Skip to content

Latest commit

 

History

History

BB

readme.md

Source code related to interval B&B

The IbexOpt solver was used. This solver accepts an optimization problem written its own Minibex language. We provide the Minibex files for the four case studies in this folder. Additionally, since IbexOpt will take a very long time (quite a few hours) to get each result, we also provide the results we have obtained here that correspond to Table 2 and Table 3 in the paper.

Source files and results of ibexopt

SDM

  • RTC France
    • Source: RTCFrance-SDM.txt
    • Result: RTCFrance-SDM.result
  • Photowatt-PWP201
    • Source: Photowatt-SDM.txt
    • Result: Photowatt-SDM.result

DDM

  • RTC France
    • Source: RTCFrance-DDM.txt
    • Result: RTCFrance-DDM.result
  • Photowatt-PWP201
    • Source: Photowatt-DDM.txt
    • Result: Photowatt-DDM.result

Other files

  • ./visualization interval BB.ipynb: used to generate the estimated curves using the parameters optimized by BB. The curves are placed in the ./img subfolder as follows.
    • ./img/BB_PW.pdf: curve for PW
    • ./img/BB_RT.pdf: curve for RT

Note

  • The direct objective for optimization is the sum of squared error (SSE), while the RMSE is computed manually after optimization finishes.
  • To set up similar scales among these model parameters, the unit for Rp in optimization is 100 Ω.

How to run

  1. First install ibex following the official instructions
  2. In the terminal, run the ibexopt command with proper options. For example, to optimize the "SDM + RTC France" case with custom relative precision (-r) and absolute precision (-a) as well as given timeout (-t), we can type
	ibexopt -t15000 RTCFrance-SDM.txt -r1e-9 -a1e-13
  1. To run ./visualization interval BB.ipynb, please follow the instructions in the DE folder.

Additional notes

  • The extension of the Minibex file is arbitrary. The Minibex file as well as the result file are just raw text files and can be opened with any text editor.
  • ibexopt can be warm started with the -i option, as we have done in our study. That is, we can first execute ibexopt for a short time, say, one hour, and see whether the result is satisfactory. If not, then we can continue right from where it stopped by specifying the COV file generated in the previous run as the input COV file into the current run.