基于BP神经网络的电力短期负荷模拟研究

Study on Power Short-term Load Simulation Based on BP Neural Network

电力负荷日常运行过程中存在随机性和不确定性,对短时负荷难以实现精准预测。为此,以BP神经网络为基础,建立网络模型进行电网短时负荷预测。由于BP算法收敛速度慢、易陷入局部最优等缺陷,引入遗传算法对网络结构的学习连接权重和阈值进行训练优化,根据适应度来选取候选解,再进行交叉、变异操作,加快网络的收敛速度,获得全局的可行解。将改进的模型应用于实际电网短时负荷预测中,结果表明改进的模型负荷预测平均误差为0.32%,最大相对误差为3.26%,平均学习次数5000~6000次之间,短时负荷的预测可靠性和准确性、运算效率大大提高,满足电力负荷预测要求。

Due to the characteristics of power load randomness and uncertainty,the accurate prediction of short-time load is difficult.BP neural network is used for short-time load prediction.The BP algorithm has slow convergence speed and is easy to fall into the local optimal solution defects,hence,the genetic algorithm is introduced to train and optimize the learning connection weights and threshold of the network.It can speed up the convergence speed of the network,and obtain global feasible solutions.The improved model is applied to the actual short-time load prediction of the power grid,the result shows that the average error of the improved model load prediction is 0.32%,the maximum relative error is 3.26%,and the average learning number is between 5000~6000 times.The prediction reliability,accuracy and operational efficiency of the short-time load are greatly improved,which meets the requirements of power load prediction.