Dynamic update of model constrains and objective function parameter

[Giacomo Pallucca]

I'm looking for some help to optimize my code execution.

I build up a Model Predictive Control and I use Gurobi QP solver to solve the QP problem on-line.

For each sample time the code create a new model, variables and constrains. The structure of the problem doesn't change during a single simulation, but change the value of coefficients of constraints and objective function.

I want to find a way to update this coefficients due to reduce the time execution of the solver.

Now more than 60% of time execution was passed in build new constrain each time.

In the code example, the QP problem is:

Objective function: x^t H x + F^t x ||| Constrains: A x <= b

F,A and b are matrix that change each time step.

import numpy as N
import gurobipy as gpy
m=gpy.Model()
H=mpcSim.H
A_cons=N.asarray(mpcSim.A_cons)
b=N.asarray(mpcSim.b)
F=mpcSim.F

n_cons,n_stato=A_cons.shape

x={}
for ii in range(n_stato):
    x[ii]=m.addVar(lb=-1e20,ub=1e20,name='x{}'.format(ii))

m.update()    

for ii in range(n_cons):
    m.addConstr(gpy.quicksum(A_cons[ii,jj]*x[jj] for jj in range(n_stato)), gpy.GRB.LESS_EQUAL, b[ii,0],name='Const{}'.format(ii))
m.update()

obj1=gpy.quicksum(gpy.quicksum(H[ii,jj]*x[ii] for ii in range(n_stato))*x[jj] for jj in range(n_stato))
obj2=gpy.quicksum(F[ii,0]*x[ii] for ii in range(n_stato))


m.setObjective(obj1+obj2)

m.optimize()

Thanks in advance,

Jack

[des...@gurobi.com]

Jack,

You don't need to create a model every time. You can create one model and then change the coefficients.

To change a coefficient in the coefficient matrix you can use the following function::

http://www.gurobi.com/documentation/7.0/refman/py_model_chgcoeff.html

To change the linear objective coefficients you have to change the obj variable attribute:

http://www.gurobi.com/documentation/7.0/refman/obj.html#attr:Obj

However, changing the coefficients of a quadratic objective is a bit more complex.

To do this you would need to do the following steps:
1. Call Model.getObjective() to get the QuadExpr

http://www.gurobi.com/documentation/7.0/refman/py_model_getobjective.html

2. Call QuadExpr.remove() to remove the term

http://www.gurobi.com/documentation/7.0/refman/py_quadexpr_remove.html

3. Call QuadExpr.addTerms() to add the new term

http://www.gurobi.com/documentation/7.0/refman/py_quadexpr_addterms.html

4. Finally, call Model.setObjective() to replace the objective

http://www.gurobi.com/documentation/7.0/refman/py_model_setobjective.html

Note that the optimization should be faster as well, as  Gurobi will do a warm start. That is, it will start from the optimal solution of the previous model instead of starting from beginning ( provided you set the parameter Method=0 or 1).

I hope this helps

Amal de Silva
Gurobi Optimization

[Rabab Haider]

Hello there,

I've been looking into creating a few Gurobi models, keeping them in memory, and then modifying & running them as needed (I have a few models I need to solve together over numerous iterations). However, I am using Matlab, and I cannot find any documentation showing if this is possible outside of the Python/C interfaces, and how to do this.

Could you provide any suggestions/direction? I am using Gurobi 7.52

[fiona H]

Hi Amal,

I got a similar problem as Jack's, what if I want to change the 'b' in constraint: Ax <= b, what function can I use to do that?

Thanks,

Fiona

在 2017年4月18日星期二 UTC+2上午12:45:01，des...@gurobi.com写道：

[Michael Winkler]

Hi Fiona,

do you want to change the right hand side (b) just for some constraint(s) or the whole matrix? For the former you might want to directly use

c1 = model.addConstr(x + y >= 1.0, "c1")
c1.rhs = 2.0

if you know your constraints. Otherwise you might want to use the setAttr() method to chnage multiple/all right hand sides.

model.setAttr("rhs", conss, values)

where conss is a list of your constraints where you want to change the right hand side (e.g.,  conss=m.getConstrs() would be all) and values is a list of new right hand side values. See here https://www.gurobi.com/documentation/8.1/refman/py_model_setattr.html.

Best,

Michael

[Robert Luce]

Hello Rabab,

you can keep as many model structs in memory as you'd like, and modify them as you need.  For example...

% Read a example model, comes along with Gurobi installation
model = gurobi_read('afiro.mps.bz2');
% Solve it, print objective value
result = gurobi(model);
disp(result.objval);
% Save a copy of original model
model_org = model;
% Fix some variable to some value
model.lb(5) = 15.0;
model.ub(5) = 15.0;
% Solve again, objective value now larger
result = gurobi(model);
disp(result.objval);

Robert

[Fiona Huang]

Hi Micheal,

Thank you! This function is super useful for me!

Best,

Fiona

--

---
You received this message because you are subscribed to a topic in the Google Groups "Gurobi Optimization" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/gurobi/89FfTQo48qg/unsubscribe.
To unsubscribe from this group and all its topics, send an email to gurobi+un...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.