I have data measured at fixed intervals in one process. It needs to be weighed with cost value that is generated in another process that is analogue and the costs it generates have jumps.

Imagine a wind turbine that is recording wind speed every 250ms with power storage and consumption. Every time storage connects/disconnects to external grid - a fixed value cost generated (say 50p).

While connected it can generate positive or negative amount which can be sampled at same 250ms intervals. Values are fraction if compared to 50p connect/disconnect penalty.

Also connects/disconnects never aligned properly with 250ms sampling.

If I just add 50p to nearest sample any RNN would be confused to how to deal with it and fitting data wont ever work properly.

Ultimately I need RNN to predict good wind pattern where cost will be minimal which is highly influenced by the fact that frequent connects drive costs high very fast.

How to process data in this case so it fits a generic RNN model?


1 Answer 1


It seems reinforcement learning is better suit this than supervised learning. The jumps can be added to rewards easily and intuitively


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.