摘要
提出了一种基于小波变换和群智能演化的神经网络集成预测新模型,对日前交易边际电价进行预测。首先利用小波变换将历史边际电价序列分解为高频和低频部分,并分别构造学习样本作为神经网络集成的输入;然后将边际电价预测问题转化为神经网络实际输出与预测输出误差最小化问题,其寻优过程采用粗—细二阶段学习算法。在第1阶段,采用粒子群优化算法把神经网络的结构和权重映射成问题空间中的粒子,通过粒子速度和位置更新方程进行粗学习,获得多个相对占优的神经网络结构和初始权重并构成神经网络集成单元;在第2阶段,采用梯度学习算法和交叉验证对神经网络集成单元的权重进行细学习,并以误差最小的神经网络集成单元的输出作为神经网络集成预测模型的输出。美国加州日前交易电力市场边际电价预测算例表明,该预测方法可以获得较高的预测精度,且优于BP神经网络方法和ARIMA预测方法。
A novel neural network ensemble （NNE） model based on wavelet transformation and swarm intelligence evolution is proposed to forecast market clearing price （MCP） in day-ahead electricity market. Firstly, MCP series is decomposed into low- frequency and high-frequency parts by wavelet transformation, and learning samples for NNE input are constructed. Then, MCP forecasting problem is converted into error minimization problem between actual output and desired output, and extensive and intensive learning algorithms are used in two stages. In the first stage, construction and weights of neural network are designed to be particles in problem space, neural network are extensively trained by particle velocity and position update equations of particle swarm optimization, and NNE units with different constructions and initial weights are constructed. In the second stage, weights in units are intensively trained by gradient learning algorithm and cross validation, and the unit output with minimal error is regarded as the output of NNE. The novel model is applied to MCP forecasting of California day-ahead electricity market. Results show that the model can achieve satisfactory forecasting accuracy and is superior to BP neural network and ARIMA forecasting model.
出处
《电力系统自动化》
EI
CSCD
北大核心
2007年第12期40-44,共5页
Automation of Electric Power Systems
关键词
电力市场
边际电价
小波变换
群智能
粒子群优化
人工神经网络
神经网络集成
electricity market
market clearing price
wavelet transformation
swarm intelligence
particle swarm optimization
artificial neural network
neural network ensemble
