The Predictive Study Of The Influences Of The Copper Slag's Grinding Parametres On The Concentrate Based On ANN

Mineral Processing is a very complicated process, there are a number of factors influencing on the results of dressing, such as ore concentration, ore particle size, concentration of milling, milling grain, and flotation condition, which have a great impact on the concentrate grade and recovery. There is a typical multi-dimensional non-linear system between these factors and the ultimate indicator of beneficiation process parameters, which is hard to describe by using traditional simple linear mathematical tools.ANN is a kind of structure simulating the human nervous system, which reveals the non-linear relationship contained in the sample data. A large number of units deal with non-linear adaptive dynamic system, which has a good self-adaptive, self-organized and very strong learning-self, fault-tolerant, lenovo and anti-jamming ability and could model on many unknown complexity factors flexiblely. By using artificial neural network pattern recognition, dressing process solves the problems in the traditional model. In view of this, based on the new artificial neural network of copper slag grinding process parameters, the forecast system of copper concentrate indicator is built by using neural network technology.According to Jiangxi Copper Guixi Smelter production process, seven parameters were chosen as the input data, including semi-autogenous grinding concentration and size, ball-milling concentrate and size, raw ore concentrate, size and grade. The neural model was built by predicting the copper concentrate grade, the grade of tailings, and the recovery rate. First of all, qualitative analysis was determined to establish network computing functions. Through the comprehensive comparison with other algorithmes in the same network model, traingdm algorithm was established to build the network model. Then the quantitative analysis was made to establish the number of hidden layer neurons in the model. By training the model and analysing the test data, the result showed that when the neuron number was 13, the network model was more stable for the traingdm algorithm. So the 7-13-3 type of network model was established and it could be implemented by Matlab toolbox.In order to create a visual and friendly platform, call of Matlab was made by using C++ Builder programming language platform and Matlab engine technology. 196 samples data were collected in this paper to verify the accuracy of the system. Taking 7 of them as sample data and comparing the test data with the verified data, average prediction accuracy rate reached 80% above. Forecasting results showed that the prediction system had some use value and it could take a guiding role in the industrial production.