基于遗传模拟退火算法改进BP神经网络的中长期电力负荷预测  

Medium and long-term power load forecasting based on BP neural network improved by genetic simulated annealing algorithm

作　　者：徐扬 张紫涛 XU Yang;ZHANG Zitao(College of Energy and Electric Engineering,Hohai University,Nanjing 211100)

机构地区：[1]河海大学能源与电气学院,南京211100

出　　处：《电气技术》2021年第9期70-76,共7页Electrical Engineering

摘　　要：针对目前中长期负荷预测方法中存在过拟合、预测精度和效率较低等问题,本文提出一种基于遗传模拟退火算法(GSA)改进BP神经网络的中长期电力负荷预测模型,即BP-GSA模型。首先建立标准三层神经网络,即输入层、隐藏层和输出层,选择国民生产总值、第二产业生产总值、市区常驻人口及月平均温度四个影响因子作为输入变量,月度负荷为输出变量。其次利用遗传模拟退火算法不断修正网络节点连接权值,以最优适应度为标准,确定最优网络节点连接权值分布。最后,代入权值最优解,通过训练样本数据,获取最小方均差预测模型。分别应用本文提出的BP-GSA模型及其他四种传统方法,对某市2020年月度负荷进行预测。误差分析表明,BP-GSA模型预测精度最高。随后将BP-GSA模型分别应用于不同年份的月度负荷预测,预测结果表明其误差稳定,证明了模型的鲁棒性。Aiming at the problems of over-fitting,low accuracy and low efficiency in current medium and long-term load forecasting methods,a novel model,which is based on improved BP neural network(BP-GSA),is proposed.Firstly,a standard three-layer neural network including the input layer,the hidden layer and the output layer is established.The paper selects GDP,secondary industry GDP,urban resident population,monthly average temperature as input variables,and monthly load as the output variable.Secondly,the genetic simulated annealing algorithm is used to continuously modify the network node connection weights until the optimal network node connection weight distribution is achieved according to the optimal fitness standard.Finally,with the optimal solution of weights substituted,the paper obtains the model that has the minimum mean square error through the training of the data.The calculation example compares the BP-GSA model proposed in the paper with the other four types of traditional methods by predicting one city’s monthly load in 2020.The error analysis shows that the BP-GSA provides the best prediction.Then the model is applied to other different years.The error remains stable,which verifies the robustness of the algorithm.

关 键 词：负荷预测 遗传算法 模拟退火 BP神经网络 

分 类 号：TM715[电气工程—电力系统及自动化] TP18[自动化与计算机技术—控制理论与控制工程]