训练一个深度网络的时候，loss和accuracy始终没有变化

Question

model = keras.models.Sequential()
model.add(layers.Bidirectional(layers.LSTM(50, return_sequences=True,activation="relu"), input_shape=(19,1)))
model.add(layers.Bidirectional(layers.LSTM(100, return_sequences=True,activation="relu"), input_shape=(19,1)))
model.add(layers.TimeDistributed(layers.Dense(32, activation= "softmax" )))
model.add(layers.Flatten())
model.add(layers.Dense(10))
model.summary()

model.compile(optimizer=keras.optimizers.Adam(lr=0.0005),
              loss='categorical_crossentropy',
              metrics=['acc'])

history=model.fit(x_train,y_train,batch_size=128, epochs=100,validation_data=(x_test,y_test))