Research And Implementation On Sintering State Prediction Method Based On SVM And PSO

In the process of producing alumina in rotary kiln equipment, exact prediction of sintering conditions in the rotary kiln determines the quality of aluminum products and the safety of rotary kiln equipment. At present, the prediction method for sintering condition in rotary kiln is mainly "artificial look at the fire", however,"artificial look at the fire" frequently requires operators to observe sintering condition through the hole in the front of the kiln, prediction result of sintering condition is vulnerable to the subjective factors of the operators, sometimes it appears that something is wrong, leading to low production of alumina products, poor quality or even safety-related incidents. Therefore, the prediction of sintering condition in rotary kiln has important research significance and application value on optimal control of the preparation process of aluminum oxide.Among the pattern recognition methods, because the method based on support vector machine (SVM) basically does not involve the definition of probability measure and the law of large numbers, the ultimate decision-making function is determined by only a few support vectors, avoiding the "dimension disaster" problem, "removing" a large number of redundant samples, modeling required less intervention from priori, so the model based on SVM has good generalization ability and robustness. However, this method leads to low prediction rate of sintering condition. In response to these issues, firstly, this thesis collects sintering condition images in rotary kiln online, and then deals images with denoising algorithm, and then separates images using Otsu Method and Fast Marching Method for the concerned regions such as the material area, the flame zone, combustion zone and the full handle area, and then proforms feature extraction and reduction feature for the concerned region. Finally, this thesis predicts the sintering condition using SVM and particle swarm(PSO) algorithm.By optimizing the parameters in SVM, we can enhance the prediction rate of sintering condition. And by experimental verification, the method proposed in this thesis can improve the correct prediction rate of sintering condition greatly.