def run_epednn(inputs, outputs, betan):
eped_inputs = {}
eped_inputs['r'] = inputs['R']
eped_inputs['ip'] = inputs['Ip']
eped_inputs['a'] = outputs['a']
eped_inputs['bt'] = inputs['Bt']
eped_inputs['kappa'] = inputs['kappa']
eped_inputs['delta'] = inputs['delta']
eped_inputs['neped'] = 10.0 * outputs[
'ne_ped'] #convert 10^20 m^-3 to 10^19 m^-3
eped_inputs['zeffped'] = inputs['zeff']
eped_inputs['betan'] = betan
mixing = 'dt'
if mixing == 'dt':
eped_inputs['m'] = 2.5
else:
eped_inputs['m'] = 2.0
model = 'eped1nn/models/EPED_mb_128_pow_norm_common_30x10.pb'
nn_inputs = np.atleast_2d([[
eped_inputs['a'], eped_inputs['betan'], eped_inputs['bt'],
eped_inputs['delta'], eped_inputs['ip'], eped_inputs['kappa'],
eped_inputs['m'], eped_inputs['neped'], eped_inputs['r'],
eped_inputs['zeffped']
]])
with btf_connect(path=model) as tf:
eped_run = tf.run(nn_inputs)
pped = eped_run[0, 4]
wped = eped_run[0, -3]
return wped, pped
def run_tglf(inputs, outputs, tglf_model, ne, Te, gridpoint, betap):
Rmaj0 = inputs['R'] * 1.e2 # m to cm
a = outputs['a'] * 1.e2 # m to cm
Bt = inputs['Bt'] * 1.e4 # T to Gauss
q0 = inputs['q0']
q95 = outputs['q95']
zeff = inputs['zeff']
kappa1 = inputs['kappa']
delta1 = inputs['delta']
r = np.linspace(0, a, 101)
dr = r[1] - r[0]
# Shafranov shift estimated as linear interpolation between R0 - (R0+betap*r*r/R), ~10% underestimated comparing to geqdsk
Rmaj = np.ones(101) * (Rmaj0 + (betap + 0.5) * r[-1] * r[-1] / Rmaj0) - (
betap + 0.5) * r * r / Rmaj0
dRmaj = np.gradient(Rmaj, dr)
# New shifted elipse estimation of dRmaj
kappa0 = 0.5 * (1. + kappa1)
kappa = kappa0 * np.ones(101) + (kappa1 - kappa0) * r**2 / a**2
delta = 0.5 * delta1 * (r / a + r**2 / a**2)
dlogTdr = np.gradient(Te, dr) / Te
dlogndr = np.gradient(ne, dr) / ne
skappa = r * np.gradient(kappa, dr) / kappa
sdelta = r * np.gradient(delta, dr)
q = q0 + (q95 - q0) * r**2 / (0.95 * a)**2
#rough estimate of psi for elongated plasma (found to be 1.11 times overestimated compared to geqdsk)
#This error could cause ~10% difference in flux when converting to Gyro-Bohm units to real units
#psi = kappa[1:]*Bt*scipy.integrate.cumtrapz(r/q,dx=dr)
#psi = np.append(0.0,psi)
#bunit = q/r*np.gradient(psi,dr)
# New Miller psi estimate based on Stacey et al 2009
bunit = np.zeros(101)
for i in range(1, 101):
theta = np.linspace(0, 2.0 * np.pi, 100)
dtheta = theta[1] - theta[0]
x = np.arcsin(delta[i])
sdelta_stacey = sdelta[i] / np.sqrt(1. - delta[i]**2)
bigtheta = np.sqrt(
np.sin(theta + x * np.sin(theta))**2 *
(1. + x * np.cos(theta))**2 + kappa[i]**2 * np.cos(theta)**2)
bigtheta *= 1.0 / (np.cos(x * np.sin(theta)) +
2.0 * dRmaj[i] * np.cos(theta) +
(skappa[i] - sdelta_stacey * np.cos(theta) +
(1. + skappa[i]) * x * np.cos(theta)) *
np.sin(theta) * np.sin(theta + x * np.sin(theta)))
line_integrand = 1.0 / (
1 + r[i] / Rmaj[i] * np.cos(theta + x * np.sin(theta))) / bigtheta
Ztmp = kappa[i] * r[i] * np.sin(theta)
Rtmp = Rmaj[i] + r[i] * np.cos(theta + x * np.sin(theta))
dRdt = np.gradient(Rtmp, dtheta)
dZdt = np.gradient(Ztmp, dtheta)
area = scipy.integrate.cumtrapz(
np.sqrt(dRdt**2 + dZdt**2) * line_integrand, theta)
bunit[i] = Bt * kappa[i] * (area[-1] / 2.0 / np.pi / r[i])
bunit[0] = bunit[1]
if tglf_model == 'nn':
input_tglf = {}
else:
input_tglf = OMFIT['TGLF_GACODE']['FILES']['input.tglf']
input_tglf['VEXB_SHEAR'] = 0.0
mixing = 'dt'
if mixing == 'dt':
input_tglf['NS'] = 5
input_tglf['MASS_1'] = 0.000272313
input_tglf['MASS_2'] = 1.
