Short-Term Electric Load Forecasting With Sparse Coding Methods

Abstract

Short-term load forecasting is a key task for planning and stability of the current and future distribution grid, as it can significantly contribute to the management of energy market for ancillary services. In this paper we introduce the beneficial properties of applications of sparse representation and corresponding dictionary learning to the net load forecasting problem on a substation level. In this context, sparse representation theory can provide parsimonial predictive models, which become attractive mainly due to their ability to successfully model the input space in a self-learning manner, by interacting between theory, algorithms, and applications. Several techniques are implemented, incorporating numerous dictionary learning and sparse decomposition algorithms, and a hierarchical structured model is proposed. The concept of sparsity in each case is embedded throughout the utilization of different regularization forms which include the 0 , 1 , 2 and tree 0 norms. The observed superiority of the proposed theory, especially the one which embeds the atoms and corresponding coefficients in a tree structure, stems from the construction of the dictionary so as to represent efficiently the ambient electricity signal space and the consequent extraction of sparse basis-vectors. The performance of each model is evaluated using real hourly load measurements from a high voltage/medium voltage (HV/MV) substation and compared with that of widely used machine learning methods. The provided analytical results, verify the effectiveness of hierarchical sparse representation in short-term load forecasting applications, in terms of common accuracy indices.INDEX TERMS Generative models, hierarchical dictionaries, load forecasting, power grid, sparse representation.

A Machine Learning Model Ensemble for Mixed Power Load Forecasting across Multiple Time Horizons

“…In order to obtain the best possible performance of each sub-model, their optimal training configuration has to be determined. Starting with the simpler methods used, a linear and an SR model are trained by least squares and fast iterative shrinkage thresholding algorithm, respectively, the latter being a faster implementation of the corresponding iterative shrinkage thresholding algorithm used for load forecasting [ 36 ]. In the case of the sparse coding approach, sparsity is induced by the norm and the regularization parameter was set by trial and error to 0.01.…”

“…A number of papers based on sparse Bayesian learning (SBL) have been published during the last decade, featuring weighted SBL [ 33 , 34 ] or combined kernels SBL [ 35 ]. More recently, a hierarchical sparsity approach has been proposed [ 36 ] for hourly load forecasting, achieving remarkable results and outperforming both well-known sparse techniques and rival linear and non-linear methodologies.…”

Cable Temperature Prediction Based on RF-GPR for Digital Twin Applications

“…The training set is used to train the prediction model, and the test set is used to validate the model and evaluate the performance. The criterion to judge the accuracy of the model is expressed by accuracy and heel mean square deviation [25][26][27].…”

Evaluation of neural networks for residential load forecasting and the impact of systematic feature identification

“…Short-term forecasts are applied to electricity grids and microgrids, power and substations, residential and office buildings, cities, provinces, and countries, using weather data and temporal features as independent variables (Li et al 2021;Xu et al 2019;Panapongpakorn and Banjerdpongchai 2019;Ahmad and Chen 2018;Ruiming 2008). Short-term forecasts are essential to determine if the load exceeds the capacity of a transformer, which can prevent power outages (Dung and Phuong 2019;Giamarelos et al 2021;Al-Rashid and Paarmann 1996).…”