基于神经网络与遗传算法的离心泵汽蚀性能优化设计

Optimization Design of Centrifugal Pump Cavitation Performance Based on Neural Network and Genetic Algorithm

为了研究离心泵发生汽蚀过程时其内部流动规律和对其汽蚀性能的优化设计,在结合传统优化方法的基础上,提出神经网络与遗传算法结合的智能优化方法。通过Plackett-Burman试验设计从离心泵的叶轮进出口直径、进出口安放角、叶片数、叶片包角等7个设计参数中筛选出显著影响的3个优化设计变量。叶片出口宽度、叶片包角和叶轮出口直径3个优化设计变量,其影响汽蚀性能的显著性由大到小。运用拉丁超立方抽样方法抽取30组设计方案,分别进行数值模拟计算,得出相应的汽蚀余量值。建立神经网络模型,结合遗传算法在规定范围内进行寻优,得到最优设计变量组合及最优汽蚀余量值。取优化后的参数进行数值模拟计算,优化后的离心泵在相同工况下汽蚀余量降低了43.1%,说明优化后的离心泵抗汽蚀性能显著提高。

In order to study the internal flow law of centrifugal pump during cavitation process and optimize its cavitation performance,this paper proposes an intelligent optimization method combining neural network and genetic algorithm on the basis of combining traditional optimization methods.Through the Plackett Burman experimental design,three optimization design variables are selected from 7 design parameters of the centrifugal pump,including impeller inlet and outlet diameter,inlet and outlet placement angle,number of blades,and blade wrap angle.The significance of the three optimized design variables’impact on cavitation performance from large to small is ranked as follows:blade outlet width>blade wrap angle>impeller inlet and outlet diameter.The Latin hypercube sampling method is used to extract 30 groups of design schemes,and the corresponding NPSH values are obtained by numerical simulation.The neural network model is established,and the optimal design variable combination and the optimal NPSH value are obtained by combining the genetic algorithm to optimize within the specified range.By taking the optimized parameters for numerical simulation calculation,the NPSH of the optimized centrifugal pump decreases by 43.1%under the same working conditions,indicating that the anti-cavitation performance of the optimized centrifugal pump was significantly improved.