Commit 71982e6a authored by Ibrahim Ahmed's avatar Ibrahim Ahmed
Browse files

controllers: Removed setpoint max clipping; add perf evaluation notebook; 

Temperature setpoints were earlier clipped if they were higher than temperature of water coming out of cooling tower. Why set point for warmer water than current? However in hot/humid settings where optimization is not possible, and setting a warmer setpoint doesn't warm water but still causes PID controlled fans to spin down, that is a power saving that we can afford.
parent 9b9fb051

src/controllers/esb.py

+1 −1
Original line number Diff line number Diff line
@@ -38,7 +38,7 @@ class Controller(SimpleFeedbackController): 


        def clip_action(self, u, X):

            u = super().clip_action(u, X)

            return np.clip(u, a_min=X['TempWetBulb'], a_max=X['TempCondOut'])

            return np.clip(u, a_min=X['TempWetBulb'], a_max=None)

src/controllers/kissam.py

+1 −1
Original line number Diff line number Diff line
@@ -33,7 +33,7 @@ class Controller(SimpleFeedbackController): 


        def clip_action(self, u, X):

            u = super().clip_action(u, X)

            return np.clip(u, a_min=X['TempWetBulb'], a_max=X['CT_1.TempCondOut'])

            return np.clip(u, a_min=X['TempWetBulb'], a_max=None)

src/notebooks/ESB-Controller-Evaluation.ipynb

0 → 100644
+337 −0
Original line number Diff line number Diff line 
%% Cell type:code id: tags:



``` python

%reload_ext autoreload

%autoreload 2



import notebook_setup

import os



import numpy as np

import pandas as pd

import matplotlib.pyplot as plt

from matplotlib import colors

from datetime import datetime, timezone, timedelta

from sklearn.preprocessing import MinMaxScaler

from tslearn.utils import to_time_series

from tslearn.metrics import dtw

from tqdm.auto import tqdm



from bdx import get_trend

from utils import get_credentials

```



%% Cell type:code id: tags:



``` python

cst = timezone(offset=-timedelta(hours=6))

cdt = timezone(offset=-timedelta(hours=5))



datadir = os.path.join(os.environ.get('DATADIR'), 'EngineeringScienceBuilding')

username, password = get_credentials()

trend = '3481'      # python setpoint trend

ch2trend = '2841'  # chiller 2 detailed trend

ch1trend = '2422'  # chiller 1 detailed trend

simcols = ['TempAmbient', 'TempWetBulb', 'Tonnage']



ch1start, ch1end = datetime(2021,5,16,6,10, tzinfo=cdt), datetime(2021,5,23,6,0, tzinfo=cdt)

ch2start, ch2end = datetime(2021,4,19, tzinfo=cdt), datetime(2021,5,3, tzinfo=cdt)

oldstart, oldend = datetime(2020,3,1, tzinfo=cst), datetime(2021,3,8, tzinfo=cst)

```



%% Cell type:markdown id: tags:



## Getting historical and benchmark data



Using Dynamic Time Warping to compare a day under controller vs a similar day under old control. Ambient and Wetbulb temperature time series are used to measure similarity.



%% Cell type:code id: tags:



``` python

chiller = 2

if chiller == 1 or chiller == '1v2':

    start, end, chtrend = ch1start, ch1end, ch1trend

    refstart, refend, reftrend = (oldstart, oldend, ch1trend) if chiller == 1 else (ch2start, ch2end, ch2trend)

elif chiller == 2 or chiller == '2v1':

    start, end, ch_trend = ch2_start, ch2end, ch2trend

    refstart, refend, reftrend = (oldstart, oldend, ch2trend) if chiller == 2 else (ch1start, ch1end, ch1trend)

