Abstract: The matching optimization design of the blade geometry of the existing hydraulic model of the impeller and guide vane is carried out in order to improve the hydraulic performance of the waterjet pump. The Latin Hypercube sampling method is used for the space sampling of parameters that affect the performance of the waterjet pumps, such as the impeller blade load, the guide vane blade load and the stacking position. The performance parameters of the pump is obtained through steady numerical simulation of each sample by using the RANS equations. The mapping response model is built between the pump design parameters and the calculated performance parameters based on the artificial neural network. The genetic algorithm is adopted to optimize the waterjet pump hydraulic model with the aim of the maximum hydraulic efficiency. The iterative cycling of the optimization process is automatically completed, which can shorten the design and development period of the hydraulic model. The results show that after optimization, the internal flow field is significantly improved with the same head of the pump, and the efficiency of the calculation design point reaches 91.8%, while the efficiency validated by model test reaches 88.9%, as well as the range of the flow rate in high efficiency region is widened by 15%.

Key words: waterjet pump, load, optimization, neural network, genetic algorithm

摘要： 为提升喷水推进泵的水力性能,对现有水力模型动叶轮和导叶体的叶片几何形状进行匹配优化设计。该文针对影响喷水推进泵性能的叶轮叶片载荷、导叶叶片载荷、积迭位置等参数,采用拉丁超立方抽样方法进行样本空间采样,应用RANS方程对各样本进行定常数值求解,得到泵的性能参数,基于人工神经网络,建立泵设计参数和性能参数之间的映射响应模型,并以效率最高为目标,采用遗传算法进行优化,获得性能优异的喷水推进泵水力模型。该过程迭代循环自动完成,可缩短水力模型的设计开发周期。结果表明：优化后,在保持泵扬程不变的情况下,内流场显著改善,计算设计点效率达到91.8%,经过试验验证后效率达到88.9%,高效区流量范围拓宽15%。

关键词: 喷水推进泵, 载荷, 优化, 神经网络, 遗传算法