基于互信息的改进RBF神经网络模型及其在短期径流预报中的应用

发布时间：2018-05-07 00:26

本文选题：RBF神经网络 + 短期径流预报　；参考：《华北电力大学(北京)》2016年硕士论文

【摘要】：随着社会和经济的发展,水资源作为人类赖以生存和社会经济发展不可或缺的物质基础,有着不可替代的作用。对水资源进行合理的开发和利用一贯是国内外研究的热点和难点,而水文预报是合理开发利用水资源的前提,也是水利工程进行合理规划、建设的关键问题。水文预报方法发展至今,已从水文统计法、成因分析法等传统方法发展到模糊数学法、人工神经网络法、灰色系统理论法、多层递阶法、最优组合预测法等,各方法均具有其各自的特点和优势。本文分析了互信息理论和最邻近聚类RBF神经网络理论的基本原理,吸取了互信息理论变量相关关系表征能力和神经网络非线性逼近功能,建立了基于互信息的改进RBF神经网络预测模型。该算法对最邻近聚类RBF神经网络算法中的不足进行改进,对神经网络输入层进行筛选,简化了神经网络结构,提高了算法的计算速度和预测精度。在此基础上,研究分析预报精度与误差特性,利用有限记忆最小二乘法对预报径流序列进行实时校正,校正后预测值的预报精度有了进一步提升。以雅砻江流域泸宁水文站为例,进行预报模型在短期径流预报中的应用研究,验证了本文所建模型及算法的合理性及有效性。实验结果表明,基于互信息的改进RBF神经网络预测模型预报精度较高,对预报结果进行实时校正后获得了更良好的预报效果,为官地水电站水库调度提供了更为可靠的基本依据。
[Abstract]:With the development of society and economy, water resources, as an indispensable material foundation for human survival and social economic development, play an irreplaceable role. The reasonable development and utilization of water resources has always been the hot and difficult point of domestic and foreign research, and hydrological forecast is the premise of rational development and utilization of water resources, and also the key problem of rational planning and construction of water conservancy projects. The hydrological forecasting method has been developed from traditional methods such as hydrological statistics and genetic analysis to fuzzy mathematics, artificial neural network method, grey system theory method, multi-level hierarchical method, optimal combination forecasting method and so on. Each method has its own characteristics and advantages. In this paper, the basic principles of mutual information theory and nearest neighbor clustering RBF neural network theory are analyzed. An improved RBF neural network prediction model based on mutual information is established. The algorithm improves the shortcomings of the nearest neighbor clustering RBF neural network algorithm, selects the input layer of the neural network, simplifies the neural network structure, and improves the calculation speed and prediction accuracy of the algorithm. On this basis, the prediction accuracy and error characteristics are studied and analyzed, the prediction runoff series is corrected in real time by using the finite memory least square method, and the prediction accuracy of the predicted value is further improved after the correction. Taking Luning hydrologic station in Yalong River basin as an example, the application of forecasting model in short-term runoff forecasting is studied, which verifies the rationality and validity of the proposed model and algorithm. The experimental results show that the prediction accuracy of the improved RBF neural network model based on mutual information is high, and the prediction results are corrected in real time. It provides a more reliable basis for reservoir operation of Guandi Hydropower Station.
【学位授予单位】：华北电力大学(北京)
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：P338

【相似文献】