A Hybrid Intelligent Soft-Sensor Method for the Rare Earth Cascade Extraction Process

The index of element component content is very important in the rare earth cascade extraction process, because it represents the product quality. But it can not be measured online, so that the optimal operation is hardly to be achieved. To deal with the problem, this paper proposes a hybrid intelligent soft-sensor method by combining a bilinear dynamic model with a neural-network-based error compensation model to predict the element component content on-line. Parameters of the default bilinear model and weights of the neural network are first initialized off-line by the least square identification algorithm and the back-propagation algorithm respectively, and then self-tuned on-line when used. Industrial experiments are conducted on a Ce/Pr extraction separation production line of La, Ce, Pr, Nd tetra-component and the results show the effectiveness of the proposed hybrid intelligent soft-sensor method by comparing with data from an industrial locale