You can analyse your data using One-way ANOVA followed by post hoc test like Tukey's test.

Time series data actually calles for a regression.

yes, you're right.

independence assumption maybe violated due to time-sries autocorrelation

you can fit an ARIMA model

http://www.ucl.ac.uk/jdi/events/int-CIA-conf/ICIAC11_Slides/ICIAC11_1E_LTompson

Use a Tiao and Box intervention model with a step type dummy variable, that is, a general linear model with a step dummy variable (1 before the change and 0 after) and an error term following an ARIMA process. If the coefficient of the dummy variable is not significant (using a t test) you will not have a change in the mean.

A change in the level of a time can be addressed using an intervention model. Significance of the model parameters indicates significance of the mean change.

A change in the level of a time series can be addressed using an intervention model. Significance of the model parameters indicates significance of the mean change.