Control Knowledge Extraction Method For Grinding Classification Process

The grinding classification process is a key sub-process in the mineral processing process.Its operational state directly affects the product quality and economic indexes of the whole concentrator.At present,many concentrators in China still rely on operators to control the process.Manual operation is subjective and hysteretic,which often results in frequent fluctuation of the production indicator.In addition,the characteristics of nonlinearity,strong coupling and time-varying of the grinding classification process make the traditional model-based optimization control technology difficult to adapt to the process.Therefore,the optimization control method of grinding classification process based on knowledge extraction and an online updating strategy is studied in this paper to improve the adaptability of production process.The main work and innovations are as follows:(1)Due to a large amount of noise existing in industrial sensor data and the large time delay characteristic of the grinding classification process,a process variable delay analysis method combining non-local mean denoising and cross-correlation function is proposed to preprocess the data.Then,aiming at the black-box characteristic of the grinding classification process,a kernel fuzzy C-means clustering algorithm based on state transition is proposed to fine classify the working conditions.The optimal working condition is determined by combining the operation characteristic curve of the ball mill and the visualization of the clustering results.These lay the foundation for the optimal control of the grinding classification process.(2)Since manual operations often do not represent optimal control strategies,the working condition and trend characteristics are combined to select the better data under manual operation and to generate an optimization database.Then,the optimization control knowledge under different working conditions is extracted based on the weighted optimized WM algorithm,in which fuzzy rules are used to represent the knowledge and define the rule confidence degree for knowledge evaluation.The nonlinear relationship models of grinding sound,sump level and cyclone feed concentration are established based on LSTM to verify the effectiveness of the optimized control rules.Finally,aiming at the working condition fluctuation caused by the changeable ore properties in the production site,an online control knowledge extraction and updating strategy based on double sliding window is proposed.The experiment results show that the online control rules have better adaptive ability than offline rules and manual control decisions.