Additive Ensemble Neural Network with Constrained Weighted Quantile Loss for Probabilistic Electric-Load Forecasting.

Manuel Lopez-Martin,Belen Carro,Luis Hernandez-Callejo,Antonio Sanchez-Esguevillas,Juan Ignacio Arribas,Juan Ignacio Arribas

doi:10.3390/s21092979

Manuel Lopez-Martin, Belen Carro + Show 4 more

Open Access

https://doi.org/10.3390/s21092979

Copy DOI

Abstract

This work proposes a quantile regression neural network based on a novel constrained weighted quantile loss (CWQLoss) and its application to probabilistic short and medium-term electric-load forecasting of special interest for smart grids operations. The method allows any point forecast neural network based on a multivariate multi-output regression model to be expanded to become a quantile regression model. CWQLoss extends the pinball loss to more than one quantile by creating a weighted average for all predictions in the forecast window and across all quantiles. The pinball loss for each quantile is evaluated separately. The proposed method imposes additional constraints on the quantile values and their associated weights. It is shown that these restrictions are important to have a stable and efficient model. Quantile weights are learned end-to-end by gradient descent along with the network weights. The proposed model achieves two objectives: (a) produce probabilistic (quantile and interval) forecasts with an associated probability for the predicted target values. (b) generate point forecasts by adopting the forecast for the median (0.5 quantiles). We provide specific metrics for point and probabilistic forecasts to evaluate the results considering both objectives. A comprehensive comparison is performed between a selection of classic and advanced forecasting models with the proposed quantile forecasting model. We consider different scenarios for the duration of the forecast window (1 h, 1-day, 1-week, and 1-month), with the proposed model achieving the best results in almost all scenarios. Additionally, we show that the proposed method obtains the best results when an additive ensemble neural network is used as the base model. The experimental results are drawn from real loads of a medium-sized city in Spain.

Highlights

Electric-load forecasting aims to predict future values of electricity consumption in a specific time horizon
We provide a comprehensive comparison between CWQFNN and a significant number of state-of-the-art (SOTA) data-driven forecasting models, some of them widely applied to time-series forecasting and others novel or rarely applied to short and medium-term load forecast (SMTLF), such as (a) classic machine learning (ML) models, e.g., linear regression and random forest [18,19,20,21], (b) multilayer perceptron [5], (c) deep learning models based on separate convolutional neural networks (CNN) and recurrent neural networks (RNN) [19,20], (d) dynamic mode decomposition (DMD) [22,23,24], (e) deep learning (DL) models based on specific combinations of CNN and RNN [25,26], (f) sequence-tosequence (Seq2seq) models with and without soft attention [27,28,29], and (g) deep learning additive ensemble models especially targeted for time-series forecasting [9]
An additional aim is to present the results obtained by classic forecasting methods together with new ous evaluation criteria) the results obtained by classic forecasting methods together with or less used methods, e.g., deep learning ensembles, Dynamic mode decomposition (DMD), deep learning models with new or less used methods, e.g., deep learning ensembles, DMD, deep learning models combinations of CNN

Summary

Introduction

Electric-load forecasting aims to predict future values of electricity consumption in a specific time horizon. Depending on the time horizon, we have short-term (range of hours to a week), medium-term (from a week to a year), and long-term (more than a year) forecasts. Depending on the expected outputs, a point forecast provides a single forecast value as the most likely estimated value of the future load. A density forecast provides an estimate of the future load probability distribution either point-wise (assigning probabilities to point-forecasts) or interval-wise (quantile forecasts for predefined probabilities). The first approach to density forecasting is based on extracting probabilities from a set of forecasts [2], and the second is based on quantile regression models [3]

Objectives

Methods

Results

Discussion

Conclusion