With the improvement of the digitalization and automation level of production line

a large number of data reflecting the operation process mechanism and operation state are produced

which provides the possibility for the analysis and prediction of production line performance. In this paper

a cycle–deep belife network (C–DBN) model is proposed for production line prediction based on the performance indexes of work in process

cycle time and throughput. Aiming at the problems of slow convergence and low accuracy of the prediction model in the traditional training method based on SGD (Stochastic gradient descent)

a production line performance prediction model training method based on AMM (Adam with momentum) algorithm was proposed. The training flow of production line performance prediction model for time series data is determined. Finally

the effectiveness of the production line performance prediction model and training method is verified by an example.