Research On The Methods Of Soft Sensorfor Burning Zone Temperature Of Alumina Rotary Kiln

There is plenty of alumyte in China, but80%of alumyte reserves is of low alumina silica ratio. Rotary kiln is the core equipment in alumina sintering process and its main function is to sinter raw material slurry to produce qualified alumina clinker. In the sintering process of alumina clinker, the burning zone temperature is one of the most important elements that determine the quality of alumina clinker, the correct measurement has important significance to raise the reliability of the kiln process control, as well as to the production quality.However, due to the nearly one hundred meters of rotary kiln length, continuous rotation and high temperature calcinations in the sintering process, this speciality of structure and complexity of sintering process bring about many integrated complexity such as multi-variables strong coupling, strong nonlinearity, big inertia as well as uncertain disturbance etc, also the burning zone temperature is fail to be continuously detected accurately online by using measurement apparatus.Alumina rotary kiln process control has depended on "man-watch" operation for a long time, depending on the experience to judge the kiln situation and carry on the operation, over-burning or under-burning usually happen,and result in low qualified rate of clinker, short lifetime of kiln lining, low running efficiency of kiln, low productivity, high energy consumption, labor intensity remains high etc.In view of the above question,the dissertation is supported by the National Hi-tech863Program named "Overall project design and key technology development of the integrated automation system of China Aluminum Corporation". This dissertation has made detailed research on the methods of soft-sensing measurement for burning zone temperature of Aluminum oxide kiln. The detailed works have been summarized as follows:1.Data pre-processing method of soft sensor for burning zone temperature is developed. Model inputs include examination variables,such as:cooler dragging electric current, kiln bow cap pressure, kiln tail temperature, air blower loose quantity and so on,as well as controlled variables stoker rotational speed.Data pre-processing consists of filter and normalization:In view of the high frequency noise gravity of the electric current and the current capacity, chosing the butterworth filter to realize the lowpass filter. In view of the teacher sample existing the phenomenon of data outliers,has used the developed absolute value filtering algorithm to eliminate the phenomenon of the big peak".2.Soft sensor method for burning zone temperature.based on PCR is proposed.Model input variables include eleven examination variables,such as:cooler dragging electric current, kiln tail temperature and so on.Using the eigenvalue Cumulative percent variance,comparison of the neighboring characteristic values and cross validation to definite principal element integer, the standard error on test samples is59.2.3. Soft sensor method for burning zone temperature based on static PLS and dynamic PLS(DPLS) are proposed:Model input variables of satic PLS are the same as PCR model. DPLS considered the dynamic relation between the process variables, increasing control variable-stoker rotational speed and the past sample values of all variables.Useing the cross validation and eigenvalue Cumulative percent variance to definite the numble of the hidden variable, the standard error on test samples is59.1, and the standard error on test samples of DPLS model is57.9. The dynamic model surpasses the static model.4. Soft sensor method for burning zone temperature based on BP neural network is proposed.Model input variables of BP are the same as PCR model.Useing the cross validation to definite the number of hidden node, the standard error on test samples is52.2.Soft sensor method for burning zone temperature based on the SVR is proposed, Model input variables are the same as PCR. Through the grid search and the cross validation techniques, choosinig penalty factor C, the loss function as well as the parameter related to the nuclear function. The nonlinear models surpass the linear models. Soft sensor method for burning zone temperature based on SVR obviously surpasses other methods, the standard error on test samples even achieves31.8.