Research On Optimal Decision-making Of Indices Of Electricity Consumption Per Ton Magnesia For Electric Smelting Furnaces

Electric smelting furnaces are the equipments to make fused-magnesia by smelting magnesite. Fused-magnesia is widely applied in the fields of melallurgy, cement and so on. The index of electricity consumption per ton magnesia (EpT for short) is the operational index of a furnace. The target value of EpT is used to guide an operational optimization control system to generate the setpoint of electrode current. Current controller adjusts the electrode current to track the setpoint that guarantees the EpT in the target value. The target value of EpT concerns the global production indices, such as total yields and energy consumption of multifurnace and ratio of high grade magnesia. Therefore, optimal decision of the target value of EpT has very important significance in maximizing the total yields, minimizing the total energy consumption and maximizing average quality (i.e., ratio of high grade magnesia).Optimal decision of EpT is to maximize the total yield and average quality of multifurnace, and minimize the energy consumption of multifurnace. The three global production indices are conflict with each other. It is difficult to model the relations between global production indices and EpT by mechanism analysis. Besides, the relations will be affected by the changing technical parameters. Meanwhile, it needs to meet the average power limits of multifurnace. The relation is dynamic between average power of a furnace and EpT, which cannot be modeled by mechanism analysis. The dynamic of two relations concerns two scales of hours and minutes. Therefore, the optimal decision problem is difficult to model and it also is a dynamic multiobjective optimization problem. It is difficult to handle the problem by the available approaches. Now, the corporations adopt a manual decision of EpT, which results in bad global production indices.Supported by National Basic Research Program of China (973 Program-2009CB320601), this dissertation is part of the project "the total control strategy and operational control approach for complex manufacturing processes" and carries out study on optimal decision methods for EpT index of each furnace to solve the problems above. The main work has been summarized as follows:1) It presents the mathematical formulation of optimal decision of EpT. In this formulation, the objectives are to maximize the total yield and average quality of multifurnace, and minimize the energy consumption of multifurnace. The decision variables consist of the EpT index of each furnace. The constraints include upper and lower limitations of the synthetical production indices, and the upper limitation of the average power of multifurnace, and the range of power consumed by each furnace. Moreover, the difficulties in solving the above problem are analyzed.2) The models of performance index (total output, average quality, and energy consumption of multifurnace) and the model of average power are built based on LSSVM techniques. In the models of performance index, it is first to figure out the function relation between the synthetical production indices of multifurnace and production indices of each furnace based on mechanism analysis. Then establish respectively the functional between the production indices of each furnace and the EpT index, measureable disturbances (ingredient of raw material, granularity parameter and so on) based on LSSVM. An adaptive mesh PSO algorithm is proposed to optimize the LSSVM parameters. The models above are verified based on the actual production data, respectively. Average power model is built based on an online LSSVM with variable parameters, which lowers the computation difficulties of intercept and vector of Lagrange multiplier. Kernel parameter and regularization coefficient of LSSVM are optimized by a controllable pseudorandom particle swarm optimization algorithm. The proposed modeling approaches are also verified with actual production data, and the experimental results show that the online LSSVM with variable parameters can shorten the online training time and the controllable pseudorandom PSO ensure the online training precision.3) It proposes a multiobjective particle swarm optimization algorithm (MOPSO) based on case-based reasoning (CBR) and fast identification of nondominated set. Case library stores Pareto optimal solutions generated by MOPSO. Once the problem to be optimized changes, CBR retrieves the case solutions that are most similar to the current problem and adds them to population of MOPSO. The efficient identification of nondominated set can improve the operation efficiency of multiobjective optimization. Three identification algorithms are proposed, i.e. fiducial-value based, fiducial-vector based and sorting-matrix based identification algorithm, which are utilized to address the two-objective optimization problem, three-objective optimization problem and many-objective optimization problem, respectively. The above algorithms are proved by mathematical reasoning and the computation complexity of each is analyzed. A series of benchmark problems are selected to test and validate the algorithms. It improves the running speed of MOPSO by adopting the proposed fast identification of nondominated set. Compared to NSGA2 and SPEA2, the proposed MOPSO has a faster convergence to Pareto optimal set.4) It adopts the proposed MOPSO above and devises an optimal decision of EpT indices method for electric smelting furnaces. In the method, case is expressed by the environmental parameters, associated solutions from expert experience and historical optimization results. According to the current environment, case solutions are retrieved and regarded as temporary decision. Initial population is obtained by digging case solutions. A MOPSO algorithm based on efficient identification of non-dominated set is applied to search the Pareto optimal set. The infeasible solutions that violate the constraint of average power limits of multifurnace are modified before assessing their target values to avoid complicated and unnecessary computation. Non-dominated solution, which is out of the range of target value, is judged by dynamic evaluation criterion to determine whether they are kept to extend Pareto fronts. One of Pareto optimal solution is chosen based on niinimization of average power of multifurnace. The optimal decision method is verified with actual production data, and the experimental results show that it is better than manual decision. Total yields and average quality of multifurnace raise 8.26% and 4.78% respectively, and total energy consumption drops 0.81%.