Research On Temperature Prediction Of Rotary Kiln Based On Improved BOA-RBF Neural Network

With the rapid increase of domestic iron and steel industrialization,the market supply of steel products will become more and more large.The production and quality control of iron ore pellets is particularly important.Domestic iron ore pellets mainly adopt the production technology of chain grate rotary kiln.The temperature inside the rotary kiln is one of the important factors affecting the quality of the pellets,Determining the temperature within the rotary kiln in real-time online is therefore crucial.Moreover,because the production of pellet involves a series of physical and chemical changes,the measurement environment inside the rotary kiln is complex and changeable.And faced with a series of problems such as poor contact of expensive tester,time delay in the measurement process,low accuracy and difficult to determine the service life,resulting in the inability to effectively and promptly measure the temperature in the kiln,which has a great impact on the quality control of pellet production.Therefore,it becomes very critical to build a soft sensor model to predict the temperature in the kiln.Therefore,this thesis uses data dimension reduction,machine learning and intelligent optimization algorithm to build a soft measurement prediction model of kiln temperature,so as to achieve real-time and effective online prediction of temperature in rotary kiln.To provide real-time and effective data reference for pellet production.(1)In view of the high dimension and information redundancy of nonlinear data,this thesis adopts the local linear embedding algorithm(LLE)for data preprocessing,and uses Mahalanobis distance to replace the Euclidean distance of the original algorithm for data measurement,and then uses the global distance to homogenize it,so that the data is homogenized in the data distribution area.The sample points distributed in sparse areas and dense areas should be avoided.If the same nearest neighbor number is adopted,the dimensionality reduction effect will be significantly different.Finally,the test data set is used to compare the dimensionality reduction effect.The results show that the improved local linear embedding algorithm has excellent dimensionality reduction ability.(2)The structure of RBF neural network is introduced,several traditional algorithms of RBF neural network are analyzed,and the defects and coping strategies of RBF neural network are pointed out,which provides a substantial theoretical basis for the optimization and construction of the model in the future.(3)This study optimizes the crucial structural parameters of the RBF neural network using an upgraded version of the Butterfly Optimization algorithm(BOA)called ZRBOA in order to increase prediction accuracy.Numerous methods are employed to enhance the BOA algorithm.First,chaotic mapping disturbs the algorithm population,dispersing the algorithm individuals over the entire collection space.The population is in a better position in the beginning than the ideal solution.The proposed soft sensing model of the temperature in the grate kiln is more precise and stable,according to experiments.Secondly,adaptive weights and Levy flight disturbance enhancement algorithm are used to avoid precocious convergence and local optimization.Finally,the sine and cosine optimization algorithm is introduced to replace the local search formula of BOA optimization algorithm,and the algorithm’s capacity for local growth is substantially enhanced.According to the findings of the standard test function,the upgraded ZRBOA has a quicker convergence time and a more precise search.(4)Create a soft sensor RBF neural network-based temperature prediction model for rotary kilns.Firstly,the abnormal data were removed and the improved local linear embedding algorithm(LLE)was used for data processing.The RBF neural network’s hidden layer nodes’ number was calculated using the sensitivity experiment.Second,The RBF neural network’s structural parameters were optimized using ZRBOA.and By using ten fold cross validation,the model is trained.Finally,the performance of the model was assessed using four precision indices,including training inspection accuracy and prediction accuracy.The proposed soft sensing model of temperature in the grate kiln is more precise and stable,according to experiments.