In March 2020, Covid-19 reached Australia and caused a nationwide lockdown. That meant most of us were home, industries, restaurants, schools, public spaces shut. This lockdown surely impacted several ways in which our world ran before, but a key part of our world running is: electricity.
So, what happened? The first thought would be that we must have seen a reduction in electricity consumption. This study tries to answer two questions briefly and is based on Victorian electricity consumption as recorded by AEMO:
- How has COVID-19 impacted the way we consume Electricity?
- Are traditional models still good to predict and forecast our future consumption needs?
Has COVID-19 impacted the way we consume electricity?
Yes and no.
From fig 1, it can be seen that there is no significant change in the amount of electricity Victoria consumes. The dips and peaks attribute to by and large, the weather. For example, the peak in July is due to the increased use of heating with July being the coldest month of the year.
However, from fig 2, a change or rather a significant dip can be observed in the electricity demand from an hourly perspective over the last nine months as compared to other years. This time window is the 8 am to 5 pm, the office hours with natural light seeping in through the windows of our homes.
These visualisations can be found on Tableau here .
Now, the real question.
Are traditional models still good to predict and forecast our future consumption needs?
The way we consume electricity is dependent on a couple of factors such as weather, business cycles, day of the week and here, we tried to understand if the lockdown had an effect on this consumption. While building the models for this study, the following factors were considered – monthly time series (from 2015–2020), average temperature in these months and a categorical variable representing if we were in a lockdown or not.
I tried a bunch of time series forecasting methods such as ETS, ARIMA, Linear Regression and Dynamic Regression to forecast the consumption in 2020. Most of these models performed identically with small variations in terms of model accuracies.
The model that comparatively outperformed others was a time series linear model with mean temp and ARIMA errors. I trained the models using data from January 2015 until February 2020 and made predictions for March 2020 to September 2020 (the months of lockdown in Victoria). More details of this model can be found here and this is what the forecast looks like.
Conclusion
So, what is all this telling us about the pandemic and electricity consumption?
Well, at least in Victoria which has been seeing a decreasing trend in electricity consumption over the last five years, the 2% decrease due to the pandemic is not difficult to predict. The electricity that was used by industries, offices and business are now redirected to our homes.
There is, however, one difference which can be observed during the office hours of the day when there is a significant dip in the electricity consumption in the months of lockdown. And this could be a scope for improvement if we could build a model that takes in this as a factor.
You can find the code and more details on this on my GitHub repo.
