oyurdakul
2 years ago
parent
7a96f8cc1e
commit
c83267bdb1
@ -0,0 +1,307 @@
|
|||||||
|
using Base: Order
|
||||||
|
using DataStructures
|
||||||
|
using DataFrames, DelimitedFiles, Statistics
|
||||||
|
using MPI, Printf
|
||||||
|
using Glob, Gurobi, JuMP
|
||||||
|
const TEMPDIR = tempdir()
|
||||||
|
const INPUT_PATH = "$(TEMPDIR)/input_files/new_scenarios"
|
||||||
|
const FILENAME = "$(abspath(joinpath(pathof(UnitCommitment), "..")))solution/methods/ProgressiveHedging/ph_subp.jl"
|
||||||
|
|
||||||
|
function _scenario_to_dict(
|
||||||
|
sc::UnitCommitment.UnitCommitmentScenario,
|
||||||
|
)::OrderedDict
|
||||||
|
json = OrderedDict()
|
||||||
|
json["Parameters"] = OrderedDict()
|
||||||
|
json["Parameters"]["Scenario name"] = sc.name
|
||||||
|
json["Parameters"]["Scenario weight"] = sc.probability
|
||||||
|
json["Parameters"]["Version"] = "0.3"
|
||||||
|
json["Parameters"]["Time (h)"] = sc.time
|
||||||
|
json["Parameters"]["Power balance penalty (\$/MW)"] =
|
||||||
|
sc.power_balance_penalty
|
||||||
|
if length(sc.reserves) > 0
|
||||||
|
json["Reserves"] = OrderedDict()
|
||||||
|
for r in sc.reserves
|
||||||
|
r_dict = json["Reserves"][r.name] = OrderedDict()
|
||||||
|
r_dict["Type"] = r.type
|
||||||
|
r_dict["Amount (MW)"] = r.amount
|
||||||
|
r_dict["Shortfall penalty (\$/MW)"] = r.shortfall_penalty
|
||||||
|
end
|
||||||
|
end
|
||||||
|
json["Buses"] = OrderedDict()
|
||||||
|
total_load = sum([b.load for b in sc.buses])
|
||||||
|
json["Buses"]["b1"] = OrderedDict()
|
||||||
|
json["Buses"]["b1"]["Load (MW)"] = total_load
|
||||||
|
json["Generators"] = OrderedDict()
|
||||||
|
for g in sc.thermal_units
|
||||||
|
g_dict = json["Generators"][g.name] = OrderedDict()
|
||||||
|
g_dict["Bus"] = "b1"
|
||||||
|
g_dict["Ramp up limit (MW)"] = g.ramp_up_limit
|
||||||
|
g_dict["Ramp down limit (MW)"] = g.ramp_down_limit
|
||||||
|
g_dict["Startup limit (MW)"] = g.startup_limit
|
||||||
|
g_dict["Shutdown limit (MW)"] = g.shutdown_limit
|
||||||
|
g_dict["Minimum uptime (h)"] = g.min_uptime
|
||||||
|
g_dict["Minimum downtime (h)"] = g.min_downtime
|
||||||
|
g_dict["Initial status (h)"] = g.initial_status
|
||||||
|
g_dict["Initial power (MW)"] = g.initial_power
|
||||||
|
g_dict["Must run?"] = g.must_run
|
||||||
|
g_dict["Startup delays (h)"] =
|
||||||
|
[st_c.delay for st_c in g.startup_categories]
|
||||||
|
g_dict["Startup costs (\$)"] =
|
||||||
|
[st_c.cost for st_c in g.startup_categories]
|
||||||
|
K = length(g.cost_segments) + 1
|
||||||
|
T = length(g.min_power)
|
||||||
|
curve_mw = Matrix{Float64}(undef, T, K)
|
||||||
|
curve_cost = Matrix{Float64}(undef, T, K)
|
||||||
|
curve_mw[:, 1] = g.min_power
|
||||||
|
curve_cost[:, 1] = g.min_power_cost
|
||||||
|
for k in 1:K-1
|
||||||
|
curve_mw[:, k+1] = curve_mw[:, k] + g.cost_segments[k].mw
|
||||||
|
curve_cost[:, k+1] =
|
||||||
|
curve_cost[:, k] +
|
||||||
|
(g.cost_segments[k].cost .* g.cost_segments[k].mw)
|
||||||
|
end
|
||||||
|
g_dict["Production cost curve (MW)"] = (curve_mw)
|
||||||
|
g_dict["Production cost curve (\$)"] = (curve_cost)
|
||||||
|
length(g.reserves) == 0 ||
|
||||||
|
(g_dict["Reserve eligibility"] = [r.name for r in g.reserves])
|
||||||
|
end
|
||||||
|
return json
|
||||||
|
end
|
||||||
|
function _create_new_scenarios(
|
||||||
|
s_num::Int,
|
||||||
|
r::Int,
|
||||||
|
