基于集成学习和卷积神经网络的电网客服短期话务量预测

doi:10.16183/j.cnki.jsjtu.2023.383

上海交通大学学报 ›› 2025, Vol. 59 ›› Issue (2): 266-273.doi: 10.16183/j.cnki.jsjtu.2023.383

• 新型电力系统与综合能源 • 上一篇下一篇

基于集成学习和卷积神经网络的电网客服短期话务量预测

覃浩¹^,², 苏立伟¹, 伍广斌¹, 蒋崇颖², 徐智鹏², 康峰¹, 谭火超¹, 张勇军²()

1.广东电网有限责任公司客户服务中心,广州 510699
2.华南理工大学电力学院,广州 510640

收稿日期:2023-08-08 修回日期:2023-09-28 接受日期:2023-11-07 出版日期:2025-02-28 发布日期:2025-03-11
通讯作者: 张勇军,博士,教授;E-mail: zhangjun@scut.edu.cn.
作者简介:覃浩(1976—),硕士,高级工程师,研究方向为电力市场营销及客户服务、电力营销大数据、智能客服.
基金资助:
国家自然科学基金(52177085);中国南方电网有限责任公司科技项目(036800KK52220003)

Short-Term Telephone-Traffic Prediction of Power Grid Customer Service Based on Adaboost-CNN

QIN Hao¹^,², SU Liwei¹, WU Guangbin¹, JIANG Chongying², XU Zhipeng², KANG Feng¹, TAN Huochao¹, ZHANG Yongjun²()

1. Customer Service Center, Guangdong Power Grid Co., Ltd., Guangzhou 510699, China
2. School of Electrical Power Engineering, South China University of Technology, Guangzhou 510640, China

Received:2023-08-08 Revised:2023-09-28 Accepted:2023-11-07 Online:2025-02-28 Published:2025-03-11

摘要/Abstract

摘要：

现代供电服务体系对用电客户服务的服务质量提出更高要求,精准的供电服务话务量预测不仅可以提高用电客户服务质量,还能有效降低客服人员成本.为此,基于集成学习和卷积神经网络提出一种电网短期话务量预测方法.首先,采用孤立森林算法进行异常数据识别,建立拉格朗日插值函数对异常数据或缺失数据进行修补;其次,利用层次分析法量化用户信息、气象信息和停电信息,采用灰色关联法分析话务量的影响因子,将影响因子作为话务量预测模型输入;然后,构建自适应增强(Adaboost)算法集成多个卷积神经网络(CNN)模型,提出一种Adaboost-CNN的话务量预测模型;最后,考虑供电服务系统增值服务,对预测结果进行修正,得到最终的话务量预测值.算例分析表明,所提预测模型较单一预测模型误差平均减少11.05个百分点、较组合预测模型误差平均减少5.32个百分点,具有更好的预测精度.

关键词: 现代供电服务体系, 话务量预测, Adaboost算法, 卷积神经网络, 孤立森林算法, 增值服务

Abstract:

The introduction of modern power supply service system has raised higher requirements for the service quality of electricity customer service. Accurate power supply service traffic prediction not only improves the quality of power customer service, but also effectively reduces the cost of customer service personnel. Therefore, this paper proposes a short-term traffic prediction method for power grid based on Adaboost and convolutional neural network (Adaboost-CNN) and a value-added service correction method. First, the isolated forest algorithm is used to identify the abnormal data, and the Lagrange interpolation function is applied to repair the abnormal data or missing data. Next, the analytic hierarchy process is employed to quantify user information, meteorological data, and power outage details. The grey correlation method is then used to analyze the influence factors of traffic volume, and these factors are incorporated as inputs to the traffic volume prediction model. An Adaboost algorithm is applied to integrate multiple CNN models, resulting in an Adaboost-CNN traffic prediction model. Finally, considering the value-added services within the power supply service system, the prediction results of the model are corrected to obtain the final traffic prediction value. The case analysis shows that the proposed forecasting model reduces prediction error by an average of 11.05 percentage points compared to a single forecasting model and by 5.32 percentage points compared to a combined forecasting model, demonstrating better forecasting accuracy.

Key words: modern power supply service system, traffic prediction, Adaboost, convolutional neural network (CNN), isolated forest algorithm, value-added services

中图分类号:

覃浩, 苏立伟, 伍广斌, 蒋崇颖, 徐智鹏, 康峰, 谭火超, 张勇军. 基于集成学习和卷积神经网络的电网客服短期话务量预测[J]. 上海交通大学学报, 2025, 59(2): 266-273.

QIN Hao, SU Liwei, WU Guangbin, JIANG Chongying, XU Zhipeng, KANG Feng, TAN Huochao, ZHANG Yongjun. Short-Term Telephone-Traffic Prediction of Power Grid Customer Service Based on Adaboost-CNN[J]. Journal of Shanghai Jiao Tong University, 2025, 59(2): 266-273.

