3 Components Data¶

In [1]:

import numpy as np
import time

from ml4pd import components
import flash_utils
from ml4pd.aspen_units import Flash
from ml4pd.streams import MaterialStream

input_molecules = flash_utils.get_mol_labels()
components.set_components(input_molecules['name'].to_list())

raw_data = flash_utils.get_benchmark_data(comp=3, date='220606')
data = flash_utils.add_name_columns(raw_data, input_molecules[['name', 'mol']])
data = flash_utils.rename_flowrate_columns(data)

import numpy as np import time from ml4pd import components import flash_utils from ml4pd.aspen_units import Flash from ml4pd.streams import MaterialStream input_molecules = flash_utils.get_mol_labels() components.set_components(input_molecules['name'].to_list()) raw_data = flash_utils.get_benchmark_data(comp=3, date='220606') data = flash_utils.add_name_columns(raw_data, input_molecules[['name', 'mol']]) data = flash_utils.rename_flowrate_columns(data)

Benchmark Data Info¶

In [2]:

input_molecules.T

input_molecules.T

Out[2]:

	0	1	2	3	4	5	6	7	8	9	10	11	12	13	14
name	acetone	2-butanone	3-pentanone	3-hexanone	4-heptanone	4-octanone	5-nonanone	5-decanone	6-undecanone	6-dodecanone	6-tridecanone	7-tetradecanone	8-pentadecanone	water	carbon dioxide
mol	C3-ONE	C4-ONE	C5-ONE	C6-ONE	C7-ONE	C8-ONE	C9-ONE	C10-ONE	C11-ONE	C12-ONE	C13-ONE	C14-ONE	C15-ONE	H2O	CO2
smiles	CC(=O)C	CCC(=O)C	CCC(=O)CC	CCCC(=O)CC	CCCC(=O)CCC	CCCCC(=O)CCC	CCCCC(=O)CCCC	CCCCCC(=O)CCCC	CCCCCC(=O)CCCCC	CCCCCCC(=O)CCCCC	CCCCCCCC(=O)CCCCC	CCCCCCCC(=O)CCCCCC	CCCCCCCC(=O)CCCCCCC	O	C(=O)=O

In [3]:

raw_data.describe().T

raw_data.describe().T

Out[3]:

	count	mean	std	min	25%	50%	75%	max
flowrate_feed_A	12096.0	0.119533	0.066156	0.025641	0.065651	0.120719	0.162065	0.253731
flowrate_feed_B	12096.0	0.440234	0.033078	0.373134	0.418968	0.439641	0.467175	0.487179
flowrate_feed_C	12096.0	0.440234	0.033078	0.373134	0.418968	0.439641	0.467175	0.487179
flowrate_vapor_A	12096.0	0.070908	0.058747	0.000000	0.023837	0.057587	0.109484	0.253731
flowrate_vapor_B	12096.0	0.236345	0.145594	0.000000	0.115852	0.246874	0.363467	0.487179
flowrate_vapor_C	12096.0	0.192747	0.145102	0.000000	0.063064	0.172253	0.313764	0.487179
flowrate_liquid_A	12096.0	0.048625	0.052188	0.000000	0.009578	0.029135	0.072137	0.253731
flowrate_liquid_B	12096.0	0.203889	0.144616	0.000000	0.079647	0.190933	0.324360	0.487179
flowrate_liquid_C	12096.0	0.247486	0.144825	0.000000	0.133180	0.266041	0.373134	0.487179
feed_pressure	12096.0	15.050000	9.383694	0.100000	7.575000	15.050000	22.525000	30.000000
flash_pressure	12096.0	15.050000	9.383694	0.100000	7.575000	15.050000	22.525000	30.000000
feed_vapor_fraction	12096.0	0.500000	0.322762	0.000000	0.250000	0.500000	0.750000	1.000000
flash_duty	12096.0	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000
temp_feed	12096.0	274.185635	84.742900	19.774817	229.775247	282.244614	337.523041	423.014277

Time Benchmark¶

In [4]:

times = []
for i in range(0, 10):
    start_time = time.time()
    feed_stream = MaterialStream(stream_type="feed")(
        vapor_fraction=data['feed_vapor_fraction'].to_list(),
        pressure=data['feed_pressure'].to_list(),
        molecules=flash_utils.get_name_columns(data).to_dict('list'),
        flowrates=flash_utils.get_flowrate_columns(data).to_dict('list'),
    )
    
    flash = Flash(
        pressure=data['feed_pressure'].to_list(),
        duty=data['flash_duty'].to_list()
    )

    vapo_stream, liqu_stream = flash(feed_stream)
    
    times.append(time.time() - start_time)

ordered_data = flash_utils.sort_targets_by_weight(data, feed_stream._mw_idx)

average = np.mean(times).round(2)
std = np.std(times).round(2)

print(f"{len(data)} data pts take {average} +/- {std} seconds to predict.")

times = [] for i in range(0, 10): start_time = time.time() feed_stream = MaterialStream(stream_type="feed")( vapor_fraction=data['feed_vapor_fraction'].to_list(), pressure=data['feed_pressure'].to_list(), molecules=flash_utils.get_name_columns(data).to_dict('list'), flowrates=flash_utils.get_flowrate_columns(data).to_dict('list'), ) flash = Flash( pressure=data['feed_pressure'].to_list(), duty=data['flash_duty'].to_list() ) vapo_stream, liqu_stream = flash(feed_stream) times.append(time.time() - start_time) ordered_data = flash_utils.sort_targets_by_weight(data, feed_stream._mw_idx) average = np.mean(times).round(2) std = np.std(times).round(2) print(f"{len(data)} data pts take {average} +/- {std} seconds to predict.")

12096 data pts take 2.45 +/- 0.09 seconds to predict.

Classifier Benchmark¶

In [5]:

ok_idx = np.array(ordered_data[ordered_data['Status'] == 'OK'].index)
flash_utils.plot_confusion_matrix(ordered_data, flash.status)

ok_idx = np.array(ordered_data[ordered_data['Status'] == 'OK'].index) flash_utils.plot_confusion_matrix(ordered_data, flash.status)

Flowrates Benchmark¶

In [6]:

flash_utils.plot_flow(
    y_true=ordered_data.loc[ok_idx],
    y_pred=liqu_stream.flow.loc[ok_idx]
)

flash_utils.plot_flow( y_true=ordered_data.loc[ok_idx], y_pred=liqu_stream.flow.loc[ok_idx] )

Temperature Benchmark¶

In [7]:

flash_utils.plot_temp(
    y_true=ordered_data,
    y_pred=liqu_stream.temperature,
    ok_idx=ok_idx
)

flash_utils.plot_temp( y_true=ordered_data, y_pred=liqu_stream.temperature, ok_idx=ok_idx )

Mean Absolute Errors¶

In [8]:

flash_utils.evaluate_flow(
    ordered_data.loc[ok_idx], 
    liqu_stream.flow.loc[ok_idx], 
    metric='mae'
)

flash_utils.evaluate_flow( ordered_data.loc[ok_idx], liqu_stream.flow.loc[ok_idx], metric='mae' )

Out[8]:

	flowrate_liquid_A	flowrate_liquid_B	flowrate_liquid_C	mean
0	0.003151	0.01498	0.008023	0.008718

In [9]:

flash_utils.evaluate_temp(
    y_true=ordered_data,
    y_pred=liqu_stream.temperature,
    ok_idx=ok_idx
)

flash_utils.evaluate_temp( y_true=ordered_data, y_pred=liqu_stream.temperature, ok_idx=ok_idx )

Out[9]:

13.880691747496204