Research On CNN Intelligent Recommendation Model And Its Application In Pellet Firing System

In the process of image processing,the Convolutional Neural Network(CNN)is prone to over fitting due to the way of pooling,which shows that it falls into the local optimal solution.Using the Beetle Antennae Search(BAS)to improve the CNN,changing the pool way in the pool layer and applying it to the best recommendation of pellet roasting system.Taking the final output characteristic diagram of the model as the research result,using these characteristics to connect the metallurgical properties and roasting system of pellet.Firstly,the framework and theoretical knowledge of CNN are introduced.The internal mechanism of the network is expounded from three aspects,such as network structure,network depth and convolution kernel selection.And the BAS is implanted into the pool layer.The BAS uses unconstrained penalty function as fitness function.According to the characteristics of the BAS,the intelligent selection of the pooling mode of the pooling layer is made,and the CNN is improved.Secondly,1,000 groups of sample data are used to train and test the original network and the improved network at the same time.The ratio of training samples and testing samples is 7:3.In the test phase,the test data is repeated five times,and the model test results illustrate the recognition accuracy and convergence speed.The average value of the test value of the five times model represents the final recognition accuracy value: the average recognition accuracy of the original model is 97.27%;the average accuracy of the improved model is 98.874%.In terms of data,the accuracy of the improved model is nearly two percentage points higher than that of the original model.However,the former avoids the over fitting phenomenon,and the convergence speed shows rapid convergence compared with the latter in the sample size of about 150.Finally,4,500 groups of ore phase samples are used for simulation.In accordance with the influence degree of roasting temperature and other indicators on the metallurgical properties of pellets,the parameter range is reduced in the early stage.At last,432 groups of high-quality samples are selected,and each parameter of these samples is in the optimal range.Then the optimal samples are substituted into the improved CNN model,and the results show that only group 253,293 and 309 show the best metallurgical properties.The three roasting systems are as follows: the roasting temperature of group I is 1263?,roasting time is 25 min,heating speed is 85?/min,oxidation content is 3.7%,cooling speed is 98?/min;the roasting temperature of group II is 1304?,roasting time is 23 min,heating speed is 88?/min,oxidation content is 2.8%,cooling speed is 95?/min;the roasting temperature of group III is 1,314?,roasting time is 29 min,heating speed is 85?/min,oxidation content is 2.8%,and cooling speed is99?/min.In the research process,the pool way of CNN is improved and the pellet roasting system is recommended intelligently,which provides a theoretical reference for the formulation of pellet roasting system in this field.Figure 28;Table 7;Reference 78...