Improved Ensemble Random Weights Neural Network Based Online Prediction System Of The Production Rate For Mineral Beneficiation Process

Beneficiation processing is a process industrial procedure to separate the raw ore into useful minerals and gangue and enrich the useful minerals at the same time via a chain of complex physical or chemical methods.It consists of a series of unit process which connects series or parallel.And each unit process has its own indexes to evaluate the state of production operation and the quality,which is denoted as technical indexes.While the global production indexes are to evaluate the entire mineral process,including concentrate production rate and concentrate production grade,which depend on the technical indexes to a great extent.The global production indexes are the most significant indexes of the beneficiation process,how to set and adjust the entire mineral process according to the global production indexes is the key to ensure the benefit of beneficiation processing enterprises.However,it is very difficult to measure the global production indexes online,it is usually gotten by statistical analysis method based on the laboratory,which is difficult to meet the requirements of real-time optimization.So,it is of great significance to establish the prediction model of production rate,which will be of great significance to the plant-wide optimization and control for beneficiation process.The input variables of the prediction model of production rate existed for beneficiation process are usually decided by the researchers,experiences,which does not avoid the noisy variables at all.In addition,it cannot show the correlation between the target variable and each input variable intuitively.While the intelligent modeling methods in the current literatures like BP neural networks are based on empirical risk minimization,which are easy to fall into overfitting.While ensemble neural networks enhance the generalization performance of the model.In addition,most of the existed method are offline model,which cannot deal with the environmental working conditions of the beneficiation process.Based on the problems abo've,supported by the National Natural Science Foundation project "Closed-loop optimal decision-making approach of technology index for complex industrial processes under dynamic environment",the research on the prediction model of concentrate production rate in the mineral process and the development of the experimental system have been carried out.The detailed work is list as follows:(1)The input variable of the prediction models of' the production rate for mineral beneficiation were usually depended on the researcher's experience.In order to solve this problem,we analyze the correlation between the concentrate production rate and the unit process indexes based-on the maximal information coefficient algorithm.Then,according to the maximal information coefficient values,we choose the unit indexes which have stronger correlation with the concentrate production rate as the input variables of the predicted model considering the actual beneficiation process.(2)We propose improved ensemble random weights neural network based online prediction model of the production rate for mineral beneficiation process.The proposed ensemble neural networks uses the RNCL to combine the individual neural networks,which takes the empirical error,diversity issue and complexity of the model into consideration.The algorithm learns the new data through online learning and updates the model at the same time.In addition,it also shows how to optimize the parameters of activation function of the hidden nodes.In order to validate the performance of the proposed model,six benchmark datasets are used in the comparative study,and the results indicate that the proposed algorithm has a better generalization performance than the existed methods.In addition,predicted model of concentrate production rate based-on MIC and the proposed ensemble random weights neural networks algorithm is established.By compared to prior methods in concentrate production rate Adaptive cPSO,PCA-MGA-LSSVM and Online-DNNE,the improvement by our method can be proved.(3)We develop the" The experimental system of the prediction model of the production rate for mineral beneficiation process" based the theoretical research on the prediction model.The system is based on the "Apache + PHP + MySQL" system architecture,and uses the "vue+ Webpack" progressive framework.In addition,the model algorithm is programmed by MATLAB.The system includes different invariable selection methods and different regression models.Based on the system,we conduct the experimental study for different prediction model.Furthermore,we can also study the effectiveness of the different parameters on the model performance.In the end,the effectiveness of the system is verified by experiments.