updating optimizations for benchmark controllers

Author: dantzert

baseline_controllers/alpha/evaluate_baseline_controllers.py

@@ -21,7 +21,7 @@
 
 # ALPHA
 # options are: 'equal-filling' and 'constant-flow' (or 'uncontrolled')
-evaluating = 'equal-filling' 
+evaluating = 'uncontrolled' 
 verbose = True
 version = "2" # options are "1" and "2"
 level = "1" # options are "1" , "2", and "3"
@@ -52,8 +52,10 @@
 
 
 for parameter in tuning_values:
-    if evaluating == "constant-flow" or evaluating == "uncontrolled":
+    if evaluating == "constant-flow":
         optimal_constant_flows = np.loadtxt(str("./v" + version + "/optimal_constant_flows.txt"))
+    elif evaluating == "uncontrolled":
+        optimal_constant_flows = np.array([1.0,1.0,1.0,1.0,1.0,0.15,0.6,0.5,0.5,0.65])
     elif evaluating == "equal-filling":
         optimal_constant_flows = np.loadtxt(str("./v" + version + "/optimal_efd.txt"))[:-1]
         optimal_efd_params = np.loadtxt(str("./v" + version + "/optimal_efd.txt"))[-1]
@@ -140,7 +142,12 @@
                 regulator_idx = [i for i in range(len(env.config['states'])) if regulator_id in env.config['states'][i]][0]
 
                 # assume linear scaling with opening percentage
-                u_open_pct[idx] = Q_desired / (Cd * Ao * np.sqrt(2*g*state[regulator_idx]))
+                if state[regulator_idx] > 0:
+                    u_open_pct[idx] = Q_desired / (Cd * Ao * np.sqrt(2*g*state[regulator_idx]))
+                else:
+                    u_open_pct[idx] = 1.0
+                if np.isnan(u_open_pct[idx]):
+                    print("nan value for u_open_pct")
                 if u_open_pct[idx] > 1.0:
                     u_open_pct[idx] = 1.0
                 elif u_open_pct[idx] < 0.0:

baseline_controllers/alpha/optimize_parameters.py

@@ -225,7 +225,7 @@ def f_efd(efd_parameters):
     CSO_penalty = 0
     for weir in CSO_flows.keys():
         CSO_penalty += sum(CSO_flows[weir])**2
-    '''
+    
     flooding_penalty = 0
     for junction in flooding.keys():
         flooding_penalty += sum(flooding[junction])*100
@@ -243,7 +243,7 @@ def f_efd(efd_parameters):
     return_value = CSO_penalty + flooding_penalty - peak_filling_reward + sum(data["final_depths"]) + np.std(data['final_depths'])
     
     return return_value
-    '''
+    
     flooding_penalty = 0
     for junction in flooding.keys():
         #flooding_penalty += sum(flooding[junction])*100
@@ -282,7 +282,7 @@ def objective(input_data):
                 flow_cost = 0.0
                 for key,value in data['data_log']['flow'].items():
                     flow_cost += sum(value)
-                objective_cost = flow_cost + sum(data['final_depths']) + np.std(data['final_depths'])
+                objective_cost = flow_cost #+ sum(data['final_depths']) + np.std(data['final_depths'])
                 flood_cost = 0.0
                 for key, value in data['data_log']['flooding'].items():
                     flood_cost += sum(value)
@@ -311,7 +311,7 @@ def constraint(input_data):
                 flow_cost = 0.0
                 for key,value in data['data_log']['flow'].items():
                     flow_cost += sum(value)
-                objective_cost = flow_cost + sum(data['final_depths']) + np.std(data['final_depths'])
+                objective_cost = flow_cost #+ sum(data['final_depths']) + np.std(data['final_depths'])
                 flood_cost = 0.0
                 for key, value in data['data_log']['flooding'].items():
                     flood_cost += sum(value)
@@ -355,7 +355,7 @@ def objective(input_data):
                 flow_cost = 0.0
                 for key,value in data['data_log']['flow'].items():
                     flow_cost += sum(value)
-                objective_cost = flow_cost + sum(data['final_depths']) + np.std(data['final_depths'])
+                objective_cost = flow_cost #+ sum(data['final_depths']) + np.std(data['final_depths'])
                 flood_cost = 0.0
                 for key, value in data['data_log']['flooding'].items():
                     flood_cost += sum(value)
@@ -385,7 +385,7 @@ def constraint(input_data):
                 flow_cost = 0.0
                 for key,value in data['data_log']['flow'].items():
                     flow_cost += sum(value)
-                objective_cost = flow_cost + sum(data['final_depths']) + np.std(data['final_depths'])
+                objective_cost = flow_cost #+ sum(data['final_depths']) + np.std(data['final_depths'])
                 flood_cost = 0.0
                 for key, value in data['data_log']['flooding'].items():
                     flood_cost += sum(value)
@@ -421,10 +421,10 @@ def create_bo_model(data):
     for i in range(10):
         if i < 5: # orifices, set constant flow rate
             lower_bounds.append(0.5)
-            upper_bounds.append(10.0)
+            upper_bounds.append(8.0)
         else: # regulators, set constant opening percentage
             lower_bounds.append(0.0)
-            upper_bounds.append(1.0)
+            upper_bounds.append(0.62)
 
