Operation Mode Optimization Of Rotary Kiln Pellet Sintering Process Based On Multi-source Heterogeneous Information Fusion

Chain grate machine-rotary kiln oxide pellets roasting process is a veryimportant key link in the roasting process of the iron ore rotary kiln. It plays adecisive role to the process of products. This paper is based on the sintering process ofa domestic pellet plant, and because of the constraints of process requirements andsite environmental, the closed-loop control and real-time monitoring is often difficultto be directly and accurately performed. So this paper uses a multi-sourceheterogeneous information fusion method to controlling and research of the optimizedprocess operating mode for rotary kiln roasting process.First, according to the texture features of flame image which coal combusts inthe rotary kiln oxide pellet sintering process, this paper utilizes GLVQ neuralnetworks to recognize the pulverized coal combusting conditions in rotary kiln. Usingthe texture characteristics which GLCM extracted from flame images, such as energy,entropy and other parameters, describe the visual characteristics of flame images; Andusing KPCA reduces the dimensionality of high dimensional input vector. In this way,the goal dimension and network scale of LVQ are greatly lower; Then using thenormalized texture feature data samples trains and recognizes GLVQ. Test resultsshow that KPCA-GLVQ method has better performance than the LVQ on trainingtime and recognition accuracy, and it can meet the requirements of real-timeidentification of the rotary kiln combustion conditions.Secondly, in order to achieve the finished pellet quality indicators (chemicalcomposition, physical properties and metallurgical properties) in rotary pelletsintering process, RBF neural network soft-measurement model of pellet qualityindicators optimized by biogeographic algorithm. Combining pellet sintering processreaction mechanism with rotary kiln thermal system, this paper selects grate materialthickness, grate machine speed, kiln temperature, rotary speed and the amount ofpulverized coal as the auxiliary variable, and the finished pellet quality indicators asthe output of soft measurement model. Thereby, establish multiple multi-inputsingle-output RBF neural network soft-measurement model, and the structuralparameters of the model optimized by the biogeographic algorithm. Simulation resultsshow that the model has good generalization results and the prediction accuracy, andit can meet the online soft measurement requirements in the real-time control of kilnpellets sintering process.Finally, in order to achieve the goal to optimize the key process indicators,integrating the hybrid intelligent modeling technology, image feature recognitionalgorithms and systems integrate information fusion method establishes conditionsrecognition model based on flame image characteristics of sintering combustion of the rotary kiln and pellet quality indicators soft measuring predictive model. Then thesemodels are combined with the rotary kiln process data to achieve multi-sourceheterogeneous information fusion, and advanced control strategies architecture of kilnroasting process optimized by the operating mode is established. Simulation resultsindicate the effectiveness of coordination optimizing control strategies of proposedrotary kiln roasting process.According to the design, research and simulation in this paper, the results showrotary kiln operation mode based on multi-source heterogeneous information fusioncan be better for kiln pelletizing process to monitor and control in real-time. Itimproves the accuracy of iron ore pellets process control and automation level, andalso increases efficiency for enterprises.