基于残差卷积神经网络的温度敏感负荷辨识方法研究

Research on temperature⁃sensitive load identification method based on residual convolutional neural network

中文关键词: 负荷特性  基准负荷比较法  相关性分析  动态负 荷特征库  负荷辨识

英文关键词: load characteristics  benchmark load comparison  correlation analysis  dynamic load feature library  load identification

中文摘要:

      地区电网负荷特性易受环境温度影响，导致负荷辨 识结果往往存在较大偏差，研究了基于残差卷积神经网络的 温度敏感负荷辨识方法，有效提高负荷辨识准确率。首先，利 用基准负荷比较法，构建了商业各企业基准日负荷曲线；其 次，利用皮尔逊相关系数法，筛选出与温度相关性强的温度敏 感负荷，同时采用多项式回归模型进一步分析温度敏感负荷 与实时温度变化的规律，量化温度因素的影响程度；最后，针 对温度敏感负荷，提出利用负荷与温度的多项式回归模型系 数构建动态温度敏感负荷特征库，作为辨识模型的输入。将 基于残差卷积神经网络的负荷辨识结果与传统卷积神经网络 负荷辨识结果进行对比，前者的辨识准确率有较大提升。

英文摘要:

      The load characteristics of regional power grids are easily affected by the environmental temperature, which often results in large deviations in the load identification results. The temperature-sensitive load identification method based on residual convolutional neural network is studied to effectively improve the accuracy of load identification. Firstly, the benchmark load comparison method is used to construct the benchmark daily load curve of commercial enterprises. Secondly, the Pearson correlation coefficient method is used to screen out temperature-sensitive loads with strong temperature correlation, and a polynomial regression model is used to further analyze the temperature -sensitive load and the law of real-time temperature changes quantifies the degree of influence of temperature factors. Finally, for temperature - sensitive loads, a polynomial regression model coefficient of load and temperature is used to construct a dynamic temperature-sensitive load feature library as the input of the identification model. Comparing the load identification results based on the residual convolutional neural network with the traditional convolutional neural network load identification results, the identification accuracy of the former is greatly improved.