     search_space = Box(lower_bounds, upper_bounds)
 
@@ -485,14 +485,14 @@ def create_bo_model(data):
             upper_bounds.append(10.0)
         else: # regulators, set constant opening percentage
             lower_bounds.append(0.0)
-            upper_bounds.append(1.0)
+            upper_bounds.append(0.65)
     # efd gain bounds
     lower_bounds.append(0.0)
     upper_bounds.append(2.0)
 
     search_space = Box(lower_bounds, upper_bounds)
 
-    num_initial_points = 20
+    num_initial_points = 120
     initial_data = observer_efd(search_space.sample(num_initial_points))
 
     initial_models = trieste.utils.map_values(create_bo_model, initial_data)
@@ -503,7 +503,7 @@ def create_bo_model(data):
     )
     rule = EfficientGlobalOptimization(eci)  # type: ignore
 
-    num_steps = 20
+    num_steps = 450
     bo = trieste.bayesian_optimizer.BayesianOptimizer(observer_efd, search_space)
 
     opt_result = bo.optimize(
@@ -545,10 +545,10 @@ def create_bo_model(data):
     for i in range(10):
         if i < 5: # orifices, set constant flow rate
             lower_bounds.append(0.5)
-            upper_bounds.append(10.0)
+            upper_bounds.append(12.0)
         else: # regulators, set constant opening percentage
             lower_bounds.append(0.0)
-            upper_bounds.append(1.0)
+            upper_bounds.append(0.6)
 
     search_space = Box(lower_bounds, upper_bounds)
 
@@ -563,7 +563,7 @@ def create_bo_model(data):
     )
     rule = EfficientGlobalOptimization(eci)  # type: ignore
 
-    num_steps = 1000
+    num_steps = 500
     bo = trieste.bayesian_optimizer.BayesianOptimizer(observer_cf, search_space)
 
     opt_result = bo.optimize(
@@ -604,17 +604,17 @@ def create_bo_model(data):
     for i in range(10):
         if i < 5: # orifices, set constant flow rate
             lower_bounds.append(0.5)
-            upper_bounds.append(10.0)
+            upper_bounds.append(12.0)
         else: # regulators, set constant opening percentage
             lower_bounds.append(0.0)
-            upper_bounds.append(1.0)
+            upper_bounds.append(0.6)
     # efd gain bounds
     lower_bounds.append(0.0)
     upper_bounds.append(2.0)
 
     search_space = Box(lower_bounds, upper_bounds)
 
-    num_initial_points = 110
+    num_initial_points = 150
     initial_data = observer_efd(search_space.sample(num_initial_points))
 
     initial_models = trieste.utils.map_values(create_bo_model, initial_data)
@@ -625,7 +625,7 @@ def create_bo_model(data):
     )
     rule = EfficientGlobalOptimization(eci)  # type: ignore
 
-    num_steps = 1200
+    num_steps = 500
     bo = trieste.bayesian_optimizer.BayesianOptimizer(observer_efd, search_space)
 
     opt_result = bo.optimize(