Nowadays, the industrial control system has become open and interconnected, and informatization also increases the risk of network attacks and damage due to frequent intrusion. Research on industrial intrusion detection is ongoing, but many current methods need to consider the characteristics of industrial control flow. Therefore, this paper proposes an industrial network intrusion detection algorithm based on IGWO-GRU: starting from the timing of industrial control network traffic, select the simple architecture of the gated recurrent unit (GRU) as the network model; in view of the problem of the number of network parameters such as neurons and the learning rate, the Grey Wolf Optimizer (GWO) is integrated with conducting autonomous learning to find the optimal parameters of the model and solve the problem of slow convergence rate caused by a large amount of data volume of the industrial control network traffic. However, due to the slow convergence speed and low optimization accuracy of the GWO algorithm and data imbalance, this paper improves an improved grey Wolf optimization algorithm (IGWO) by improving the nonlinear convergence factor and weight adjustment strategy to increase the convergence rate of the algorithm further and avoid falling into the local optimal solution. With the data set of the natural gas pipeline control system, the intrusion detection system is simulated for classifying abnormal flow attacks. The experimental results show that the IGWO-GRU algorithm has obvious advantages in accuracy, false alarm rate, and false report rate, which improves the safety protection ability of industrial control systems.