input_tglf['MASS_3'] = 1.5
input_tglf['MASS_4'] = 2.
input_tglf['MASS_5'] = 10.
for i in range(1, 6):
input_tglf['VPAR_' + str(i)] = 0.0
input_tglf['VPAR_SHEAR_' + str(i)] = 0.0
input_tglf['RLNS_' + str(i)] = -a * dlogndr[gridpoint]
input_tglf['RLTS_' + str(i)] = -a * dlogTdr[gridpoint]
input_tglf['TAUS_1'] = 1.
input_tglf['TAUS_2'] = inputs['Tratio']
input_tglf['TAUS_3'] = inputs['Tratio']
input_tglf['TAUS_4'] = inputs['Tratio']
input_tglf['TAUS_5'] = inputs['Tratio']
input_tglf['ZS_2'] = 1
input_tglf['ZS_3'] = 1
input_tglf['ZS_4'] = zimp1 = 2
input_tglf['ZS_5'] = zimp2 = 10
impurityFraction2 = (zeff - 1.) / (
5 * zimp1**2 + zimp2**2 - 5 * zimp1 - zimp2
) # setting He impurity density to 5 times Neon
impurityFraction1 = 5.0 * impurityFraction2
input_tglf['AS_1'] = AS_1 = 1.
input_tglf['AS_2'] = AS_2 = 0.5 * (1. - zimp2 * impurityFraction2 -
zimp2 * impurityFraction2)
input_tglf['AS_3'] = AS_3 = 0.5 * (1. - zimp2 * impurityFraction2 -
zimp2 * impurityFraction2)
input_tglf['AS_4'] = AS_4 = impurityFraction1
input_tglf['AS_5'] = AS_5 = impurityFraction2
else:
input_tglf['NS'] = 3
input_tglf['MASS_1'] = 0.000272313
input_tglf['MASS_2'] = 1.
input_tglf['MASS_3'] = 6.0
for i in range(1, 4):
input_tglf['VPAR_' + str(i)] = 0.0
input_tglf['VPAR_SHEAR_' + str(i)] = 0.0
input_tglf['RLNS_' + str(i)] = -a * dlogndr[gridpoint]
input_tglf['RLTS_' + str(i)] = -a * dlogTdr[gridpoint]
input_tglf['TAUS_1'] = 1.
input_tglf['TAUS_2'] = inputs['Tratio']
input_tglf['TAUS_3'] = inputs['Tratio']
input_tglf['TAUS_4'] = inputs['Tratio']
input_tglf['TAUS_5'] = inputs['Tratio']
zimp1 = 6.0
impurityFraction1 = (zeff - 1.) / (zimp1 * (zimp1 - 1.))
input_tglf['AS_1'] = AS_1 = 1.
input_tglf['AS_2'] = AS_2 = 1. - zimp1 * impurityFraction1
input_tglf['AS_3'] = AS_3 = impurityFraction1
AS_4 = AS_5 = 0.0
e = 4.8e-10
k = 1.6e-12
me = 2.0 * 1.6e-24 * input_tglf['MASS_1']
mi = 2.0 * 1.6e-24 * input_tglf['MASS_2']
c = 3.e10
c_s = np.sqrt(k * Te / mi)
input_tglf['BETAE'] = 8. * np.pi * ne[gridpoint] * k * Te[
gridpoint] / bunit[gridpoint]**2
loglam = 24.0 - np.log(np.sqrt(ne[gridpoint] / Te[gridpoint]))
input_tglf['XNUE'] = r[gridpoint] / c_s[gridpoint] * e**4 * np.sqrt(
2.) * np.pi * ne[gridpoint] * loglam / (np.sqrt(me) *
(k * Te[gridpoint])**1.5)
input_tglf['ZEFF'] = zeff
rho_s = c_s / (e * bunit / (mi * c))
input_tglf['DEBYE'] = 7.43e2 * np.sqrt(Te[gridpoint] /
(ne[gridpoint])) / rho_s[gridpoint]
input_tglf['RMIN_LOC'] = r[gridpoint] / a
input_tglf['RMAJ_LOC'] = Rmaj[gridpoint] / a
input_tglf['ZMAJ_LOC'] = 0.
