欢迎访问《草业学报》官方网站,今天是 分享到:

草业学报 ›› 2023, Vol. 32 ›› Issue (12): 77-89.DOI: 10.11686/cyxb2023064

• 研究论文 • 上一篇    下一篇

基于烟花算法的BP神经网络预测含根土抗剪强度

刘俊麟1(), 郝勇1(), 刘春艳2, 邵严1, 丁琅1   

  1. 1.长江大学城市建设学院,湖北 荆州 434000
    2.长江大学园艺园林学院,湖北 荆州 434000
  • 收稿日期:2023-03-02 修回日期:2023-05-04 出版日期:2023-12-20 发布日期:2023-10-18
  • 通讯作者: 郝勇
  • 作者简介:Corresponding author. E-mail: 518004@yangtzeu.edu.cn
    刘俊麟(1998-),男,山西临汾人,在读硕士。E-mail: 2021720738@yangtzeu.edu.cn
  • 基金资助:
    茶树生物学与利用国家重点实验室开放基金(SKLTOF20200122);湖北省教育厅“百校百县-高效服务新农村振兴科技支撑行动”(BXLBX0296);国家自然科学基金(32102315)

Prediction of shear strength of root-containing soil by a back-propagation neural network optimized with the fireworks algorithm

Jun-lin LIU1(), Yong HAO1(), Chun-yan LIU2, Yan SHAO1, Lang DING1   

  1. 1.College of Urban Construction,Yangtzu University,Jingzhou 434000,China
    2.College of Horticulture and Landscape Architecture,Yangtzu University,Jingzhou 434000,China
  • Received:2023-03-02 Revised:2023-05-04 Online:2023-12-20 Published:2023-10-18
  • Contact: Yong HAO

摘要:

为了解决BP神经网络在对含根土抗剪强度的预测中存在的预测精度低,计算收敛速度较慢,容易陷入局部极值等问题,本研究通过直剪试验、团聚试验、根系分形分析试验等探究了不同因素对含根土抗剪强度的影响,并对各因素进行相关性分析,从中选取了大团聚体含量(R0.25)、平均重量直径(MWD)、几何平均直径(GMD)、土壤分形维数(Dd)、根表面积、平均直径6个影响含根土抗剪强度的因素作为模型输入层节点,含根土的抗剪强度作为输出层节点。参考FangfaGorman 理论公式、Kolmogorov理论公式以及一种经验公式分别计算,并对结果进行讨论,确定了本研究中神经网络的最佳隐含层节点数量为13。建立6∶13∶1的BP神经网络模型,并引入了烟花算法(FWA)对BP神经网络进行优化。结果显示,BP神经网络、粒子群算法(PSO)优化的BP神经网络、FWA-BP神经网络的预测值与期望值的最大相对误差分别为11.12%、9.06%、7.44%,平均相对误差分别为4.60%、3.24%、1.96%,相较于BP神经网络和PSO-BP神经网络,FWA-BP神经网络预测误差值有明显降低;对比引入的统计参数,均方根误差(RMSE)、平均绝对误差(MAE)、平均绝对百分比误差(MAPE),FWA-BP神经网络对应的数值分别为0.779353、0.625993、5.6679%,均小于作为对比的普通BP神经网络和PSO-BP神经网络,显示出FWA-BP神经网络在实际应用中的优越性。

关键词: 人工神经网络, 含根土, 抗剪强度, 烟花算法

Abstract:

There are various problems with using a back propagation (BP) neural network model to predict the shear strength of root-containing soil, such as low prediction accuracy, a slow calculation convergence rate, and a tendency to generate local extreme values. In this study, we analyzed the influence of different factors on the shear strength of root-bearing soil through direct shear tests, agglomeration tests, root fractal analysis, and other tests, and carried out correlation analyses for each factor. Six factors affecting the shear strength of root-bearing soil, namely the proportion of large aggregates (R0.25), mean weight diameter (MWD), geometric mean diameter (GMD), fractal dimension of soil (Dd), root surface area, and root average diameter were selected as input layer nodes for the model, while the shear strength of root-bearing soil was selected as the output layer node. The FangfaGorman theory formula, Kolmogorov theory formula, and an empirical formula were applied in the model. The optimal number of hidden layer nodes in the neural network was determined to be 13. A BP neural network model of 6∶13∶1 was established, and the fireworks algorithm (FWA) was introduced to optimize the BP neural network. The maximum relative errors of the predicted value and the expected value determined using the BP neural network, BP neural network optimized by particle swarm optimization (PSO), and FWA-BP neural network were 11.12%, 9.06%, and 7.44%, respectively, and the average relative errors were 4.60%, 3.24%, and 1.96%, respectively. Compared with the BP neural network and the PSO-BP neural network, the FWA-BP neural network had a smaller prediction error. For the FWA-BP neural network, the root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) values were 0.779353, 0.625993, and 5.6679%, respectively, all of which were lower than their corresponding values for the other network models. Thus, the FWA-BP neural network shows greater adaptability and will be superior to the other models in practical applications.

Key words: artificial neural network, root-bearing soils, shear strength, fireworks algorithm