Optimization Design Of Bidirectional Through-flow Tubular Turbine Runner Blade Based On Deep Belief Network

Tubular turbine unit is one of the commonly used units in tidal power station.As one of the main components of the turbine,the performance of the runner blade has an indispensable influence on the efficiency and stability of the turbine.CFD numerical simulation and neural network are commonly used in the optimization design of the main components of the turbine,but the former has a long calculation cycle,and the latter has low accuracy in dealing with nonlinear multi-objective problems.Therefore,this paper proposes an innovative deep belief network optimization model based on NSGA-Ⅱ genetic algorithm.The model can optimize the multi-objective design of the runner blade under both positive and negative operating conditions of the two-way tubular turbine.The performance parameters of the optimized blade obtained by CFD calculation are compared with the performance parameters of the original blade.The optimization model greatly shortens the experimental time while ensuring the design accuracy.The main contents of this paper are as follows:(1)Mathematical model for optimum design of runner blades of bidirectional tubular turbine is established.Firstly,three-dimensional blades are expanded into two-dimensional airfoil by conformal transformation,and the airfoil skeleton is parameterized by Bezier curve parameterization.The geometry of the blades is changed by changing the design variables of the blades.The Euclidean distance method is introduced to reduce the dimension of multi-objective optimization problems,reduce the experimental cycle and reduce data errors.The optimum Latin hypercube sampling technology is used to uniformly sample the runner blade sample points and generate the predicted sample points of DBN model.(2)The NSGA II-DBN prediction model is established.Based on ISIGHT simulation software,a method is proposed to correct and reduce the model prediction error by updating the network parameters.Based on the NSGA-Ⅱ algorithm,the original leaves are optimized by the established multi-dimensional network model,and the Pareto solution set of the optimized leaves is obtained.According to the demand that the two sets of objective functions are reduced,a set of optimized blades are selected for CFD numerical simulation calculation,and the optimized blade performance parameters are obtained and compared with the original blade performance parameters.The comparison results show that the efficiency performance of the optimized blade under both positive and negative operating conditions exceeds 80%,which proves the feasibility of the optimization method.(3)A classification method based on a set of objective functions is proposed.Two clustering methods,SOM self-organizing map neural network and Kmeans mean clustering,are used to cluster the optimized design sample points based on a single objective function.Onedimensional and two-dimensional PNN neural network verification models are established.The one-dimensional PNN verification model verifies the classification results of the above two clustering algorithms.The two-dimensional PNN verification model is used to verify the feasibility of the proposed classification method.(4)In order to study the influence of blade shape on performance parameters,the blade geometry with different performance preferences is studied and analyzed.According to the proposed classification method,the optimized blades with different performance preferences are obtained,and the performance parameters of the selected optimized runner blades are calculated by CFD numerical simulation.Taking the reverse operation condition as an example,taking the center of the circle corresponding to the blade radius as the origin,and taking the blade radius and the wrap angle between the edge of the inlet and outlet end of the blade and the center of the blade as the horizontal and vertical coordinates respectively,the change curve of the inlet and outlet end of the optimized blade is extracted and compared with the change curve of the inlet and outlet end of the original blade.The following conclusions are drawn:the geometry of the inlet end of the blade will affect the efficiency performance of the blade,and the geometry of the outlet end of the blade will affect the cavitation performance of the blade.