基于神经网络集成算子的混合模型的研究与应用

发布时间：2018-03-28 18:08

本文选题：时间序列预测　切入点：神经网络　出处：《兰州大学》2015年硕士论文

【摘要】：随着世界上风力发电的快速发展,风电在电网中所占的比例不断增大。然而,准确的风速预测仍然是目前大规模的风力发电的一个挑战。因此,准确地预测风速对于风力发电、实现电力系统的优化运行和调度是非常重要的。自回归求和移动平均时间序列模型(ARIMA模型)可以通过对原序列的差分得到平稳的时间序列,相应的非线性时间序列的建模方法,广义自回归条件异方差模型(GARCH模型)在ARIMA模型的基础上,考虑了预测残差的条件异方差,增加了条件方差方程,而滚动灰色模型(RGM模型)是在灰色建模的基础上,不断引入新数据的同时,淘汰对现在的时刻影响较小的老数据,通过循环迭代实现数据更新。三者均在风速预测上有着广泛的应用。人工智能神经网络方法由于其良好的非线性拟合能力和泛化能力,用以代替传统的时间序列预测方法。其中,较为常用的有BP神经网络和小波神经网络。BP神经网络是一种前馈型的神经网络,可以有效的根据误差不断的调节网络结构参数,从而达到理想的预测效果；而小波神经网络则是在BP神经网络的基础上,结合小波分析的优点,引起了研究者极大的重视。与包括ARIMA模型、GARCH模型和RGM模型在内的三种传统时间序列模型以及单一的BP神经网络和小波神经网络相比,本文以神经网络为基础,针对其结构初始化的随机性问题,结合集成预测的思想,给出了集成BP神经网络和集成小波模型风速预测模型,可以避免因为不确定的权重导致的缺陷。在构造的平均数集成算子、中位数集成算子和众数集成算子三种集成算子下,集成网络的预测输出与单一的神经网络输出相比较有一定的改善。同时为了进一步提高预测的准确性,本文分别利用粒子群优化(PSO)算法和布谷鸟搜索(CS)算法对计算众数算子网络输出的核密度估计中的窗宽参数进行优化。在内蒙古的风速预测研究中表明,本文提出的基于神经网络集成算子的混合预测模型在对中国内蒙古中部风速的模拟预测中优于其他五种传统单一的预测模型。
[Abstract]:With the rapid development of wind power generation in the world, the proportion of wind power in power grid is increasing. However, accurate wind speed prediction is still a challenge for large-scale wind power generation. It is very important to realize the optimal operation and dispatch of power system. The autoregressive summation moving average time series model (ARIMA model) can obtain the stationary time series by the difference of the original sequence and the corresponding modeling method of the nonlinear time series. The generalized autoregressive conditional heteroscedasticity model (GARCH) is based on the ARIMA model, which considers the conditional heteroscedasticity of the prediction residual and increases the conditional variance equation, while the rolling grey model is based on the grey model. At the same time that new data are being introduced, old data that has less impact on the present moment will be eliminated. The three methods are widely used in wind speed prediction. Artificial intelligence neural network method has good nonlinear fitting ability and generalization ability. The BP neural network and the wavelet neural network. BP neural network is a kind of feedforward neural network, which can adjust the network structure parameters continuously according to the error. The wavelet neural network is based on BP neural network and combines the advantages of wavelet analysis. Compared with three traditional time series models, including ARIMA model and RGM model, as well as a single BP neural network and wavelet neural network, this paper is based on neural network. Aiming at the randomness of the structure initialization, the integrated BP neural network and the integrated wavelet model wind speed prediction model are presented, combined with the idea of integrated prediction. The defects caused by uncertain weights can be avoided. Under three kinds of integration operators, the mean integration operator, the median integration operator and the mode integration operator, The predictive output of the integrated network is better than that of the single neural network, and in order to further improve the accuracy of the prediction, In this paper, the particle swarm optimization (PSO) algorithm and the cuckoo search algorithm (CSC) are used to optimize the window width parameters in the kernel density estimation of the output of the modular operator network. In this paper, the hybrid forecasting model based on neural network ensemble operator is superior to the other five traditional single forecasting models in forecasting wind speed in the middle of Inner Mongolia.
【学位授予单位】：兰州大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP183;TM614

【相似文献】