کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

Electricity Demand Forecasting Using a Consumption Pattern Extraction Model Based on Stacked LSTM Networks from Short-Term and Long-Term Time Series Consumption Data

نویسندگان :

Reyhaneh Taghizadeh khankook¹ Mojtaba Banifakhr² Vahid Baghshani³

1- شرکت توزیع برق خراسان رضوی 2- دانشگاه یزد 3- شرکت توزیع برق خراسان رضوی

کلمات کلیدی :

Electricity consumption prediction،LSTM،Energy management

چکیده :

In recent years, the success of intelligent management and operation systems for power grids relies not only on monitoring real-time consumption data but also significantly on the capability to accurately predict consumption patterns for future time intervals. Consequently, numerous studies have been conducted in the field of short-term and long-term electricity demand forecasting. However, a substantial portion of this research has primarily focused on employing machine learning algorithms for modelling historical consumption data, without effectively incorporating the inherent dynamics and temporal dependencies of consumption patterns into the learning process. In this study, by analysing the distribution of electricity consumption time series, we attempt to identify behavioural patterns related to peak and off-peak consumption periods and, accordingly, train distinct models based on LSTM recurrent neural networks for each type of consumption behaviour. The salient feature of the proposed method is that even in the absence of explicit temporal information, the algorithm can detect consumption trend changes by examining the data distribution pattern. This approach enables the model to be aware of the context and nature of the training data, reducing its likelihood of converging to local optima. To evaluate the performance of the proposed method, experiments were conducted on real-world data, including a dataset derived from household electricity consumption in Panama City for short-term and long-term intervals. The obtained results indicate the high efficiency, stability, and scalability of the LSTM-based model compared to other common forecasting methods.