基于LSTM神经网络的高拱坝混凝土温升阶段温度预测

Prediction of Concrete Temperature in Temperature Rising Stage for High Arch Dam Based on LSTM Neural Network

为实时掌控高拱坝混凝土温度变化,及时制定合理控温措施,防止产生温度裂缝,深入分析了混凝土温升阶段温度影响因素,并选取初始浇筑温度、环境气温、通水温度、通水流量、绝热温升等5个主要因素作为LSTM神经网络的输入因素,建立了基于LSTM神经网络的高拱坝混凝土温升阶段温度预测模型,同时采用最大误差、平均绝对误差(M_(MAE))、对称平均百分比误差(S_(SMAPE))等评价指标检验模型精度,最后以白鹤滩高拱坝为例,对大坝混凝土温升期的温度进行预测。结果表明,所建预测模型的最大绝对误差为0.58℃,M_(MAE)、S_(SMAPE)分别为0.30℃、1.35%,预测精度较高,可操作性强,能为高拱坝混凝土温度控制提供决策支撑。

In order to control the development and change of concrete temperature of high arch dam in real time,reasonable temperature control measures should be formulated in time to prevent temperature cracks in concrete.Based on the depth analysis of the influencing factors of concrete temperature rise stage,five main factors such as initial pouring temperature,ambient air temperature,water temperature,water flow and adiabatic temperature rise were selected as the input factors of the LSTM neural network,and the temperature prediction model of high arch dam concrete temperature rise stage was established based on LSTM neural network.Meanwhile,the maximum error,mean absolute error and symmetric mean percentage error were used to test the accuracy of the model.Finally,taking Baihetan high arch dam as an example,the temperature of dam concrete in temperature rise period was predicted.The results show that the maximum absolute error of the proposed model is 0.58℃,M_(MAE) and S_(SMAPE) are 0.30℃and 1.35%,respectively,and it has high prediction accuracy,which can provide decision support for concrete temperature control of high arch dam.