Montenegro net position forecast model
In the context of AI4CZC project, this model forecasts the Montenegro net position (difference between generation and load) for the next 48h. The model was trained with data from 2019 to 2021 and tested on 2022 data. It performs with a MAE of 75.8 and a R2 of 0.7309.
Model Card
This models predicts Montenegro net position (difference between generation and load) for the next 48 hours. It uses an input context of the last 12 hours and produces 48 outputs.
This model was built and runs with Inceptive MLE, the training block of Igloo. The model is deployed on AI4CZC platform.
More details about how the model was trained is here : A4CZC model building - Session 1.
Model Inputs
All the inputs are linear values. This is the list of the model inputs:
- load_week_ahead_forecast_max_RS : Week ahead forecast load of Serbia
- ME_outaged_nominal_gen : Total nominal generation that is currently in outage in Montenegro
- IT-CS_generation_hydro_run_river : Italy Center-South bidding zone generation of Hydro run of river.
- ME_generation_brown_coal : Montenegro brown coal generation
- ME_generation_hydro_reservoir : Montenegro hydro reservoir generation
- RS_generation : Total Serbia generation
- ME_generation : Total Montenegro generation
- BA_generation : Total Bosnia generation
- ME_net_position : Montenegro net position
The model considers that every input is separated by an hour. If there a time gap bigger, please provide separate files for each continuous time series.
A valid input is provided on the file sample_entry.zip with 36 entries. Other valid inputs are here.
Model outputs
For an entry of k values, the model will produce k-12 rows of 48 outputs each. This is because the model considers for an entry i, the values of i, i-1 to i-11 and concatenate them as entry. So it can not produce entries for the first eleven values.
Following this logic, our sample file "sample_entry" that has 36 entries will only produces 24 entries.
The model then generates 48 outputs per entry, named output_0 to output_47. They represent the net position for the next 48h.
So for the entry of id i, with a time_stamp of t, output 0, will be the output value for time t+1 and output_47 will be the output for time t+48
Training performances
The model was built with data from 2019, 2020 and 2021. When tested on 2022 data the performances were :
- MAE: 75.8
- R2: 0.7309
- sMAPE: 78.28%
- RMSE: 100.49
Manual
This model is delivered as a Docker image and takes a CSV file as input and writes the results as CSV file. Sample_entry.zip file in this page provides a ready to use sample file to test the model.
Getting the model
We provide this model as a Docker image. To retrieve and run the model, install Docker Engine.
Then get the image :
docker pull abelalonso1305/tech-inceptive-models:ai4czc_1
Running the model
To provide the CSV file you are going to use, you will mount the folder containing the input file on the /data folder of the image. Then the docker image will run and create an output file on the same folder.
If your input file is named sample_entry.csv and it is located in your current directory and you want to save the predictions on output.csv, run :
docker run --rm -v "$(pwd)":/data abelalonso1305/tech-inceptive-models:ai4czc_1 -i /data/sample_entry.csv -o /data/output.csv
CLI Manual
Is is possible to specify some aspects of the provided CSV.
-h,--help Displays this help message.
-i,--input-file Input CSV file path.
-o,--output-file Output file path to print the output CSV.
-id,--id CSV ID column name, default is _id.
-s,--separator CSV column separator, default is ','.
-ec,--escape-char CSV escape character, default is '"'.
If you want to run with a CSV in the current directory, specify the that the CSV separator will be a ";" and the id column will be "id", then run :
docker run --rm -v "$(pwd)":/data inceptive/modelcli:1 -i /data/input.csv -o /data/output.csv --id id -s ;
About this model
This model was done in the context of AI4CZC project.
This ML model is part of a project that has received funding from the European Union’s Horizon 2020 research and innovation programme under I-NERGY grant agreement No 101016508.
