At present, the network data under the environment of industrial information physical system is larger and more complex. Traditionally, feature extraction by machine learning is cumbersome and computation-intensive, which is not conducive to anomaly detection of industrial network data. To solve the above problems, this paper proposes a stacked denoised auto-encoding network based on kernel principal component analysis for industrial cyber physical systems intrusion detection. Firstly, a novel kernel principal component analysis method is used to reduce the data feature dimension and obtain a new low-dimension feature data set. Then, a multi-stacked denoised auto-encoding network model is used to classify and identify the data after dimensionality reduction by voting. Experimental results show that the proposed method has better classification performance and detection efficiency by comparing the state-of-the-art intrusion detection methods.