图/表 7

表1

图1

图2

表2

表3

图3

表4

参考文献 23

[1]	谭刚, 陈聿, 彭云竹. 融合领域特征知识图谱的电网客服问答系统[J]. 计算机工程与应用, 2020, 56(3): 232-239. doi: 10.3778/j.issn.1002-8331.1907-0385
	TAN Gang, CHEN Yu, PENG Yunzhu. Hybrid domain feature knowledge graph smart question answering system[J]. Computer Engineering and Applications, 2020, 56(3): 232-239. doi: 10.3778/j.issn.1002-8331.1907-0385
[2]	李玮, 李树国, 喻玮. 基于停电事件分析的区域性话务峰涌预测[J]. 自动化技术与应用, 2024, 43(4): 9-13.
	LI Wei, LI Shuguo, YU Wei. Prediction of regional traffic rush based on blackout event analysis[J]. Techniques of Automation and Applications. 2024, 43(4): 9-13.
[3]	彭渤. 呼叫中心排班优化研究—以电力公司呼叫中心为例[D]. 天津: 天津大学, 2020.
	PENG Bo. Study on scheduling optimization of call center—Take the call center of Power Company as an example[D]. Tianjin: Tianjin University, 2020.
[4]	XIAO X P, DUAN H M, WEN J H. A novel car-following inertia gray model and its application in forecasting short-term traffic flow[J]. Applied Mathematical Modelling, 2020, 87: 546-570.
[5]	孙同川, 王振岭, 孙建设, 等. 基于Kalman滤波的原子时算法研究[J]. 计算机测量与控制, 2023, 31(3): 294-299.
	SUN Tongchuan, WANG Zhenling, SUN Jianshe, et al. Research on atomic time algorithm based on Kalman filter[J]. Computer Measurement & Control, 2023, 31(3): 294-299.
[6]	SONG Y, WANG H R. Real-time adjustment way of reservoir schedule forecasting projects based on improved variable oblivion factor least square arithmetic coupling Kalman filters[J]. Energy Reports, 2022, 8: 555-562.
[7]	秦艳辉, 马晓磊, 吴鑫, 等. 基于灰色滚动预测模型的多接口变换器功率直接控制方法[J]. 工业仪表与自动化装置, 2023(1): 62-66.
	QIN Yanhui, MA Xiaolei, WU Xin, et al. Direct power control method of multi interface converter based on grey rolling prediction model[J]. Industrial Instrumentation & Automation, 2023(1): 62-66.
[8]	LI Y, BAI X, LIU B. Forecasting clean energy generation volume in China with a novel fractional Time-Delay polynomial discrete grey model[J]. Energy and Buildings, 2022, 271: 112305.
[9]	万安平, 杨洁, 缪徐, 等. 基于注意力机制与神经网络的热电联产锅炉负荷预测[J]. 上海交通大学学报, 2023, 57(3): 316-325. doi: 10.16183/j.cnki.jsjtu.2021.346
	WAN Anping, YANG Jie, MIAO Xu, et al. Boiler load forecasting of CHP plant based on attention mechanism and deep neural network[J]. Journal of Shanghai Jiao Tong University, 2023, 57(3): 316-325.
[10]	HAN R, JIA Z H, QIN X Z, et al. Application of support vector machine to mobile communications in telephone traffic load of monthly busy hour prediction[C]// 2009 Fifth International Conference on Natural Computation. Tianjian,China: IEEE, 2009: 349-353.
[11]	赵龙, 周源, 李飞, 等. 基于XGBoost算法的坐席话务量预测[J]. 现代信息科技, 2021, 5(22): 86-88.
	ZHAO Long, ZHOU Yuan, LI Fei, et al. Call center seat telephone-traffic volume prediction based on XGBoost algorithm[J]. Modern Information Technology, 2021, 5(22): 86-88.
[12]	JALAL M E, HOSSEINI M, KARLSSON S. Forecasting incoming call volumes in call centers with recurrent Neural Networks[J]. Journal of Business Research, 2016, 69(11): 4811-4814.
[13]	黄雪婷. 基于SARIMA和CNN-LSTM组合模型的呼叫中心日话务量预测研究[D]. 南京: 南京邮电大学, 2022.
