پیش بینی استفاده از نیرو مبتنی بر یادگیری عمیق
Abstract
1-Introduction
2-Our Approach
3-Evaluation
4-Conclusion
5-Acknowledgement
6-References
Abstract
The growing Internet of Things (IoT) provides significant resources to be integrated with critical infrastructures to enable cyber-physical systems. More specifically, the deployment of smart meters for electricity usage monitoring in the smart grid can provide granular and detailed information from which power load forecasting can be carried out. However, the accurate prediction of long-term power usage remains a challenging issue. In light of many recent advances, deep learning has the potential to significantly improve the ability to assess data and make predictions, and is already rapidly changing the world we live in. As such, in this paper, we consider the use of deep learning, via Recursive Neural Network (RNN) and Long Short-Term Memory layers, for the long-term prediction of localized power consumption. In particular, we consider the optimization of both data feature sets and neural network models, developing three model-feature combinations to maximize prediction accuracy and minimize error. Through detailed experimental evaluation, our results demonstrate the ability to achieve highly accurate predictions over periods as large as 21 days through the integration of correlated features.
Introduction
In the context of the smart home and smart grid, the ability to absorb and analyze massive amounts of big data provide for unprecedented levels of command and control for electrical load distribution, industrially and by individual users alike, supporting a new level of situation awareness. Moreover, emerging tools for machine learning have been used to achieve increasingly accurate predictions on continuous time-series data in a variety of applications. As the state-of-the-art to achieve highly accurate data analysis, deep learning applies significant coalitions of computing neurons to approximate dataset distributions. What’s more, deep learning has been applied to abstract problems and exceeded even human capabilities, and has been used to extract meaningful and hidden information from massive datasets [4, 5].