Abstract: control of industrial big data multi-source domain，this paper proposes a multi-source domain fine-grained security access control optimization method, which is divided into three parts: dynamic policy optimization，lightweight dynamic attribute encryption and cross-domain attribute mapping. The optimal multi-source domain access decision is obtained by modeling the multi-source domain access decision through markov decision process，improving deep q-network，and optimizing the MDP model with the improved DQN；in the ciphertext-policy attribute-based encryption，the optimal multi-source domain access decision is obtained by optimizing the MDP model. Policy Attribute-Based Encryption algorithm to introduce morton filters-based policy hiding，and use the improved MF-CP-ABE encryption algorithm to improve the security and real-time performance of fine-grained access to multi-source domains； and add cross-domain attribute consistency checking，to improve the security and efficiency of the method in this paper. security and efficiency of this paperʹs method. Experiments show that：the delay of multi-source domain fine-grained secure access control for industrial big data in this paper is reduced to 12ms，and the access success rate is increased to 95%，which makes this paperʹs method effective and reliable.