Flotation is a process with nonlinear, large lag, strong coupling and other characteristics, concentrate grade is an important economic index to test the flotation production process, and it is difficult to establish accurate mathematical model, accurate prediction of concentrate grade is needed, which is of great significance to the quality prediction of flotation process.Support vector machine regression prediction was used as basic model, and modified basic artificial fish swarm algorithm(AFSA) was used to optimize parameters about the model, and GAFSA-SVM model was established to predict flotation concentrate grade.The initial value, crowding factor and visual field of the AFSA are improved, and the elimination mechanism is added.By simulating the historical data, the improved gafsa-svm model can predict the flotation concentrate grade well, effectively improve the prediction accuracy and speed of concentrate grade, and meet the requirements of field practice.
IntroductionFlotation is the method of ore dressing in most of China's concentrators. Flotation has a wide range of application fields, fine particle size of flotation ore, high production index and a variety of flotation concentrates, and has a good development prospect. However, there are many physical and chemical changes in the process of flotation, with complex mechanism of action and many influencing factors, such as ore particle size, concentration of pulp, dosage of agents, concentration of foam and temperature. Concentrate grade is not only a parameter of the flotation process, but also an important index to test the flotation efficiency. Soft measurement method can obtain some physical quantities that cannot be measured by instruments, and predict some physical quantities that cannot be measured by establishing mathematical models. . In recent years, some progress has been made in the application of soft-sensing method to flotation prediction. Now there are many soft-sensing modeling methods based on intelligent algorithms. TongXi [1] by optimal ` ESN chaos time series RBF neural network to flotation prediction analysis of economic indicators. Zhang Xong, Zhu Jing [2], proposed the economic and technical index prediction of flotation process based on the combination of chaotic ant colony and least square method and simultaneously training neural network.Li Xiaolei proposed a new method -AFSA that about animal autonomy [3], which is an application of the idea of swarm intelligence and a bottom-up optimization method. AFSA includes random behavior, foraging behavior, clustering behavior and rear-end behavior, through which optimization is carried out. At present, the research of artificial fish swarm algorithm has penetrated into various application fields and obtained good practical application .Cui Qiang[4] et al. applied the AFSA to fault diagnosis of transformer. By improving the AFSA to train the neural network, the accuracy of transformer fault diagnosis was effectively improved. Zhu Xuhui[5] et al. optimized the SVM's parameters by improv...