Beta
Future price levels for electricity in Melbourne and battery storage to buy low, sell high
Executive Summary
Future prices
- The electricity prices in Melbourne varied by more than a factor of two in the 5.5 years of data analyzed here. While there are seasonal and weekly regularities in the price and demand of electricity, these seasonal effects only account for a small fraction of the variation in electricity prices. This is likely due to the price of electricity following the prices of fossil fuels, which are still used to generate the majority of the electricity in the world. The price of fossil fuels on the other hand is tied to things like geopolitics and the state of the global economy, which are much more difficult to predict than the changing of the seasons.
- I expect electricity prices to return to at least 2017-2019 levels as the world economy recovers from the COVID pandemic. Thus we can expect an average price of 102 AUD/ MWh for Melbourne once the COVID pandemic subsides.
- There are two sesonal peaks in demand and price for electricity in Melbourne:
- The mean demand for electricity is highest during the Australian winter, presumably due to increased electricity use in heating (the average low in June/July in Melbourne is around 8 C / 46 F). This is also reflected in the prices: at roughly 20% higher than the yearly average, the mean electricity price shows a peak in June and July compared to the fall and spring months
- The highest peak demand occurs during the heat of summer, when people crank up the air conditioning on days with highs of more than 40 degrees celcius. The electricity production in Melbourne doesn't always seem to be able to keep up with demand on these days, and prices skyrocket. This results in the prices in January being approximately 50% higher than the yearly average, and with large daily variance in both demand and prices. This is good news from the point of view of solar arrays, since their production will be at a maximum on sunny summer days.
- The expected mean prices for each month are shown in the plot below. The black lines on each bar are proportional to the variance in prices for each month.
- Demand and price also vary by day of week: on weekdays the demand and price are higher than during the weekend or on holidays. For demand, the difference is approximately 10%, for prices around 25%. The variation in prices is shown in the figure below.
- For estimating the prices for the next 12-18 months, I recommend using the average price of 102 AUD/ MWh that was seen before the pandemic, and to scale that price by the monthly relative price. If one is interested in a particular day, whether or not the day is a weekend/holiday or a working day should be taken into account.
- Since the seasonal effects explain only a small part of the price variation, to truly make accurate long term predictions we would need more data. The most important would probably be something that indicates long term pricing of fossil fuels, such as oil or gas futures.
- The return-on-investment for solar arrays Victoria seems very good: somewhere between 12-38% based on a quick estimate.
Battery storage to buy low, sell high
- Battery storage is not financially viable at current electricity prices, battery costs and battery lifetimes:
- 70 MWh lithium-ion battery would cost 12 million AUD at a minimum
- With perfect information on tomorrow's electricity prices, the battery system could make a profit of 300,000 AUD per year, assuming we sell at mean daily prices.
- Yearly ROI is then 2.6% - much more profitable to spend money on more solar panels rather than batteries
- Battery system would take almost 40 years to pay back for itself. Lithium-ion batteries can last up to 3,500 charge-discharge cycles which would be just under 10 years if done daily. In our case the cycling might only occur every two days or so, but the batteries would still not be able to pay for themselves.
- With minimal tuning, a machine learning model (LightGBM, a gradient boosted tree regressor) can predict price differences between today's and tomorrow's prices well enough to make 48% of the maximum theoretical profit. With some effort this can surely be improved, so that predicting when to store and when to sell electricity should not be a problem.
- Once the price of battery capacity comes down by a factor of four, it may start to become more profitable than solar panels
- Analysis of battery storage here assumed electricity is sold at mean daily price - profitability may be much improved if the electricity is sold during peak hours. I still would not expect this to be more than a factor of two increase in the yearly profit, making investing in solar arrays the better option for now.
Import packages
%%capture
# Install LightGBM for predicting daily energy prices
!pip install lightgbm
from typing import Tuple
from lightgbm import LGBMRegressor
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import seaborn as sns
from scipy.stats import sem
import seaborn as sns
from sklearn.linear_model import LinearRegression
from sklearn.metrics import mean_absolute_percentage_error
plt.style.use('seaborn-deep')
plt.rcParams['figure.figsize'] = (16,9)
plt.rcParams['axes.labelsize'] = 16
plt.rcParams['axes.titlesize'] = 18
plt.rcParams['legend.fontsize'] = 14
plt.rcParams['xtick.labelsize'] = 14
plt.rcParams['ytick.labelsize'] = 14Import and preprocess data
Electricity prices in Melbourne Jan 1 2015 - Oct 6 2020
The plot below shows the weekly median RRP for electricity in Melbourne. The range of the price (max - min) is shown by the shaded blue area. Some things to note:
- The price of electricity increased significantly during the first quarter of 2017, and stayed at this level for the next 2.5 years. The increase in price coincides with the closure of the Hazelwood coal power station, shown by the dashed line in the figure. It seems likely that the higher price is due to switching from coal to natural gas which is cleaner but more expensive.
- The price of electricity started going down already before the COVID pandemic started (start of lockdown in Victoria is shown by the dot-dashed line in the figure), and is expected to stay low while economic activity is decreased due to the COVID lockdown. Once the pandemic subsides, the price is likely to go up again as oil, gas and coal producers struggle to keep up with demand, having decreased production during the pandemic.
- Seasonal variations in electricity prices are not easily discernible in the data. The seasons do have an effect as shown later, but it is masked by the effect of other factors (probably variation in the price of fossil fuels).
- There are a number of days in the Australian summer (mostly in January) where the price is 10s of times the median weekly price. This seems to indicate that on very hot days, electricity production in Melbourne is not able to keep up with the demand for air conditioning units working on overdrive. This is good nows from the point of view of solar cell arrays, since they will work very well during sunny summer days.
Hidden code
Get the average price in during 2017-2019:
print(f"Mean price during 2017, 2018 and 2019: {df['2017-01-01':'2019-12-31'].price.mean():.1f} AUD/MWh")Seasonal variation in prices
- Though overshadowed by the effects of other variables, such as fossil fuel prices, the price of electricity does also show variations by season, and also by day of the week
- As shown below, these changes in price seem to be driven by variations in demand: higher demand results in higher prices as one might expect based on basic economics.
- Average demand is highest during the winter, but peak daily demand is reached during the summer.
- The peak prices and demand occur during the hottest days in the data so they are likely due to electricity demand from air conditioning outstripping supply.
- Solar arrays produce the most electricity during the summer when they get more sunexposure, so they would be able to supply electricity when it is at its most valuable during hot, sunny days.
Variation in demand
The plot below shows the seasonal variation in demand for electricity. The average demand is highest during the Australian winter (peaking around June and July), presumbaly due to electrical heating. The highest peak demand, however, is during the summer on very hot days where the need for air conditioning causes a surge.
Hidden code
Correlation between demand and price
The plot below shows the correlation between electricity price and demand. Higher demand results in a higher price, as one might expect based on basic economics. The demand explains roughly 25% of the variance in the prices. Since the correlation is fairly weak, the seasonal variations are difficult to see by eye when plotting the price over time, even though the demand has clear seasonal variation.
Hidden code
Seasonal variation in relative price
To better see the variation in price by season, we can plot the average monthly prices normalized by the average price for the year, as done in the barplot below.
Observations:
- The mean price of electricity is the highest during January. As mentioned before, this is due to days of extreme heat, during which the demand for electricity (probably due to air conditioning) outstrips supply causing prices to surge.
- During the coldest part of the year (June-July), the mean price is also slightly higher than the rest of the year. This is likely due to heating increasing demand for electricity.
Hidden code