基于Bi-LSTM和多通道注意力的电网施工中违章量预测方法

Prediction method for quantity of violations in power grid construction based on Bi- LSTM and multi-channel attention

针对配电网施工中违章现象频发导致监管工作中出现的频繁调度问题,提出一种基于双向长短时记忆(Bi-LSTM)和多通道注意力机制的电网施工作业违章量预测方法。首先,通过Word-Embedding将文本转换为词向量;其次,通过Bi-LSTM网络来捕捉违章数量影响因素的浅层语义和时序特征;最后,设计多通道注意力机制,通过多组权重向量来学习多个影响因素以及各因素之间的复杂关系,从而提高模型的特征学习能力。以云南地区真实数据为算例开展实验,所提模型在一天预测中的平均绝对百分比误差(MAPE)与均方根误差(RMSE)两个指标分别达到了2.58%和5.2,从而验证了模型在预测精度和算法鲁棒性方面的优越性。该方法不依赖于任何具体情境,在实际工况环境中具有较好的推广价值。

In view of the frequent dispatch problems in supervision work caused by frequent violations in the construction of distribution networks,a method for predicting the quantity of violations in power grid construction based on Bi-directional Long Short-Term Memory(Bi-LSTM)and multi-channel attention mechanism was proposed.Firstly,text was transformed into word vectors by word-embedding.Secondly,the Bi-LSTM network was used to learn the shallow semantic and temporal characteristics of the factors affecting the number of power grid violations.Finally,a multi-channel attention mechanism was designed to learn multiple influencing factors and the complex relationships among them through multiple sets of weight vectors,so as to improve the feature learning ability of the model.Taking the real data of Yunnan as an example,the experimental results show that the average absolute percentage error and root mean square error of the proposed model reaches 2.58%and 5.2 respectively in daily prediction,which verifies the superiority of the model in prediction accuracy and algorithm robustness.The proposed method does not depend on any specific situation and has a good popularization value in the actual working conditions.