The characters of the production systems for cold rolling mills were analyzed. A cold rolling lot sizing and scheduling problem was abstracted as multi-stage, multi-item lot sizing and scheduling one with buffers.  A mathematic model is presented, by which the relationship between inventory level and the number of setup could be coordinated by adjusting the penalties of inventory costs and setup costs. 　A hybrid algorithm is proposed based on binary particle swarm optimization and local search method.  The validity and flexibility of the model and algorithm were verified by calculating the data from production practices numerically.