Hybrid Forecasting Model for Short-Term Electricity Market Prices with Renewable Integration

Gerardo Osório,Vasco Campos,Miadreza Shafie-Khah,Mohamed Lotfi,João Catalão

doi:10.3390/su11010057

Abstract

In recent years, there have been notable commitments and obligations by the electricity sector for more sustainable generation and delivery processes to reduce the environmental footprint. However, there is still a long way to go to achieve necessary sustainability goals while ensuring standards of robustness and the quality of power grids. One of the main challenges hindering this progress are uncertainties and stochasticity associated with the electricity sector and especially renewable generation. In this paradigm shift, forecasting tools are indispensable, and their utilization can significantly improve system operation and minimize costs associated with all related activities. Thus, forecasting tools have an essential key role in all decision-making stages. In this work, a hybrid probabilistic forecasting model (HPFM) was developed for short-term electricity market prices (EMP) combining wavelet transforms (WT), hybrid particle swarm optimization (DEEPSO), adaptive neuro-fuzzy inference system (ANFIS), and Monte Carlo simulation (MCS). The proposed hybrid probabilistic forecasting model (HPFM) was tested and validated with real data from the Spanish and Pennsylvania-New Jersey-Maryland (PJM) markets. The proposed model exhibited favorable results and performance in comparison with previously published work considering electricity market prices (EMP) data, which is notable.

Highlights

In competitive and liberalized markets with prominent renewable integration, the natural renewable stochasticity is echoed in all market players’ decisions, bringing additional challenges in the way of achieving a sustainable, profitable, and reliable operation of the electricity structure [1]
From the similar definition resulted from mean absolute percentage error (MAPE), WthTe duinreccetriotaninty of the Probabilistic Hybrid Forecasting Model (PHFM) is com“pruowte”d considering the error variance: (Re) Decomposition level
To analyze the forecasting electricity market prices (EMP) results obtained by the proposed PHFM algorithm with other available and published ones, which use the same input historic EMP sets, the mean absolute percentage error (MAPE) measure was used

Summary

Introduction

In competitive and liberalized markets with prominent renewable integration, the natural renewable stochasticity is echoed in all market players’ decisions, bringing additional challenges in the way of achieving a sustainable, profitable, and reliable operation of the electricity structure [1]. In Reference [9], a hybrid forecasting model was presented, combining wavelet transforms (WT), hybrid particle swarm optimization (DEEPSO), and adaptive neuro-fuzzy inference system (ANFIS) methods to forecast the EMP series for the Spanish market (2002, 2006), and Pennsylvania-New Jersey-Maryland (PJM) markets (2006), with different forecasting windows (i.e., between 24 h and 168 h ahead) with a 1-h time-step. In Reference [14], a hybrid forecasting approach was presented considering a pre-processing technique combining particle swarm optimization (PSO) and fuzzy neural networks techniques to forecast and classify the EMP of the Spanish electricity market. The proposed model uses a combination of WT, as a pre-processing data technique, with DEEPSO in order to reduce the overall forecasting error by tuning ANFIS parameters. The hybrid particle swarm optimization (DEEPSO), due to the differential evolutionary process and hybrid enhancements, brings better capabilities to the ANFIS structure to reduce the forecast error through the tuning of ANFIS membership functions, providing a first-stage result. It MAPE sahnoduledDrrEboEerPnvSoaOtreiadntcheactrtihteerSaiatSvatertaotrriantaiggnvegsoPwEsiodpMaprrneumPauladismtsipinouregnersaiaescdcidazcinlee(glfieonprrsarttboaiocobentishlsiteyleccaturisceitdywmhaernkep111t6––sr844iucensdteernadntaolybseisz) eforor [16]

Case Studies and Results

Forecasting Validation

Conclusions