Comparison of Recurrent Neural Network Model for Future Electric Power Prediction

• Main Authors: Balasubramaniyan, S., Duraipandy, P., Kalyani, N., Shanmugaraja, P., Shobana, D., Vimala, M
• Format: Conference Proceeding
• Language: English
• Subjects: Analytical models; Convolutional neural networks; Data Cleaning; Data models; Electricity; Forecast; Loss; Power demand; Predictive models; Recurrent neural networks; RNN model; Time series data; Veins
• ISSN: 2768-0673
• DOI: 10.1109/I-SMAC55078.2022.9986504

The increasing demand for electricity can be directly attributed to the exponential rise in both the global population and the sophistication of modern technologies. Because electric power is utilized simultaneously with its generation at the power station, it is crucial to precisely anticipate electric power consumption in advance to ensure a reliable power supply. Estimating future electricity needs is critical to the success of any electricity provider. Having accurate predictions of electricity use has several practical and financial benefits, including ensuring the security and stability of the power grid. This research employs a Recurrent Neural Network (RNN) model based on a Long Short-Term Memory (LSTM) network to anticipate electricity usage, aiming to resolve the discrepancy between the need for precise prediction and the limitations of conventional methods. The model incorporates conventional LSTM and hybrid Convolutional Neural Networks and Long Short-Term Memory (CNN-LSTM). First, we'll examine the Kaggle data we collected on electricity consumption. Second, we utilize an RNN model to forecast future power consumption as a whole. Finally, we will utilize loss and Mean Square Error (MSE) to qualitatively and statistically assess the performance of the model. According to the research conducted in the experiments, the CNN-LSTM prediction model outperforms traditional LSTM models in terms of reliable prediction.