Author: mim

Tracking AEMO data using PowerBI

I was looking for the Power Production  of a particular solar farm, and I couldn’t find any public dashboard that show this level of details, all I could find was high level aggregated data (Later after I built the dashboard I found this excellent resources Nemlog)

The dashboard is published here  https://djouallah.github.io/AEMO-POWERBI/  , it is refreshed every day at 5 AM

Capture

How it works

Australian Energy Market Operator (AEMO) publish all kind of datasets,  one I believe is real time (require a  subscription ) but for my particular use case, I m interested in this dataset

http://www.nemweb.com.au/#daily-reports

there are two folders :

  1. Current, last 60 days of data ( current day not included, Updated at 4 AM)
  2. Archive : the last 13 Months of data ( current month not included, Updated Monthly)

Pulling data from a website and building a dashboard in PowerBI is straightforward,  it took me a couple of hours on a weekend to do it, the problem is how to maintain it.

Ideally, you build a dashboard and all the refresh is done by the service, which was not the case here

  • Pulling the data directly from the archive is very slow, it takes nearly 3 hours ( unzip, filters only the data we are interested in), and is not sustainable as the earliest month will be removed from the website, I like to keep the history, and it is really bad practise to download the same data every day
  • To keep the history, we need to save the archive somewhere else, too easy , just save it on a local laptop
  • History issues solved, now we created a new problem, on-premise data require a gateway, basically you need to install a software on your laptop, and obviously the laptop must be on when you do the refresh

After playing around of some options, I come up with this workflow

  • Create a local folder that contains all the archive files.
  • Create a PowerBI data model on the desktop just to process the archive data
  • Export to clean tables ( price and Production ) to CSV using DAX studio !!!!!!
  • Load the CSV to azure blob storage ( to get rid of the gateway)
  • Load the current zip files from the web site , it does not require the gateway, but you need the following consideration                                                                                 – Use relative path in web.content functions ( see Chris Blog) and @TheBIccountant 

Web.Contents("http://www.nemweb.com.au/REPORTS/CURRENT/",[RelativePath = "Daily_Reports/" ])

Don’t use Web.Page function but parse it using XML or csv , ( Thanks Reda Rad for the advise)

so you can use something like this

Table.FromColumns({Lines.FromBinary(Web.Contents("http://www.nemweb.com.au/REPORTS/CURRENT/",[RelativePath = "Daily_Reports/" ]), null, null, 65001)})

  • Append the data from azure blob storage and the current folder from the web site, the refresh is now very fast, as PowerBI just read the csv without any transformation
  • Publish to web

Good so far, I manage to get rid of the gateway, the dashboard is refreshed automatically in the service, no maintenance for 60 days.

as the current folders contains data for the current 60 days only, you need to update the initial CSV files.

  • Download the pbix from the service, export the csv , and upload to blob storage, you need to do that only once every 60 days.

PRO

  • PowerBI Publish to web is an amazing service and it is totally free
  • Powerful solution without writing any codes
  • PowerBI free license is free 🙂

Cons

  • Publish to web is not suitable for real time, as it takes nearly 1 hours to propagate the update to the web site, that’s why I can’t publish the current day data, which is updated every 5 minutes.
  • Publish to web does not include export data from the visual
  • pricing for azure blob storage can be tricky : storage itself is very cheap, data upload is free, download in the same region is free ( for example blob to PowerBI service), but when you read data from the blob to PowerBI desktop you incurs charges, so just be careful, it is not your Onedrive model, where download is free.

we showed here a simple workflow using PowerBI free license and azure blob storage (Dropbox), it is very easy but with one inconvenient you need manual operation once every two months, that’s a bit annoying.

edit 23-June-2019 :after I published this blog, I got an excellent feedback from Maxim Zelensky, actually using PowerBI dataflows ( require a PRO license), we can fully automated the whole process, as with dataflows we can have a self reference query, I am not going to repeated here, go and read it

edit 24-June-2019: as it is a personal project, and the data is public, I am not really excited about using a paid service to host the CSV files, I moved the two csv files from blob storage to dropbox, it is totally free, so the whole dashboard infrastructure is free, Good work Microsoft

edit 26-June-2019 : a proper solution will be to save the raw data in a data lake, see here

Edit: 7 -Sept-2019 : I was a bit tired of workaround, now using Google Cl0ud functions and BigQuery , see blog here

Edit : 18-August-2023 : Finally we Have Fabric, This pattern is solved using Dataflow Gen2

Using Or Conditions between Slicers to filter Primavera Schedule using DAX

The Interactive report is published here,  you can download the PBIX here.

