基于人工蜂群算法优化的改进高斯过程模型

高斯过程(GP)的非线性特征导致其对大样本的训练时间复杂度过高,而且其超参数的选取是否适当直接影响高斯过程回归模型的预测精度。提出采用人工蜂群(ABC)算法优化改进GP以减小时间复杂度和提高预测精度。改进GP通过选取训练样本的子样本进行模型学习,以降低训练过程的时间复杂度。ABC通过优化改进GP的超参数,提升预测精度。选取训练样本的子样本构建改进GP回归(GPR)模型,采用ABC算法搜寻改进GPR的最优超参数,并用得到的超参数构建最优的改进GPR模型,输入测试样本进行预测并输出预测精度。将该模型应用于解决海上远程精确打击(LPSS)体系作战效能评估问题中,通过MATLAB仿真实验,与常见的多种优化方法相比较,验证了该模型的有效性。

Optimized Improved Gaussian Process Model based on Artificial Bee Colony Algorithm

Gaussian Process(GP) is characterized by non-linear, which leads to too high training time complexity for a large sample. And hyper-parameters directly affect the prediction accuracy of Gaussian Process. For two issues above, improved Gaussian Process Regression (GPR) model optimized by artificial bee colony (ABC) algorithm is proposed. Improved GP constructs model by selecting a sub-sample of training samples to reduce training time. The conjugate gradient method has the disadvantages of difficult determination of iteration steps, too strong dependence of optimization effect on initial values,easily falling into local optimum in the process of searching for the optimal hyper-parameters. Artificial Bee Colony algorithm was used to optimize the hyper-parameters of the GPR. First improved GPR model is constructed by selecting a sub-sample of training samples, followed by ABC algorithm searching the optimal hyper-parameters of improved GPR, finally using the test sample to predict and output prediction accuracy.The model is applied to solve maritime long-range precision sea strike (LPSS) system-of-systems operational effectiveness evaluation issues, and MATLAB simulation experiments verify the validity of the model by comparing with other evolution algorithm.