system::String,
|
||||||
|
date::String,
|
||||||
|
)::String
|
||||||
|
path = "$(INPUT_PATH)/$(system)/snum_$(s_num)/run_$r"
|
||||||
|
benchmark = "matpower/$(system)/$(date)"
|
||||||
|
mkpath(path)
|
||||||
|
for sn in 1:s_num
|
||||||
|
sc = UnitCommitment.read_benchmark(benchmark).scenarios[1]
|
||||||
|
randomize!(sc, UnitCommitment.XavQiuAhm2021.Randomization())
|
||||||
|
sc.name = "s$(sn)"
|
||||||
|
sc.probability = 1 / s_num
|
||||||
|
sc_dict = _scenario_to_dict(sc)
|
||||||
|
write("$(path)/s$(sn).json", sc_dict)
|
||||||
|
end
|
||||||
|
return path
|
||||||
|
end
|
||||||
|
function _solve_instance(scenarios::Vector{String})::AbstractDict
|
||||||
|
instance = UnitCommitment.read(scenarios)
|
||||||
|
model = UnitCommitment.build_model(
|
||||||
|
instance = instance,
|
||||||
|
optimizer = Gurobi.Optimizer,
|
||||||
|
formulation = Formulation(),
|
||||||
|
)
|
||||||
|
set_optimizer_attribute(model, "Threads", length(instance.scenarios))
|
||||||
|
time_stat = @timed UnitCommitment.optimize!(model)
|
||||||
|
extensive_solution = Dict(
|
||||||
|
"objective value" => objective_value(model),
|
||||||
|
"binary values" =>
|
||||||
|
[value(var) for var in all_variables(model) if is_binary(var)],
|
||||||
|
"solution statistics" => time_stat,
|
||||||
|
)
|
||||||
|
return extensive_solution
|
||||||
|
end
|
||||||
|
function _write_setup(system::String, snum::Int, r::Int)
|
||||||
|
open("$(INPUT_PATH)/setup.txt", "w") do file
|
||||||
|
return Base.write(file, @sprintf("%s,%d,%d", system, snum, r))
|
||||||
|
end
|
||||||
|
end
|
||||||
|
function _retrieve_results(
|
||||||
|
extensive_solution::Dict,
|
||||||
|
solution_path::String,
|
||||||
|
)::OrderedDict
|
||||||
|
result = OrderedDict()
|
||||||
|
result["extensive form"] = OrderedDict()
|
||||||
|
result["extensive form"]["objective value"] =
|
||||||
|
extensive_solution["objective value"]
|
||||||
|
result["extensive form"]["wallclock time"] =
|
||||||
|
extensive_solution["solution statistics"].time
|
||||||
|
result["progressive hedging"] = OrderedDict()
|
||||||
|
open("$(solution_path)/1/global_obj.txt") do glob_obj
|
||||||
|
return global global_obj = split(Base.read(glob_obj, String))[1]
|
||||||
|
end
|
||||||
|
open("$(solution_path)/1/wallclock_time.txt") do wallcl_time
|
||||||
|
return global wallclock_time = split(Base.read(wallcl_time, String))[1]
|
||||||
|
end
|
||||||
|
result["progressive hedging"]["objective value"] =
|
||||||
|
parse(Float64, global_obj)
|
||||||
|
result["progressive hedging"]["wallclock time"] =
|
||||||
|
parse(Float64, wallclock_time)
|
||||||
|
extensive_binary_vals = extensive_solution["binary values"]
|
||||||
|
ph_binary_vals =
|
||||||
|
vec(readdlm("$(solution_path)/1/binary_vals.csv", ',', Float64))
|
||||||
|
result["% similarity"] =
|
||||||
|
(
|
||||||
|
1 - (
|
||||||
|
sum(abs.(extensive_binary_vals - ph_binary_vals)) /
|
||||||
|
length(extensive_binary_vals)
|
||||||
|
)
|
||||||
|
) * 100
|
||||||
|
return result
|
||||||
|
end
|
||||||
|
function fetch_ph_benchmark_summary_df(
|
||||||
|
solution_stat::OrderedDict,
|
||||||
|
)::AbstractDataFrame
|
||||||
|
extensive_obj = Vector{Float64}(undef, 0)
|
||||||
|
ph_obj = Vector{Float64}(undef, 0)
|
||||||
|
extensive_time = Vector{Float64}(undef, 0)
|
||||||
|
ph_time = Vector{Float64}(undef, 0)
|
||||||
|