In PowerBI or Tableau or any BI solution by defaults when you have multiple slicers, the filter conditions is always, AND, in this particular case, the user want to filter a schedule based on a start date  OR the end date OR the activity status, basically how to reproduce the following Primavera Filter using DAX

Primavera

 

 

Using the normal slicers will not work as it will simply show the activities where all the conditions met, instead we will use some DAX techniques (disconnected slicers and filter using the result of a measure) to get a new behavior.

 

Let’s build a simple data mode.

1-The main table is task: it contains Activity ID, Activity Status, Start, Finish

2- add a date Table, Start_Date, don’t link it to any table, use it in a slicer , the table is generated using Powerquery,

3- add another Date Table, Mstdate, don’t link it either, use it in a slicer

model

 

4- add Activity status Table, don’t link it, , use it in a slicer

5- add this measure

Filter_measure = CALCULATE(COUNTROWS(task),

FILTER(task, ([Start] <= LASTDATE(Start_Date[Date])

&& [Start]>= FIRSTDATE(Start_Date[Date]))

||

([finish] <= LASTDATE(Finish_Date[Date])

&& [finish]>= FIRSTDATE(Finish_Date[Date]))

||

IF(NOT(ISFILTERED(‘Activity status'[Activity status])),BLANK(),’task'[Activity status]IN ALLSELECTED(‘Activity Status'[Activity Status]))))

 

Add this measure to the matrix visual as a filter only, and make it not equal null, and voila

result

Load polygons geometry into PowerBI data model

Starting a new project that involve plotting a relatively big map ( 3000 polygons), it should be straightforward, as I have done it before, but a limitation in my prefered PowerBI custom visuals had an interesting result.

1-  synoptic panel: in my view it is the best visual to show custom map, ( floor plan, general layout etc), unfortunately not this time, my new map has 3000 polygons, when I tried to plot some attributes, I get the equivalent of Windows screen of death for PowerBI, too many values !!!

1

That’s bad, I filled a bug report, the author was kind enough to reply that it will be fixed in a next release, (something to do with incremental loading) fine let’s try other options

2- Shape map : (it did not even render the polygons correctly, and to be honest the visual did not receive any update since the first time I use it, 2 years ago)

3- Mapbox: showstopper, to load your own shape file and keep it private, you need a business plan.

I am stuck, I can’t plot a 3000 polygons map in PowerBI, let’s try Rstats.

4-Leaflet: is very versatile mapping engine, right now I use it in other projects (outside of PowerBI) and it plot 60K points and nearly 500 polygons in sub 2 seconds, but there is a problem, Leaflet output is HTML which is not supported under PowerBI, I never manage to make decent screenshot, something to do with the zoom.

5-ggplot2 + SF : quick google and got this code

library(sf)

library(tidyverse)

map_shp <- read_sf(‘C:/Users/mimoune.djouallah/test.geojson’) map =dataset %>%

left_join(map_shp, ., by = c(‘id’ = ‘id’)) %>%

ggplot() +

geom_sf(aes(fill = type),lwd = 0)

map

2

Great, I have my map now ( this is only a portion, I can show the whole layout as it is proprietary), there is only a little problem, code take 25 seconds to render, it is very annoying, especially as in PowerBI you touch anything, and all the visuals render again.

my first thought maybe the join between PowerBI dataset ( dataframe) and the geojson is slow !!! the beauty of R integration in PowerBI is: you click on R icon, a new window open with an empty dataset, then you add the columns you want to analyse, yes that simple no ETL, no messing with data, you have to the mighty data model.

