Loss function improvement method and system for power grid data

The invention discloses a loss function improvement method and system for power grid data, and the method comprises the steps: marking a data category, and forming a power grid data classification sample; classifying the classification samples into a training set, a verification set and a test set in proportion; and designing a variant maximum Markov center loss function, inputting data into the model for deep learning training, and continuously iterating and controlling the training process. According to the method, the distance between the categories is increased, and meanwhile, the distance between the samples in the same category is reduced, so that the distinction degree between the categories is increased, and meanwhile, the sample features are reflected into a relatively balanced feature space. 一种面向电网数据的损失函数改进方法及系统，对数据类别进行标注，形成电网数据分类样本。将分类样本按比例分为训练集，验证集和测试集。设计变体最大马氏中心损失函数，将数据输入模型进行深度学习训练，不断迭代并控制训练过程。本发明拉大各类别之间的距离，同时也缩小同类别中样本之间的距离，以增大类别之间区分度，同时使样本特征反映到相对平衡的特征空间中。