```



%% Cell type:code id: tags:



``` python

# Download data under controller script used for comparison

data_stpt = get_trend(trend, username, password,

                 start=start,

                 end=end)

data_stpt = data_stpt[['CDWTPythonSetpt', 'JCI Cooling_Tower_Water_Setpoint']]

data_ch = get_trend(ch_trend, username, password,

                 start=start,

                 end=end)



valid = (data_ch['RunChi']==True) & (data_ch['PowChi'] > 0) & (data_ch['Tonnage'] > 0)

data_ch = data_ch[valid]

data_ch.index = data_ch.index.tz_convert(cdt)

data_stpt = data_stpt[valid]

data_stpt.index = data_stpt.index.tz_convert(cdt)

```



%% Cell type:code id: tags:



``` python

# Download reference setpoint data for comparison

ref_stpt = get_trend(trend, username, password,

                    start=refstart,

                    end=refend)

ref_stpt = ref_stpt[['CDWTPythonSetpt', 'JCI Cooling_Tower_Water_Setpoint']]



# Reference chiller data for comparison

ref_ch = get_trend(reftrend, username, password,

                   start=refstart,

                   end=refend)



ref_stpt = ref_stpt[(ref_ch['RunChi']==True) & (ref_ch['PowChi'] > 0)]

ref_ch = ref_ch.loc[(ref_ch['RunChi']==True) & (ref_ch['PowChi'] > 0)]



# Save data to disk for later use

ref_stpt.to_csv(os.path.join(datadir, '%s_Chiller%s_setpoint_eval.csv' % (trend, str(chiller))))

ref_ch.to_csv(os.path.join(datadir, '%s_Chiller%s_eval.csv' % (chtrend, str(chiller))))

ref_stpt.index = ref_stpt.index.tz_convert(cdt)

ref_ch.index = ref_ch.index.tz_convert(cdt)

```



%% Cell type:code id: tags:



``` python

# Or Load reference data from disk

ref_stpt = pd.read_csv(os.path.join(datadir, '%s_Chiller%s_setpoint_eval.csv' % (trend, str(chiller))), index_col='time', parse_dates=True)

ref_ch = pd.read_csv(os.path.join(datadir, '%s_Chiller%s_eval.csv' % (chtrend, str(chiller))), index_col='time', parse_dates=True)

ref_stpt.index = ref_stpt.index.tz_convert(cdt)

ref_ch.index = ref_ch.index.tz_convert(cdt)

```



%% Cell type:markdown id: tags:



## Processing data



%% Cell type:code id: tags:



``` python

starttime = data_ch.index[0].timetz()  # start TIME of controller

endtime = data_ch.index[-1].timetz()   # end TIME of controller



# Chiller agnostic referencing for analysis

refdata = ref_ch # reference data

data = data_ch  # controller data

```



%% Cell type:code id: tags:



``` python

# Find most similar time period in reference data

# Find the start time in reference data which is the same as the controller data:

refstart = None

refend = refdata.index[-1]

for t in refdata.index:

    if t.timetz() == starttime:

        refstart = t

        break



scaler = MinMaxScaler()

scaler.fit(refdata[simcols])

refsegments = []

similarities = []

span = pd.Timedelta(days=1)

# Run a window of size `span` by shifting it by a freq=1D

# to find similarities between segments using dynamic time warping

date_ticks = pd.date_range(data.index[0], data.index[-1], freq='1D')

rdate_ticks = pd.date_range(refstart, refend, freq='1D')

similarities = np.zeros((len(date_ticks), len(rdate_ticks))) + np.inf

for i, date in enumerate(tqdm(date_ticks, leave=False, desc='Data spans')):

    segment = data.loc[date:date + span]

    for j, rdate in enumerate(tqdm(rdate_ticks, leave=False, desc='Reference spans')):

        rsegment = refdata.loc[rdate:rdate + span]

        refsegments.append(rsegment)

        if len(rsegment)==0: continue

        similarities[i, j] = dtw(

                to_time_series(scaler.transform(rsegment[simcols])),

                to_time_series(scaler.transform(segment[simcols])))



# Rank reference segments by similarity against data segments

ranked = np.argsort(similarities, axis=1)

mostsim, dates = [], []

for row in ranked:

    for idx in row:

        if refsegments[idx]['PowChi'].mean() > 0:

            mostsim.append(refsegments[idx])

            dates.append(refsegments[idx].index.to_series())

            break

# Create a dataframe of the most similar reference segments for each segment in data

mostsim = pd.concat(mostsim, ignore_index=True)

mostsim['dates'] = pd.concat(dates, ignore_index=True)

