【作者】 肖爽；

【导师】 李扬； 刘娟；

【作者基本信息】 东南大学 ， 电气工程（专业学位）， 2022， 硕士

【摘要】 本文基于国网客服中心数据,进行了话务预测模型方面的研究,研究内容主要包括以下几个方面:1、研究了话务量的理论基础,包括话务量的定义、话务预测的原理、话务预测的分类、话务预测工作的步骤及话务预测结果的误差分析方法;2、分析了电力话务量的分类,确认了影响话务量的主要关键因素,即欠费停电户数、故障停电数、时间、气温四方面;3、搭建了长短期记忆网络模型(Long-short Term Memory,LSTM),并创新性结合了卷积神经网络(Convolutional Neural Networks,CNN),构建了卷积长短期记忆网络模型(Convolutional Long-short Term Memory,CLSTM)模型,然后引入自注意力机制,最终形成了(Convolutional Long-short Term Memory Based On Self-Attention,ACLSTM)模型。选取业务差异显著的湖北、江苏作为典型省份进行算例分析,将ACLSTM模型的预测结果与CLSTM、LSTM模型进行比较,证实ACLSTM模型可以高准确度进行话务量预测,同时预测性能显著优于CLSTM和LSTM模型;4、将话务量预测ACLSTM模型与Erlang-A公式相结合,建立了呼叫中心人员调控模型,并证明了模型对呼叫中心运营管理的重要作用;5、对未来研究方向进行了展望,主要包括ACLSTM模型在台风等特殊事件时的预测效果验证等方面。综上所述,本研究是深度学习在话务预测领域的又一创新应用,同时该模型具有较强泛化性,对深度学习在各领域的应用提供重要的参考。更多还原

【Abstract】 Based on the data from State Grid Customer Service Center,this thesis conducts a research on traffic forecasting models,which mainly includes the following aspects:1.The theoretical basis of traffic is studied,including the definition of traffic,the principle of traffic forecasting,the classification of traffic forecasting,the steps of traffic forecasting and the error analysis method of traffic forecasting results.2.The classification of electrical traffic is analyzed,and the main key factors affecting the traffic,namely,the number of customers who are in power failure due to electricity arrears,the number of fault outages,time and temperature are confirmed.3.A Long-short Term Memory(LSTM)model is built,and innovatively Convolutional Neural Networks(CNN)is combined to construct a Convolutional Long-Short Term Memory(CLSTM)model.Then the self-attention mechanism is considered,and finally the Convolutional Long-short Term Memory Based On Self-Attention(ACLSTM)model is formed.Selecting Hubei and Jiangsu with significant business differences as typical provinces to analyze,compare the prediction results of the ACLSTM model with the CLSTM and LSTM models,it is confirmed that the ACLSTM model can predict traffic with high accuracy,and the prediction performance is significantly better than CLSTM And LSTM model.4.Combining the traffic forecast ACLSTM model with the Erlang-A formula,a call center personnel regulation model is established,and it proves that the model plays an important role in the operation and management of the call center.5.Prospects for future research directions mainly include the verification of the ACLSTM model’s prediction effect during special events such as typhoons.In summary,this research is another innovative application of deep learning in the field of traffic forecasting.At the same time,the model has strong generalization and provides an important reference for the application of deep learning in various fields.更多还原