Improving load forecast in energy markets during COVID-19

Abstract

The abrupt outbreak of the COVID-19 pandemic was the most significant event in 2020, which had profound and lasting impacts across the world. Studies on energy markets observed a decline in energy demand and changes in energy consumption behaviors during COVID-19. However, as an essential part of system operation, how the load forecasting performs amid COVID-19 is not well understood. This paper aims to bridge the research gap by systematically evaluating models and features that can be used to improve the load forecasting performance amid COVID-19. Using real-world data from the New York Independent System Operator, our analysis employs three deep learning models and adopts both novel COVID-related features as well as classical weather-related features. We also propose simulating the stay-at-home situation with pre-stay-at-home weekend data and demonstrate its effectiveness in improving load forecasting accuracy during COVID-19. CCS CONCEPTS• Computing methodologies → Neural networks; • Hardware → Power and energy.

Residential Short-Term Load Forecasting during Atypical Consumption Behavior

“…A prerequisite for developing an accurate forecasting model under atypical consumption behavior or power load uncertainty is a trigger that announces the decision factors for atypical consumption behavior to occur. This knowledge concerning the behav-2 of 15 ior of the load curve is determined by correlation between the influence factors, consumer data and statistical analysis of past consumption [3][4][5][6][7].…”

“…In addition, we can see in [3] that the CNN method had higher maximum errors than the SOM. A database of New York (NY) consumption for the same atypical COVID-19 lockdown consumption event was analyzed in [4] by deploying three forecasting methods, namely Fully Connected Deep Neural Network (FCDNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU), along with Auto-Regressive Integrated Moving Average (ARIMA) which did not produce meaningful results on their database and therefore was not considered. The MAPE results in [4] were best in GRU with 4.04%, followed by FCDNN with 4.08% and lastly LDTM with 4.26%, all under the 5.35% benchmark for the NY database.…”

“…A database of New York (NY) consumption for the same atypical COVID-19 lockdown consumption event was analyzed in [4] by deploying three forecasting methods, namely Fully Connected Deep Neural Network (FCDNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU), along with Auto-Regressive Integrated Moving Average (ARIMA) which did not produce meaningful results on their database and therefore was not considered. The MAPE results in [4] were best in GRU with 4.04%, followed by FCDNN with 4.08% and lastly LDTM with 4.26%, all under the 5.35% benchmark for the NY database. The Jordanian National Electric Power Company (NEPCO) power database was used in [5] to evaluate, also during the COVID-19 lockdown period, the forecast efficiency of Autoregressive Integrated Moving Average with Exogenous (ARIMAX) and Artificial Neural Network (ANN).…”

“…The modeled characteristics of the consumer to be analyzed [1] are an essential indicator of the health of the forecast, seen even more during unpredictable power load events, as the previous research states [3][4][5][6][7]. Power consumers absorbing load in similar socio-economic and weather/climate areas usually have similar consumer behavior, and consumption forecast models developed for a type of consumer can easily be adapted for forecasting the consumption of other consumers in the same conditions.…”

“…The advantages of this database are that it contains a huge number of consumers (households) and that the residential consumers are the ones that have the best correlation to the consumption influence factors, e.g., weather. Previous research was conducted [3][4][5][6][7] mainly on national or international databases containing all consumption, including residential, commercial, industrial, transport, etc. This means that the nonresidential consumers, which have the obligation to forecast their own consumption, accounted for more than half of the power consumption forecasted.…”