Prediction and optimization of isentropic efficiency of vortex pump under full operating conditions in Organic Rankine Cycle waste heat recovery system based on deep learning and intelligent algorithm
Prediction and optimization of isentropic efficiency of vortex pump under full operating conditions in Organic Rankine Cycle waste heat recovery system based on deep learning and intelligent algorithm
复制标题
DOI：
10.1016/j.seta.2020.100898
复制
发表时间：
2020-11
影响因子：
8
通讯作者：
Xu Ping;Fubin Yang;Hongguang Zhang;Wujie Zhang;Gege Song;Yuxin Yang
中科院分区：
文献类型：
--
作者：
Xu Ping;Fubin Yang;Hongguang Zhang;Wujie Zhang;Gege Song;Yuxin Yang
在有机朗肯循环（ORC）系统中，旋涡泵的等熵效率对系统的性能有重要影响。本文将旋涡泵实验数据与深度学习相结合，构建了实验数据驱动的旋涡泵全工况等熵效率预测模型。模型中的激活函数和隐层节点数采用嵌套筛选技术进行筛选。通过双线性插值算法，分析了旋涡泵运行参数对等熵效率的影响。此外，在四维空间中选择优化边界。最后，将深度学习预测模型与线性递减惯性权重粒子群优化算法（LDIWPSO）相结合，对涡旋泵全工况下的最大等熵效率进行了预测和优化。用平均绝对误差（MAE）、平均绝对百分误差（MAPE）、均方根误差（RMSE）和决定系数（R-square）综合评价模型的预测精度。预测和优化结果表明，旋涡泵的最大等熵效率可达22.89%。深度学习和LDIWPSO的结合可以高精度地预测和优化旋涡泵的最大等熵效率。同时也为理论分析和数值模拟中旋涡泵等熵效率的最大值提供了参考。
The isentropic efficiency of vortex pump in Organic Rankine Cycle (ORC) system has an important influence on the performance of the system. In this paper, vortex pump experimental data and deep learning are combined to construct an experimental data-driven isentropic efficiency prediction model of vortex pump under full operating conditions. The activation function and the number of hidden layer nodes in the model are filtered by nested screening technique. Through bilinear interpolation algorithm, the influence of vortex pump operation parameters on the isentropic efficiency is analyzed. In addition, the optimization boundary is selected in the four-dimensional space. Finally, the deep learning prediction model is combined with Linear Decreasing Inertia Weight Particle Swarm Optimization (LDIWPSO) to predict and optimize the maximum isentropic efficiency of vortex pump under full operating conditions. Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE) and Coefficient of Determination (R-square) are combined to evaluate the prediction accuracy of the model. The prediction and optimization results show that the maximum isentropic efficiency of vortex pump can reach 22.89%. The combination of deep learning and LDIWPSO can predict and optimize the maximum isentropic efficiency for vortex pump with high precision. It also provides a reference for the maximum value of vortex pump isentropic efficiency in theoretical analysis and numerical simulation.