基于时间序列LSTM算法对渗流条件下地埋管间距的优化

Optimization of buried pipe spacing under seepage condition based on time series LSTM algorithm

利用长短期记忆(LSTM)神经网络模型预测考虑地下渗流条件下的合理布井间距。建立数值对流传热模型,基于变参数模拟构建训练数据库,搭建LSTM神经网络模型,包括两层LSTM层,第一层128个单元,第二层64个单元,以4个样本为一个训练批次,每次迭代15次,预测得到热影响半径的稳定值为12 m。预测损失(以均方差损失MSE为主)均在10-4数量级以下,满足工程精度要求,可应用在地源热泵工程设计与预测中。

The long short-term memory(LSTM)neural network model has been used to predict the reasonable well spacing under the condition of underground seepage.A numerical convection heat transfer model is first established,while the training database is built based on variable parameter simulation.The LSTM neural network model is then constructed,including two layers of LSTM.The first layer has 128 units,and the second layer has 64 units.Four samples are taken as a training batch,and each iteration is 15 times.The predicted stable value of the thermal impact radius is 12 m.The predicted losses(mainly mean square error loss MSE)are all below the 10-4 order of magnitude,meeting the requirements of engineering accuracy,thus can be used in the design and prediction of ground source heat pump projects.