摘要
为了精准识别变速恒频风电机组的运行状态,提出基于mRMR-GRU的变速恒频风电机组图像识别算法。通过分析变速恒频风电机组机械动态结构,获取整体变速恒频风电机组运行的重要控制部分;基于物联网技术,集成智能终端传感设备在特定结构中采集变速恒频风电机组图像;分析风电机组特定结构,生成变速恒频风电机组原始图像;基于直方图均衡方法对变速恒频风电机组图像进行增强处理,输出高分辨率采集图像;基于mRMR算法,定义特征变量的互信息,获得最具代表性和相关性的特征子集;融合GRU单元与mRMR值生成mRMR-GRU算法,优化图像识别网络结构,输出的风机图像状态识别结果。实验结果表明:设计算法MAPE值为4.46,RMSE值为3.57,图像识别准确度最高值为0.98,能够有效调节图像的特征变量,有效识别变速恒频风电机组图像存在的异常情况,且具有明显的图像像素点均衡效果与图像增强识别效果。
In order to accurately recognize the operating state of variable-speed constant-frequency wind turbines,a variable-speed constant-frequency wind turbine image recognition algorithm based on mRMR-GRU was proposed.By analyzing the mechanical dynamic structure of variable-speed and constant-frequency WTGs,the important control parts of the overall variable-speed and constant-frequency WTGs operation were obtained;based on the Internet of Things(IoT)technology,the integrated intelligent terminal sensing equipment collected the variable-speed and constant-frequency WTGs imaged in a specific structure;analyzing the specific structure of WTGs generated the original images of the variable-speed and constant-frequency WTGs;based on the histogram equalization method,the image of the variable-speed and constant-frequency WTGs was enhanced processing,the output of high-resolution acquisition of images;based on the mRMR algorithm,the definition of the mutual information of feature variables,to obtain the most representative and relevant feature subset;fusion of GRU units and mRMR values to generate mRMR-GRU algorithm,optimize the image recognition network structure,the output of the wind turbine image state recognition results.The experimental results show that:the designed algorithm has a MAPE value of 4.46,an RMSE value of 3.57,and the highest value of image recognition accuracy is 0.98,which is enough to effectively regulate the feature variables of the image,effectively identify the abnormalities that exist in the image of variable-speed and constant-frequency WTGs,and it has obvious image pixel-point equalization effect and image enhancement recognition effect.
作者
贾洪岩
亢涵彬
刘玉龙
郭龙雨
闫志强
JIA Hongyan;KANG Hanbin;LIU Yulong;GUO Longyu;YAN Zhiqiang(State Grid Jibei Zhangjiakou Wind and Solar Energy Storage and Transportation New Energy Co.,LTD.,Zhangjiakou Hebei 075000,China;Dongfang Electric New Energy Technology(Chengdu)Co.,LTD.,Chengdu 610036,China)
出处
《微电机》
2024年第9期56-62,共7页
Micromotors
关键词
变速恒频风电机组
图像识别
机械动态结构
图像采集
直方图均衡方法
variable-speed constant-frequency wind turbines
image recognition
mechanical dynamic structure
image acquisition
histogram equalization methods