blindness of kernel function parameters selection in kernel independent component analysis, the kernel function parameters are optimized by particle swarm optimization arithmetic to reduce feature dimension. Finally, the oil monitoring data of self propelled gun engine is used for dimension reduction. Testing results show the feasibility and effects of the proposed method.