A novel dynamic modeling methodology for meteorological sensitive load forecasting of smart grids is proposed by using Fisher information theory and online support vector regression (OSVR) technology to improve load prediction accuracy in this paper. According to the changes in the operating state of the smart grids and streaming data characteristics, the OSVR model is employed to implement accurate online training algorithms to avoid retraining the entire training data whenever a sample is added to or removed from the training set. On the other hand, the paper originally utilizes Fisher Information theory to address the introduction of weather factors into the meteorological sensitive load prediction model and feature selection for it. We also present a practical and concise implementation of the proposed methodology. We demonstrate the application of the proposed methodology in the meteorological sensitive load prediction for the local utilities in different periods and compared it with the traditional method. The results indicate that the forecast model constructed by the proposed methodology can obtain the superior prediction performance among the conventional SVR models.

利用Fisher信息论和在线支持向量回归（OSVR）技术，提出了一种新颖的动态建模方法，用于智能电网的气象敏感负荷预测，以提高负荷预测的准确性。根据智能电网的运行状态和流数据特性的变化，采用OSVR模型来实现精确的在线训练算法，从而避免在将样本添加到训练集中或从训练集中删除样本时重新训练整个训练数据。另一方面，本文最初利用Fisher信息理论解决了天气因素在气象敏感负荷预测模型中的引入及其特征选择。我们还提出了一种实用而简洁的方法。我们证明了所提出的方法在不同时期对当地公用事业的气象敏感负荷预测中的应用，并将其与传统方法进行了比较。结果表明，所提出的方法构建的预测模型可以在常规的SVR模型中获得较好的预测性能。