基于改进SAX算法与贝叶斯超参数优化的配电网负荷-馈线智能匹配方法

An intelligent load-feeder matching method of distribution networks based on an improved SAX algorithm and Bayesian hyperparameter optimization

新型电力系统下配电网运行方式调整愈来愈频繁,配电网负荷-馈线匹配面临采样数据高维异构且价值密度低、现有匹配算法对负荷物理特征依赖度高、参数设置灵活性弱等难点,为此提出一种基于改进SAX(符号聚合近似)算法与贝叶斯超参数优化的配电网负荷-馈线智能匹配方法。首先,建立面向离散符号化时间数据序列的数据价值提升模型,将高维异构的数据近似表示为低维统一的符号,修正和填充异常数据、空白数据。其次,构建改进CNN-LSTM(卷积神经网络-长短期记忆)混合神经网络,对负荷数据进行所属馈线匹配分类训练,利用多头注意力机制深入挖掘负荷数据的潜在数学关系,降低对负荷物理特征的依赖度。然后,引入贝叶斯超参数优化算法对神经网络训练参数进行逐次更新,提高馈线拓扑变化时神经网络模型的灵活性与适应性。最后,对某地区100条馈线进行负荷匹配实验验证,结果证明所提方法较传统方法具有更高的匹配精度。

As the operation mode of distribution networks is adjusted more and more frequently in the new-type power system,the load-feeder matching of distribution networks meets with difficulties such as high-dimensional heterogeneity and low value density of sampled data,high dependence of existing matching algorithms on physical characteristics of load,and poor flexibility of parameter settings,etc.To this end,an intelligent load-feeder match⁃ing method of distribution networks based on an improved SAX(symbolic aggregation approximation)algorithm and Bayesian hyperparameter optimization is proposed.Firstly,a data value enhancement model for discrete symbolic time data series is established to approximate the high-dimensional heterogeneous data into low-dimensional uniform symbols to correct and fill the abnormal data and blank data.Secondly,an improved CNN-LSTM(convolutional neu⁃ral network-long short-term memory)hybrid neural network is constructed to train the load data as per the matching,and the multi-headed attention is employed to explore the potential mathematical relationships between load and data to curtail dependency on physical features of load.Thirdly,a Bayesian hyperparameter optimization algorithm is introduced to update the neural network training parameters one by one to improve the flexibility and adaptability of the neural network model in the case of feeder topology changes.Finally,the proposed method is experimentally validated by matching between load and 100 feeders in a region.The results prove that the proposed method is supe⁃rior to the traditional method in matching accuracy.