Accelerating Multi-layer Perceptron based short term demand forecasting using Graphics Processing Units

Abstract

Load forecasting plays a vitally important role in the operation and planning of the power system in a deregulated electricity market. A large variety of methods have been proposed for load forecasting. In this paper, we introduce the Graphics Processing Units (GPU) based computing to accelerate the short term load forecasting with Multi-layer Perceptron (MLP). The proposed method is tested with the Queensland electricity market demand series. The result shows that the GPU based computing largely reduce the computational cost.Index Terms-GPU, Multi-Layer Perceptron, Neural Network, Load Forecasting.T. He is pursuing her PhD at the University of Queensland, Australia. Her research interest is in power system stability and computational methods application in power system analysis. Z.Y. Dong is with the Department of Electrical Engineering, Hong Kong Polytechnic University, Hong Kong. He held academic positions with the University of Queensland, Australia. He also worked with Powerlink Queensland and Transend Networks Tasmania, Australia. His research interest includes power system planning and stability, power market analysis and power system computation K. Meng is with the EE Dept, Hong Kong Polytechnic University. His research interests include power system stability and control, evolutionary computing, and power market analysis. H. Wang is a research fellow with the University of Queensland. Y.T. Oh is a Professor of Electrical Engineering with the KoreaUniversity of Technology and Education, Korea. His research interests include power system analysis and power quality research.

A deep learning short-term load forecasting method for extreme scenarios

“…The trend extrapolation method is to find the change law of the forecast object according to its own historical data, while the law of related factors is to study the relationship between load and other factors, and judge the load change trend in the future according to the changes of other factors in the future. With the development of artificial intelligence technology, there are more and more short-term prediction methods based on deep learning, such as MLFNN [1][2][3][4], LSTM [5][6][7][8][9][10], WNN [11][12][13][14], ELMNN [15,16], SDA [17], etc. However, the depth neural network is a method that depends on the volume of data.…”

A GPU accelerated hybrid GA-SVM for large scale datasets: Cu-GA-SVM

“…Cao ve arkadaşları [6], bir küme sisteminde MPI ile uygulanan bir Paralel Sıralı Minimal Optimizasyon algoritması geliştirmiştir. Ting He ve arkadaşları [7], CUDA işlevlerini kullanarak sinir ağlarını eğitmek için GPU standart yeteneklerini kullandılar ve GPU üzerinde matris çarpımlarının ve vektör işlemlerinin çoğunu yaparak CPU'ya kıyasla 5.21X'lik bir hızlanma faktörü elde ettiler. Wang ve arkadaşları [8], en yakın komşu bölümleme (NNP) yönteminin CUDA tabanlı optimizasyonunu gerçekleştirdiler.…”