This GRU model overfits heavily . Is there a way to improve it?

Question

model = Sequential()
model.add(Masking(mask_value=-1, input_shape=(None, feature_shape)))
model.add(GRU(128, return_sequences=True, activation='tanh' , dropout = 0.3 , recurrent_dropout = 0.3, input_shape = (None , feature_shape)))
model.add(GRU(128, return_sequences=False, activation='tanh' , dropout = 0.3 , recurrent_dropout = 0.3))
model.add(BatchNormalization())
model.add(Dense(256, activation='relu'))
model.add(Dense(128, activation='relu'))
model.add(Dense(64, activation='relu'))
model.add(Dense(actions.shape[0], activation='softmax'))
model.summary()

Things I have changed but didn't show much results at overcoming overfitting:

1. increasing the dropout and recurrent_dropout
2. adding batch normalization between dense layers.
3. Using L2 regularization underfitted the GRU model when done on dense layers.