Neural Network-based Smart Meter Demand Response Analysis: A Case Study of Addis Ababa Power System
No Thumbnail Available
Date
2021-10
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Most African countries including Ethiopia used the old way of going door to door to record usages of electricity. This results lots of guesswork, which has direct impact on consumers, especially in billing.
In case of Ethiopian Electric Utility (EEU) getting real-time information about power interruption, maximum power consumption in the grid is difficult to maintain and implement especially when going down towards the end consumers. The concepts of Smart Meters are introduced to address problems associated in electricity transmission throughout the usual traditional grid, which allows a bidirectional communication between the household smart meters and the supplier.
This study aims to explore demand response analysis of smart meters using available recorded information by training Neural Network method to identify maximum demand response, type of power interruption and identify theft, by means of Artificial Neural Networks (ANNs) with Feedforward backpropagation algorithm for the selected cases in Addis Ababa as a case study. Four districts in the EEU Addis Ababa City were used for collecting quantitative data.
The result for theft identification purpose consists of 169,296 samples, 25 neurons, two outputs. The best validation performance is 0.003124, and the overall correctly predicted percentage becomes 99.7%. In power fluctuation classifications, the model data sets consist of 3,596 sample sizes with 30 hidden neurons. The best validation performance is 0.03197. Moreover, the overall percentage of correctly predicted values is 97.2%. Finally, for the maximum power demand the percentage of correctly predicted values is 100%. The data analysis highly affected the performance of the NN system. Lastly the study recommends, further improvements can be achieved by process real-time data from millions of smart metering, more efficient modeling can lead to higher prediction accuracy.
Description
Keywords
Smart Grid, Demand Response, Smart Meter, Data, Neural Network, Fault detection Theft detection, MATLAB