It's a fairly well-used statistical model based on aggregating the quality of chances a team generates / concedes and their probability of scoring from each of them. So to give an example, Ebere Eze banging a fk in off the bar from 30 yards looks great, but 9 times out of 10 it'll miss. A gameplan built around doing that every game would not generate many points. Similarly, expecting your keeper to have a worldie every game isn't a sustainable plan for a season.
Looking at the current stats, some things jump out:
Liverpool are conceding more and scoring fewer than they should - if this reverts to the mean (as expected) they'll win the league. Spurs are over-performing on both metrics atm, which isn't sustainable if there isn't an underlying reason for it.
You guys have scored ~4 more goals than you should from the chances you've had, but also conceded ~2 more from the chances given away. So you're in round about the right place.
Palace are over-scoring and under-conceding, suggesting they'll drop down the table as the season goes on. Both Fulham and Burnley are under-scoring and over-conceding - meaning they're far from relegation certs given their performances.
West Brom however are exactly where their xG says they should be - 6 scored, 18 conceded. Meaning that's a good bet they'll stay there.
It's an interesting model to analyse how teams are actually performing not just see their scores. All available on InfoGol.