【作者】 杨成慧；

【导师】 王爽心；

【作者基本信息】 北京交通大学 ， 检测技术与自动化装置， 2012， 硕士

【摘要】 小世界网络是国内外关注的前沿课题及学术热点,近年来已广泛应用于社会关系网络、互联网络、生物工程、交通网络等众多领域。然而现有对小世界神经网络的研究仍有不足,小世界优化算法在神经网络中的应用也鲜有提及。本文针对现有小世界神经网络的缺陷,提出了基于层连优化的多层前向小世界神经网络；另外,将小世界优化算法应用于RBF神经网络中,并将两种网络应用于风电功率的实时预报中。论文主要开展了如下几方面的工作：1.对小世界网络理论及其重要参数做了较为全面、系统的分析和阐述。通过对现有小世界神经网络进行深入分析和总结,对其目前存在的问题作了较为深入的探讨,这些工作有助于小世界神经网络的改进和创新研究。2.对现有小世界神经网络进行改进,提出了基于层连优化的多层前向小世界神经网络。通过函数逼近及交通流量预测数据对新型小世界神经网络与现有小世界神经网络的仿真对比表明,新型网络在网络逼近性能、逼近速度、运算时间上要全面优于现有小世界神经网络。3.对现有应用于RBF神经网络优化算法的优缺点加以对比分析,探讨将小世界优化算法应用于RBF神经网络的可行性,并将其用于RBF神经网络的参数选取中。实验验证了小世界优化算法在针对RBF神经网络的参数选取优化中要优于其他优化算法。4.针对风电功率实时预报中的难点问题,对现有风电功率预报技术进行了总结并分析其不足之处,分析了多种基于神经网络的统计预测方法并进行选择；通过对风电历史资料的相关性分析,找出其历史资料与预测功率相关性最强的项,令其作为神经网络的输入；对数据进行预处理,并分别应用新型多层前向小世界神经网络以及经小世界优化算法优化的RBF神经网络对风电功率进行提前20min至4h的预测。仿真结果表明,两种模型可达到所需预测精度,并实现较好的实时预测效果。更多还原

【Abstract】 Small-world network is one of forefront subjects and academic focus in the research area, which has been widely used in many fields such as social networks, Internet networks, biological products, transport networks and so on. On one side, small-world neural network has been proved to have a better performance than regular neural network, but many defects still exist; on the other side, application of small-world optimization algorithm in neural network are still rarely reported. Aiming at what is mentioned above, the main contributions of this thesis are listed as follows:1. Small-world network theories and its important parameters were expatiated fully and systematically. By analyzing and summarizing currently available small-world neural networks, generally existed problems in them were discussed in depth. These efforts were helpful for further study on the improvement and innovation of small-world neural network.2. A novel multilayer feedforward small-world neural network based on connecting optimizition was proposed. Simulation results showed that novel network model presented a better performance of fast convergence rates, small iteration times and strong stability on comparison with different kinds of existed small-world neural networks.3. Optimization algorithms in RBF neural network were analyzed, and method of using small-world optimization algorithm in RBF neural network was proposed. Experiments verified a better performance in parameter researching of RBF neural network.4. Aiming at problems of wind power forecasting, existed techniques were summarized and statistical prediction methods based on neural network were analyzed in order to choose a better one. Based on correlation analysis, history data which had a strong correlation were extracted as input of neural network. Small-world neural network model and RBF neural network using small-world optimization algorithm were operated in20min to4h wind power real-time forecasting respectively. Simulation showed good performance of two models mentioned above.更多还原