作者：廖清阳1, 王军1, 胡凯强2, 宋尧1, 宗志亚1, 范俊秋1
作者单位：1. 贵州电网有限责任公司贵安供电局，贵州 贵安 550025;
2. 华南理工大学，广东 广州 510599
关键词：电负荷;热负荷;联合预测;卷积神经网络;双向长短期记忆网络
摘要：
为提升电、热负荷预测能力，适应能源互联网的多能源互联状态，设计由双向长短期记忆网络和并行卷积神经子网络组成的深层次电、热负荷联合预测模型，以便联合捕获强依赖性、多维度、抽象的电、热负荷特征信息。经仿真可知，该联合预测模型的综合负荷误差均值比低于串行网络模型约3%；其联合预测方式的综合负荷误差均值比低于单负荷预测模型约3%，同时其训练时长和预测时长均约为单负荷预测模型的一半；与同领域RNN-Net、LSTM-Net、CNN-Net、Shi-Net相比，其电负荷、热负荷、综合负荷误差均值比最低，分别为0.0315、0.0301和0.0311。说明本模型与串行网络相比，其并行网络有利于捕获多层次负荷特征；与单负荷预测方式相比，其联合预测方式可采用较高效率捕获电、热负荷的互联信息；它具备较优的电、热负荷联合预测性能，适用于电负荷、热负荷联合预测任务。
Joint forecasting model of electrical load and thermal load based on deep parallel CNN-BiLSTM in energy internet
LIAO Qingyang1, WANG Jun1, HU Kaiqiang2, SONG Yao1, ZONG Zhiya1, FAN Junqiu1
1. GuiAn Power Supply Bureau of Guizhou Power Grid Co., Ltd., Guian 550025,China;
2. South China University of Technology, Guangzhou 510599,China
Abstract: In order to improve the power and heat load forecasting ability and adapt to the multi energy interconnection state of the energy internet, a deep-seated electric and thermal load forecasting model composed of bidirectional long-term memory network and parallel convolution neural sub network is designed to capture the strong dependence, multi-dimensional and sampled characteristic information of electric and thermal load. The simulation results show that the average percentage of comprehensive load error of the joint forecasting model was about 3% lower than that of the serial network model. The average percentage of the comprehensive load error of the joint forecasting model was about 3% lower than that of the single load forecasting model, and its training time and prediction time are about half of that of the single load forecasting model. Compared with RNN-Net, LSTM-Net, CNN-Net and Shi-Net in the same field, the average error percentage of electric load, heat load and comprehensive load in this model was the lowest, which are 0.0315, 0.0301 and 0.0311, respectively. It shows that compared with the serial network, the parallel network of this model is advantageous to capture the multi-level load characteristics. Compared with the single load forecasting mode, its joint forecasting mode can capture the interconnection information of electric and thermal loads with higher efficiency. The model has better joint forecasting performance of electric and thermal loads, and is suitable for joint forecasting tasks of electric and thermal loads.
Keywords: electrical load;thermal load;joint forecasting ;convolution neural network;bidirectional long short term memory network
2022, 48(4):146-153 收稿日期: 2020-08-20;收到修改稿日期: 2020-11-20
基金项目: 贵州电网有限责任公司科技项目（061000KK52180003）
作者简介: 廖清阳（1984-），男，贵州贵安人，高级工程师，主要研究方向为电力系统保护与控制
参考文献
[1] 付学谦, 周亚中, 孙宏斌, 等. 园区农业能源互联网在线安全分析: 评述与展望[J/OL]. 中国电机工程学报: 1-8[2020-05-03]. https://doi.org/10.13334/j.0258-8013.pcsee.200108.
[2] 朱赫炎, 张明理, 郭笑林, 等. 计及分布式光伏电源出力影响的母线净负荷预测[J/OL]. 电测与仪表: 1-7[2020-05-03]. http://kns.cnki.net/kcms/detail/23.1202.TH.20200429.1548.008.html.
[3] 祝学昌. 基于IFOA-GRNN的短期电力负荷预测方法研究[J/OL]. 电力系统保护与控制: 1-7[2020-05-03]. https://doi.org/10.19783/j.cnki.pspc.190760.
[4] 吕海灿, 王伟峰, 赵兵, 等. 基于Wide&Deep-LSTM模型的短期台区负荷预测[J]. 电网技术, 2020, 44(2): 428-436
[5] 刘晓悦, 魏宇册. 基于改进灰色关联分析的BA-BP短期负荷预测[J]. 科学技术与工程, 2020, 20(1): 223-227
[6] 王浩林, 张勇军, 毛海鹏. 基于时空特征变量数据分析的共享汽车充电负荷预测方法[J]. 电力自动化设备, 2019, 39(12): 169-175
[7] 陈培垠, 方彦军. 基于卡尔曼滤波预测节假日逐点增长率的电力系统短期负荷预测[J]. 武汉大学学报(工学版), 2020, 53(2): 139-144
[8] 代强, 乔焰, 程曦, 等. 基于深度学习的农作物病害叶片的图像超分辨率重建[J]. 黑龙江八一农垦大学学报, 2020, 32(2): 82-90
[9] 董亚超, 刘宏哲, 徐成. 基于显著性多尺度特征协作融合的行人重识别方法[J/OL]. 计算机工程: 1-14[2020-05-05]. https://doi.org/10.19678/j.issn.1000-3428.0057938.
[10] 齐玉东, 丁海强, 赵锦超, 等. 基于biRNN的海军军械不均衡文本数据集处理方法[J]. 计算机与现代化, 2019(12): 21-26
[11] 闫昆鹏. 基于BiRNN的学生手写文字识别系统设计与实现[D]. 北京：北京工业大学, 2019.
[12] 帕丽旦·木合塔尔, 买买提阿依甫, 杨文忠, 等. 基于BiRNN的维吾尔语情感韵律短语注意力模型[J]. 电子科技大学学报, 2019, 48(1): 88-95
[13] 胡庆芳, 曹士圯, 杨辉斌, 等. 汉江流域安康站日径流预测的LSTM模型初步研究[J]. 地理科学进展, 2020, 39(4): 636-642
[14] 曾安, 赵恢真. 融合了LSTM和PMF的推荐算法[J/OL]. 计算机工程与应用: 1-11[2020-05-03]. http://kns.cnki.net/kcms/detail/11.2127.TP.20200422.1614.016.html.
[15] 勾玄, 肖先勇. 基于经验模式分解与LSTM神经网络的短期电价预测模型[J/OL]. 西安理工大学学报: 1-8[2020-05-03]. http://kns.cnki.net/kcms/detail/61.1294.N.20200421.1719.019.html.
[16] 戴邵武, 陈强强, 刘志豪, 等. 基于EMD-LSTM的时间序列预测方法[J/OL]. 深圳大学学报(理工版): 1-6[2020-05-03]. http://kns.cnki.net/kcms/detail/44.1401.N.20200417.1011.002.html.
[17] 郑延斌, 韩梦云, 樊文鑫. 基于2DPCA与CNN的手写体汉字识别[J/OL]. 计算机应用: 1-8[2020-05-03]. http://kns.cnki.net/kcms/detail/51.1307.TP.20200429.1626.006.html.
