GAMS条件参数TWC(j)实现求助：基于HS(j)的分段取值

Hey there! Let's sort out this conditional logic issue you're hitting in GAMS. The problem with your current approach is that using HS.l(j) (the level value from a prior iteration) in a $ condition doesn't work for endogenous variables like HS(j)—since HS(j)'s value isn't known when the model starts solving, you can't base a parameter or constraint on it that way. Instead, we need to model this piecewise condition explicitly within the optimization constraints. Here are two reliable methods to implement what you need:

Method 1: Binary Variable + Big-M Constraints (Most Compatible) #

This approach uses a binary variable to track whether HS(j) meets the 1000 threshold, then links that binary to the cost term in your objective.

1. Define a binary variable:

   binary variable y(j);
   

   y(j) will be 1 when HS(j) ≥ 1000, and 0 when HS(j) < 1000.

2. Add constraints to link y(j) and HS(j):
   Choose a scalar M that's larger than the maximum possible value HS(j) could take (e.g., 1e6 if your HS values don't go beyond that):

   scalar M / 1000000 /;
   
   HS_Threshold_Low(j).. HS(j) >= 1000 * y(j);
   HS_Threshold_High(j).. HS(j) <= 1000 + M * (1 - y(j));
   

   • The first constraint ensures that if y(j)=1, HS(j) must be at least 1000.
   • The second constraint ensures that if y(j)=0, HS(j) can't exceed 1000 (the M*(1-y(j)) term drops out when y(j)=1, so no upper bound is enforced there).

3. Update your objective function:
   Replace the TWC(j)*HS(j) term with 3.21*y(j)*HS(j). This automatically handles the conditional logic:

   • When HS(j) ≥1000, y(j)=1, so the term becomes 3.21*HS(j)
   • When HS(j) <1000, y(j)=0, so the term equals 0

Method 2: Piecewise Linear Function (For Solvers That Support It) #

If you're using a solver that handles piecewise linear functions (like CPLEX or Gurobi), you can define the conditional cost directly as a piecewise linear term:

1. Define the piecewise linear points:
   We need to map HS(j) values to the corresponding cost term (TWC(j)*HS(j)):

   parameter pwl_points(4,2);
   pwl_points(1,1) = 0;      pwl_points(1,2) = 0;       # HS=0 → cost=0
   pwl_points(2,1) = 1000;   pwl_points(2,2) = 0;       # HS=1000 (below threshold) → cost=0
   pwl_points(3,1) = 1000;   pwl_points(3,2) = 3210;    # HS=1000 (at threshold) → cost=3.21*1000=3210
   pwl_points(4,1) = 1e6;    pwl_points(4,2) = 3.21e6;  # Max HS → cost=3.21*maxHS
   

2. Integrate into your objective:
   Use GAMS' piecewise function to include this in your objective equation:

   equation obj_def;
   obj_def.. z =e= [your other objective terms] - piecewise(pwl_points, HS(j));
   

Why Your Initial Approach Failed #

Just to clarify why your original equation didn't work: HS.l(j) refers to the solution value of HS(j) from a previous solve, not the value it will take in the current optimization run. GAMS evaluates $ conditions before solving the model, so it can't use an unknown variable value to set a parameter or constraint dynamically. We have to model this logic as part of the problem's constraints instead of relying on iteration-level values.

内容的提问来源于stack exchange，提问作者Gigi Botte