PQ Issues Dataset Using Short Time Fourier Transform

Power Quality (PQ) is now one of the most important issues of a modern power system because of the increased reliance on sensitive electrical and electronic equipment, integration of renewable energy, and sophisticated industrial loads. In order to identify the type of PQ issues in the system using AI techniques, there is a need for data. The dataset required to train AI models is developed using the Short Time Fourier Transform (STFT) technique. While developing the PQ dataset using STFT, we consider a total of 5 issues, i.e., sag, swell, interruption, transients, and harmonics. For each issue, 100 samples of data with 2091 features are extracted using STFT. This is the original dataset. Later, we reduced the number of features based on standard deviation, and that reduced dataset has a total of 792 features for each issue. Further, the dataset is reduced in terms of features based on correlation. And that optimal dataset has 500 samples with 44 features. Each PQ issue is represented with a numeric value. 0: Sag, 1: Swell, 2: Harmonics, 3: Interruption, 4: Transients电能质量（Power Quality, PQ）如今已成为现代电力系统最为关键的议题之一，这源于电力系统对敏感电气电子设备依赖度的提升、可再生能源并网的推进，以及复杂工业负荷的广泛应用。若要借助人工智能技术识别系统内的电能质量扰动类型，便需要配套的数据集支撑。用于训练人工智能模型的数据集，通过短时傅里叶变换（Short Time Fourier Transform, STFT）技术构建得到。在通过短时傅里叶变换构建电能质量数据集时，本次研究共涵盖5类电能质量扰动，即电压暂降、电压暂升、电压中断、暂态冲击与谐波畸变。针对每一类扰动，我们通过短时傅里叶变换提取了100组数据样本，每组样本包含2091个特征，此为原始数据集。后续，我们基于标准差对特征维度进行了裁剪，经此处理后的精简数据集，每类扰动对应的样本特征总数降至792个。进一步地，我们又基于特征相关性对数据集进行了特征维度精简，最终得到的最优数据集包含500组样本，每组样本仅含44个特征。每类电能质量扰动均通过数值进行标注：0代表电压暂降，1代表电压暂升，2代表谐波畸变，3代表电压中断，4代表暂态冲击。