| In recent years,with the continuous development of marine energy,tidal current turbines have become more and more widely used.Runner blades are the main components of tidal energy turbines,and the performance of the blades is closely related to the hydrofoil.Therefore,the development of the hydrodynamic optimization design of hydrofoil has a great significance for improving the performance of tidal energy turbines.In the current optimization design methods of hydrofoils and impeller blades,still focusing on the combination of optimization algorithms and CFD numerical simulations.This method takes too long to calculate the performance of a single sample of CFD,which makes the time period of the entire optimization process far beyond the acceptable range of the designer.For this reason,some studies have used shallow neural networks to establish a nonlinear response model between hydrofoil geometric parameters and hydrofoil performance,and replaced the traditional CFD process in hydrofoil optimization.However,due to the large number of target performance parameters in hydrofoil optimization process,shallow neural networks lack accuracy when constructing complex nonlinear response relationships.Therefore,it is necessary to adopt a new neural network to construct the response relationship between the geometric parameters and performance parameters of hydrofoil,in order to improve the reliability and efficiency of the optimization design method of hydrofoil.This paper proposes a multi-condition optimization design method of hydrofoil combining deep belief network and NSGA-Ⅱ algorithm to improve the efficiency of hydrofoil optimization.The main contents are as follows:(1)Bezier curve is used to parameterize hydrofoil NACA63-815.The Euclidean distance in the metric space is introduced to unify the lift-drag ratio and cavitation performance parameters of the hydrofoil under different attack angles(0°,6° and 12°),thereby greatly reducing the number of objective functions of the optimization problem.Finally,the optimal Latin hypercube experimental design method is used to obtain the hydrofoil samples required for training the neural network,and the performance parameters of each hydrofoi samples are calculated by numerical simulation.(2)On the basis of the traditional deep belief network,a set of constrained variables,including the drag coefficient and lift coefficient of hydrofoil under different conditions;two sets of target variables,including the lift-to-drag ratio gap and the cavitation performance gap of hydrofoil,a total of three sets of variables are the output variables of the three sub-networks(DBN1,DBN2,and DBN3),and taking the ordinate values of the upper and lower profile of hydrofoil as input variables,a multi-dimensional deep belief network suitable for the performance parameters of hydrofoil studied in this paper is constructed.Based on the optimal Latin hypercube sampling design method,1600 hydrofoil samples are generated in the specified design space for DBN training,and 100 samples are re-extracted to test the accuracy of DBN.In order to improve the prediction accuracy of deep belief network,the NSGA-Ⅱ algorithm is used to optimize the network parameters of DBN1,DBN2 and DBN3 respectively.In addition,the accuracy and stability of deep learning prediction are verified by comparing the accuracy of predicting hydrofoil parameters with BP neural network of the same structure.(3)Taking the lift-drag ratio gap and cavitation performance gap of hydrofoil as the objective function,the hydrofoil surface area and the drag coefficient and lift coefficient under the attack angles 0°,6° and 12°as the constraint function,the multi-dimensional DBN and NSGA-Ⅱ genetic algorithm of optimization process to carry out the hydrfoil optimization design.The entire optimization design process saves about 77.3% of time compared with directly using CFD to evaluate individual samples,and improves the efficiency of the overall optimization design work.Next,CFD is used to analyze the geometric shape,pressure coefficient,wall entropy production rate and turbulent kinetic energy of the optimized hydrofoil and the original hydrofoil.Comparing the calculation results,it can be seen that the lift-drag ratio performance of optimized hydrofoil is better than the original hydrofoil,and the cavitation phenomenon has also been improved,which proves the effectiveness and feasibility of the optimization method. |
