Research On The Detection Method Of Abnormal Data On The Power Generation Side Of New Energy

Renewable power generation assumes an important power supply task in the smart grid.However,due to the intermittency,instability and human factors of renewable power generation,the resulting data may be abnormal.The injection of abnormal data will disturb the stability of the data,thus affecting the grid dispatch with security constraints,and thus endangering the power system security.Especially on renewable power generation side,some malicious users upload abnormally forged data in order to obtain high “feed-in-tariff” subsidies.Traditional anomaly detection methods are difficult to effectively detect such patterns of anomalies.The study of anomaly data detection methods on the renewable generation side based on deep learning is of great significance for the security and stability of smart grids.In order to improve the detection ability of abnormal data on the renewable generation side,this paper analyzes the real power generation environment,and then combines machine learning and deep learning ideas to design a complete data anomaly detection framework,which is divided into four stages.The first is data preprocessing,which mainly includes data normalization,vector to sequence,and division of training and test sets.It also includes designing 4 different attack functions to generate malicious data.Secondly,a new data rebalancing method based on Wasserstein Generative Adversarial Network（WGAN）is proposed to supplement the imbalanced samples in the training set to achieve data rebalance.Thirdly,a method for feature extraction of conditional power generation data based on Autoencoder network is proposed.The extracted conditional power generation features reflect the overall power generation capability and can assist the classifier to detect abnormal data.Finally,an anomaly detection classifier based on Convolutional Neural Network（CNN）combined with Gated Recurrent Unit（GRU）is proposed,which can detect new patterns of abnormal data.This paper uses the real data set of the Ausgrid to conduct a comparative experiment and result analysis of different anomaly data detection methods.The experimental results show that using the proposed data balancing algorithm and anomaly data detection classifier can achieve high accuracy and low false alarm rate.