| The submerged arc furnace (SAF) is used to convert the high-voltage electric energy to arc heat and charge resistance heat to meet the heat needs of silicon-manganese ferroalloy smelting. The power supply system model consists of a transformer, short network, three-phase electrode, electric-arc, charge resistors and so on. Because of the position adjustment and the corrosion of three-phase electrode, as well as the load coupling, the power supply system model has the characteristics of nonlinearity, timevayring and uncertainty, which make it difficult to establish its precise mathematical model. Moreover, the silicon-manganese smelting process is a process with physical and chemical reactions of hign temperature and multi-composition, which has such characteristics as complex mechanism, multivariable and large delay. The process indexes are extremely difficult to be measured and the operating parameters can't be optimized, which leads to fluctuant product quality, poor productivity and high power consumption.In this paper, a calculation model for the power supply system of the SAF was developed and the properties of three-phased asymmetric loads and the reasons of unbalanced three phase electrode effective power for different electrode control strategies were analyzed. An on-line calculation method of three phase electrode effective power was proposed by designing an electrical zero in the primary circuit and a neutral point in the bottom of the bath. Then, on the basis of the process mechanism analysis of the SAF, the process indexes prediction models based on least squares support vector machine were proposed. The proposed methods were validated by using the production data of a12500KVA submerged arc silicomanganese furnace. Main research work and innovation achievements are as follows:(1) Aiming at the unbalanced effective power of three-phase electrode caused by three-phased asymmetric loads of the SAF under constant current control strategy, an effective power balanced control strategy based on electrode current ratio was proposed. The equivalent circuit equations for the the power supply system were first established. Then, the equivalent electrical parameters were calculated by solving the nonlinear circuit equations using Newton method. At last, a three phase electrode effective power balanced control strategy based on electrode current ratio was obtained. The tests results show that the unbalanced degree of three-phase electrode effective power was restrained within±3%by using the proposed method.(2) Considering the difficulties in the measurement of three-phase electrode effective power and the problems in balance of three phase electrode effective power, an on-line calculation method of three phase electrode effective power was proposed. An electrical zero in the primary circuit and a neutral point in the bath bottom were designed, where the phase voltage, phase current, phase active power in the primary side of transformer as well as the electrode voltage in the secondary side can be measured. Then a mechanism model based on Energy Conservation Method of three-phase electrode effective power was proposed and an improved BP neural network was presented to compensate the estimated error. The simulation results show that the proposed method has good performance and its relative error is only±5.71%.(3) Aiming at the problem that the silicon-manganese alloy composition is difficult to be measured, an on-line prediction model for silicon-manganese alloy composition based on adaptive recursive least squares support vector machine was proposed. Three recursive including methods increased memory algorithm, fixed memory algorithm and decreased memory algorithm, are employed to train the prediction model. To improve the solution speed and prediction precision of the proposed model, an error minimizing function was set to adjust the structure and parameters of the prediction model adaptively. The simulation results show its effectiveness.(4) Aimed at the large-lag of manual examination in the submerged arc smelting process, an on-line prediction method for silicon-manganese slag composition based on simplified least squares support vector machine was proposed. Firstly, a slag composition mechanism model and a slag basity mechanism model based on mass balance were constructed from special chemical reactions under ideal production state, which reflects the change tendency of slag composition. However, for the simplification of the process reactions and condition hypothesis in the process modeling, the accuracy of the mechanism model could not completely meet the technical requirement of industrial production process. Then, a simplified least square support vector machines (SLS-SVM) based slag composition prediction model and a least square support vector machines with kernel principal component analysis (KPCA-LSSVM) based slag basity prediction model are implented. In the SLS-SVM prediction model, the sample data are mapped to the high dimensional feature space using Schmitt orthogonalized to obtain the kernel matrix of the space. Then, the regression parameters were identified by Direct Kernel PLS, and the kernel function parameters in the SLS-SVM were optimized by adaptive differential evolution algorithm. In the KPCA-LSSVM prediction model, the kernel principal component analysis is used to denoise the input data and capture the high-ordered nonlinear principal components among the input data space to improve the model's performace. The experiment results show that the accuracy of the proposed method can meet the requirement of slag composition determination in the industrial process. |
