基于BPNN-GA的泵站前池整流底坎参数优化

为了减少泵站进水结构内的不良流态，提高泵组运行效率，该研究基于计算流体动力学以及BPNN-GA(Back Propagation Neural Network-Genetic Algorithm)算法对泵站前池内底坎的结构设计参数进行优化。为便于计算遗传算法的适应度，在轴向流速分布均匀度和速度加权平均角基础上提出了前池水流流态的综合评价指标F。以综合评价指标F为目标参量，通过遗传算法优化训练好的BPNN模型，得出最优底坎结构设计参数，并同正交试验设计的最优方案的数值模拟结构对比分析。研究结果表明，在1、2、4号水泵机组运行情况下，相比于正交试验设计的最优方案泵站前池水流流态， 1号水泵进水流道的轴向流速分布均匀度提高了16.58个百分点，速度加权平均角增加了4.66°；2号水泵进水流道的轴向流速分布均匀度提高了0.49个百分点，速度加权平均角下降了2.81°；4号水泵进水流道的轴向流速分布均匀度提高了8个百分点，速度加权平均角增加了7.81°，综合评价指标F为1.16，表明前池流态得到较大幅度的提高。基于BPNN-GA算法对泵站前池整流底坎参数进行优化，克服传统方法陷入局部最优的缺陷，可在满足设计要求的范围内选出当轴向流速分布均匀度和速度加权平均角最优时的底坎设计参数，为计算智能在泵站优化水力设计方面提供参考。

Parameter optimization of rectification sill in the forebay of pumping station using BPNN-GA algorithm

In the field of hydraulic engineering, the design of pumping stations played an essential role in ensuring the efficient and reliable operation of water supply systems. However, due to various factors such as asymmetric pump configurations and non-straight alignment of the forebay and inlet pool center lines, the flow patterns in the forebay could be poor, which could lead to problems such as cavitation and reduced pump efficiency. In order to improve the unfavorable flow patterns within the pump station''s inlet structure and enhance the efficiency of the pump operation, this study focuses on optimizing the structural design parameters of the forebay''s sill using computational fluid dynamics and the BPNN-GA (Back Propagation Neural Network-Genetic Algorithm) model. To facilitate the genetic algorithm''s fitness calculation, a comprehensive evaluation index F for the flow patterns in the forebay is proposed based on the uniformity of axial velocity distribution and velocity-weighted average angle. By taking the comprehensive evaluation index F as the objective parameter, the BPNN is optimized using the genetic algorithm, leading to the determination of the optimal sill design parameters, which are then compared and analyzed against the numerically simulated structures from the orthogonal experimental design.The research findings indicate that, under the operation of pumps No.1,No. 2, and No.4, compared to the optimal design from the orthogonal experimental design, the uniformity of axial velocity distribution of No.1 pump inlet flow increased by 16.58 percentage points, and the velocity-weighted average angle increased by 4.66°. For No.2 pump, the uniformity of axial velocity distribution increased by 0.49 percentage points, while the velocity-weighted average angle decreased by 2.81°. As for No.4 pump, the uniformity of axial velocity distribution improved by 8 percentage points, and the velocity-weighted average angle increased by 7.81°, resulting in a comprehensive evaluation index F of 1.16, indicating a significant improvement in the flow patterns of the forebay.The optimization of the pump station''s forebay sill parameters using the BPNN-GA algorithm overcomes the drawback of traditional methods being trapped in local optima. It enables the identification of the optimal sill design parameters for the uniformity of axial velocity distribution and velocity-weighted average angle within the required design range, providing reference for the application of computational intelligence in optimizing hydraulic design in pump stations.