基于神经网络的闸门水封止水性能预测模型研究

Research on a prediction model for gate water seal performance based on neural networks

针对闸门水封止水性能研究计算量大、效率低的问题,基于神经网络算法建立一种闸门水封止水性能预测模型。首先,对P型水封止水装置进行有限元仿真,得到了106个工况下P型水封的接触宽度、接触应力及封头偏移量;其次,基于BP神经网络算法建立输入特征为预压量及库水压力,输出特征为接触应力、接触宽度及封头偏移量的预测模型;最后,以20个工况下水封仿真算例为样本对预测模型进行训练并采用训练后的模型对其余工况下水封的仿真结果进行预测。模型预测结果较好,对水封接触宽度、接触应力、封头偏移量预测平均误差分别为0.48%,0.63%,0.71%。结果表明,基于神经网络的闸门水封止水性能预测模型仅需少量训练样本即可对水封止水性能取得较好的预测效果,能够显著减少闸门水封性能研究的工作量。

To address the issues of large computational workload and low efficiency in the study of gate water seal performance,a prediction model for gate water seal performance based on a neural network algorithm was established.First,a finite element simulation of the P-type water seal device was conducted,resulting in the contact width,contact stress,and seal head displacement under 106 working conditions.Next,a prediction model was built based on the BP neural network algorithm,with input features being preload and reservoir water pressure,and output features being contact stress,contact width,and seal head displacement.Finally,the prediction model was trained using water seal simulation examples from 20 working conditions,and the trained model was then used to predict the simulation results for the remaining working conditions.The model's prediction results were satisfactory,with average prediction errors of 0.48%,0.63%and 0.71%for contact width,contact stress,and seal head displacement,respectively.The results indicate that the gate water seal performance prediction model based on neural networks can achieve good prediction results with only a small number of training samples,significantly reducing the workload in the study of gate water seal performance.