Anomaly prediction in electricity consumption using Machine Learning Techniques

Electricity usage has been increasing globally in recent years, primarily as a result of population growth, technology advancements and the increased use of electronic devices and appliances. Thus, reducing electricity usage is becoming ever more significant, with predicted anomaly electricity usage being one of the important topics since it allows users to take precaution and avoid high electricity consumption or inefficiencies. The use of machine learning techniques in the development of predictive models to forecast power consumption anomalies has recently grabbed the attention of researchers and data scientists, but one of the main challenges in the machine learning approach is to accurately label the anomalous power consumption data points. This is due to the fact that anomalous behavior can take many different forms and may not always be easy to identify or distinguish from normal usage. In this research, a machine learning-based predictive model on anomalous power consumption is built. In addition to predicting the anomalous power consumption, the suggested work addresses the issue of data labelling by automatically classifying power consumption data as normal or abnormal using the Enhanced Isolation Forest (IF) algorithm. The suggested Enhanced IF algorithm extends the IF algorithm by proposing a threshold to improve the algorithm's capability to recognize anomalies caused by spikes in power consumption. For the predictive model, two supervised learning approaches, Random Forest (RF) and Decision Tree (DT), are proposed to predict abnormal electricity consumption patterns. The model performance was tested based on varying lengths of the input sequence, known as 'm', that ranges from 3 to 7. The size of m represents the number of instances. Each instance accounts for a 30- minute interval power consumption reading. The experimental results on the Irish social data illustrated that the performance of the RF algorithm and DT algorithm performance in predicting abnormal electricity consumption patterns are similar. The model's accuracy increased as m increased, indicating that as the size of input instances increased, the model could more accurately reflect the underlying patterns in the data. Simulation results also showed that the enhanced IF algorithm increased data labelling accuracy and reduced the incidence of false positive and false adverse outcomes. The study’s findings point to the potential of applying machine learning algorithms to predict anomalous energy consumption, which would assist the energy providers and consumers in reducing the total energy consumption and creating a more sustainable energy system. The model's accuracy increased as m increased, indicating that as the size of input instances increased, the model could more accurately reflect the underlying patterns in the data