Deep neural network approach for load forecasting

In this project, a fast and accurate hybrid electric load forecasting (FA-HELF) framework is proposed. The proposed FA-HELF is an integrated framework of three modules. First, random forest and relief-F algorithms are fused together to propose a hybrid feature selection technique

Project Title

Deep neural network approach for load forecasting

Project Area of Specialization

Artificial Intelligence

Project Summary

In this project, a fast and accurate hybrid electric load forecasting (FA-HELF) framework is proposed. The proposed FA-HELF is an integrated framework of three modules. First, random forest and relief-F algorithms are fused together to propose a hybrid feature selection technique for the purpose to eliminate redundancy. Second, the kernelbased principle component analysis is introduced for feature extraction in order to overcome the problem of dimensionality reduction. Finally, to perform fast and accurate load forecasting heuristic based optimizer is integrated with a support vector machine (SVM) based forecaster. The proposed FA-HELF framework shows significant improvements than other existing forecasting models in terms of forecast accuracy and convergence rate.

Project Objectives

(1) To minimize the operating cost, electric supplier will use forecasted load to control the number of running generator unit.

 (2) Tell the Future!

 (3) A central problem in the operation and planning of electrical power generation.

Project Implementation Method

In this project, we focus on short-term electricity load forecasting, which mainly refers to the prediction period for few hours, one day, one week and one month. The electricity load data (www.aemo.com.au) and temperature data (www.bom.gov.au) from Australian, and the electricity load data (www.nyiso.com) and temperature data (www.weather.gov) from New York City and also load data from Mardan and Malakand Pakistan are used to evaluate the performance of the proposed model. Further we will design deep neural network framework in MATLAB to tackle the load forecasting problem. Fractional order DPSO is also implemented in MATLAB to optimize the weights for DNN.

Benefits of the Project

Extreme volatility of wholesale electricity prices, which can be up to two orders of magnitude higher than that of any other commodity or financial asset, has forced market participants to hedge not only against volume risk but also against price movements. A generator, utility company or large industrial consumer who is able to forecast the volatile wholesale prices with a reasonable level of accuracy can adjust its bidding strategy and its own production or consumption schedule in order to reduce the risk or maximize the profits in day-ahead trading. Yet, since load and price forecasts are being used by many departments of an energy company, it is very hard to quantify the benefits of improving them. A rough estimate of savings from a 1% reduction in the mean absolute percentage error (MAPE) for a utility with 1GW peak load is:

• $500,000 per year from long-term load forecasting,
• $300,000 per year from short-term load forecasting,
• $600,000 per year from short-term load and price forecasting.

Technical Details of Final Deliverable

In this project we will minimize errors in the actual data and the predicted data in electricity load forecasting problem using MATLAB.  We will combine deep neural network (DNN) and Fractional Order Darwinian Particle Swarm Optimization algorithm (FO-DPSO) to minimize the errors. The aim of the optimization algorithm is to optimize the weights use in the deep neural network and obtain minimum residual error.

Final Deliverable of the Project

Software System

Core Industry

IT

Other Industries

Energy

Core Technology

Artificial Intelligence(AI)

Other Technologies

Big Data

Sustainable Development Goals

Responsible Consumption and Production

Required Resources

If you need this project, please contact me on contact@adikhanofficial.com