【摘要】 文章针对水电仿真系统中水轮发电机机组的非线性动态数学模型建模复杂问题,提出了一种基于信息融合思想的神经网络模型。通过现场设置的多个异质传感器采集数据,作为该神经网络模型的输入训练样本数据,网络训练中动态修改网络权值和阈值,从而完成复杂的非线性建模功能。同时采用了具有较强全局寻优能力的遗传算法在训练中修改网络结构,从而避免神经网络训练速度慢、容易陷入局部极值的缺点,在现场在线数据预测测试中准确率可达95.8%以上,可以满足仿真模型需要。更多还原

【Abstract】 Aiming at problematic complexity of the nonlinear dynamic mathematical modeling of generator in the hydro-electric simulation system,a neural network based on information fusion is brought forward.Take the data gathered by multi-sensors on spot as input sample of the neural network model,and the weight and threshold are dynamic modified by network training in order to accomplish complex nonlinear modeling.Synchronously genetic algorithm which has ability of global optimize is adopted to modifiy the structure of network and eleminate rate tardiness of neural network training and relapsing into local extremum.The veracity of the model reached96.5%online in spot and can meet the need of hydro-electric simulation.更多还原