Abstract: Radial basis function (RBF) neural network is widely
applied in short-term load forecasting because of its strong
generalization ability and fast convergence speed. However,
the traditional method of using K-means and self-organizing
map (SOM) for the training of the radial basis center of RBF
possess certain shortcomings. Due to the weak global searching
capability, this method falls into local optimal solution easily,
which seriously restricts the improvement of the precision of
load forecasting of RBF. To relieve the restriction, an improved
RBF based on reinforcement learning (RL) is proposed for
short-term load forecasting. The proposed method dramatically
enhances the global searching capability of SOM by applying
the feedback-correction mechanism of RL in SOM, which drives
it to approach to the optimal radical basis center. As a result, the
precision of short-term load forecasting of RBF is improved. To
verify the proposed method, simulation case is carried out based
on the load data of a certain area in UK from May to September
2016.Comparing with K-means method and SOM method, the
simulation results show that the average relative error is notably
reduced by using proposed method, which demonstrates the
correctness and superiority of the proposed method.
Keywords: short-term load forecasting; reinforcement learning;
RBF neural network; self-organizing map; radial basis center
摘 要:径向基(radial basis function,RBF)神经网络因
其泛化能力强、收敛速度快的特点广泛应用于负荷预测。
但传统采用K-means和自组织映射(self-organizing map,
SOM)训练RBF径向基中心的方法因其全局搜索能力偏
弱,仍然存在容易陷入局部最优解的问题,严重制约了RBF
预测精度的提高。针对此问题,提出了一种基于强化学习
(reinforcement learning,RL)改进的RBF短期负荷预测方法。
强化学习通过环境的反馈不断完善搜索策略,具有非常突出
的全局搜索能力。所提方法将强化学习以环境反馈修正搜索
策略的机制应用于SOM,大幅增强了SOM的全局搜索能力,
使其获得逼近最优的径向基中心,提高RBF负荷预测精度。
以英国某地区2016年5~9月的负荷数据进行仿真实验。结果
显示,与采用K-means和SOM方法训练径向基中心的RBF相
比,所提的强化学习改进RBF方法的负荷预测平均相对误差
分别由4.58%和4.37%降低至3.30%。
关键词:短期负荷预测;强化学习;径向基人工神经网络;
自组织映射;径向基中心
0 引言
作为电力系统运行与规划的基础课题,短期负荷
预测为经济调度、电力系统安全分析、电力市场交易
等提供不可或缺的重要依据
[1]
。因此,精确的负荷预
测技术一直受到学术界的广泛关注。
目前,众多方法已被应用到短期负荷预测,主要
可分为统计类方法和元启发式方法两类。统计方法
基于历史数据,应用概率统计、聚类和小波分析等
方法进行负荷预测,主要包括:时间序列、模糊聚
类、分类回归和小波分析等方法
[2-4]
。统计方法难以
准确模拟多种影响因素和负荷之间的函数关系,制
约了预测精度的提高。元启发式学习方法可从气象、
负荷等历史数据中挖掘温度等关键气象因素和负荷
的耦合关系,是目前负荷预测的主要方法和研究热
点。元启发式学习预测方法的主要代表方法是人工神
经网络(artificial neural network,ANN)和支持向量
机(support vector machine,SVM)
[5-6]
。其中,径向基
(radial basis function,RBF)神经网络因其泛化能力强、
基于强化自组织映射和径向基神经网络的短期负荷预测
黄乾,马开刚,韦善阳,黎静华
(广西电力系统最优化与节能技术重点实验室(广西大学),广西壮族自治区 南宁市 530004)
A Short-term Load Forecasting Method Based on Reinforcement Self-organizing Map and
Radial Basis Function Neural Network
HUANG Qian, MA Kaigang, WEI Shanyang, LI Jinghua
(Guangxi Key Laboratory of Power System Optimization and Energy-saving Technology (Guangxi University),
Nanning 530004, Guangxi Zhuang Autonomous Region, China)
基金项目:
国家重点研发计划(2016YFB0900100)。
National Key Research and Development Program of China
(2016YFB0900100).
全球能源互联网
Journal of Global Energy Interconnection
第 2 卷 第 1 期
2019 年 1 月
Vol. 2 No. 1
Jan. 2019
文章编号:2096-5125
(
2019
)
01-0070-08 中图分类号:TM714;TP183 文献标志码:A
DOI:10.19705/j.cnki.issn2096-5125.2019.01.009