面向电网海量数据的复杂人工智能模型小型化  

Miniaturization of Complex Artificial Intelligence Models for Massive Data in the Power Grid

作　　者：英晓勇 阙波 曹刚 李永欢 YING Xiaoyong;QUE Bo;CAO Gang;LI Yonghuan(Nanjing Suyi Industrial Co.,Ltd.,Nanjing 210000,China)

机构地区：[1]南京苏逸实业有限公司,南京210000

出　　处：《自动化与仪表》2025年第4期157-161,共5页Automation & Instrumentation

基　　金：国网江苏省电力有限公司省管产业单位科技项目(JC2024101)。

摘　　要：为更好处理电网海量数据,提出面向电网海量数据的复杂人工智能模型小型化方法。构建面向电网海量数据的小型化人工智能模型,将标准卷积替换为深度可分离卷积,减少深度学习模型的参数和计算量,实现主体结构小型化处理;并引入小型化残差注意力模块和小型化多尺度特征融合模块,提升网络对海量电网数据中有价值特征的提取效果,融合多尺度特征促进电网海量数据分析处理精度。实验结果显示,该方法可以降低约85%以上的参数量,规模缩减程度约在86%,在不影响模型电网异常检测精度的条件下可以显著提升模型实时性,降低内存占用及能耗。In order to handle the massive data of the power grid,a method for miniaturizing complex artificial intelligence models targeting massive data of the power grid is proposed.Building a miniaturized artificial intelligence model for massive data in the power grid,replacing standard convolutions with depthwise separable convolutions,reducing the parameters and computational complexity of deep learning models,and achieving miniaturization of the main structure.Introduce a miniaturized residual attention module and a miniaturized multi-scale feature fusion module to improve the network's ability to extract valuable features from massive power grid data,and integrate multi-scale features to promote the accuracy of power grid data analysis and processing.The experimental results show that this method can reduce the number of parameters by more than 85%,with a scale reduction of about 86%.It can significantly improve the real-time performance of the model,reduce memory usage and energy consumption without affecting the accuracy of power grid anomaly detection.

关 键 词：电网海量数据 人工智能模型 深度学习 小型化 深度卷积 注意力机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]