AI Challenge HIT 2022 - Team 10.

---

Table of Contents

1. The Challenge
2. Repository Usage
   • Depencencies
   • Run the code
   • Structure

---

1. The Challenge

MeteoTrentino is a structure of the Autonomous Province of Trento founded in 1997. It counts more than 100 hundred weather stations all over the Trentino region, collecting a variety of meteorological data:

• temperature
• humidity
• wind intensity
• precipitation abundance
• snow levels

The data collected from MeteoTrentino are of interest not only for weather forecasting but also for monitoring and research for example in meteorology, nivology and glaciology, therefore maintaining good quality archived data is of crucial importance.

The relevant number of weather station and the critical conditions that these encounter (abundant precipitations, critical temperatures, damage by animals and vegetation) make the task of validating the data quite cumbersome. Damage, malfunctioning or anomalous conditions of the sensors may result in untruthful data.

Intervening to restore the sensors to proper functioning may consists in a remote reset but often it may require a manual intervention on the weather station site. Such intervention understandably take several hours or even days, therefore detecting anomalies in the shortest possible time is essential in order to garantee continuity of good quality stored data.

Currently, anomaly detection is performed “manually” by an operator of MeteoTrentino, based on a protocol which generates alerts about which sensor/weather station encountered an anomaly. At the moment, the alerts are generated by a simple signal analysis code based on threshold levels of the acquired signals. The process is overall too slow due to the huge load of incoming data and the low performance of the code, which is often not able to detect anomalies that remain unnoticed for long times.

The challenge proposed by MeteoTrentino is to use Artificial Intelligence to make up a code able to effectively detect anomalies in the real-time data and be integrated in the protocol to generate the alerts, thus supporting the operator in charge of data quality control in his work.

---

2. Repository Usage

• Dependecies

Pytorch

# Python 3.x
pip3 install torch torchvision

os-sys

pip install os-sys

Easiest way is to install the rest of the libraries/modules is Anaconda [here]

otherwise

Numpy

pip install numpy

Matplotlib

python -m pip install -U matplotlib

Pandas

pip install pandas

---

• Run the code

clone the repo with

git clone https://github.com/eliazonta/AI-Challenge-HIT.git

• Structure