you can use some smoothing method for data processing.

I am puzzled by what you mean by error here. Electricity bills and consumption figures are obtained from instruments and calculated using precise algorithms. If your electricity supply company can't do this without introducing " error data which have value larger than our standard " then I suggest you change your supplier. Like NOW.

We encountered a similar problem a little while ago. There was a discrepancy between data collected from a number of data loggers and the 'point of entry' meter. It arose from where the readings were taking.

As the energy flowed from the inverter, line losses and other diversions of energy created a discrepancy. They are miniscule but over time the cumulative effect is noticable. There are also error parameters in every meter so one might be on the high side and another on the low side, but both within required specifications.

I found that with sub-meters in water distribution networks trying to reconcile to mains meter reading is impossible. Might I suggest as long as the discrepancies are within tolerable limits, and reasonably constant - ie no obvious leaks between the two monitoring stations, just note the discrepancy and live with it? Trying to match all data to the nth degree will drive you insane. Mind you, I am not a scientist.

Australian electrcity meter parameters are set by the National Measurement Institute and Australian Standards

. That might help.

My bet is that you are monitoring from two points and trying to reconcile one with the other. Losses can be measured and/or calculated. Be also aware of Tony Maine's answer.

Thanks for all answer and suggestion. I already find the way by comparing it to the standard value from electric company. And remove some data which have too large value from the standard value. I also implement some standard error to remove some data which has strange value.