input_tglf['DRMINDX_LOC'] = 1.
input_tglf['DRMAJDX_LOC'] = dRmaj[gridpoint] / a
input_tglf['DZMAJDX_LOC'] = 0.
input_tglf['Q_LOC'] = q[gridpoint]
input_tglf['KAPPA_LOC'] = kappa[gridpoint]
input_tglf['S_KAPPA_LOC'] = skappa[gridpoint]
input_tglf['DELTA_LOC'] = delta[gridpoint]
input_tglf['S_DELTA_LOC'] = sdelta[gridpoint]
input_tglf['ZETA_LOC'] = 0.0
input_tglf['S_ZETA_LOC'] = 0.0
press = ne * Te * (AS_1 + AS_2 + AS_3 + AS_4 + AS_5)
dpdr = np.gradient(press, dr) / press
dqdr = np.gradient(q, dr)
input_tglf['P_PRIME_LOC'] = q[gridpoint] / (
r[gridpoint] /
a)**2 * r[gridpoint] * input_tglf['BETAE'] / (8. * np.pi) * (
(AS_1 + AS_2 * input_tglf['TAUS_2'] + AS_3 * input_tglf['TAUS_3'] +
AS_4 * input_tglf['TAUS_4'] + AS_5 * input_tglf['TAUS_5']) *
dpdr[gridpoint])
s = r[gridpoint] / q[gridpoint] * dqdr[gridpoint]
input_tglf['Q_PRIME_LOC'] = q[gridpoint]**2 * a**2 / r[gridpoint]**2 * s
q_gb = ne * k * Te * c_s * (rho_s / a)**2 * 1.0e-7
nn_inputs = np.atleast_2d([[
input_tglf['AS_2'], input_tglf['AS_3'], input_tglf['BETAE'],
input_tglf['DEBYE'], input_tglf['DELTA_LOC'],
input_tglf['DRMAJDX_LOC'], input_tglf['KAPPA_LOC'],
input_tglf['P_PRIME_LOC'], input_tglf['Q_LOC'],
input_tglf['Q_PRIME_LOC'], input_tglf['RLNS_1'], input_tglf['RLNS_2'],
input_tglf['RLNS_3'], input_tglf['RLTS_1'], input_tglf['RLTS_2'],
input_tglf['RMAJ_LOC'], input_tglf['RMIN_LOC'],
input_tglf['S_KAPPA_LOC'], input_tglf['TAUS_2'],
input_tglf['VEXB_SHEAR'], input_tglf['XNUE'], input_tglf['ZEFF']
]] * 2)
if (tglf_model == 'nn'):
model = 'tglfnn/models/nn_SAT0_mb_1024_abs_reg_common_stair2x2x6.pb'
with btf_connect(path=model) as tf:
qturb = sum(tf.run(nn_inputs)[0, 0:2]) * q_gb[gridpoint]
else:
#OMFIT['TGLF_GACODE']['input.tglf']=OMFITgacode(input_tglf)
OMFIT['TGLF_GACODE']['SCRIPTS']['runTGLF'].run()
qturb = sum(OMFIT['TGLF_GACODE']['FILES']['gbflux']['data']
['Q/Q_GB']) * q_gb[gridpoint]
return qturb
import sys
sys.path.append("neural")
import numpy as np
from brainfusetf import btf_connect
import time
model = 'eped1nn/models/EPED_mb_128_pow_norm_common_30x10.pb'
#model = 'tglfnn/models/nn_SAT0_mb_1024_abs_reg_common_stair2x2x6.pb'
input = np.atleast_2d([[
5.45338e-01, 4.17925e-02, 7.21986e-03, 1.24956e-01, -1.37899e-01,
1.58491e+00, -4.20101e-03, 1.55640e+00, 8.36932e+00, 1.02569e+00,
2.05921e+00, -4.45231e-01, 3.00670e+00, 2.06023e+00, 2.38220e+00,
7.66336e-01, 3.20824e-01, 1.14110e+00, 3.21049e-01, 3.36619e-01,
1.87324e+00
]] * 2)
print input
t0 = time.time()
with btf_connect(path=model) as tf:
print tf.info()
#with btf_connect(path=model) as tf:
# print tf.run(input=input)
#print tf.run(input=input)[0,0:2]
# time.sleep(1)
#print tf.run(input=input*3)
#print(time.time()-t0)