The method is suitable for typical power utilization mode extraction method of massive types of unbalanced load data

Abstract

The invention discloses a typical power consumption mode extraction method suitable for massive category imbalance load data. The method comprises the steps of (S1) processing load data by adopting aBorderline-SMOTE training sample category imbalance processing method; (S2) decomposing the load data by using MODWT to obtain wavelet coefficients and scale coefficients, and forming a frequency domain characteristic matrix by using the wavelet coefficients and the scale coefficients; (S3) carrying out modeling processing on a frequency domain characteristic matrix obtained after decomposition based on a load classification model of a deep LSTM network; and (S4) carrying out structure parallelization on the load classification model based on Spark. Through the above scheme, the above scheme is adopted, according to the invention, the classification precision of the morphological similarity curve is improved by means of frequency domain decomposition, sample oversampling processing, distributed calculation and the like; the classification precision of the load data with the class imbalance problem is improved, the calculation efficiency of typical power utilization mode extraction of massive load data is improved, and the method has very high practical value and popularization value.

Description

technical field [0001] The invention belongs to the technical field of power consumption, and in particular relates to a typical power consumption pattern extraction method suitable for massive types of unbalanced load data. Background technique [0002] Extracting power consumption behavior patterns from user load data is of great significance for improving the reliability of power system operation, improving the utilization efficiency of power grid assets, improving the economic benefits of enterprises, and saving energy. With the development of power Internet of Things technology and the improvement of power consumption information collection system, the load data presents the characteristics of massive, diversified and unbalanced. Traditional methods for extracting typical load consumption patterns often have problems such as low learning efficiency and low classification accuracy when faced with massive unbalanced load data. [0003] The traditional typical electricity...

AI Technical Summary

Problems solved by technology

[0004] However, when faced with massive load data, the traditional typical power consumption pattern extraction method is often inefficient due to too large training samples in the learning process, and due to the randomness and diversity of users' power consumption behavior, the load data has serious category inconsistencies. Balance problem, the number of loadings of some categories is much less than that of other categories, and the model's ability to distinguish minority samples will be greatly affected

On the other hand, although models such as deep LSTM networks in deep learning have good time-series data learning capabilities, they cannot effectively grasp the frequency-domain characteristics hidden in load data, so they cannot accurately identify the time-domain Euclidean distance. The load data information with large fluctuation characteristics in the frequency domain

Images