基于模拟退火的Gauss-Newton算法神经网络在短期负荷预测中的应用

Application of ANN to short-term power load forecasting basedon simulated annealing Gauss-Newton algorithm

针对一般BP网络存在的一些缺陷,首次提出了利用基于模拟退火的Gauss-Newton算法的神经网络预测电力系统短期负荷,并编制了通用程序.在相同的初始条件下,用基于模拟退火的Gauss-Newton算法的神经网络和自适应学习率附加动量法神经网络进行了比较,得出前者的特点和优点:一次性求解权值和偏差,收敛快,精度高,收敛于全局最优解.在算例中,基于人工神经网络的非线性特点进行了负荷预测,通过和真实值的比较说明本方法预测结果精度很高,从而更进一步验证了该方法应用于短期负荷预测的可靠性和优势.

Aimed at some limitation of ordinary BP neural network, an algorithm based on simulated annealing Gauss-Newton algorithm applied to neural network which predicts short-term power load forecasting originally and describes universal procedure. On the same initial conditions, the paper compares the neural network based on simulated annealing Gauss-Newton algorithm with the neural network based on adaptive learning ratio and additive momentum algorithm, and explains the characteristic and excellence of the former., getting weight and bias one-off, quicker convergence, higher precision and global optimization. As an example, based on the non linear characteristic of ANN, the method is applied to power load forecasting. Compared with the actual value, prediction error of our method is acceptable, which illuminates the reliability and superiority of the algorithm ap plied to short term power load forecasting.