基于LSTM与XGBoost融合的养殖水质pH值预测方法研究

郭方一, 刘明剑, 王刚, 张思佳, 单渤林, 刘通

大连海洋大学学报 ›› 2024, Vol. 39 ›› Issue (6) : 1021-1031.

PDF(10010 KB)
PDF(10010 KB)
大连海洋大学学报 ›› 2024, Vol. 39 ›› Issue (6) : 1021-1031. DOI: 10.16535/j.cnki.dlhyxb.2024-093

基于LSTM与XGBoost融合的养殖水质pH值预测方法研究

  • 郭方一,刘明剑*,王刚,张思佳,单渤林,刘通
作者信息 +

Prediction method of pH value in aquaculture water quality based on the integration of LSTM and XGBoost

  • GUO Fangyi,LIU Mingjian*,WANG Gang,ZHANG Sijia,SHAN Bolin,LIU Tong
Author information +
文章历史 +

摘要

为了确保水产养殖生态系统平衡及水生动物的健康,提出了一种融合长短期记忆网络(LSTM)和XGBoost算法的养殖水质pH值预测方法(PCA-ES-LSTM-BSO-XGBoost,PELBX)。首先,通过主成分分析(PCA)对水质数据进行降维处理,以简化参数复杂性并提高模型训练的效率与精度;其次,利用LSTM网络捕获水质参数随时间的动态变化,并采用早停法避免过拟合,确保模型对未见数据具有较高的预测准确度;此外,通过BSO算法并行优化XGBoost模型的参数,提高pH值预测的精确度;最后,将LSTM与XGBoost模型的预测结果进行加权集成,有效结合了时间序列分析与非线性学习的优势,显著提高了预测准确度。结果表明,PELBX模型在pH值预测方面表现优越,具体表现为0.115的均方根误差、0.088的平均绝对误差、1.066%的平均绝对百分比误差,以及0.747的决定系数;相较于消融试验中表现最佳的PCA-LSTM-BSO-XGBoost模型,性能分别提升了8.73%、8.33%、8.26%和7.64%;与同领域中表现最好的BiLSTM-GRU预测模型相比,性能分别提升了10.16%、1.12%、0.56%和8.73%。研究表明,本研究中提出的PELBX模型在提升水质pH值预测的准确性和稳定性方面表现出明显的优势,验证了该方法的有效性和可行性。

Abstract

To ensure equilibrium of aquaculture ecosystem and health of aquatic animals, a pH prediction method for aquaculture water quality, designated as PCA-ES-LSTM-BSO-XGBoost (PELBX), was established. In the PELBX, principal component analysis (PCA) was firstly applied to reduce the dimensionality of water quality data, simplifying parameter complexity and enhancing the efficiency and accuracy of model training. Subsequently, the Long Short-Term Memory (LSTM) network was utilized to capture the dynamic changes in water quality parameters over time, employing early stopping to prevent overfitting and to ensure high prediction accuracy for unseen data. Moreover, the parameters of the XGBoost model in parallel were optimized by the BSO algorithm to improve the precision of pH predictions. Finally, the predictions from the LSTM and XGBoost models were weighted and combined, effectively integrating the advantages of time series analysis and nonlinear learning, significantly enhancing prediction accuracy. Experimental results showed that the PELBX model outperformed in pH prediction with a root mean square error of 0.115, mean absolute error of 0.088, mean absolute percentage error of 1.066%, and a coefficient of determination of 0.747. Compared to the best-performing PCA-LSTM-BSO-XGBoost model in ablation studies, the performance parameters above were improved by 8.73%, 8.33%, 8.26%, and 7.64% respectively; and relative to the best model in the field, BiLSTM-GRU, performances were improved by 10.16%, 1.12%, 0.56%, and 8.73% respectively. The finding demonstrates that the PELBX model significantly enhances the accuracy and stability of water pH value prediction, validating the effectiveness and feasibility of the proposed method.

关键词

LSTM / XGBoost / PCA / PELBX模型 / 水质pH值预测

Key words

LSTM / XGBoost / PCA / PELBX model / water quality pH prediction

引用本文

导出引用
郭方一, 刘明剑, 王刚, 张思佳, 单渤林, 刘通. 基于LSTM与XGBoost融合的养殖水质pH值预测方法研究[J]. 大连海洋大学学报, 2024, 39(6): 1021-1031 https://doi.org/10.16535/j.cnki.dlhyxb.2024-093
GUO Fangyi, LIU Mingjian, WANG Gang, ZHANG Sijia, SHAN Bolin, LIU Tong. Prediction method of pH value in aquaculture water quality based on the integration of LSTM and XGBoost[J]. Journal of Dalian Fisheries University, 2024, 39(6): 1021-1031 https://doi.org/10.16535/j.cnki.dlhyxb.2024-093
中图分类号: TP 183|X 524   

基金

国家自然科学基金(61802046);辽宁省属本科高校基本科研业务费专项资金资助项目(2024JBQNZ007);辽宁省教育厅基本科研项目(LJ21241058018)

PDF(10010 KB)

280

Accesses

0

Citation

Detail

段落导航
相关文章

/