Do you have one for shots conceded, because although this is true: -
...I think the rate we've been conceding has been going up.
Not according to this but you have to appreciate that these graphs are very robust models (are not sensitive to fluctuations) due to the way they are calculated:
Sum(Total Shots conceded)/Games Played
So that is why you're likely to see relatively large fluctuations over the first few games as the sample size is quite small but this soon stabilises as the sample size grows (similar to how if you flip a coin, it won't necessarily be 50/50 for the first 5 tosses, but will most certainly tend towards it after 50). So if we are conceding a few more shots per game in the last couple of games, overall you won't see that in effect. That might sound useless but it can actually be quite useful.
To add to this conversation, here's the xGA (xG Against) over the same time period. Again we're actually quite level but you can see how the same does not apply to Man City: