How to solve infinite horizon problem under Makovian policy graph in SDDP.jl?

odow · May 28, 2025, 8:36pm

Can the infinite horizon problem instance be solved under a Markovian Policy graph using SDDP.jl?

Yes.

See Create a general policy graph · SDDP.jl for instructions on how to create an arbitrary graph. I would start with a SDDP.MarkovianGraph and then add the cyclic edges.

Here’s the code for the model in https://doi.org/10.1002/net.21932:

github.com/odow/SDDP.jl

papers/policy_graph/paper.jl

b072e859b


      
          # ===== Markovian Graph =====
          transition = Array{Float64,2}[]
          for transition_matrix in data["transition"]
              push!(
                  transition,
                  convert(
                      Array{Float64,2},
                      Base.reshape(
                          vcat(transition_matrix...),
                          length(transition_matrix[1]),
                          length(transition_matrix),
                      ),
                  ),
              )
          end
          graph = SDDP.MarkovianGraph(transition)
          for markov_state = 1:size(transition[end], 2)
              SDDP.add_edge(
                  graph,
                  (data["number_of_weeks"], markov_state) => (1, 1),

This file has been truncated. show original

Topic		Replies	Views
Policy Graph in SDDP.jl package Optimization (Mathematical) question , optimization	1	334	June 8, 2023
End-of-horizon effect, Infinite horizon policy graph, SDDP.jl Optimization (Mathematical) question , sddp	11	260	June 14, 2024
Tons of nodes in the Policy Graph, SDDP.jl package Optimization (Mathematical) question , gurobi	20	616	November 8, 2023
Cyclic policy graph with fix number of cycles - SDDP.jl Optimization (Mathematical) question , sddp	1	98	July 13, 2024
SDDP simulation command Optimization (Mathematical) question , optimization	5	243	April 8, 2023

How to solve infinite horizon problem under Makovian policy graph in SDDP.jl?

Related topics