基于时间序列神经网络的温度场预测分析

Characteristics of freezing temperature field under water-heat coupling and prediction by intelligent method

为合理将人工冻结法应用于地下工程建设中,确保冻结壁的稳定性。通过对人工冻结试验过程中的温度场进行预测分析,利用神经网络对样本进行学习,并与实测数据进行对比,表明该方法可以较为准确地对未知温度场进行预测。通过对西南某地区泥炭土进行冻结试验,试验结果表明:在封闭不补水条件下人工冻结试样冷端温度越低,土体的降温速率越快,温度场稳定后值越小。以实测温度场构成时间序列,基于神经网络,通过建立时间序列神经网络预测模型对泥炭土的温度变化进行预测,对比实测值和预测值,平均绝对误差为0.0668,均方根误差为0.0347,整体误差较小,该预测模型能够较为精确地预测温度场变化规律。

The artificial freezing method is reasonably applied to underground engineering construction to ensure the stability of frozen wall.By predicting and analyzing the temperature field in the process of artificial freezing test,the neural network is used to learn the samples,and compared with the measured data,it shows that this method can accurately predict the unknown temperature field.Through the frost heaving test of peat soil in a certain area of Southwest China,the test results are analyzed.The results show that the lower the temperature of the cold end of the frozen specimen under the condition of closed without water supply,the faster the cooling rate of the soil,and the smaller the value of the temperature field after stabilization.The measured temperature constitutes a time series.Based on neural network,a time series neural network prediction model is established to predict the temperature change of peat soil.Compared with the experimental value and the model prediction value,the average absolute error is 0.0668,the root mean square error is 0.0347,and the overall error is small.The prediction model can accurately predict the temperature field change rule.