基于数据驱动的组合电价预测研究  

Research on Combined Price Forecasting Based on Data Driven

作　　者：张亮 王洪涛 邹斌[1] Zhang Liang

机构地区：[1]上海大学机电工程与自动化学院,上海200444 [2]宁德师范学院信息与机电工程学院,福建宁德352100

出　　处：《工业控制计算机》2019年第11期61-63,共3页Industrial Control Computer

基　　金：国家电网科技项目：电力金融产品设计与定价技术研究;福建省自然科学基金项目(2019J01845)

摘　　要：在解除管制的电力市场中,电价预测是核心。在不考虑电力市场本身情况及影响电价的诸多因素的基础上,单纯从数据驱动角度出发,以历史负荷和电价数据作为输入来预测电价。首先构造Lasso、随机森林、Gradient Boosting、SVM、BP神经网络和LSTM六种单算法电价模型,然后再构建组合六种算法的Lasso、BP神经网络和LSTM组合模型,以及组合三种算法的BP神经网络模型,最后以澳大利亚昆士兰州电力市场历史数据进行仿真。实验结果表明:单算法电价模型中,LSTM模型精度最高,MAE为5.468;BP神经网络适合用于组合单算法电价模型。First Lasso,random forests and Gradient Boosting the SVM and BP neural network model and LSTM six kinds of single algorithm electricity price models are construct,and then the combination model of Lasso,BP neural network and LSTM combining six algorithms and the combination model of BP neural network model combining three kinds of algorithm are constructed.The actual electricity price and load data from Queensland were used for simulation,and the simulation results shows that:Among the single algorithmic electricity price models,the LSTM model has the highest accuracy,MAE is 5.468.BP neural network is suitable for combining the single algorithm electricity price model.The BP neural network model with the combination of the six algorithms has the highest accuracy.

关 键 词：电价预测 LSTM 组合预测 

分 类 号：F42[经济管理—产业经济]