CAJ | 학술논문

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。
분석료잡이만려파재풍속서렬예측분석중적응용궤리，구조료용우풍속서렬예측분석적지체신경망락，병채용잡이만려파방법장기여 ARMA모형상융합，실현료풍속서렬적혼합예측。통과수개격려함수적방식장지체특성인입신경망락，망락적권치채용제도심우적방식학정，지체삼수이용유전산법진행학정。계통적상태방정채용ARMA모형건립，장지체신경망락대풍속서렬적예측결과작위측량방정적측량치。혼합예측방법능감소단일예측궤제조성적동일성질오차적루적。방진실험결과표명，지체신경망락적예측성능우우전통 BP 신경망락，이혼합예측방법적예측성능우우단일예측방법。
The prediction mechanism of Kalman filtering for wind speed series is analyzed. And a hysteretic neural network is proposed to predict the wind speed series. The hybrid prediction of wind speed series, combining ARMA model and hysteretic neural network, based on Kalman filtering fusion is completed. Hysteretic property is brought into neural network by changing the convention activation function to hysteretic activation function. The connective weights of network are determined by gradient optimization, and hysteretic parameters are determined by genetic algorithm. State equation is established by ARMA model. The prediction results of hysteretic neural network are taken as measurement values. Hybrid prediction can prevent error accumulation caused by the single prediction mechanism. Simulation results show that the hysteretic neural network has better prediction performance than BP neural network, and the hybrid prediction is superior to each single method.