摘要：

为解决XGBoost模型参数调整复杂和预测水质数据准确率低的问题，设计了一种基于灰狼算法（GWO）优化XGBoost的预测算法。首先利用灰狼优化算法全局寻参收敛能力强的优点，使用GWO优化XGBoost的超参数（最大生成树数目、学习率以及树的最大深度）进行优化，获取XGBoost的最佳预测表现。其次通过对水质关键数据的预处理提高数据的可靠性，使用多种算法进行对比分析实验。结果表明：与LSTM、未经GWO优化参数的XGBoost相比，采用灰狼算法优化后的XGBoost模型具有较好的非线性预测能力，模型的决定因数R ²达到0.85以上。

关键词: 工业污水, XGBoost, 灰狼优化算法, 参数调整

Abstract:

In order to solve the complex adjustment of XGBoost model parameters and further improve the accuracy of water quality data prediction，a predictin algorithm based on Gray Wolf Optimization（GWO） to optimize XGBoost. First，taking advantage of the strong convergence ability of the grey wolf optimization algorithm for global parameter search，we use GWO to optimize the hyperparameters of XGBoost（the maximum number of spanning trees，the learning rate，and the maximum depth of the tree） to obtain the best prediction performance of XGBoost. Secondly，the data reliability was improved by pre-processing the critical water quality data，and various algorithms are used for comparative analysis experiments. The results showed that compared with LSTM and XGBoost without GWO optimization parameters，the optimized XGBoost model had better nonlinear prediction ability，and the determination factor R ² of the model reached above 0.85.

Key words: industrial wastewater, XGBoost, the wolf optimization algorithm, parameter adjustment

中图分类号: 

• X84

• TD672