Loading Geometry data into PowerBI data model ?

 

After reading the SF documentation, I learned all I need is a dataframe  with a column that store geometry data, too easy just use QGIS , save as csv and make sure to select geometry, WKT

Capture

Now I loaded the csv file to the data model and join it with the attribute

3

Now here is the new code, notice, we don’t load any external file, all data is from the data model

library(sf)

library(tidyverse)

map_file= st_as_sf(dataset, wkt = “geometry”)

map = ggplot(map_file) +

geom_sf(aes(fill = type),lwd = 0)

map

basically, you need just to tell SF which column represent the geometry , unfortunately it is still slow maybe now it is 24 seconds

Another round of googling and I find the issue ggplot2 is very slow to render, no problem, this is R after all, there are a least a couple dozen of packages that do the same thing 🙂

6-Pure SF

 Turn out SF is not only to manage Geometric dataframe but it can plot too,( who would thought that)

library(sf)

map = plot(st_as_sf(dataset, wkt =geometry”)[“type”])

map

 render in less than 2 seconds, happy dance  

 

Bonus point, Cross-filing geometry

 My purpose to load geometry was just to speed ggplot2, but hold on, the geometry is in the data model, not only the attribute change ( install quantities, category etc), I can even filter polygons on the fly !!!, this is quite amazing, I can load all the layers and filter out any shape as I want.

4

HTML map using R, Part 1

I always tried to learn R, but I never succeed, I did not like it, after enjoying the GUI of PowerQuery, using R was like a self-inflicted torture. But obviously I was wrong as usual, R is not only for data cleaning.

in a PowerBI usergroup , I heard someone talking about R, and I made a mistake of telling him what R can do that PowerBI can not do ?, all I can say is I wish if I did not ask him this question, anyway, in defending R, he said something about generating HTML file and boom, that’s exactly what I was looking for, a file that anyone can read and it has interactivity and it works offline.

my problem is, I build very nice interactive map visualization, but for large distribution, I just print it to pdf, which obviously fail the purpose, not everyone has access to PowerBI service, and for external parties ( client, subcontractor), you can’t reasonably expect them to subscribe to PowerBI service just to see a map :), using Excel 2016 was an option but unfortunately some functionalities are still missing (can’t show a label) and honestly the integration of 3D map with Excel looks like it was a second thought by the developers.

The good news is, R has an excellent integration of PowerBI, so instead of recreating the wheel, just use R for the stuff that can not be done natively in PowerBI.

I hope to publish multiple blogs to document my workflow.

In my real case scenario which I can not publish for obvious reason, we need to view the status (Completed, or not) of nearly 50,000 Piles, show all the piles will clutter the view, instead, I found by accident that a library called Leaflet (the Engine used by MapBox), has the ability of showing layer only when you zoom.

I will use a dummy data, (literally a foundation on a mountain in a New Zealand as an example),

Capture

the data source is an excel file with the pile location and topojson file to show the foundations location all generated using the Excellent QGIS.
the R code is

library(readxl)
library(leaflet)
library(htmlwidgets)
#read Excel Data
data <- read_excel("pile.xlsx", col_types = c("text","text", "numeric", "numeric"))
#read json
map_data <- geojsonio::geojson_read("foundation.json",what = "sp")
######################################
map <-leaflet(map_data) %>%
addTiles(urlTemplate = "https://mts1.google.com/vt/lyrs=s&hl=en&src=app&x={x}&y={y}&z={z}&s=G", attribution = 'Google') %>%
addPolygons()%>%
addCircleMarkers(lng = data$x, lat = data$y,radius = 3)
map
saveWidget(map,file="foundation.html",selfcontained=TRUE,title = "foundation")

the result is a self contained html, at the moment, it is show only 2 foundations and some piles, in the next blog post, we will add zoom, and how to filter layers, and show labels and stuff.

the result is here, and the source data is located in this github repository.