厂顶溢流式水电站厂房结构振动响应预测研究

发布时间：2018-07-17 14:55

【摘要】：水电站厂房结构极其复杂，引起结构振动的振源种类更是多种多样，致使电站运行过程中的厂房结构振动问题非常普遍。考虑到厂房结构振动对仪器设备、工作人员健康以及建筑物运行稳定性和安全可靠性的影响，，利用较少的监测数据达到全面掌握和控制水电站振动的目的成为新的研究课题。本文运用智能算法与神经网络混合的预测方法，不考虑结构精确的数学和精准的力学模型，而是依据尾水脉动和机组振动的观测数据，非线性的映射出水电站结构的振动特性，达到预测未知工况和未观测部位的结构振动响应状况的目的。结合某厂顶溢流式水电站原型观测实验，运用果蝇算法优化广义回归神经网络平滑参数P，构建FOA-GRNN网络模型。同时结合反向传播神经网络（BP）、局部回归神经网络（ELMAN）展开对比预测研究。最终得出：FOA-GRNN网络在预测能力、学习速度上明显优于BP和ELMAN网络。说明运用FOA-GRNN神经网络预测厂房结构振动响应的可行性和优越性。为弥补基本粒子群优化算法易陷入局部最优、收敛性差的缺陷，提出了优胜劣汰、步步选择粒子群优化算法—SSPSO，通过典型测试函数证明SSPSO具有很强的寻优能力。并运用SSPSO对广义回归神经网络平滑参数进行优化，充分利用SSPSO寻优能力强及径向基函数调整参数少的优点，建立厂房结构的振动响应预测模型，展开厂坝结构振动响应预测研究。证明了：SSPSO算法的寻优能力很强；基于SSPSO优化的广义回归神经网络与其他网络相比，在预测精度，收敛性能，泛化能力等各个方面得到了很大提升。运用粒子群优化算法、遗传算法和果蝇优化算法分别对径向基神经网络进行优化，建立最优的PSO-RBF、GA-RBF、FOA-RBF网络模型，展开泄流诱发厂房结构振动响应的预测研究。结果表明：PSO-RBF、GA-RBF和FOA-RBF预测效果均良好，适合运用于泄流诱发水电站厂房结构振动响应预测研究中，其中FOA-RBF稳定性及泛化能力最强。综上所述：智能算法与神经网络构建的混合模型不仅易于理解、掌握，而且精度很高。非常适合于厂顶溢流式水电站厂房结构的振动响应的预测研究，为厂房结构振动响应预测提供了新的方法和思路，增强了厂房结构的智能监测水平。
[Abstract]:The structure of hydropower plant is extremely complex, and the types of vibration sources causing structural vibration are more and more varied, which makes the vibration problem of powerhouse structure very common in the process of operation of hydropower station. Considering the influence of structural vibration on instrument and equipment, staff health, operation stability and safety and reliability of buildings, it is a new research subject to make use of less monitoring data to master and control the vibration of hydropower station in an all-round way. In this paper, the hybrid forecasting method of intelligent algorithm and neural network is used. The accurate mathematical and mechanical model of the structure is not taken into account, but the vibration characteristics of the hydropower station structure are nonlinear mapped according to the observation data of the tail water pulsation and the vibration of the unit. The objective is to predict the vibration response of the structure under unknown working conditions and unobserved positions. Combined with the prototype observation experiment of a plant roof overflow hydropower station, the FOA-GRNN network model was constructed by using Drosophila algorithm to optimize the smoothing parameters of generalized regression neural network (PNN). At the same time, combining back propagation neural network (BP) and local regression neural network (ELMAN), a comparative prediction study was carried out. Finally, it is concluded that the prediction ability and learning speed of the FOA-GRNN network are obviously better than those of BP and ELMAN networks. The feasibility and superiority of using FOA-GRNN neural network to predict the vibration response of powerhouse structure are illustrated. In order to remedy the defect that the basic particle swarm optimization (PSO) is easy to fall into the local optimum and the convergence is poor, the survival of the fittest is proposed, and the particle swarm optimization (SSPSO) algorithm is selected step by step. It is proved by the typical test function that SSPSO has a strong searching ability. The smooth parameters of generalized regression neural network are optimized by using SSPSO, and the prediction model of vibration response of powerhouse structure is established by making full use of the advantages of strong searching ability of SSPSO and less adjustment parameters of radial basis function. Research on prediction of vibration response of plant dam structure is carried out. It is proved that the optimization ability of the SSPSO algorithm is very strong, and that the generalized regression neural network based on SSPSO optimization has been greatly improved in prediction accuracy, convergence performance, generalization ability and so on. Particle swarm optimization (PSO) algorithm, genetic algorithm (GA) and Drosophila optimization algorithm were used to optimize radial basis function neural network (RBNN), and the optimal PSO-RBFN GA-RBFFFOA-RBF neural network model was established to predict the vibration response of powerhouse structure induced by discharge. The results show that the prediction results of GA-RBF and FOA-RBF are good, and it is suitable for the prediction of vibration response of powerhouse structure induced by discharge. Among them, FOA-RBF has the strongest stability and generalization ability. To sum up, the hybrid model constructed by intelligent algorithm and neural network is not only easy to understand and master, but also has high accuracy. It is very suitable for predicting the vibration response of powerhouse structure of roof overflow hydropower station. It provides a new method and train of thought for predicting the vibration response of powerhouse structure and enhances the level of intelligent monitoring of powerhouse structure.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TV731.3;TV312

【参考文献】