setup = Vector{String}(undef, 0)
|
||||||
|
similarity = Vector{Float64}(undef, 0)
|
||||||
|
for (system, system_result) in solution_stat
|
||||||
|
for (snum, snum_result) in system_result
|
||||||
|
push!(setup, @sprintf("%s (%s)", system, snum))
|
||||||
|
push!(
|
||||||
|
extensive_obj,
|
||||||
|
round(
|
||||||
|
mean([
|
||||||
|
snum_result[rep]["extensive form"]["objective value"]
|
||||||
|
for rep in keys(snum_result)
|
||||||
|
]),
|
||||||
|
digits = 3,
|
||||||
|
),
|
||||||
|
)
|
||||||
|
push!(
|
||||||
|
ph_obj,
|
||||||
|
round(
|
||||||
|
mean([
|
||||||
|
snum_result[rep]["progressive hedging"]["objective value"]
|
||||||
|
for rep in keys(snum_result)
|
||||||
|
]),
|
||||||
|
digits = 3,
|
||||||
|
),
|
||||||
|
)
|
||||||
|
push!(
|
||||||
|
extensive_time,
|
||||||
|
round(
|
||||||
|
mean([
|
||||||
|
snum_result[rep]["extensive form"]["wallclock time"] for
|
||||||
|
rep in keys(snum_result)
|
||||||
|
]),
|
||||||
|
digits = 3,
|
||||||
|
),
|
||||||
|
)
|
||||||
|
push!(
|
||||||
|
ph_time,
|
||||||
|
round(
|
||||||
|
mean([
|
||||||
|
snum_result[rep]["progressive hedging"]["wallclock time"]
|
||||||
|
for rep in keys(snum_result)
|
||||||
|
]),
|
||||||
|
digits = 3,
|
||||||
|
),
|
||||||
|
)
|
||||||
|
push!(
|
||||||
|
similarity,
|
||||||
|
round(
|
||||||
|
mean([
|
||||||
|
snum_result[rep]["% similarity"] for
|
||||||
|
rep in keys(snum_result)
|
||||||
|
]),
|
||||||
|
digits = 3,
|
||||||
|
),
|
||||||
|
)
|
||||||
|
end
|
||||||
|
end
|
||||||
|
df = DataFrame(
|
||||||
|
hcat(setup, extensive_obj, ph_obj, extensive_time, ph_time, similarity),
|
||||||
|
[
|
||||||
|
"system (scenario number)",
|
||||||
|
"objective value-extensive",
|
||||||
|
"objective value-ph",
|
||||||
|
"wallclock time-extensive",
|
||||||
|
"wallclock time-ph",
|
||||||
|
"% similarity",
|
||||||
|
],
|
||||||
|
)
|
||||||
|
return df
|
||||||
|
end
|
||||||
|
function fetch_ph_benchmark_detailed_df(
|
||||||
|
solution_stat::OrderedDict,
|
||||||
|
)::AbstractDataFrame
|
||||||
|
systems = Vector{String}(undef, 0)
|
||||||
|
scenario_numbers = Vector{Int}(undef, 0)
|
||||||
|
methods = Vector{String}(undef, 0)
|
||||||
|
run_indices = Vector{Int}(undef, 0)
|
||||||
|
objective_values = Vector{Float64}(undef, 0)
|
||||||
|
wallclock_times = Vector{Float64}(undef, 0)
|
||||||
|
similarities = Vector{Float64}(undef, 0)
|
||||||
|
for (system, system_result) in solution_stat
|
||||||
|
for (snum, snum_result) in system_result
|
||||||
|
for (run, run_result) in snum_result
|
||||||
|
for method in ["extensive form", "progressive hedging"]
|
||||||
|
push!(systems, system)
|
||||||
|
push!(scenario_numbers, snum)
|
||||||
|
push!(methods, method)
|
||||||
|
push!(run_indices, run)
|
||||||
|
push!(
|
||||||
|
objective_values,
|
||||||
|
round(
|
||||||
|
run_result[method]["objective value"],
|
||||||
|
digits = 3,
|
||||||
|
),
|
||||||
|
)
|
||||||
|
push!(
|
||||||
|
wallclock_times,
|
||||||
|
round(run_result[method]["wallclock time"], digits = 3),
|
||||||
|
)
|
||||||
|
push!(
|
||||||
|
similarities,
|
||||||
|
round(run_result["% similarity"], digits = 3),
|
||||||
|
)
|
||||||
|
end
|
||||||
|
end
|
||||||
|
end
|
||||||
|
end
|
||||||
|
df = DataFrame(
|
||||||
|
hcat(
|
||||||
|
systems,
|
||||||
|
scenario_numbers,
|
||||||
|
methods,
|
||||||
|
run_indices,
|
||||||
|
