Short answer is yes, you can use Conv layers in a model for regression. However, to do this, you’ll need to process the data differently. Time series data means that information is contained in the order of the data points, but just the data points themselves. To capture this, you’ll either need to use Sequence Model (think RNN/LSTM/GRU), or chop up the data into labeled subsequences.
There’s a great, lightweight course about how to do this by Lawrence Moroney and Andrew Ng on Coursera. Courses are free for the first 7 days, and I don’t believe you get charged if you finish the course before those 7 days are up. I’ve been through this course, and you can do the entire thing in about 3-4 hours in total. The code samples from this are all you need, and the videos explaining the additional preprocessing come with working notebooks that’ll make this a cinch for you. The final model it has you build uses Conv layers and LSTM layers.
https://www.coursera.org/learn/tensorflow-sequences-time-series-and-prediction