The thickness and tension system in the cold tandem rolling process has the characteristics of multivariable,strong coupling and uncertainty.In order to reduce the coupling effect of the two and improve the response speed and anti-interference ability of the system,a PID control strategy based on the neural network inverse system decoupling principle was proposed.The change coefficient of the rolling force relative to the tensile stress was considered,a dynamic coupling model of the thickness and tension system was established,and the Interactor algorithm was used to prove the reversibility of this model.The decoupling principle of the inverse system of BP neural network was used to realize the decoupling of the thickness and tension system,the coupling effect of thickness and tension was weakened.Aiming at the problem of particle swarm optimization easily falling into local optimization,an optimization algorithm was proposed in which particle swarm optimization was combined with bacterial foraging algorithm,and the PID parameters were adjusted.The results show that compared with the quasi-diagonal recurrent neural network (QDRNN) multivariable PID decoupling method,the decoupling degree,model anti-interference ability and system response speed of the proposed method are greatly improved.