Research And Application Of Twin Support Vector Regression Based On Improved Fireworks Optimization Algorithm In BOF Steelmaking

BOF steelmaking is one of the most important steel production modes in the world.The endpoint judgment and prediction of the carbon content and temperature is the most significant link in the steelmaking production process.Due to the instability of material composition and the complex physical or chemical reaction in the converter oxygen blowing process,it is difficult to establish an accurate prediction model.An accurate endpoint prediction model plays an important role in Optimization of steelmaking process control.Therefore,In order to improve the prediction accuracy of endpoint carbon content and temperature,TSVR algorithm is proposed to establish prediction model,in view of the important parameters selection problem that affecting prediction precision and generalization ability in TSVR,FWA and EFWA algorithms with strong search ability are used.However,there are problems existing in FWA and EFWA,accordingly,the research in this paper is described as follows:First,in order to solve the problem of the waste of resources due to the small explosion amplitude with best fitness(even close to 0)in the conventional FWA and the problem of the relatively weak local search ability owing to the novel minimal explosion amplitude check method used in EFWA,two strategies are provided to improve the performance of the EFWA.Firstly,an adaptive dynamic explosion amplitude update strategy based on the heuristic information of distance between the current best firework and other fireworks is introduced to balance the global and local search of the EFWA.And then,Lévy flight strategy with strong randomness is used to generate explosion sparks to enhance the diversity of local search.The experimental results on twelve standard benchmark functions and their shifted functions indicate that the proposed algorithm ALEFWA outperforms EFWA in terms of optimization accuracy and convergence abilities on complex high-dimension optimization.Second,In view of the important parameters selection problem that affecting prediction precision and generalization ability in TSVR,a novel prediction model based on FWA optimizing the parameters of TSVR is proposed.Firstly,the factors influencing the basic oxygen furnace endpoint prediction accuracy are selected as input variables to establish end-point prediction model of the carbon content and temperature based on TSVR.And then FWA,EFWA and ALEFWA is applied to optimize the penalty coefficient and Gauss kernel width coefficient of the TSVR prediction model.Finally,simulations are implemented by using different actual production data from basic oxygen furnace to test the performance of the proposed model.The results show that three proposed model reduce the error of endpointprediction model,and increases the endpoint hitting ratio.In particular,the model based on ALEFWA optimizing the parameters of TSVR owns the higher prediction accuracy.