Electricity Anomaly Point Detection using Unsupervised Technique Based on Electricity Load Prediction Derived from Long Short-Term Memory

Nur Shakirah Md Salleh, Mulyana Saripuddin, Azizah Suliman, Bo Norregaard Jorgensen*

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review

Abstract

Electricity theft caused a major loss for electricity power provider. The anomaly detection helps to predict the abnormal load usage of a consumer. Usually, the classification method used in anomaly detection. This research paper proposed to identify the potential anomaly points by using threshold and outliers. The prediction in time-series applied Long Short-Term Memory (LSTM) algorithm. The historical electricity load dataset of a single industrial consumer was used to generate the prediction of electricity load. There were five optimizers used to produce the model: Adam, Adadelta, Adagrad, RMSProp, and Stochastic gradient descent (SGD). The prediction model was evaluated using mean squared error (MSE) and mean absolute error (MAE). The best model among all five models was generated by Adadelta optimizer with the error rate value of 0.091982 for MSE and 0.018433 for MAE. The prediction values were generated by this model. The anomaly point was detected by using threshold and outliers. The threshold value was 0.218983. One week in August 2019 was chosen to detect any anomaly load occurrences. There were 24 outliers were found within the selected week. The study shall expand on the electricity usage trend during COVID-19 pandemic period.

Original languageEnglish
Title of host publication2021 2nd International Conference on Artificial Intelligence and Data Sciences, AiDAS 2021
Number of pages5
PublisherIEEE
Publication date2021
ISBN (Electronic)9781665417266
DOIs
Publication statusPublished - 2021
Event2nd International Conference on Artificial Intelligence and Data Sciences, AiDAS 2021 - Ipoh, Malaysia
Duration: 8. Sept 20219. Sept 2021

Conference

Conference2nd International Conference on Artificial Intelligence and Data Sciences, AiDAS 2021
Country/TerritoryMalaysia
CityIpoh
Period08/09/202109/09/2021

Keywords

  • anomaly detection
  • Electricity load
  • long short-term memory
  • regression

Fingerprint

Dive into the research topics of 'Electricity Anomaly Point Detection using Unsupervised Technique Based on Electricity Load Prediction Derived from Long Short-Term Memory'. Together they form a unique fingerprint.

Cite this