The thought process behind some of the visualizations from the CREG's Monitoring Report 2022

As every year, the CREG - Commission for Electricity and Gas Regulation - Belgium publishes its "Monitoring Report" (link). The most recent edition contains 68 figures and 25 tables, summarizing some of the most striking observations and evolutions we observed on the Belgian, regional or European electricity markets between 2015 and 2022.

This article is not about the analyses or the messages in the report: for those of you interested in that, feel free to download it and go through it - if you have any questions you are welcome to reach out.

In this post I will rather provide a personal selection of a number of remarkable figures from the report and explain a bit f the thought process behind creating them. They are remarkable, not because of the content of the messages in the charts, but more because of the way they are created, the way information is encoded in them and the way in which the message is visualized.

Electricity generation mix

Showing the evolution of categorical variables is always tricky - either you loose sense of the evolution of the sum of all categories, or you don't show the evolution of all of the individual categories. In the case of data on power generation, we have a typical stacked bar chart, with the colors encoded to the different fuel types. Adding to that, we decided to use a "waterfall chart", allowing for an easy comparison of the total generated volumes of 2022 and the previous year, 2021, as well as the individual categories.

A waterfall chart shows a basic mathematical equation: adding or subtracting values from some initial value to produce a final amount. It is essentially a bar chart, but each subsequent bar starts where the previous one left off, showing how they accumulate across the graph. (from "Better Data Visualizations" by J. Schwabish)

Assigning different colors to positive and negative values allows an easy identification of which fuel types have shown a decrease since last year (in red: "nuclear" and "other") and which ones have increased (in blue: "gas", "solar" and "wind"). The sum of the increases and decreases corresponds to the difference in generation in 2021 (93,4 TWh) and 2022 (89,9 TWh): a net decrease of 3,5 TWh.

This is not a new idea. As an excellent example: Ember publishes a series of these figures in their "European Electricity Review 2023" (can be accessed here: link). In this figure, the colors conveniently align across the report with the different fuel categories under consideration.

Cross-border flows

This is by far my favorite figure of the report, yet probably the most challenging one to create (especially given that it's 100% R / ggplot2: there's no visual post-processing involved), the figure below is a combination of an area chart (left) and a Sankey diagram (right).

Area charts typically don't differ a lot from line graphs, with the exception that the area below the lines is filled in. In the case of import and export flows, it can conveniently be used to show the total cross-border flows (as the sum of the positive/export and negative/import exchanges). The left x-asis shows time, measured in years, yet could also be days, weeks, months, etc.

The export flows could, in the area chart, be shown in different shades of blue (or different colors altogether) to distinguish between the different countries, and similarly for the red import flows.

The chosen alternative allows to compare the import and export flows from and to each country in 2022, as a sort of "net balance" of Belgium with each of its four neighbors.

In line with the creative nature of the visualization, is the name coined by its inventor: the "Frankensankey". This chart type has been developed by Financial Times journalist Alan Smith in a piece about migration in Europe (Twitter).

In his list of "weird but (sometimes) useful charts", Maarten Lambrechts more precisely called it a "stacked area alluvial diagram". If these types of visualizations make you curious, definitely check out Xenographics (link), an incredibly rich repository of innovative visualizations.

Price for long-term capacity and day-ahead spreads

Chapters 4 and 5 of the Monitoring Report deal with long-term and day-ahead markets, respectively. Often however, it makes sense to consolidate data which belong to different themes to describe a certain practice, as we did in the section where the auction price of long-term transmission rights is compared against the day-ahead spread between the two corresponding bidding zones. We do this to understand what a TSOs' income from long-term capacity allocation is, and how it relates to the pay-outs to the holders of these rights, calculated on the basis of the day-ahead prices.

This comparison leads to positive values (when the income exceeds the pay-out) or negative values (when the pay-out exceeds the income). This could be shown as a simple diverging bar chart, where one bar represents one auction. However, this only shows the unit prices (for income and pay-outs) per MWh. Taking into account the volumes allocated, we can assess and compare the total monetary flows (positive: towards the TSO; negative: towards the rights' holders). This is done by encoding the allocated volumes horizontally: the bars become rectangles, as both their height (unit price in €/MWh) and width (MWh) matter, and the resulting surface area becomes a monetary flow (€/MWh x MWh = €).

This is an example of a "Marimekko" or "mosaic chart", and its main difference with a normal bar chart is that it does not have standard widths for the bars, but they are transformed into rectangles which encode height and width separately.

This figure is inspired by a yearly publication from the Joint Allocation Office, which does this calculation for all European borders on which long-term rights are allocated (link to 2022 report).

Another brilliant example of a Marimekko plot comes from the same Ember report mentioned before (the European Electricity Review 2023). In this figure, the share of the total generation of each fuel source (%) is shown on the x-axis, while the total generation (TWh) is shown vertically. Typically, this kind of information is shown only one way in a regular bar chart: either you see the share per type (summing to 100%) or you get the total volume, without seeing the shares per type. By showing the information this way, you can easily see both aspects and the square of each colored surface (obtained by multiplying lenght / x and height / y) represents the total volume per fuel type, per country.

