基于监督式学习的自适应竞价预测模型研究  

Study on adaptive bidding prediction algorithm based on supervised learning

作　　者：初日辉 胡秦然[2] 时翔 李鹏 Chu Rihui;Hu Qinran;Shi Xiang;Li Peng(Nanjing SAC Power Grid Automation Co.,Ltd,Nanjing 211106;SEAS,Harvard University,Cambridge UK 02138;State Grid Qingdao Electric Power Company,Qingdao,Shandong 266002)

机构地区：[1]南京国电南自电网自动化有限公司,南京211106 [2]哈佛大学约翰.A.保尔森工程与应用科学学院,剑桥英国02138 [3]国网青岛供电公司,山东青岛266002

出　　处：《电气技术》2018年第10期1-5,9,共6页Electrical Engineering

基　　金：2017年江苏省工业和信息产业转型升级专项资金资助项目(苏经信综合[2017]378号)

摘　　要：电力市场改革初期,月度竞价由于可供研究的数据少、环境变化大、市场参与者行为不确定等特点难以分析预测。本文将监督式机械学习算法与当前电力市场规则和竞价者行为特征相结合,提出了一种具有自适应能力的竞价预测方法,引入遗忘机制和惯性机制来模拟真实市场参与者竞价行为,设计了自我验证机制修正不合理的预测结果,改进了正则化参数,有效避免了过拟合的发生。本文实验算例采用广东月度竞价的实验市场数据,验证了所提方法的有效性。Aimed at the characteristics of competitive bidding in the early stage of electricity market, which is less data available for research, the environment changes greatly and the user behavior is uncertain, a set of strategies for adaptive bidding supervised learning algorithm based on time series is proposed. The strategy combines the traditional mechanical learning methods, current electricity market rules and user behavior characteristics, the forgetting mechanism is used to simulate the maturing market behavior of users and the inertia mechanism is used to simulate the delayed response of users to the market. The self-verification mechanism is designed to correct the unreasonable forecasting errors. The regularization parameters avoid overfitting. The simulation results of the experimental example predict the supply-side curve and the demand-side curve, and verify the correctness and effectiveness of the proposed study based on the clearing result of the Guangdong electricity market.

关 键 词：自适应 监督式学习 电力市场 竞价 

分 类 号：F426.61[经济管理—产业经济]