在Pymoo优化中如何为目标函数输出值添加约束？

Got it, let's walk through how to build this repair function step by step. First, let's clarify the core logic: your target function is f1 = -1 * (constant1 * (variable1 - constant2)), and you want to enforce 10 ≤ f1 ≤ 20 using a Repair function instead of traditional constraint checks.

Step 1: Reverse-engineer the variable adjustment #

Since we need to fix f1 to stay within the valid range, we can rearrange your target function to solve for variable1 when f1 goes out of bounds:

• If f1 < 10: Set f1 = 10, then calculate variable1 = constant2 - f1 / constant1
• If f1 > 20: Set f1 = 20, then calculate variable1 = constant2 - f1 / constant1
• If 10 ≤ f1 ≤20: Leave variable1 unchanged

Step 2: Full Code Implementation #

Here's a complete, runnable example with explanations:

from pymoo.core.problem import Problem
from pymoo.core.repair import Repair
from pymoo.algorithms.soo.nonconvex.ga import GA
from pymoo.optimize import minimize

# Replace these with your actual constant values
CONSTANT1 = 5
CONSTANT2 = 3

# Define your optimization problem
class MyProblem(Problem):
    def __init__(self):
        super().__init__(
            n_var=1,  # Only variable1 as input
            n_obj=1,  # Single objective f1
            n_constr=0,  # We're using Repair instead of explicit constraints
            # No variable bounds set per your request
        )

    def _evaluate(self, X, out, *args, **kwargs):
        # X is a 2D array where each row is an individual's variable1 value
        variable1 = X[:, 0]
        # Calculate f1 using your formula
        f1 = -1 * (CONSTANT1 * (variable1 - CONSTANT2))
        out["F"] = f1.reshape(-1, 1)

# Define the custom Repair function
class F1RangeRepair(Repair):
    def _do_repair(self, problem, X, **kwargs):
        # Iterate over every individual in the population
        for i in range(len(X)):
            variable1 = X[i, 0]
            # Compute the current f1 value for this individual
            current_f1 = -1 * (CONSTANT1 * (variable1 - CONSTANT2))
            
            # Adjust variable1 if f1 is outside the allowed range
            if current_f1 < 10:
                # Force f1 to the minimum valid value, then reverse-calculate variable1
                new_f1 = 10
                X[i, 0] = CONSTANT2 - new_f1 / CONSTANT1
            elif current_f1 > 20:
                # Force f1 to the maximum valid value, then reverse-calculate variable1
                new_f1 = 20
                X[i, 0] = CONSTANT2 - new_f1 / CONSTANT1
            # No action needed if f1 is already in range
        
        return X

# Run the optimization process
if __name__ == "__main__":
    # Initialize problem and algorithm with the repair function
    problem = MyProblem()
    algorithm = GA(pop_size=20, repair=F1RangeRepair())
    
    # Execute minimization (adjust the objective direction if you need to optimize f1 further)
    res = minimize(problem,
                   algorithm,
                   termination=('n_gen', 10),
                   verbose=True)
    
    # Verify that all results meet the f1 constraint
    print("\nOptimized variable1 values:", res.X)
    print("Corresponding f1 values:", -1*(CONSTANT1*(res.X - CONSTANT2)))

Key Details to Note: #

1. Problem Setup: We set n_constr=0 because we're handling the constraint via Repair instead of Pymoo's built-in constraint validation.
2. Repair Logic: The _do_repair method checks each individual's f1 value. If it's outside [10,20], it adjusts variable1 to force f1 back into the valid range using the reversed formula.
3. Algorithm Integration: The repair function is passed directly to the GA (or any other Pymoo algorithm) via the repair parameter.
4. Validation: The final print statements let you confirm that all resulting f1 values fall within your desired range.

If you need to optimize f1 further (e.g., maximize or minimize it while keeping it in bounds), you can adjust the objective function's direction in the minimization call or tweak the problem's objective definition.

内容的提问来源于stack exchange，提问作者Karthick Mohanraj