SVR-based Modeling And Energy Consumption Optimization For Rotary Dryer Process

Rotary dryers are widely used in industrial processes, particularly in the non-ferrous metal production. But how to reduce the high-energy consumption of rotary dryer during production process is still a business problem. Its drying process not only involves complex heat transfer and mass transfer mechanism, but also closely relates to the properties of dry material. The rotary dryer is a control object with large inertia, large time delay, and time-varying parameters, and what's more, the main parts of the dryers are in a rotating state, the key technological parameter can not be measured accurately and timely. Therefore, it is difficult to describe the thermal state of closed dryer quantitatively. The ambiguity of parameter information makes modeling very difficult, so the traditional optimization control strategy is hard to implement.In the paper, taking a zinc concentrate ore rotary dryer as the background, a support-vector-regression(SVR)-based model for rotary dryer production process is proposed based on the deep study of the drying process mechanism. A SVR-based fuzzy model is adopted to solve the problem of determing the model parameters drying rate. In order to improve prediction accuracy, using the multi-phased-distribution feature of rotary drying process, the multi-phased and multi-SVR cascade model is proposed and drying rate is determined by utilizing multi-phased and multi-SVR fuzzy model. While for the optimization of hyper parameters of SVR fuzzy model, an improved genetic algorithm and an immune genetic algorithm are proposed respectively to avoid the lack of experience in choosing parameters. In order to improve the computation speed of the model, an improved variable-length particle swarm optimization algorithm based reduced support vector is proposed. Based on the obtained model, the chaos disturbence particle swarm constained optimazation algorithm is proposed to optimize the thermal parameters of drying process. Experimental data, simulations and applications have proved the effectiveness of this method. The main research and innovative achievements of the paper are as follows: (1) Rotary dryer production process first-principle model based on energy conservation and mass conservation law is established. The model consists of combustion chamber and rotary dryer body first-principle model. The volume and temperature of flue gas are obtained by the combustion chamber model. The heat utilization and the continuously changing situations of material and drying medium in the dryer along the axial direction are gotten by the rotary dryer body model. Thermal balance testing experiments, parameters testing experiments and drying rate testing experiments are designed and simulated, which lays a foundation for follow-up study. The simulation results show that this model can reflect the trend of the actual process.(2) The SVR-based rotary dryer production process model is proposed. First-principle model is used to describe the drying process and determine the structure of the model. Fuzzy modeling method is employed to determine the drying rate. In regard to the problem of obtaining the drying rate fuzzy rules, a method using SVR model is adopted. In order to compensate errors between experimental samples and actual production samples of drying rate, a support vector residual compensation method is proposed. This method can compensate for the errors effectively.(3) An improved genetic algorithm and an immune genetic algorithm are used respectively to solve the problem of optimization of hyper parameters of SVR fuzzy model. Features of the former include:retain the global search ability of genetic algorithm, and in order to overcome the slow convergence of genetic algorithm, integrate local linear search in the latter part of the algorithm to speed up the convergence rate. For the latter, in order to overcome the "premature" problem of genetic algorithm during process of searching for the optimal solution, genetic algorithm is improved. Dividing the search space of algorithm into small niches, and according to the concentration of individual niche, immune evolutionary algorithm is used to amend the fitness values. The results show that both methods can find the optimal model hyper parameters effectively, while the improved genetic algorithm has faster convergence speed.(4) A rotary dryer production process cascade model based on multi-drying-phased and multi-SVR is proposed. For the key problem that input sample space is very difficult to divide during modeling, an improved fuzzy C means algorithm (FCM) based on space entropy rules is proposed. Meanwhile, in order to overcome the problem that FCM algorithm is very sensitive to noise data, the possibility FCM clustering algorithm (PFCM) is introduced, the Kernel-based clustering algorithm (KPFCM) is discussed, to make the accuracy of input sample space better. The experimental results show that the cascade model can improve the prediction accuracy, but increase the complexity of model.(5) For the problems of increasing computation speed caused by too many support vectors, a new particle swarm optimization algorithm based reduced support vector method is proposed. Using variable length particle coding to predict accuracy, letting it as a performance index directly, and obtaining the reduced support vector set by particle swarm optimization algorithm. The results show that the algorithm can reduce the number of support vectors effectively on the basis of maintaining the original prediction accuracy, and thus improve the model speed.(6) Based on rotary dryer production process model, for the optimization problems with constraints in engineering, the chaos disturbence particle swarm constained optimization algorithm is proposed. Taking the range of parameters in drying process as constraint, with the highest thermal efficiency as the goal, make particles out of local minima and non-fixed point by chaotic disturbance to obtain the optimal technological parameters, and thus save energy and reduce consumption. The results and practical application have proved the effectiveness of this method.