基于宽度学习的发电功率智能时间序列预测算法  

Intelligent Time Series Prediction Algorithm for Generating Power Based on Width Learning

作　　者：汪涛 袁晓鹏 申少辉 关英宇 WANG Tao;YUAN Xiaopeng;SHEN Shaohui;GUAN Yingyu(Beijing Kedong Power Control System Co.,Ltd.,Beijing 100194,China)

机构地区：[1]北京科东电力控制系统有限责任公司,北京100194

出　　处：《微型电脑应用》2024年第7期114-117,共4页Microcomputer Applications

摘　　要：发电功率预测受气象数据的影响较大,这可能导致功率预测值与实际值之间存在一定的偏差,为准确预测发电功率,提出基于宽度学习的发电功率智能时间序列预测算法。根据不同类型形成相应的数据集,分别对预测模型进行训练;使用模糊宽度学习替代原始的稀疏自动编码,利用时间序列模型进行非线性变换,利用非线性变换,形成增强节点层,并通过构造目标函数来建立发电功率预测模型;结合气象数据和宽度学习模型生成更可靠的数字孪生体功率预测结果。实验结果表明,该方法进行发电功率预测的归一化平均绝对误差为0.687%,归一化均方根误差为0.634%,相关系数为0.976,整体拟合程度较好,发电功率接近真实值,能够准确预测光伏发电功率,提供有价值的参考和决策支持。The prediction of generation power is greatly influenced by meteorological data,which may lead to a certain deviation between the power prediction value and the actual value.In order to accurately predict the generation power,an intelligent time series prediction algorithm for generation power based on width learning is proposed.Corresponding data sets are formed according to different types and are used to train the prediction model.The fuzzy width learning is used to replace the original sparse automatic coding,the time series model is used to perform nonlinear transformation to form enhanced node layer,and establish a generating power prediction model by constructing an objective function.By combining meteorological data and width learning model,a more reliable digital twin power prediction result can be obtained.The experimental results show that the normalized average absolute error of the power prediction is 0.687%,the normalized root mean square error is 0.634%,and the correlation coefficient is 0.976.The overall fitting degree is good,and the generating power is close to the true value,which can accurately predict the photovoltaic power,and provide valuable reference and decision support.

关 键 词：宽度学习 发电功率 时间序列 智能预测 SOM神经网络 聚类分析 

分 类 号：TM614[电气工程—电力系统及自动化]