Problem with trainable variable while transforming Tensorflow network to Keras version

Question

I want to rewrite a network in DDPG as follows using Keras, and I think what this code does is put state and action into a Dense layer together using Relu, then connect it w