基于改进型鲸鱼优化算法和最小二乘支持向量机的炼钢终点预测模型研究

终点碳含量是决定钢质量的关键因素，是转炉炼钢过程中需要控制的核心变量之一.本文建立了一种基于莱维飞行的鲸鱼优化算法（Levy Whale Optimization Algorithm，LWOA）和最小二乘向量机（Least Squares Support Vector Machine，LSSVM）的钢水终点碳含量综合预测模型.通过莱维飞行代替了传统鲸鱼优化算法（Whale Optimization Algorithm，WOA）参数的随机选择，优化了鲸鱼算法中跳出局部最优的能力；借助改变鲸鱼算法的系数向量收敛方式明显提高了鲸鱼优化算法的泛化能力、预测精度和收敛速度.数据仿真结果表明，所提出的LWOA-LSSVM预测模型，不仅能够克服局部寻优获取全局最优解，而且具有快速的收敛速度和更高的预测精度，得出预测结果的均方根误差、平均绝对误差和平均绝对百分比误差与遗传算法BP神经网络、遗传算法最小二乘支持向量机和传统鲸鱼算法最小二乘支持向量机相比均有着明显提高.同时，通过调整目标命中率和训练输入样本量验证了预测模型具有更好的鲁棒性.

Research on Prediction Model of Steelmaking End Point Based on LWOA and LSSVM

The final carbon content is the key factor in determining the quality of steel,and is one of the core variables to be controlled in the process of converter steel-making.Based on the Levy whale optimization algorithm (LWOA) and least squares support vector machine (LSSVM),a comprehensive prediction model of carbon content at the end of the steel-making process is established.When the random selection of the parameters of the traditional whale optimization algorithm (WOA) is replaced with the Levy flight algorithm,the ability to jump out of the local optimum is optimized.Changing the method of coefficient vector convergence results in improvements to the generalization ability,prediction precision and convergence speed of the WOA.Data simulation results show that the proposed LWOA-LSSVM forecasting model not only overcomes the local optimization to obtain the global optimal solution,but also achieves faster convergence speed and higher prediction accuracy.Prediction results of the model,concerning root mean square error,mean absolute error,and mean absolute percentage error,show noticeable improvements when compared to those of the genetic algorithm and back propagation (BP) neural network,the genetic algorithm and LSSVM,and the traditional WOA and LSSVM.At the same time,through adjustments of the target hit ratio and the number of training sample entries,the prediction model is proven to be more robust than the aforementioned algorithms.