Quick Learn Approach for load forecasting during COVID 19 lockdown

Abstract

be applied everywhere. If one model works well for a utility, it might not work well for other utility.Recently, the pandemic of COVID 19 has affected the entire world. At the initial stages, in order to suppress and mitigate the spread, the Government of India took a stringent step of imposing lockdown in the entire country which to led to a massive shutdown of all the industrial and commercial activities. This severely affected the load consumption profiles and decreased the consumption level. Since then, the load consumption patterns are changing because of continuously changing policies and regulations imposed by the government. The distribution utilities are facing challenges in their scheduling operations as the load forecasts obtained using conventional techniques are no longer accurate. They are suffering from huge economic losses because of bad forecasts.The lockdown measures in India started from 22 nd March 2020 when a curfew was announced for one day. The first phase of complete lockdown 'Lockdown 1.0' started on 24 th March 2020 and continued for 21 days till 14 th April 2020. The consumption pattern drastically changed during this period. Fig. 1: Load consumption profiles in Lockdown 1.0.We have considered an urban utility which has dominant industrial and commercial load than the residential load. The change in load of this utility is shown in 1. It can be seen that the load consumption suddenly decreased to a very low value from 22 nd March 2020. If we study the statistics of load in Table. I, average load during lockdown decreased by almost 46% while the peak decreased by 54%. If we observe Abstract-Load forecasting is extremely important for day to day operations and trading of any distribution utility. In recent times due to pandemic situation, the imposition of lockdown has caused all the commercial and industrial activities to shutdown. This has profoundly impacted the entire nation's electrical demand consumption. The consumption levels as well as the profiles have changed drastically. This has affected the forecast accuracy of the utilities exposing them to financial and technical risks. This paper presents and analyses the impact of COVID 19 on load consumption of a urban distribution utility. It also proposes a new forecasting model called 'Quick Learn Approach' (QLA) developed specifically for such lockdown period. Its main aim is to quickly learn and adapt to the constantly changing load consumption pattern while giving fairly accurate forecasts. The performance of the QLA model is compared with several other standard forecasting models so as to ensure its effectiveness.

Design and development of Residential Sector Load Prediction model during COVID-19 Pandemic using LSTM based RNN

“…Weather conditions like temperature, rainfall, humidity etc., and also type of days (week days and weekends) generally affects the consumer load demand [ 60 , 61 ]. Even though the load variation during pandemic period is mostly due to lockdown imposed and work from home policy adopted, [4] the temperature parameter is also considered for load prediction as it significantly effects residential power usage. The load data is categorized as training, validation and testing datasets.…”

“…Novel Corona virus (COVID-19), the global epidemic (as declared by WHO) that was first detected in Wuhan city, China [3] also had a menacing effect on living styles of world population including India. The spreading rate and terrific situations of COVID-19 enforced the Indian government to impose complete lockdown from 24 th March 2020 till 7 th June 2020 [ 2 , 4 ] in different phases ( Table. 1 ).…”

“…The authors also proposed load forecasting using conventional machine learning methods and claimed superiority of ANN results. Lokhande et.al [4] in addition to analysis of Covid-19 effect on energy consumption, proposed a load forecasting model for industrial and commercial dominated urban utility using quick learning approach. Alavi et.al [40] presented the variations in sectorial load demand of Bangladesh and proposed nonlinear autoregressive neural network for 30 minute ahead load prediction which proved to be inaccurate for longer predictions.…”

“…The proposed prediction model aims at predicting load demand during lockdown period especially for residential sector, whose electricity profile gets affected due to the work from home / online classes due to pandemic. Contrary to the other predictions [ 4 , 40 ], the work uses real-time load data collected from pure residential feeders during lockdown period. Because of lockdown conditions, no load demand variations were observed based on day of a week so, the data collected was not segregated into week days and week-ends for prediction.…”