Multi RBF Model Based Component Content Prediction For Pr/Nd Extraction Process

With the increasingly expanding application fields of rare-earth element, rare-earth element has become a kind of irreplaceable key material. In 2015, China's rare earth production has reached to 105,000 tons. However, the domestic rare-earth production level still remains in a half-automation state, process parameters, such as flow, need to adjust manually, the component content of rare-earth solution relies on manual sampling detection, this kind of production condition decrease production efficiency and product quality. Therefore, establishing rare-earth component content extraction process detection online is the key to achieve rare-earth extraction automation.In the rare-earth extraction separation system which has ionic color feature, process parameters can be adjusted through the change of “ion characteristic color zone” in extraction tank, however, manual observation method is subjective, which could increase the extraction separation process control's randomness. This paper based on machine vision technology to extract the rare-earth solution's color feature, using data-driven algorithm to build the rare-earth element component content soft sensor model in the extraction process. Specific contents are as follows:1?Took Pr/Nd extraction process as an example, firstly,using machine vision technology to collect the rare-earth solution's image information in the extraction tank, secondly, choosing a more suitable color characteristic components to describe the rare-earth solution information according to analyze the Pr/Nd solution's performance in different color space.2?Builded the rare-earth element component content RON model by using data-driven algorithm. Firstly, analyzing REE solution color features through PCA algorithm, and finding out the color characteristic components that have great relationship with the REE component content. Then, establishing the RON model, and optimizing the network structure according to the recently period's prediction error.3? Proposed a Multi-RBF model algorithm for the large-scale fluctuations of REE component content. Firstly, using subtraction cluster algorithm to analyze the sample data, and establishing sub- model for each cluster of data based on RBFNN, then getting the REE compo nent content prediction model by following a certain rules to assemble the sub- models. According to the model's parameters correction strategy, archive the model parameters self-adaptive correction when the extraction environment or object properties changed. Through the Pr/Nd extraction field data tested, the result shows this method has better adaptability, is suitable for REE component content's rapidly and accurately prediction...