基于一维卷积神经网络多任务学习的电能质量扰动识别方法

传统电能质量识别需要先用信号处理技术提取信号特征，且已有的多分类和多标签分类建模方式没有很好地反映多重扰动和单扰动之间的标签关联性，使得复合扰动分类的鲁棒性和抗噪性能不理想。针对这些问题，提出了一种基于多任务学习的一维卷积神经网络模型来识别各种电能质量扰动。此结构去除了传统方法的信号特征提取阶段，将扰动分类任务分成四个子任务，设计了相应的标签编码方案，最后输出一个10维标签向量完成多任务分类。仿真结果表明，该方法在不同信噪比时均具有较好的识别准确率，表明此模型具有较强的鲁棒性和抗噪声能力。同时，多任务分类相比One-hot多分类和多标签分类准确率更高，表明了该建模方式的有效性。

Power quality disturbance recognition method based on multi-task learning and one-dimensional convolutional neural network

Traditional power quality identification needs to use signal processing technology to extract signal features first, and the existing multi-class and multi-label classification modeling methods do not reflect the label correlation between multiple disturbances and single disturbances well, making the composite disturbance classification robust. The stickiness and noise resistance are not ideal. In response to these problems, a one-dimensional convolutional neural network model based on multi-task learning is proposed to identify various power quality disturbances. This structure removes the signal feature extraction stage of the traditional method, divides the disturbance classification task into four sub-tasks, designs the corresponding label coding scheme, and finally outputs a 10-dimensional label vector to complete the multi-task classification. Simulation results show that the method has good recognition accuracy in each signal-to-noise ratio, which shows that this model has strong robustness and anti-noise ability. At the same time, multi-task classification is more accurate than One-hot multi-class and multi-label classification, indicating the effectiveness of this modeling method.