```



%% Cell type:code id: tags:



``` python

plt.figure(figsize=(7,12))

plt.subplot(2,1,1)

_, xedges, yedges, _ = plt.hist2d(mostsim[simcols[0]], mostsim[simcols[1]], density=True)

plt.title('Most similar period conditions')

plt.xlabel('Ambient')

plt.ylabel('Wet bulb')

plt.colorbar()

plt.subplot(2,1,2)

plt.hist2d(data[simcols[0]], data[simcols[1]], bins=[xedges, yedges], density=True)

plt.title('Controller conditions')

plt.xlabel('Ambient')

plt.ylabel('Wet bulb')

plt.colorbar()

```



%% Cell type:markdown id: tags:



## Comparison plots of benchmarks



%% Cell type:code id: tags:



``` python

# Plot

fig, axes = plt.subplots(7, 1, figsize=(8,42), facecolor='white')

axes = axes.ravel()

# fig.suptitle('Benchmark')



mostsim[simcols[:2]].plot(ax=axes[0])

# ref_stpt[['JCI Cooling_Tower_Water_Setpoint']].plot(ax=axes[0], ls=':')

# ref_stpt[['CDWTPythonSetpt']].plot(ax=axes[0], ls=':')

axes[0].set_ylim(20,80)

axes[0].set_ylabel('Temperature /F')

axes[0].set_title('Ambient Conditions, Temp: %.1f+/-%.1fF, Wetbulb: %.1f+/-%.1fF' % \

                  (mostsim['TempAmbient'].mean(), mostsim['TempAmbient'].std(),

                   mostsim['TempWetBulb'].mean(), mostsim['TempWetBulb'].std()))



ctefficiency = (mostsim['TempCondOut'] - mostsim['TempCondIn']) / \

               (mostsim['TempCondOut'] - mostsim['TempWetBulb'])

ctefficiency.plot(ax=axes[1])

axes[1].set_ylim(0,1)

axes[1].set_title('Cooling Tower Efficiency %.2f +/- %.2f' % (ctefficiency.mean(), ctefficiency.std()))



mostsim[['PowChi', 'Tonnage']].plot(ax=axes[2])

kw = mostsim['PowChi'] / mostsim['Tonnage']

axes[2].set_title('Average chiller efficiency %.2f +/- %.2fkW/Ton' % (kw.mean(), kw.std()))

axes[2].set_ylabel('Power /kW')

axes[2].set_ylim(0, 500)



mostsim['PerFreqFanA'].plot(ax=axes[3])

axes[3].set_ylabel('Percentage frequency /%')

axes[3].set_title('Fan frequency')

axes[3].set_ylim(0, 105)



mostsim[['TempCondIn', 'TempCondOut']].plot(ax=axes[4])

# ref_stpt[['JCI Cooling_Tower_Water_Setpoint']].plot(ax=axes[4], ls=':')

axes[4].set_ylim(50, 90)

deltat = (mostsim['TempCondOut'] - mostsim['TempCondIn'])

axes[4].set_ylabel('Temperature /F')

axes[4].set_title('Condenser Temps, Avg change %.2f +/- %.2fF' % (deltat.mean(), deltat.std()))



scatterdf = pd.DataFrame({

    'Cooling Tower Efficiency': ctefficiency,

    'Chiller load /%': mostsim['Tonnage'] * 100/800,

})

corr = scatterdf.corr().values[0,1]

scatterdf.plot.scatter('Cooling Tower Efficiency', 'Chiller load /%', ax=axes[5])

axes[5].set_title('Efficiency vs Load, correlation %.2f' % corr)

axes[5].set_xlim(0, 1)

axes[5].set_ylim(0, 100)



scatterdf = pd.DataFrame({

    'Chiller load /%': mostsim['Tonnage'] * 100/800,

    'kw/Ton': mostsim['PowChi'] / mostsim['Tonnage'],

    'Returning Water Temp /F': mostsim['TempCondIn']

})

scatterdf.plot.scatter('Chiller load /%', 'kw/Ton', c='Returning Water Temp /F',

                       colormap='coolwarm', norm=colors.Normalize(50, 90), ax=axes[6])

mean, std = scatterdf.mean().values, scatterdf.std().values

axes[6].set_title('Average kw/Ton %.2f+/-%.2f, Chiller load %.2f+/-%.2f, Ret. temp %.2f+/-%.2f)' % \

                  (mean[1], std[1], mean[0], std[0], mean[2], std[2]))

axes[6].errorbar(mean[0], mean[1], xerr=std[0], yerr=std[1])

axes[6].set_xlim(0, 100)

axes[6].set_ylim(0, 1)





for ax in axes:

    ax.set_facecolor('whitesmoke')

    ax.grid(True, 'both')

plt.tight_layout()

