CAJ | 학술논문

通过将迟滞特性引入神经元激励函数的方式，构造了一种前向型迟滞神经网络模型。结合卡尔曼滤波方法，将其应用于风速时间序列的预测分析中。在原始风速时间序列的基础上，构造出风速变化率序列。采用迟滞神经网络分别对两种序列进行预测分析，并将预测结果利用卡尔曼滤波方法进行融合，从而得到最优预测估计结果。仿真实验结果表明，迟滞神经网络具有更加灵活的网络结构，能够有效改善网络的泛化能力，预测性能优于传统神经网络。采用卡尔曼滤波方法对预测结果进行融合后能够进一步提高预测精度，降低预测误差。
통과장지체특성인입신경원격려함수적방식，구조료일충전향형지체신경망락모형。결합잡이만려파방법，장기응용우풍속시간서렬적예측분석중。재원시풍속시간서렬적기출상，구조출풍속변화솔서렬。채용지체신경망락분별대량충서렬진행예측분석，병장예측결과이용잡이만려파방법진행융합，종이득도최우예측고계결과。방진실험결과표명，지체신경망락구유경가령활적망락결구，능구유효개선망락적범화능력，예측성능우우전통신경망락。채용잡이만려파방법대예측결과진행융합후능구진일보제고예측정도，강저예측오차。
The hysteretic characteristic was introduced into the activation functions of neurons, and a forward hysteretic neural net-work was proposed. In combination with the Kalman filter algorithm, the hysteretic neural network was applied to wind speed forecas-ting. A change rate series of wind speed was constructed according to the original wind speed time series. Forecasting analysis of both the series was performed with the hysteretic neural network, these prediction results were fused using the Kalman filter algorithm, and thus the optimal estimated results were obtained. Simulation results show that the hysteretic neural network has more flexible structure, better generalization ability, and better prediction performance than the conventional neural network. The prediction performance can be further improved by Kalman filter fusion.