ddpg

## CSTR MODEL ## 23/01/21 ## import gym from gym import spaces from gym.utils import seeding import numpy as np from os import path import numpy as np import random from scipy.integrate import odeint class cstr_model(gym.Env): def __init__(self): self.max_Ca = 0.3 self.max_Qchange = 10000 self.rho = 780 #kg/m3 self.Cp = 3.25 #kJ/kgK self.EaA = 41570 #Kj/kmol self.EaB = 45727 #Kj/kmol self.F = 0.0025 #m3/min self.DH = 4157 #Kj/kmol self.k0A = 50000 # 1/min self.k0B = 100000 #1/min self.R = 8.314 #Kj/kmolK self.V = 0.2 #m3 self.Ca_ss = 0.3 self.Cb_ss = 0.0 self.T_ss = 300 self.Tf = 300 high = np.array([1.,1.,self.max_Ca],dtype=np.float32) self.action_space = spaces.Box( low = -self.max_Qchange, high= self.max_Qchange, shape=(1,), dtype=np.float32 ) self.observation_space = spaces.Box( low=-high, high=high, dtype=np.float32 ) self.seed() def seed(self, seed=None): self.np_random, seed = seeding.np_random(seed) return[seed] def time(self, j, max_steps): self.t = j self.max_steps = max_steps def befstate(self, s): self.s = s #step def step(self, a): if self.t == 0 : self.Q = a else: self.Q = self.Q + a self.Q = np.clip(self.Q, -self.max_Qchange, self.max_Qchange)[0] def cstr(x,t,Q,Tf,Caf): #self??? Ca, Cb, Temp = self.state self.state[0] = x[0] self.state[1] = x[1] self.state[2] = x[2] self.ka = self.k0A * np.exp(-self.EaA / (self.R * Temp)) self.kb = self.k0B * np.exp(-self.EaB / (self.R * Temp)) self.rA = self.ka * Ca * self.V self.rB = self.kb * Cb * self.V # Calculate concentration derivative dCadt = (self.F * (self.Ca_ss - Ca) - self.rA + self.rB) / self.V dCbdt = ((self.F * -Cb) + self.rA - self.rB) / self.V # Calculate temperature derivative dTempdt = ((self.F * self.rho * self.Cp) * (self.T_ss - Temp) + self.Q - self.DH * (self.rA - self.rB)) / (self.rho * self.Cp * self.V) # Return xdot: xdot = np.zeros(3) xdot[0] = dCadt xdot[1] = dCbdt xdot[2] = dTempdt return xdot self.x0 = np.empty(3) self.x0[0] = self.s[0] self.x0[1] = self.s[1] self.x0[2] = self.s[2] ts = [self.t,self.t+1] y = odeint(cstr,self.x0,ts,args=(self.Q,self.Tf,self.Ca_ss)) newCa = y[-1][0] newCb = y[-1][1] newTemp = y[-1][2] self.x0[0] = newCa self.x0[1] = newCb self.x0[2] = newTemp # calculate the reward for the state if self.t == (self.max_steps-1): costs = newCb else: costs = 0 #the last concentration in the episode... # store the new observation for the state self.state = np.array([newCa, newCb, newTemp]) # check if the episode is over and store as "done" return self._get_obs(), costs, False, {} def _get_obs(self): Ca, Cb, Temp = self.state return np.array([Ca, Cb, Temp]) # reset method def reset(self): high = np.array([0.3, 0.0, 300]) low = np.array([0.0, 0.0, 0]) self.state = self.np_random.uniform(low=low, high=high) self.last_u = None return self._get_obs() # render def render (self): data = np.vstack(([j],self.Q,Temp)) # vertical stack data = data.Temp # transpose data np.savetxt('data_doublet.txt',data,delimiter=',',\ header='Time,Q,T',comments='')