With the rapid growth of electric vehicle adoption,electric vehicle charging behavior has a significant impact on the stability of power grids and the charging load prediction.A short-term electric vehicle charging load combination prediction model is proposed based on K-means clustering and a backpropagation neural network with a decision tree regressor(BP-DTR).First,the K-means clustering method is used to cluster the charging load.Second,BP and DTR are employed to predict each group of the clustered data separately.Finally,the optimized method is utilized to combine the weighted prediction results of each group,which are then summed up.The experiments are based on real data,and the Monte Carlo method is used to obtain the electric vehicle charging load.The results show that this prediction model can consider the influencing factors of the charging load,effectively improving the accuracy of electric vehicle charging load predictions,and providing a reference for the optimal operation and planning of the power grid.
To improve the efficiency of transmission line target detection in unmanned aerial vehicle(UAV)inspections of power grids,an improved Res2Net-YOLACT transmission line detection method is proposed.The YOLACT model with Res2Net as the backbone is used to quickly and accurately extract features of transmission lines.Then,the detection accuracy and speed of the model are increased by improving the non-maximum suppression algorithm of the model.Experimental results based on a real public dataset show that the improved model achieves higher detection accuracy and better speed compared to the original model,and has better performance in transmission line detection.
针对居民家用电器使用环境安全问题,采用非侵入式负荷监测技术进行负荷监测.负荷识别方法是负荷监测的主要难题,采用优化卷积神经网络对负荷识别方法进行研究以克服负荷识别这一难题.基于U-I轨迹负荷特征,以卷积神经网络为基础,分析了卷积神经网络对U-I轨迹图自学习式提取特征识别过程;并采用粒子群优化算法,对卷积神经网络超参数配置进行优化,解决传统人工操作导致分类性能不确定性的问题,从而使卷积神经网络分类性能趋近于最优.研究表明:优化卷积神经网络具有很好的分类性能,可精确识别电器类型,完善负荷监测技术,为家用电器提供安全的使用环境,保障居民生命财产安全.
以实现图像特征定位为目的,设计了一种适用于三维非刚体图像的特征挖掘定位算法.利用摄像机拍摄目标的三维非刚体图像,通过重建非刚体三维结构获取运动目标的三维轨迹信息,再利用直线段检测算法挖掘运动目标三维轨迹的直线特征.将挖掘到的直线特征作为输入信息,利用图像消影点方法处理图像内目标的视觉里程数据,并推算三维非刚体图像内的目标位置,从而实现特征挖掘定位.结果表明:利用该算法构建的三维非刚体目标结构较为精确,提取直线特征信息时不易受图像模糊程度的影响,能够有效定位三维非刚体图像内的目标,定位误差数值较小.该方法具有较好的特征挖掘定位效果.
锂电池被广泛应用于工业生产,针对传统方法在检测锂电池极片缺陷时存在漏检和误检率高等问题,提出一种以YOLOv5s模型为基础网络进行改进的检测算法.首先,针对锂电池极片图像分辨率低且缺陷目标较小的问题,在主干网络引入空间深度层(SPD)和无步长卷积层(Conv),以提升小目标缺陷的检测性能;其次,将通道维度(SE)注意力机制嵌入C3 模块中,并将原YOLOv5s模型网络中的C3 模块替换为新的C3SE模块,以强化模型的特征表达能力;最后,在特征融合网络的最后一层引入注意力机制,以提高模型的特征融合能力.实验结果表明,改进模型与原YOLOv5s模型相比,平均精度均值达 96.8%,提升了 5.5%,能够更好地满足锂电池极片制造工业的需求.
