If you have partial sequences for protein coding genes, it is likely that they do not start at the first codon position. If that is the case, and you count the triplets from the very first nucleotide, it will lead to a frameshift in the codons, and stop codons are likely to occur. One way of obtaining the right frame is to compare your partial sequence to a fully sequenced gene in a closely related species and see if you can determine the position of the first nucleotide in your partial sequence. Once that is determined, you can then obtain the right coding frame in your sequence and you should then be able to avoid getting any premature (internal) stop codons.

It struck me as well when recently doing PAML analyses. Some of the programs ignore such internal stop codons and mark them as ???.

Also you might want to check when these sequences were uploaded. Normally NCBI says it checks for internal stop codons:

https://www.ncbi.nlm.nih.gov/genbank/genomesubmit_annotation/

Maybe these submissions were made before this policy, and if not reported, they are not curated. That would make them a theoretical CDS sequence, not a real one, or a real one, but with an actual error, that should be reported

If they were my sequences, I would be concerned that I had inadvertently sequenced a pseudogene. That's happened to me before; I got a stop codon where there shouldn't be one in the middle of the coding sequence for cytb.  I then tried a different set of primers with the same samples and got moderately different cytb sequences with no premature stop codon.  The two cytb sequences where ~10% divergent, but only one could've been the functional copy of cytb. 

I'd also make certain I was using the correct translation table and that the sequence I was submitting was indeed the coding strand and not the template strand.  I've run into older sequences on GenBank that are labeled cox1 but you have to reverse complement them to align with other more recent cox1 sequences.   If you still want to submit your sequences as is with the internal stop codons and none of those other things are the problem, then you may be able to get away with calling them cytb-like or cox1-like (non-coding).  I've seen that on GenBank sequences before.  However, I'm not sure how useful the sequences will be to other researchers if you go that route.

Thank you all for responding. @ Artur Burzynski !  my question is related to No. 2. I have problems with my submission as they give internal stop codons. I am saying that my sequence give internal stop codon, even when my nucleotide sequence are the 100% same as already existing in genbank from others. I think that perception of Donald B Shepard is prevailing in my case.

"This happened even if the nucleotide sequences are same as of other closely related species already submitted to the gene bank."

I agree with Artur. If already submiited entries with the the same sequence motifs as yours are properly translated into aa in GenBank I would suggest that your problem is clearely related to one or more of the items mentioned by Artur above. In my own experience it was often the default choosen translation table (nuc/mito) or the wrong codon start point.

If you have partial sequences for protein coding genes, it is likely that they do not start at the first codon position. If that is the case, and you count the triplets from the very first nucleotide, it will lead to a frameshift in the codons, and stop codons are likely to occur. One way of obtaining the right frame is to compare your partial sequence to a fully sequenced gene in a closely related species and see if you can determine the position of the first nucleotide in your partial sequence. Once that is determined, you can then obtain the right coding frame in your sequence and you should then be able to avoid getting any premature (internal) stop codons.

Thank you for guidance, I got your point. @ Artur Burzynski, Dietmar Lieckfeldt, Sudhindra R Gadagkar.  

I have the same problem months ago, the genbank submissions staff send me, a link to figure out the problem

https://ncbiinsights.ncbi.nlm.nih.gov/2015/10/02/troubleshooting-genbank-submissions-annotating-the-coding-region-cds/

in this web page of Troubleshooting GenBank Submissions: Annotating the Coding Region (CDS) said " Blastn will always use the Standard Genetic Code. With blastx, you can specify the appropriate genetic code the error the stop codons appear when the genetic code is not the universal”.

My sequin file was free of errors, then, went I submit my sequences I included a note to inform the genetic code that I used for those sequences and BLASTX CDS to read those,

I hope this works for you.

@Obdulia Segura-Leon. Thank you very much. It will really help me.

You can try different translation frames for plus or minus strand during the submission.

check the CDS features while you submit your sequence. Basically you need to change the CDS (coding region) start. Check "specific spans" button of the "nucleotide interval span item" and adjust the start and the stop boxes. I am pretty sure that this will resolve your problem.