Skyline will always choose the peak which overlaps with the Explicit Retention Time regardless of the quality of the peak that is found there.
The only way to prevent Skyline from doing this is to leave the Explicit Retention Time blank.

In the "Candidate Peaks" grid in your screenshot, you can see that the score in the "Identified Count" column is 20, which is the bonus that Skyline assigns to a peak which overlaps with the Explicit Retention Time.

The recommended thing to do is to provide the predicted retention time to Skyline in a different way such as by using iRT.
You can learn about iRT in this tutorial:
https://skyline.ms/wiki/home/software/Skyline/page.view?name=tutorial_irt

Another thing you might be able to do would be to tell Skyline to use a peak scoring model which does not include the "Identified Count" feature.
You can learn about training peak scoring models here:
https://skyline.ms/wiki/home/software/Skyline/page.view?name=tutorial_peak_picking

Telling Skyline to use a peak scoring model which does not include the "Identified Count" feature would be a little tricky.
If you send me your Skyline document I could send you some screenshots showing how to do it.
In Skyline you can use the menu item:
File > Share
to create a .zip file containing your Skyline document and supporting files including extracted chromatograms.
Files which are less than 50MB can be attached to these support requests. You can always upload larger files here:
https://skyline.ms/files.url

I am not 100% sure that my idea about creating a peak scoring model without the Identified Count feature would actually work, but I will try to figure it out after I see your Skyline document.
-- Nick

Hi Nick,

Thank you very much. Attached please find the document file.

Heyi

If you want Skyline to not use the "Identified Count" feature when scoring peaks, then what you can do is:
1. Refine > Reintegrate
2. Choose "<Add...>" in the "Peak scoring model" dropdown.
3. In the "Edit Peak Scoring Model" dialog, choose "Default" in the "Choose model" dropdown, uncheck the box that says "Use decoys" and check the box that says "Use second best peaks".
4. Uncheck the box next to "Identified count" in the "Available feature scores" grid.
5. Push the "Train model" button
It should look like the attached screenshot "NoIdentifiedCountPeakScoringModel.png".
6. Type something into the "Name" box in the Edit Peak Scoring Model dialog.
7. Push OK on the Edit Peak Scoring Model dialog.
8. The new model should now be selected in the "Peak scoring model" dropdown of the Reintegrate dialog. Press OK to have Skyline choose peaks with the new peak scoring model which does not include the Identified count feature score.

-- Nick

Thank you very much, Nick and it works. One more question is about model, what is difference between mProphet and default? Heyi

When Skyline trains an mProphet scoring model, Skyline uses a support vector machine (SVM) to figure out what weights to apply to each of the individual features in order to achieve the best separation between the scores of the targets and the decoys.

In contrast, when Skyline trains a "Default" model, the relative weights of all of the features will remain constant. Skyline is only trying to figure out what to multiply and offset the combined score by so that the combined score can accurately be interpreted as a Z-score. This stretching and offsetting of the combined score has no impact on which peak Skyline ends up choosing, because the scores of all of the candidate peaks have been stretched and offset by the exact same amount. For this reason, you do not actually need decent decoys, and you can get by just using the "second best peaks".

The other thing about the Default scoring model is you can use all of the available features, even if some of the peptides in the document do not have a valid values. If a particular peptide is missing a feature for a particular replicate, Skyline treats the score as zero.

When you are training an mProphet scoring model, Skyline only allows you to use features that have values for all of the peptides in the document.

If you would like to learn more about adding decoys to your document and training an mProphet scoring model you can look at the Advanced Peak Picking Models tutorial:
https://skyline.ms/wiki/home/software/Skyline/page.view?name=tutorial_peak_picking

-- Nick