User Tools

Site Tools


Sidebar

Translations of this page:
offsprings:calibration:raptor

Calibration of Raptor Platform Xunxing Diao, Hongling Shi and Kun Mean Hou

Introduction

This document describes the different steps to calibrate Raptor platform before its deployment.

Environment and tools

The calibration is done on server. The server environment is Ubuntu 16.04.4 LTS. The calibration component is implemented in a Debian-based docker container. The data storage uses MongoDB. The main language is Python 2.7, and the main Python packages are Numpy and Pandas.

Procedure

The procedure to calibrate the Raptor platform is as follows:

Handling reference data

  1. Convert the reference data from different formats to a uniform format.
  2. Store it to database collections
  3. Link the collections to the related collections containing raw data, and setup the calibration window

Handling raw data

  1. Get the related raw data in the calibration window
  2. Basic data combining and converting
  3. Smooth data
  4. Get the average value (1 or ½ hour depending on reference data)

Calibration (by multiple linear regression)

  1. Choose variables depending on the available data
  2. Loop:
    1. divide data in training group and test group
    2. loop testing until finding the best parameters of linear model (by ordinary least squares method)
  3. Apply the parameters on all available data and get the R2. If R2 greater than 0.6, the parameters are accepted (normally the R2 of the O3 sensor in Raptors are greater than 0.9 ).

Output

  1. A periodic task on server applying the parameters to the real-time data
  2. Store calibrated data to collections
  3. Provide APIs for CaptorAir and other applications on raptor.im
offsprings/calibration/raptor.txt · Last modified: 2018/09/24 12:33 by roger