	HUANG Xueting. Research on call center daily traffic prediction based on SARIMA and CNN-LSTM combination model[D]. Nanjing: Nanjing University of Posts and Telecommunications, 2022.
[14]	林珊, 王红, 齐林海, 等. 基于条件生成对抗网络的短期负荷预测[J]. 电力系统自动化, 2021, 45(11): 52-60.
	LIN Shan, WANG Hong, QI Linhai, et al. Short-term load forecasting based on conditional generative adversarial network[J]. Automation of Electric Power Systems, 2021, 45(11): 52-60.
[15]	谢小瑜, 周俊煌, 张勇军, 等. 基于W-BiLSTM的可再生能源超短期发电功率预测方法[J]. 电力系统自动化, 2021, 45(8): 175-184.
	XIE Xiaoyu, ZHOU Junhuang, ZHANG Yongjun, et al. W-BiLSTM based ultra-short-term generation power prediction method of renewable energy[J]. Automation of Electric Power Systems, 2021, 45(8): 175-184.
[16]	龙干, 黄媚, 方力谦, 等. 基于改进多元宇宙算法优化ELM的短期电力负荷预测[J]. 电力系统保护与控制, 2022, 50(19): 99-106.
	LONG Gan, HUANG Mei, FANG Liqian, et al. Short-term power load forecasting based on an improved multi-verse optimizer algorithm optimized extreme learning machine[J]. Power System Protection and Control, 2022, 50(19): 99-106.
[17]	曾国治, 魏子清, 岳宝, 等. 基于CNN-RNN组合模型的办公建筑能耗预测[J]. 上海交通大学学报, 2022, 56(9): 1256-1261. doi: 10.16183/j.cnki.jsjtu.2021.192
	ZENG Guozhi, WEI Ziqing, YUE Bao, et al. Energy consumption prediction of office buildings based on CNN-RNN combined model[J]. Journal of Shanghai Jiao Tong University, 2022, 56(9): 1256-1261.
[18]	WANG Q, BU S Q, HE Z Y, et al. Toward the prediction level of situation awareness for electric power systems using CNN-LSTM network[J]. IEEE Transactions on Industrial Informatics, 2021, 17(10): 6951-6961.
[19]	朱吉然, 张帝, 张志丹, 等. 基于AHP和BP-Adaboost 的低压电力用户价值评价方法[J]. 电力科学与技术学报, 2022, 37(5): 155-163.
	ZHU Jiran, ZHANG Di, ZHANG Zhidan, et al. A value evaluation method of power user based on AHP and BP-Adaboost algorithms[J]. Journal of Electric Power Science and Technology, 2022, 37(5): 155-163.
[20]	游文霞, 申坤, 杨楠, 等. 基于AdaBoost集成学习的窃电检测研究[J]. 电力系统保护与控制, 2020, 48(19): 151-159.
	YOU Wenxia, SHEN Kun, YANG Nan, et al. Research on electricity theft detection based on AdaBoost ensemble learning[J]. Power System Protection and Control, 2020, 48(19): 151-159.
[21]	李国成, 陆俊, 王赟, 等. 基于Bagging二次加权集成的孤立森林窃电检测算法[J]. 电力系统自动化, 2022, 46(2): 92-100.
	LI Guocheng, LU Jun, WANG Yun, et al. Isolated-forest electricity theft detection algorithm based on Bagging secondary weighted ensemble[J]. Automation of Electric Power Systems, 2022, 46(2): 92-100.
[22]	杨少瑜, 黄国栋, 林星宇, 等. 基于拉格朗日插值法的概率建模方法及其在概率潮流分析中的应用[J]. 现代电力, 2021, 38(4): 378-385.
	YANG Shaoyu, HUANG Guodong, LIN Xingyu, et al. A Lagrange interpolation based probabilistic modeling method and its application in probabilistic power flow analysis[J]. Modern Electric Power, 2021, 38(4): 378-385.
[23]	马莉, 陈应雨, 田钉荣, 等. 基于改进层次分析法的多级电压暂降严重程度评估[J]. 电力系统保护与控制, 2023, 51(17): 49-57.
	MA Li, CHEN Yingyu, TIAN Dingrong, et al. Severity evaluation of multistage voltage sag based on an improved analytic hierarchy process[J]. Power System Protection and Control, 2023, 51(17): 49-57.

