基于卡尔曼滤波的风速序列短期预测方法
发布时间：2018-09-11 21:43
【摘要】：分析了卡尔曼滤波在风速序列预测分析中的应用机理,构造了用于风速序列预测分析的迟滞神经网络,并采用卡尔曼滤波方法将其与ARMA模型相融合,实现了风速序列的混合预测。通过修改激励函数的方式将迟滞特性引入神经网络,网络的权值采用梯度寻优的方式确定,迟滞参数利用遗传算法进行确定。系统的状态方程采用ARMA模型建立,将迟滞神经网络对风速序列的预测结果作为测量方程的测量值。混合预测方法能减小单一预测机制造成的同一性质误差的累积。仿真实验结果表明,迟滞神经网络的预测性能优于传统BP神经网络,而混合预测方法的预测性能优于单一预测方法。
[Abstract]:The application mechanism of Kalman filter in wind speed series prediction and analysis is analyzed. A hysteretic neural network is constructed for wind speed series prediction and analysis, and the Kalman filter method is used to fuse it with ARMA model. The mixed prediction of wind speed series is realized. The hysteresis characteristic is introduced into the neural network by modifying the excitation function, the weights of the network are determined by gradient optimization, and the hysteresis parameters are determined by genetic algorithm. The ARMA model is used to establish the state equation of the system, and the prediction result of the hysteresis neural network to the wind speed series is taken as the measured value of the measurement equation. The mixed prediction method can reduce the accumulation of the same property error caused by a single prediction mechanism. Simulation results show that the prediction performance of hysteresis neural network is better than that of traditional BP neural network, and that of hybrid prediction method is better than that of single prediction method.
【作者单位】： 天津工业大学电工电能新技术天津市重点实验室;天津工业大学电气工程与自动化学院;北京科技大学数理学院;
【基金】：国家自然科学基金资助项目(61203302)
【分类号】：TM614;TP183
[Abstract]:The application mechanism of Kalman filter in wind speed series prediction and analysis is analyzed. A hysteretic neural network is constructed for wind speed series prediction and analysis, and the Kalman filter method is used to fuse it with ARMA model. The mixed prediction of wind speed series is realized. The hysteresis characteristic is introduced into the neural network by modifying the excitation function, the weights of the network are determined by gradient optimization, and the hysteresis parameters are determined by genetic algorithm. The ARMA model is used to establish the state equation of the system, and the prediction result of the hysteresis neural network to the wind speed series is taken as the measured value of the measurement equation. The mixed prediction method can reduce the accumulation of the same property error caused by a single prediction mechanism. Simulation results show that the prediction performance of hysteresis neural network is better than that of traditional BP neural network, and that of hybrid prediction method is better than that of single prediction method.
【作者单位】： 天津工业大学电工电能新技术天津市重点实验室;天津工业大学电气工程与自动化学院;北京科技大学数理学院;
【基金】：国家自然科学基金资助项目(61203302)
【分类号】：TM614;TP183
【参考文献】
相关期刊论文 前10条
1 刘兴杰;米增强;杨奇逊;樊小伟;;一种基于EMD的短期风速多步预测方法[J];电工技术学报;2010年04期
2 罗文;王莉娜;;风场短期风速预测研究[J];电工技术学报;2011年07期
3 潘迪夫;刘辉;李燕飞;;基于时间序列分析和卡尔曼滤波算法的风电场风速预测优化模型[J];电网技术;2008年07期
4 孙国强;卫志农;翟玮星;;基于RVM与ARMA误差校正的短期风速预测[J];电工技术学报;2012年08期
5 武峰雨;乐秀t，
本文编号：2237941
本文链接：https://www.wllwen.com/kejilunwen/dianlilw/2237941.html
