论文部分内容阅读
为了进一步改进负荷预测模型,文章首先分析了BP网络模型的S激活函数和线性激活函数,接着就算法进行了增加动量项、可变学习速度的反向传播算法和学习速率的自适应调节的算法预测改进,在Matl AB验证BP神经网络负荷预测中,以某城市7月26日~7月31日之间的气象特点为负荷预测数据,仿真结果表明traingdm方法训练得出的结果准确度高,训练速度快,训练次数少。平均误差较小的,得出某些点的误差范围在0.04%~5.92%之间,平均误差是2.74%,在允许的误差范围内,这一研究对于BP神经网络的进一步预测应用具有一定的意义。
In order to further improve the load forecasting model, the paper first analyzes the S-activation function and the linear activation function of the BP network model. Then the algorithm is used to increase the momentum term, the variable propagation speed of the backpropagation algorithm and the learning rate adaptive adjustment algorithm In the prediction of BP neural network load forecasting by Matl AB, the meteorological characteristics of a city from July 26 to July 31 are load forecasting data. The simulation results show that the traingdm method has the advantages of high training accuracy, Training fast, less training times. The average error is smaller, the error of some points is obtained in the range of 0.04% ~ 5.92%, and the average error is 2.74%. In the range of the allowable error, this research has certain application in the further prediction of BP neural network significance.