编辑推荐 0

Metrics

阅读次数

全文

299

HTML			PDF

最新录用	在线预览	正式出版	最新录用	在线预览	正式出版
0	0	4	224	0	71

来源	本网站	其他网站

次数	47	252
比例	16%	84%

摘要

136

最新录用	在线预览	正式出版

81	0	55

来源	本网站	其他网站

次数	48	88
比例	35%	65%

影响因素	量化值	影响因素	量化值
工作日	1	雷电橙色预警	3
周末	2	雷电红色预警	5
其他节假日	3	晴天	1
不停电	0	多云	2
计划检修停电	2	阴天	3
故障停电	4	小雨	4
无雷电	0	中雨	5
雷电蓝色预警	1	大雨	6
雷电黄预警	2	雪天	7

方法	异常数据识别准确率/%	方法	异常数据修正准确率/%
标准差法	85.6	平均值法	76.3
箱线图法	88.5	多重填补法	84.7
聚类法	93.3	线性插值	88.6
本文方法	100.0	本文方法	91.2

影响因素	相关性	是否为影响因子
最高温	0.643 1	是
最低温	0.447 2	否
天气	0.328 7	否
雷电	0.578 9	是
停电	0.863 4	是
风级	0.214 5	否
节假日	0.572 3	是

模型	e_MAE/通	e_MAPE/%
ARIMA	2 224.1	32.78
ELM	1 792.6	26.42
LSTM	1 380.1	19.34
Adaboost	995.4	14.67
CNN	1 302.0	15.21
单一预测模型平均误差	1 538.8	21.68
LSTM-CNN	1 176.9	17.35
EMD-CNN	974.6	14.37
组合预测模型平均误差	1 075.6	15.86
所提方法(不考虑增值服务)	850.8	12.54
所提方法(考虑增值服务)	721.2	10.63

基于集成学习和卷积神经网络的电网客服短期话务量预测

Short-Term Telephone-Traffic Prediction of Power Grid Customer Service Based on Adaboost-CNN

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 7

参考文献 23

相关文章 15

编辑推荐 0

Metrics

本文评价

[1]	Sahaya Anselin Nisha1, NARMADHA R.1, AMIRTHALAKSHMI T. M.2, BALAMURUGAN V.1, VEDANARAYANAN V.1. LOBO优化的深度卷积神经网络用于脑肿瘤分类[J]. J Shanghai Jiaotong Univ Sci, 2025, 30(1): 107-114.
[2]	蒋伊琳1, 2, 李向1, 2, 张昊平3. 基于卷积神经网络和接收信号强度的波束状态感知方法[J]. J Shanghai Jiaotong Univ Sci, 2024, 29(6): 1017-1022.
[3]	李明爱1, 2, 魏丽娜1. 基于朴素卷积神经网络和线性插值的运动想像分类[J]. J Shanghai Jiaotong Univ Sci, 2024, 29(6): 958-966.
[4]	陆文安, 朱清晓, 李兆伟, 刘辉, 余一平. 基于卷积神经网络的新型电力系统频率特性预测方法[J]. 上海交通大学学报, 2024, 58(10): 1500-1512.
[5]	詹可, 朱仁传. 一种CNN-LSTM船舶运动极值预报模型[J]. 上海交通大学学报, 2023, 57(8): 963-971.
[6]	朱昶胜1，康亮河1,3，冯文芳2. 基于AdaBoost-AAFSA-Elman模型及CEEMDAN算法的股市网络舆情收盘价预测[J]. J Shanghai Jiaotong Univ Sci, 2023, 28(6): 809-821.
[7]	李擎, 皇甫玉彬, 李江昀, 杨志方, 陈鹏, 王子涵. UConvTrans:全局和局部信息交互的双分支心脏图像分割[J]. 上海交通大学学报, 2023, 57(5): 570-581.
[8]	万安平, 杨洁, 缪徐, 陈挺, 左强, 李客. 基于注意力机制与神经网络的热电联产锅炉负荷预测[J]. 上海交通大学学报, 2023, 57(3): 316-325.
[9]	王者蓝, 赵宏杰, 赵凡, 沈晨晨, 吴佳伟. 基于卷积神经网络与滤波融合算法的某惯导系统剩余寿命预测模型建立[J]. 空天防御, 2023, 6(1): 70-77.
[10]	. 基于锥型体素建模和单目相机的鸟瞰图语义分割和体素语义分割[J]. J Shanghai Jiaotong Univ Sci, 2023, 28(1): 100-113.
[11]	曾国治, 魏子清, 岳宝, 丁云霄, 郑春元, 翟晓强. 基于CNN-RNN组合模型的办公建筑能耗预测[J]. 上海交通大学学报, 2022, 56(9): 1256-1261.
[12]	全大英, 陈赟, 唐泽雨, 李世通, 汪晓锋, 金小萍. 基于双通道卷积神经网络的雷达信号识别[J]. 上海交通大学学报, 2022, 56(7): 877-885.
[13]	吴庶宸, 戚宗锋, 李建勋. 基于深度学习的智能全局灵敏度分析[J]. 上海交通大学学报, 2022, 56(7): 840-849.
[14]	赵勇, 苏丹. 基于4种长短时记忆神经网络组合模型的畸形波预报[J]. 上海交通大学学报, 2022, 56(4): 516-522.
[15]	陶海红, 闫莹菲. 一种基于GA-CNN的网络化雷达节点遴选算法[J]. 空天防御, 2022, 5(1): 1-5.