基于改进灰狼优化算法的支持向量回归预测

Support Vector Regression Prediction Based on Improved Grey Wolf Optimization Algorithm

摘要:

为了提高支持向量回归(SVR, Support Vector Regression)进行数据驱动预测的精度，针对SVR存在的参数优化问题，通过引入Tent混沌映射进行种群初始化、改进收敛方式、并结合模拟退火算法，改进了传统的灰狼优化算法(GWO, Grey Wolf Optimization)来优化SVR超参数，并基于改进后的GWO算法提出了一种IGWO-SVR预测模型。将提出的IGWO-SVR模型应用于NASA锂电池数据集仿真SOH预测以及实际生产中的车灯电流预测实验后，实验结果表明IGWO-SVR预测模型在NASA锂电池数据集上进行预测的误差相较GWO-SVR模型降低了23％，相较粒子群算法和遗传算法优化的SVR模型均存在明显优势，误差分别降低了39%和51%；在实际工作中使用IGWO-SVR模型进行车灯电流预测也取得良好效果，与实测值之间的相对误差达到2.67%，相较GWO-SVR模型误差降低了近7个百分点，证明了模型在实际应用中的具有良好的价值。

Abstract:

In order to improve the accuracy of data-driven prediction by Support Vector Regression (SVR), the traditional Grey Wolf Optimization (GWO) algorithm is improved to optimize the SVR hyperparameters by introducing Tent chaotic mapping for population initialization, improving the convergence method, and combining with simulated annealing algorithm for the parameter optimization problem of SVR. And an IGWO-SVR prediction model is proposed based on the improved GWO algorithm. After applying the proposed IGWO-SVR model to the simulated SOH prediction of NASA lithium battery dataset and the actual production lamp current prediction experiments, the experimental results show that the prediction error of the IGWO-SVR prediction model on the NASA lithium battery dataset is reduced by 23% compared with that of the GWO-SVR model, and there is a significant advantage over both the particle swarm algorithm and the genetic algorithm optimized SVR model. In practice, the IGWO-SVR model has also achieved good results in predicting the lamp current, with a relative error of 2.67% compared to the measured value, which is nearly 7 percentage points lower than that of the GWO-SVR model, proving the value of the model in practical applications.