双向Elman神经网络在卫星电池阵功率预测中的应用研究

Application of Bi-directional Elman Neural Networks in Satellite Battery Array Power’s Prediction

针对神经网络预测电池阵功率存在的模型阶数难以确定及预测精度低下的问题 ,提出一种基于改进的Elman神经网络的双向预测模型。该模型利用关联层动态神经元的反馈连接 ,将未来预测网络和过去预测网络的信息进行融合 ,使网络对时间序列特征信息的记忆得到加强 ,从而提高预测精度。用该文提出的双向预测模型对电池阵功率进行预测 ,输入层仅需一个节点 ,不需事先对模型进行定阶。仿真预测表明 ,预测精度比单向模型明显提高 ,且网络具有较好的泛化能力。

In order to resolve the problem of determination of the number of model order and predictive accuracy being low in predicting battery array power by ANN, the paper presents a bi directional predictive model based on improved Elman. Information of future and past prediction system is fused and memory of network for time series feature’s information is strengthened by this model which depends on the feedback of connective layer. The method effectively improves the predictive accuracy. The model presented by the paper needs not to determine the number of model order in predicting battery array power. The simulation results indicate that its predictive accuracy is greatly improved by comparison with unidirectional model, and that the model has a better generalization ability.