I have a project with an Lcd counter. it increase by One, every
second more or less because millis is not accurate
As I understand it from your previous threads there is something blocking the timely execution of the one second millis() based timer. I don't see how constantly querying the i2c bus over and over to determine if a second as passed yet is going to help solve your problem. In fact, that overhead is likely to add to it.
I think the DS3231 has the capability to output a 1 hz square wave on one of it's pins. I would think about using that.
I agree with @mixographer
But I don't see that millis() "inaccuracy" is the root cause of the failure of the software timer's ability to run exactly every second. Using another, more accurate measure of time, will not solve the problem of blocking code and sensor measurement time.