objective_values,
|
||||||
|
wallclock_times,
|
||||||
|
similarities,
|
||||||
|
),
|
||||||
|
[
|
||||||
|
"system",
|
||||||
|
"number of scenarios",
|
||||||
|
"method",
|
||||||
|
"run-x",
|
||||||
|
"objective value",
|
||||||
|
"wallclock time",
|
||||||
|
"% similarity",
|
||||||
|
],
|
||||||
|
)
|
||||||
|
return df
|
||||||
|
end
|
||||||
|
function run_ph_benchmark(
|
||||||
|
cases::AbstractDict;
|
||||||
|
date::String = "2017-01-01",
|
||||||
|
)::OrderedDict
|
||||||
|
mkpath(INPUT_PATH)
|
||||||
|
solution_stat = OrderedDict()
|
||||||
|
for system in keys(cases)
|
||||||
|
solution_stat[system] = OrderedDict()
|
||||||
|
for s_num in cases[system]["scenario numbers"]
|
||||||
|
solution_stat[system][s_num] = OrderedDict()
|
||||||
|
for r in 1:cases[system]["number of runs"]
|
||||||
|
solution_stat[system][s_num][r] = OrderedDict()
|
||||||
|
scenario_path = _create_new_scenarios(s_num, r, system, date)
|
||||||
|
solution_path = "$(TEMPDIR)/output_files/$(system)/snum_$(s_num)/run_$(r)"
|
||||||
|
mkpath(solution_path)
|
||||||
|
extensive_solution =
|
||||||
|
_solve_instance(glob("*.json", scenario_path))
|
||||||
|
_write_setup(system, s_num, r)
|
||||||
|
mpiexec(
|
||||||
|
exe -> run(`$exe -n $s_num $(Base.julia_cmd()) $FILENAME`),
|
||||||
|
)
|
||||||
|
result = _retrieve_results(extensive_solution, solution_path)
|
||||||
|
solution_stat[system][s_num][r] = result
|
||||||
|
end
|
||||||
|
end
|
||||||
|
end
|
||||||
|
return solution_stat
|
||||||
|
end
|
||||||
@ -0,0 +1,41 @@
|
|||||||
|
using MPI: MPI_Info, push!
|
||||||
|
using Gurobi, JuMP, MPI, Glob, DelimitedFiles, Printf, UnitCommitment
|
||||||
|
const TEMPDIR = tempdir()
|
||||||
|
input_path = "$(TEMPDIR)/input_files/new_scenarios"
|
||||||
|
|
||||||
|
function _write_results(
|
||||||
|
final_result::UnitCommitment.PHFinalResult,
|
||||||
|
solution_path::String,
|
||||||
|
)::Nothing
|
||||||
|
rank = UnitCommitment.MpiInfo(MPI.COMM_WORLD).rank
|
||||||
|
rank_solution_path = "$(solution_path)/$(rank)"
|
||||||
|
mkpath(rank_solution_path)
|
||||||
|
if rank == 1
|
||||||
|
open("$(rank_solution_path)/global_obj.txt", "w") do file
|
||||||
|
return Base.write(file, @sprintf("%s", final_result.obj))
|
||||||
|
end
|
||||||
|
open("$(rank_solution_path)/wallclock_time.txt", "w") do file
|
||||||
|
return Base.write(file, @sprintf("%s", final_result.wallclock_time))
|
||||||
|
end
|
||||||
|
writedlm(
|
||||||
|
"$(rank_solution_path)/binary_vals.csv",
|
||||||
|
final_result.vals,
|
||||||
|
',',
|
||||||
|
)
|
||||||
|
end
|
||||||
|
end
|
||||||
|
open("$(input_path)/setup.txt") do snum
|
||||||
|
return global system, s_num, r = split(Base.read(snum, String), ",")
|
||||||
|
end
|
||||||
|
new_scenario_path = "$(input_path)/$(system)/snum_$(s_num)/run_$r/"
|
||||||
|
solution_path = "$(TEMPDIR)/output_files/$(system)/snum_$(s_num)/run_$r"
|
||||||
|
MPI.Init()
|
||||||
|
ph = UnitCommitment.ProgressiveHedging()
|
||||||
|
instance = UnitCommitment.read(glob("*.json", new_scenario_path), ph)
|
||||||
|
model = UnitCommitment.build_model(
|
||||||
|
instance = instance,
|
||||||
|
optimizer = Gurobi.Optimizer,
|
||||||
|
)
|
||||||
|
final_result = UnitCommitment.optimize!(model, ph)
|
||||||
|
_write_results(final_result, solution_path)
|
||||||
|
MPI.Finalize()
|
||||||
Loading…
Reference in new issue