基于Bagging算法和遗传BP神经网络的负荷预测

Load Forecasting Based on Bagging Method and GABP Neural Network

负荷预测是电力规划的基础,传统的神经网络预测方法存在对初始网络权值设置敏感、收敛的速度慢、容易陷入局部极小值等缺点。文中引入遗传算法先对神经网络的初始值进行优化,再通过神经网络进行学习和训练,得出的结果再经Bagging方法集成,目的是提高其准确率。通过Matlab仿真进行实验,结果表明,基于Bagging算法集成遗传神经网络,能够克服传统BP神经网络的缺点,可较快收敛又不易陷入到局部极值中,具有较强的泛化能力,同时也大大提高了网络的预测精度。

The load forecast is the electric power plan foundation. However, there are lots of disadvantages in the traditional neural net- work prediction ways, including be sensitive to the initial network weights, easy to run into the local minimum point, etc. Brings forward genetic algorithm to the BP neural network, optimizing the initial network weights. In order to improve the accuracy, use the Bagging method integrated the results. Through the simulation experiment on Matlab, found out that by our research not only the Bagging method and genetic neural network can avoid the disadvantages in the traditional BP network and inherit its good learning and training a- bilities, but also have stronger generalization ability, and improve the prediction precision.