摘要
针对当前电力工程数据分析处理过程中过于依赖人工且数字化程度较低的问题,文中基于改进的神经网络结构,提出了一种电力工程数据分析算法。该算法对基础卷积神经网络进行了改进,并利用多尺度卷积核增强了神经网络的感知野。而对于多尺度神经网络收敛速度慢的问题,在粗细尺度网络之间加入了残差网络,同时通过加入多维时空卷积注意力机制增强了数据的编码能力,进一步提高了模型的收敛速度。仿真测试结果表明,由迭代实验确定出最佳迭代次数后,所提算法的平均预测准确率和运行时间分别为97.5%及38.2 s,在对比方法中均为最优,综合性能较为理想,可以实现对电力工程数据的合理分析与准确预测。
In order to solve the problem of over reliance on manual work and low digitization in the current process of power engineering data analysis and processing,a power engineering data analysis algorithm based on the improved neural network structure is proposed in this paper.The algorithm improves the basic convolutional neural network,and enhances the perceptual field of the neural network by using multi-scale convolution kernel.For the problem of slow convergence of multi-scale neural networks,residual networks are added between coarse and fine-scale networks,and the multi-dimensional spatiotemporal convolution attention mechanism is added to enhance the coding ability of data,further improving the convergence speed of the model.The simulation test results show that the average prediction accuracy and running time of the proposed algorithm are 97.5%and 38.2 s respectively after the optimal number of iterations is determined by the iterative experiments,which are the best in the comparison methods.The comprehensive performance is relatively ideal,and the reasonable analysis and accurate prediction of power engineering data can be realized.
作者
王琼
吕征宇
薛礼月
WANG Qiong;LV Zhengyu;XUE Liyue(Economic and Technology Research Institute of State Grid Shanghai Electric Power Company,Shanghai 200233,China)
出处
《电子设计工程》
2024年第9期129-133,共5页
Electronic Design Engineering
基金
国家自然科学基金(71804045)。
关键词
多尺度卷积神经网络
残差网络
多维注意力机制
电力工程数据
造价管理
数据分析
multi-scale convolution neural network
residual network
multi-dimensional attention mech-anism
power engineering data
cost management
data analysis
