National Key R&d Program of China (2021YFF0602404), National Natural Science Foundation of China (61873053, 61621004)
Aiming at the problem that the flotation process is difficult to establish an accurate identification model due to Insufficient fault condition information, which leads to the delay in adjusting the flotation production conditions and the failure to run statically in the normal state, this paper proposes a feature transfer learning method based on cross-domain manifold regularization(CDMRFTL). In this method, the rich and complete working condition information accumulated by the existing similar complete flotation process is transferred as the source domain to the target domain of the incomplete unmodeled flotation process. First, extracting the common features between the complete working conditions and the incomplete working conditions and mapping them to the common subspace through the largest domain class density and partial manifold regularization, to constraint discriminant information and retain the original unchanged in neighborhood domain structure information, Secondly narrowing the difference of distribution between source domain and target domain by the largest average differences, and the classification and recognition model was established, Next D-S evidence theory was combined to fuse the apparent characteristics and depth characteristics of the foam in the flotation process to improve the generalization ability of incomplete flotation process condition recognition and ensure a better recognition and classification effect. Finally, the effectiveness of the proposed method was verified by simulation experiments.