Data consistency method of heterogeneous power IOT based on hybrid model

Area of Science:

• Electrical Engineering
• Data Science
• Internet of Things (IOT)

Background:

• Traditional data preprocessing in power IOT systems struggles with high-error, time-sensitive data and heterogeneous power inconsistencies.
• Existing methods lack the capability to identify and correct data errors that exhibit specific timing characteristics, hindering accurate quantitative modeling.
• The transfer of data from Infrastructure as a Service (IaaS) to Software as a Service (SaaS) layers presents challenges in maintaining data integrity.

Purpose of the Study:

• To investigate the phenomenon of heterogeneous power inconsistencies in IOT data and its underlying physical mechanisms.
• To develop a data-driven hybrid model for correcting erroneous electricity meter data from different power IOT systems.
• To improve the accuracy of power data analysis by reducing deviations caused by data inconsistencies.

Main Methods:

• A data-driven approach was employed to construct a hybrid model for data correction.
• The study utilized electricity meter data from both sides of a commercial building transformer, sourced from distinct power IOT systems.
• A combined method integrating Linear Regression (LS), Differential Evolution (DE), and Extreme Learning Machine (ELM) was applied to revise low-voltage side data based on high-voltage side data.

Main Results:

• The proposed hybrid model effectively addresses high-error data with timing characteristics, a limitation of general preprocessing methods.
• The research demonstrated the difficulty in building forward quantitative models due to these data inconsistencies.
• The LS + DE + ELM hybrid method reduced data deviation from approximately 4% to 1%, outperforming purely neural network-based correction methods.

Conclusions:

• A novel hybrid model combining LS, DE, and ELM offers a superior solution for correcting heterogeneous power data in IOT systems.
• The developed method significantly enhances data accuracy, reducing deviations to 1% compared to traditional approaches.
• This research provides a robust framework for improving the reliability of power system data analysis through advanced data-driven techniques.