The main goal is to mine association rules from data where each instance has attributes that can be single valued or time series.

I think we are going to stay away from predicting the next value in a time series.

Let's concentrate first on associating single valued attributes with a single time series attribute.
Example: Given a=1 & b=3, the time series s={3, 5, 2, 6} (or whatever form our time series data takes)

This can be expanded to use other time series attributes on the left hand side of the rule.

Another goal could be to explain the observed attributes through association rules. So, on the right hand side would be the attribute and value pair we want the explained and on the left the "cause". This is a super set of the approaches mentioned above... I think I'll change all the wording to follow this explanation of behaviour idea.

Rules Mined from

1. Multiple single valued attributes & a single a time series attribute
2. Multiple single valued attributes & multiple time series attributes