Chemical production is a high-security industry, which often causes huge losses to production. Effective process monitoring and troubleshooting methods are designed to prevent accidents. Most of the supervised learning algorithms used in the current chemical industry generally need to obtain more labeled samples, so the cost is higher. Aiming at the problems existing in fault detection and fault diagnosis, the algorithm proposed by the process control software for fault diagnosis. The algorithm can perform complex function approximations. The accuracy of the original features can be greatly improved. The proposed method was verified in the chemical process. The experimental results verify the effectiveness of the proposed algorithm.
