def evaluate_fitness(self, params):
"""
Evaluates fitness of specified individual.
:return: Fitness of the individual
"""
network_weights = params[0]
seed = params[1]
network = Network(self._input_shape, self._output_shape, seed, nn_width=self.nn_width, initializer="zeros")
network.set_weights(network_weights)
gym = GymEnvironment(self._env_name, seed=seed)
state, done = gym.reset(), False
# start_time = time()
step = 0
equal_steps = 0
rewards = []
while not done:
if self._render_each and step % self._render_each == 0:
gym.render()
state = np.expand_dims(state, 0)
action = network(state).numpy()[0]
next_state, reward, done, _ = gym.step(action)
if self._min_equal_steps > 0:
if np.allclose(state, next_state):
equal_steps += 1
else:
equal_steps = 0
rewards.append(reward)
state = next_state
step += 1
if step >= self._max_episode_len:
done = True
elif self._min_equal_steps > 0 and equal_steps >= self._min_equal_steps:
done = True
# add expected reward if we waited till the episode would end
rewards.append((self._max_episode_len - step) * np.mean(rewards[-self._min_equal_steps:]))
# print(f"Total steps {step}: {time() - start_time:.4f}")
total_reward = np.sum(rewards)
return total_reward
self.SPMe.__init__(init_soc=self.SOC_0)
# self.state_of_charge = (np.random.uniform(.65, .99, 1)).item()
# self.SPMe.__init__(init_soc=self.state_of_charge)
# self.state_of_charge = self.SOC_0
# self.SPMe.__init__(init_soc=self.SOC_0)
self.sim_state = self.SPMe.full_init_state
[
bat_states, new_sen_states, outputs, sensitivity_outputs, soc_new,
V_term, theta, docv_dCse, done
] = self.SPMe.SPMe_step(full_sim=True,
states=self.sim_state,
I_input=0)
self.sim_state = [bat_states, new_sen_states]
self.state = self.unpack_states(bat_states, new_sen_states, outputs,
sensitivity_outputs)
self.steps_beyond_done = None
return np.array(self.state)
if __name__ == '__main__':
gym = SPMenv()
gym.reset()