Generally speaking you don't gain much by filtering genes using microarray data. You could do some filtering, but I never felt it was necessary or helpful to do so. If the batches are orthogonal to the treatment (condition, whatever), then you can simply add a batch effect to your model. If the batches aren't orthogonal (e.g., one or more treatments/conditions is found only in a subset of the batches), then there will be confounding between batch and treatment and that's mostly insurmountable.

As a simple example, let's say you want to do a study to compare weights of people in two different states. In one state, the person getting the weights is using an old scale that uses springs and is in poor condition. In the other state they are using one of those new electronic scales. The weights in this situation are nested in scale (e.g., all the weights from state A are on the old scale, and all the weights from state B are on the more accurate electronic scale), and it is impossible to know if any differences are due to some technical differences between the scales, or are actual differences in weight. You could fix that by getting the scales together and testing with a set of reference weights.

But in the case of microarrays that were processed in different batches in different labs using different reagents, it's not possible to adjust for the lab/batch specfic differences, unless each batch has all the same treatments and controls.

And I gave you advice on simple gene filtering 3 days ago.