基于深度混合专家模型的电力气象识别技术

Power meteorological recognition technology based on deep mixture-of-expert model

对电能传输和设施运行产生严重干扰的天气现象称为高影响天气现象,然而目前基于人工观测高影响天气现象的方法存在着时效低、耗费人力等缺陷。针对此问题,提出一种基于混合专家模型深度学习的电力图像高影响天气现象识别算法。通过对密集连接网络的预训练,实现了图像判别式特征的提取,采用混合专家模型对天气现象进行初步分类,联合优化两部分模型参数,设计生成了对电力图像中的高影响天气现象识别模型。基于高影响天气现象识别数据集对所提模型进行测试验证,算例结果表明,所提方法的图像识别耗时缩减了约75%,同时在模型稳定性和识别准确程度方面也具有一定优势。

The weather phenomenon that seriously interferes with the transmission of electricity and the operation of facilities is called a high impact weather phenomenon.However,current methods based on manual observation of high impact weather phenomena have shortcomings such as low efficiency and labor-intensive.In response to this issue,a hybrid expert model based deep learning algorithm for identifying high impact weather phenomena in power images is proposed in the article.Through the pre training of the dense connection network,the feature extraction of the image discriminant is realized.The mixed expert model is used to preliminarily classify the weather phenomena,and the parameters of the two parts of the model are jointly optimized to design and generate a recognition model for the high impact weather phenomena in the power image.The proposed model was tested and validated based on a high impact weather phenomenon recognition dataset.The numerical results show that the proposed method reduces the image recognition time by about 75%,and also has certain advantages in model stability and recognition accuracy.