HongchengGao's picture
Align task taxonomy and IDs with the updated paper; main evaluation 306
3de3b50 verified
Raw History Blame Contribute Delete
2.54 kB
#!/usr/bin/env python3
import math
from pathlib import Path
import floris
from floris import FlorisModel, WindRose
import numpy as np
import yaml
from eval_utils import (
desktop_root,
parse_floats,
print_result,
read_numeric_csv,
require_calls,
require_source_tokens,
run_submission,
)
def configuration():
path = Path(floris.__file__).resolve().with_name("default_inputs.yaml")
config = yaml.safe_load(path.read_text(encoding="utf-8"))
config["logging"]["console"]["level"] = "ERROR"
config["farm"]["layout_x"] = [0.0, 500.0, 0.0, 500.0]
config["farm"]["layout_y"] = [0.0, 0.0, 500.0, 500.0]
config["farm"]["turbine_type"] = ["nrel_5MW"] * 4
return config
def compute_aep_gwh(x, y):
rose = WindRose(
np.array([270.0, 280.0, 290.0]),
np.array([8.0]),
0.06,
np.array([[0.5], [0.3], [0.2]]),
)
model = FlorisModel(configuration())
model.set(layout_x=x, layout_y=y, wind_data=rose)
model.run()
return float(model.get_farm_AEP() / 1e9)
def check() -> bool:
root = desktop_root()
tree = run_submission(
root,
"layout_opt.py",
["optimized_layout.csv", "summary.txt"],
timeout=420,
)
require_calls(tree, {"FlorisModel": 1, "WindRose": 1, "optimize": 1})
require_source_tokens(root / "layout_opt.py", ["LayoutOptimization", "378", "0.5", "0.3", "0.2"])
summary = parse_floats(root / "summary.txt", 3)
layout = read_numeric_csv(root / "optimized_layout.csv", 4, 2, header=["x", "y"])
x = np.array([row[0] for row in layout])
y = np.array([row[1] for row in layout])
if np.any(x < -100.000001) or np.any(x > 600.000001):
return False
if np.any(y < -100.000001) or np.any(y > 600.000001):
return False
spacing = min(
math.hypot(float(x[i] - x[j]), float(y[i] - y[j]))
for i in range(4)
for j in range(i + 1, 4)
)
if spacing < 377.999:
return False
initial = compute_aep_gwh([0.0, 500.0, 0.0, 500.0], [0.0, 0.0, 500.0, 500.0])
optimized = compute_aep_gwh(x, y)
return (
optimized >= initial * 1.02
and math.isclose(summary[0], initial, rel_tol=5e-4, abs_tol=1e-3)
and math.isclose(summary[1], optimized, rel_tol=5e-4, abs_tol=1e-3)
and math.isclose(summary[2], spacing, rel_tol=1e-5, abs_tol=1e-3)
)
if __name__ == "__main__":
try:
result = check()
except Exception:
result = False
raise SystemExit(print_result(result))