主动配电网获得的海量电能质量数据需要进行高效、低损压缩，以降低通信与存储压力。电能质量信号可通过周期间相似性测度进行高效压缩，但易受噪声干扰。为提高测度压缩方法在配电网内抗噪能力，提出基于单类支持向量机（OCSVM）与归一化距离测度的电能质量信号压缩方法。首先，通过仿真，获取高噪声环境下故障样本，训练以1/4周期信号归一化距离测度和信号信噪比为输入的 OCSVM，并改进 OCSVM 的误差限 v 和径向基函数（RBF）核函数宽度参数 c，以提高扰动检测能力；之后通过噪声估计方法，估计待压缩的信号信噪比。如信号信噪比较高，则采用相邻两周期内对应1/4周期信号归一化距离测度阈值，进行周期化压缩。否则，采用 OCSVM，判定低信噪比信号内是否新发生扰动并开展压缩。仿真与实测配电网电能质量信号证明，新方法能够在不同噪声环境下，有效地压缩电能质量信号。%In order to reduce the communication and storage costs, the massive power quality (PQ) data obtained from the active distribution network needs to be compressed efficiently with low loss. Although the PQ signal can be efficiently compressed by the similarity of different cycle, it is sensitive to noise and difficult to be applied under high-noise environments. Aim to improve the noise immunity, a PQ signal compression method based on one class support vector machine (OCSVM) and normalized distance (ND) was proposed. Firstly, PQ signals with high noise were simulated to extract features. The feature set composed with 1/4 periodic ND and signal-noise-ratio (SNR) of PQ signals was used to train the OCSVM. The characters of OCSVM including bound on error (v) and the width of RBF Kernel function (c) were optimized to improve the disturbance detection ability. Then, the SNR of PQ signals were estimated to noise estimation method. The periodic compression of PQ signals with high SNR were carried out by using ND threshold from the adjacent 2 cycle corresponding to the 1/4 cycle signal. Otherwise, the OCSVM was adopted to determine whether a new disturbance occurs in the signals with low SNR and carry out data compression. Experiments using simulation and real PQ signals in distribution network show that the new method can effectively compress the PQ signals in different noise environments.
展开▼
