Presentation Title

Load Forecasting using Wavelet Transforms and Neural Networks

Faculty Mentor

Ankita Mohapatra

Start Date

23-11-2019 9:30 AM

End Date

23-11-2019 9:45 AM

Location

Markstein 105

Session

oral 1

Type of Presentation

Oral Talk

Subject Area

engineering_computer_science

Abstract

In 2016, California was the second largest consumer of energy in the country, with a net consumption of 7830 trillion Btu. In the same year, California’s in-state energy production was 2431 trillion Btu, of which 9.65%, 43.8%, 8.13% were generated by natural gas, crude oil and nuclear plants, and the rest by renewable energy sources such as biofuel. According to the California Renewables Portfolio Standard established in 2002, it is desired that by 2020 33% of the electricity retail sales must be fulfilled by renewable sources and 50% by 2031. To accommodate these ambitious goals, it is imperative that the projected energy demands be also known accurately. In this research, we used a combination of wavelet transforms and neural networks to predict the future electricity load trend, based on the past pattern. A comparison between different neural network learning mechanisms and level of signal decomposition were done and the most optimum combination with the lease Mean Absolute Percentage Error (MAPE) is also presented.

This document is currently not available here.

Share

COinS
 
Nov 23rd, 9:30 AM Nov 23rd, 9:45 AM

Load Forecasting using Wavelet Transforms and Neural Networks

Markstein 105

In 2016, California was the second largest consumer of energy in the country, with a net consumption of 7830 trillion Btu. In the same year, California’s in-state energy production was 2431 trillion Btu, of which 9.65%, 43.8%, 8.13% were generated by natural gas, crude oil and nuclear plants, and the rest by renewable energy sources such as biofuel. According to the California Renewables Portfolio Standard established in 2002, it is desired that by 2020 33% of the electricity retail sales must be fulfilled by renewable sources and 50% by 2031. To accommodate these ambitious goals, it is imperative that the projected energy demands be also known accurately. In this research, we used a combination of wavelet transforms and neural networks to predict the future electricity load trend, based on the past pattern. A comparison between different neural network learning mechanisms and level of signal decomposition were done and the most optimum combination with the lease Mean Absolute Percentage Error (MAPE) is also presented.