Abstract: To improve the width control precision of hot strip during rough rolling for prediction, the PSO (particle swarm optimization) algorithm is used to train neural network with the actually measured data in PANSTEEL. Then, the fuzzy clustering analysis is used to preprocess the data with reasonable sampling for self-learning so as to solve the problem of superfluous samples and slow learning speed. The calculation results of measured data showed that this method can control the prediction precision of hot strip width during rough rolling within ±6 mm with optimized structure of neural network and learning speed greatly improved. So, it is regarded as highly potential in application.