基于不同PSO-ELM模型的碾压黏土抗剪强度预测方法研究

碾压黏土的抗剪强度直接影响碾压土石坝的质量和使用寿命。为得出碾压黏土抗剪强度的最优预测模型,通过粒子群算法优化极限学习机模型(PSO-ELM),分别以Sine函数、radbas函数和hardlim函数3种激活函数为基础,构建PSO-ELM_sin,PSO-ELM_rad和PSO-ELM_hard3种模型,并将模型结果与ELM模型、广义回归神经网络模型(GRNN)、随机森林模型(RF)和BP神经网络模型进行了对比。结果表明:在黏聚力和内摩擦角的拟合结果中,PSO-ELM_sin模型精度最高,其拟合方程斜率分别为1.005,1.032; 在月值模拟中,PSO-ELM_sin模型与实测值的拟合度最高,相对误差仅在6.0%～9.3%; PSO-ELM_sin模型在黏聚力模拟中RMSE,RRMSE,MAE,E_ns和R²分别为0.776 kPa,1.80%,0.641 kPa,0.993和0.997,该模型在内摩擦角模拟中RMSE,RRMSE,MAE,E_ns和R²分别为1.635°,6.98%,1.616°,0.983和0.998,模型精度均排名第一。因此,PSO-ELM_sin模型在所有模型中精度最高,可作为碾压黏土抗剪强度预测的标准模型使用。

Research on Prediction Methods of Shear Strength of Rolled Clay Based on Different PSO-ELM Models

The shear strength of rolled clay directly affects the quality and service life of roller compacted earth-rock dams. In order to obtain the optimal prediction model for the shear strength of rolled clay, we used particle swarm optimization to optimize the extreme learning machine model(PSO-ELM), which is based on the three activation functions of Sine function, radbas function and hardlim function to construct PSO-ELM_sin, PSO-ELM_rad and PSO-ELM_hard. We compared the accuracy with ELM model, generalized regression neural network model(GRNN), random forest model(RF)and BP neural network model. The results show that: in the fitting results of cohesion and internal friction angle, the PSO-ELM_sin model has the highest accuracy, and the slopes of the fitting equations are 1.005 and 1.032, respectively, which are closer to the standard value ‘1'; in the monthly simulation, the PSO-ELM_sin model has the relative error of between 6.0% and 9.3%; RMSE, RRMSE and MAE of the PSO-ELM_sin model in the conhesion simulation of the PSO-ELM_sin model, the RMSE, RRMSE, MAE, E_ns and R² are 0.776 kPa, 1.80%, 0.641 kPa, 0.993 and 0.997, respectively. In the internal friction angle simulation, the RMSE, RRMSE, MAE, E_ns and R² are 1.635°, 6.98%, 1.616°, 0.983 and 0.998, respectively. The accuracy of PSO-ELMsin model ranks first. respectively. Therefore, the PSO-ELM_sin model has the highest accuracy among all models and can be used as a precise model for predicting the shear strength of rolled clay.