Research Of Mill Load Parameters Soft Sensor Based On Transfer Learning

Ball mills is a key device in the process industry,particularly in the chemical and metallurgical industries.However,due to the time-varying parameters,the influence of large inertia and physical conditions on the operation of the ball mill,there is no effective measurement method for critical load parameters(charge volume ratio,material to ball volume ratio,pulp density).Therefore,accurate detection of mill load parameters has important research significance for the optimization operation and control of the whole process of mineral processing.Soft measurement is based on automatic control technology and is reasonably applied to the production practice process.Based on the mathematical relationship between the easy-to-measure variable and the process variable to be measured that is difficult to measure directly,various mathematical calculations and estimation methods are used to measure Measured variable.In view of this,soft-measurement technology is widely used in mill load detection,but in the actual grinding production process,it is often unavoidable that there are multiple working conditions,so that the data distribution between historical conditions and the conditions to be tested Migration occurs,leading to a decline in predictive performance of traditional soft-measurement modeling methods.The problem that it is difficult to establish a predictive dominant variable model caused by variable working conditions in the beneficiation process,the paper introduces an effective learning strategy-migration learning,and studies the soft-sensor modeling of mill load parameters under multi-case and variable working conditions.Method,the key work of the paper is summarized as follows:(1)For the problem that the test condition only contains a small number of labeled samples,this paper studies the method based on semi-supervised domain adaptive distribution adaptation.By making full use of the label information of a small number of labeled samples in the test condition,compared with the traditional soft measurement method of load parameters,good prediction results are obtained,and the effectiveness of the method is verified by ball mill experiments.(2)For the problems that the label samples in the mill to be tested are difficult to obtain and the information correlation and complementarity between multiple historical conditions are not fully considered,this paper studies the multi-source adaptation method based on the maximum mean difference.By using integrated modeling ideas and making full use of the correlation and complementarity of information between historical conditions,the prediction accuracy of the model is further improved.(3)In order to adapt the historical conditions and the distribution of the conditions to be tested,the geometry of the data is not fully considered.The paper studies the domain adaptation algorithm based on data structure preservation,mainly through popular domain adaptation and joint low rank sparseness.The representation of the domain is described in terms of these two aspects,and the effectiveness of the proposed method is verified by ball mill experiments.