摘要
为了有效提高线损计算精度,本文提出一种基于模态分解和小波包的台区线损改进变权组合预测模型。首先使用完整集合经验模态分解讲与实践序列相关的电气特征参数分解成多个不同的子序列,通过傅立叶变换计算子序列的幅值方差,结合小波包阈值去噪对含有噪音的高频信号进行降噪处理,提取代表性变量作为预测输入。最后改进标准遗传算法,利用遗传算法的全局优化能力,对广义回归神经网络和门控循环单元构成的组合预测模型的权重系数进行样本移动自适应变权求解,将得到的权重系数进行加权求和,得到了最终线损值。以某市具体台区数据为例进行计算验证,对比结果表明,本文所提的改进变权组合预测模型比单一模型及传统变权组合预测模型具有更高的计算精度。
In order to improve the accuracy of line loss calculation,an improved variable weight combination prediction model based on modal decomposition and wavelet packet is proposed in this paper.Firstly,the complete set empirical mode decomposition is used to decompose the electrical characteristic parameters related to the practical sequence into several different subsequences.The amplitude variance of the subsequences is calculated by Fourier transform,and the high-frequency signal with noise is denoised by wavelet packet threshold denoising,and representative variables are extracted as prediction input.Finally,the standard genetic algorithm is improved,and the weight coefficient of the combination prediction model composed of generalized regression neural network and gating cycle unit is solved by using the global optimization ability of genetic algorithm.The weight coefficients obtained are weighted and summed to obtain the final line loss value.The first mock exam is conducted in a specific station area of a city.The results show that the improved variable weight combination forecasting model has higher accuracy than single model and traditional variable weight combination forecasting model.
作者
周彬
李宜伦
张异殊
王国栋
蔡娇彧
牛俊
ZHOU Bin;LI Yilun;ZHANG Yishu;WANG Guodong;CAI Jiaoyu;NIU Jun(Dandong Electrice Power Supply Company of State Grid Liaoning FElectric Power Supply Co.,Ltd,Dandong 118000 Liaoning,China)
出处
《电力大数据》
2020年第12期18-28,共11页
Power Systems and Big Data
关键词
台区线损计算
预测模型
深度学习
神经网络
电气特征参数
line loss calculation of transformer district
prediction model
deep learning
artificial neural networks
electrical characteristic parameters