Negative prices in day-ahead market coupling

Negative prices are a particularly relevant topic today (given the market coupling results in week 22 of 2023). The easiest way to show evolution throughout time of the occurrence of negative prices (spanning several years) is to summarize the number of hours (intuitively as a total sum) per year or per month, and plot that timeseries. Showing a more detailed granularity (per hour or per day) would be unreadable, as such hours only occur in about 1 - 2% of the 8.760 hours in a year.

The solution we have chosen is to show the cumulative number of hours per year, and show each year as a separate line, moving from the left (1 January) to the right (31 December). The right-most observations then shows the total number of hours per year, still showing during which parts of the year the increases were highest (mostly in between March and May, with also a slight increase in the Christmas break).

This approach reveals patterns which cannot be seen by plotting the sum of hours per year beween 2015 and 2022 (which is done, along with other countries, separately in the report).

As far as I know, this visualization - at least for the subject of showing negative prices' occurrences - was 'invented' (or at least popularized) by Martien Visser , who regularly publishes this figure for the Netherlands, along with many other highly interesting and engaging figures on his Twitter account.

Available margins on critical network elements

The next figure has appeared in different studies and presentations made by the CREG, and relates to a point on which the CREG has positioned itself quite strongly over the last years: the low margins available for cross-zonal exchanges. A lot of studies and analyses have been made, assessing to which extent transmission system operators (TSOs) respect - or not - the minimum margins for capacity calculation (either 20% in the Core DA FBMC, or 70% in the Clean Energy Package).

These studies typically treat these continuous margins as a categorical value (by binning them, for example [0 - 20%[; [20 - 50%[, etc). Alternatively, density plots are constructed, showing which margins are observed more frequently than others.

What we have chosen to do, however, is show the full distribution, by plotting all the observations (i.e. every margin observed on every network element), grouped per TSO. The combination of this continuous variable (available margin or RAM) and a discrete variable (TSO), lends itself perfectly for a jitter plot. Jitter plots include a degree of randomness, in a certain direction (in this case: vertically), to avoid that all points on the same horizontal line would overlap too much and become unreadable.

Including one bigger, distinct point per TSO allows to present another key summary indicator - in this case the average RAM across all of its network elements.

Given the complexity of this figure, we have relied on some in-chart annotations to present more context and understanding, and increase the accessibility of our chart.

This is - again - not a new type of graph, just one less frequently used. The idea came from a very instructive, clear and engaging tutorial created by Cédric Scherer . In this blog post, he explains how to transform a rather basic pair of boxplots (which you could also apply to our example above) to a beautiful "combination of a jittered dot strip plot and a lollipop plot". You can check it out here: link.

Loop flows

The final figure shows how loop flows (for more information about this concept, see also the following article) are managed by Elia, the Belgian transmission system operator. Phase-shift-transformers (PSTs) are used to "push back" loop flows, hence reducing their impact on the Belgian grid. Elia calculates these loop flows before and after the PSTs have been optimized.

Intuitively it would make most sense to plot the difference between the loop flows before and after the PST optimization, where the horizontal axis shows time and the vertical axis shows the difference. This would 1) result in a very volatile line graph, and 2) the reader would no longer be able to assess the absolute level of loop flows, as only the difference is shown.

As the time aspect is not that important, but we want to focus on the difference and absolute levels, we have chosen to present a scatter plot with both variables (before and after). The difference is implicitly shown via the red "identity line" or "1:1 line". Values above the line are - as the annotations show - hours where the loop flows have increased following the optimization, while the values below are those where the loop flows decreased.

The accompanying text specifies some intuitive suggestions from the figure: about 90% of the values are below the identity line (indicating that the PST optimization successfully reduced loop flows), leading to an average decrease with 273 MW.

Concluding reflection

Just as reading it, writing a report containing only bar charts and line graphs is rather dull and not very interesting. In order to make our Monitoring Report a bit more, we have started - since a couple of years - to think more closely about the way in which we analyse, present and contextualize data and figures. Visualizing data helps is in identifying problems and opportunities, as well as communicating our interpretations of these problems and opportunities to a wider public.

Yes, the figures presented above and throughout our report are more complex and sometimes take more time to interpret, but at the same time - I truly hope - more pleasant to look at, more engaging and have a higher amount of information encoded in them, often leading to a stronger - yet also more nuanced - message. And that, in the end, is the aim of the report: explaining how electricity markets function, by looking at very complex information and presenting it in a way which is correct, beautiful and clear. This idea is best elaborated in "Powerful Charts" from Koen Van den Eeckhout , from which I borrowed the following overview. If you're interested in the theory behind effective data visualization, along with a very broad repository of practical examples, I highly recommend to have a look at this book.

That's it - if you have any questions or remarks, or suggestions for improvements: feel free to reach out. We are always looking for new ways to process and analyze our datasets into engaging visualizations, so any input is highly welcomed!