高斯过程回归(Gaussian process regression,GPR)是一种基于高斯过程的非参数化贝叶斯回归方法,其可以灵活适应不同类型数据,用于建模和预测数据之间的复杂关系,具有拟合能力强、泛化能力好等特点.针对海量用户场景下用户量实时预测问题,提出一种基于GPR的用户量预测优化方法.在滑动窗口方法处理数据的基础上,选择合适的核函数,基于k折交叉验证得到最佳超参数组合以实现GPR模型训练,完成在线用户量的实时预测并进行性能评估.实验结果表明,相比于采用训练集中输出数据方差的50％作为信号噪声估计量的传统方案,所提方法具有较高的预测准确度,并且在测试集均方根误差(root mean square,RMS)、平均绝对误差(mean absolute error,MAE)、平均偏差(mean bias error,MBE)和决定系数R2这4个评估指标方面均有提升,其中MBE至少提升了 43.3％.
Optimization method of user quantity prediction based on GPR model
Gaussian process regression(GPR)is a non-parametric Bayesian regression method based on Gaussian processes.It is flexible in adapting to different types of data,and it is used to model and predict complex relationships between different types of data.It has strong fitting capabilities and good generalization abilities.A user quantity prediction optimization method based on GPR is proposed to tackle the problem of real-time user quantity prediction in the context of massive user scenario.Building upon the sliding window method for data processing,the method selects a suitable kernel function and uses k-fold cross-validation to determine the optimal hyperparameter combination for training the GPR model,which enables the real-time prediction of online user quantity.Finally,the performance of the model is evaluated.The experimental results demonstrate that compared with the traditional approach that uses half of the variance of the output data in the training set as the signal noise estimator,the proposed method has higher prediction accuracy and improvements in the four following evaluation metrics of root mean square(RMS),mean absolute error(MAE),mean bias error(MBE)and determination coefficient R2 on the test set.Specifically,the MBE shows an improvement of at least 43.3％.
Gaussian process regression(GPR)user quantity predictionsliding windowcross-validationhyperparameter optimization
NETL
NSTL
万方数据
