基于小波分析和PSO优化神经网络的短期风电功率预测  

Short-term wind power prediction based on wavelet analysis and PSO optimized neural network

作　　者：叶小岭[1,2] 刘波[1,2] 邓华[1,2] 肖寅[1,2] 

机构地区：[1]南京信息工程大学信息与控制学院,江苏南京210044 [2]南京信息工程大学气象灾害预报预警与评估协同创新中心,江苏南京210044

出　　处：《可再生能源》2014年第10期1486-1492,共7页Renewable Energy Resources

基　　金：公益性行业(气象)科研专项资助项目(GYHY201106040);中国气象局软科学研究课题(SK20120146);江苏省六大人才高峰项目资助(WLW-021);南京市产学研资金项目(2012t026)

摘　　要：针对风电场风速和风电功率序列起伏波动大、无明显变化规律等特点以及传统神经网络收敛速度慢、易陷入局部极小值等缺陷,提出了基于小波分析和改进粒子群算法优化神经网络的短期风电功率预测方法。首先,通过小波方法将用于神经网络训练的历史风速和风电功率序列进行分解,再针对风速和风电功率的各个分量分别建立相应的神经网络模型,采用分期变异粒子群算法对各个分量的神经网络学习算法进行优化,最后将各个分量的预测值进行小波重构得到风电功率预测结果。江苏如东某风电场风电机组的实验结果证明预测精度较传统神经网络方法有较大提高,验证了所提出方法的有效性。Because there are characteristics such as stochastic fluctuations of wind speed, irregular wind power sequences of wind farm, and wind power prediction of traditional neural network has disadvantages such as low converge speed and easily falls into minimum point, a method of short-term wind power predic- tion based on wavelet analysis and improved particle swarm optimization （PSO）neural network is proposed. First of all,the wavelet method is used to decompose history wind speed and wind power sequences for neural network training later, then corresponding neural network models are respectively established based on different wind speed and wind power sequence components and the learning algorithm of neural networks of different components are optimized by staging mutation PSO. Finally,wind power prediction result is ob- tained from wavelet reconstruction of all different components. Taking a wind farm of Rudong in Jiangsu province as an example, experimental results show that prediction accuracy is improved a lot compared to traditional neural network, which verifies the effectiveness of proposed method.

关 键 词：小波分析 改进粒子群算法 神经网络优化 短期风电功率预测 

分 类 号：TK89[动力工程及工程热物理—流体机械及工程]