The security of network control system has captured worldwide attention. For the frequent overflow caused by the disparity between buffer setting and practical requirement in the Modbus TCP protocol fuzzing test, a new data structure of improved application data unit capable to preventing the message length information from being lost is presented in this paper. Then principal component analysis (PCA) is used to reduce the dimensionality explosion risk aroused by the massive identical information which comes from the data relativity. At last, a probabilistic neural network (PNN) is deployed to estimate the vulnerabilities detecting possibility of the malformed data to be input, which makes the fuzzing test more efficient. The analysis and comparison to the experiment result denote that the input data is reduced by 8.60% via using the method presented in this paper.