Oil field power load prediction based on LSTM under abnormal data
cleaning technology
Abstract
In view of the massive load data ofoilfield distribution network
contains various types of outliers, which is not conducive to load
prediction, electric energy decision-making, dispatching and production,
it is very necessary to identify and correct abnormal loads to improve
the validity and reliability of load data and establish a safe,
efficient and sustainable power system.Based on the above reasons, this
paper proposes a data anomaly identification and scene generation method
combining boxplot and generative adversance network (WGAN). This method
firstly uses boxplot method to complete anomaly identification of active
power, reactive power, current and other data of oilfield grid.Then,
wasserstein-based generative Adversation network (WGAN) algorithm was
used to achieve data fitting and generation, which provided data support
for subsequent long-and-short Term Memory (LSTM) based load prediction
model. Finally, the effectiveness of the proposed algorithm and model
was verified by an example of an oilfield power grid.Through the
intelligent identification, sequence generation and load prediction of
oil field load data, the purpose of deep mining and analysis of oil
field production behavior is realized.