```



%% Cell type:code id: tags:



``` python

# Plot

fig, axes = plt.subplots(7, 1, figsize=(8,42), facecolor='white')

axes = axes.ravel()

# fig.suptitle('Controller')



data[simcols[:2]].plot(ax=axes[0])

data_stpt[['CDWTPythonSetpt']].plot(ax=axes[0], ls=':')

axes[0].set_ylim(20,80)

axes[0].set_ylabel('Temperature /F')

axes[0].set_title('Ambient Conditions, Temp: %.1f+/-%.1fF, Wetbulb: %.1f+/-%.1fF' % \

                  (data['TempAmbient'].mean(), data['TempAmbient'].std(),

                   data['TempWetBulb'].mean(), data['TempWetBulb'].std()))



ctefficiency = (data['TempCondOut'] - data['TempCondIn']) / \

               (data['TempCondOut'] - data['TempWetBulb'])

ctefficiency.plot(ax=axes[1])

axes[1].set_ylim(0,1)

axes[1].set_title('Cooling Tower Efficiency %.2f +/- %.2f' % (ctefficiency.mean(), ctefficiency.std()))



data[['PowChi', 'Tonnage']].plot(ax=axes[2])

kw = data['PowChi'] / data['Tonnage']

axes[2].set_title('Average chiller efficiency %.2f +/- %.2fkW/Ton' % (kw.mean(), kw.std()))

axes[2].set_ylabel('Power /kW')

axes[2].set_ylim(0, 500)



data['PerFreqFanA'].plot(ax=axes[3])

axes[3].set_ylabel('Percentage frequency /%')

axes[3].set_title('Fan frequency')

axes[3].set_ylim(0, 105)



data[['TempCondIn', 'TempCondOut']].plot(ax=axes[4])

data_stpt[['CDWTPythonSetpt']].plot(ax=axes[4], ls=':')

axes[4].set_ylim(50, 90)

deltat = (data['TempCondOut'] - data['TempCondIn'])

axes[4].set_ylabel('Temperature /F')

axes[4].set_title('Condenser Temps, Avg change %.2f +/- %.2fF' % (deltat.mean(), deltat.std()))



scatterdf = pd.DataFrame({

    'Cooling Tower Efficiency': ctefficiency,

    'Chiller load /%': data['Tonnage'] * 100/800,

})

corr = scatterdf.corr().values[0,1]

scatterdf.plot.scatter('Cooling Tower Efficiency', 'Chiller load /%', ax=axes[5])

axes[5].set_title('Efficiency vs Load, correlation %.2f' % corr)

axes[5].set_xlim(0, 1)

axes[5].set_ylim(0, 100)



scatterdf = pd.DataFrame({

    'Chiller load /%': data['Tonnage'] * 100/800,

    'kw/Ton': data['PowChi'] / data['Tonnage'],

    'Returning Water Temp /F': data['TempCondIn']

})

scatterdf.plot.scatter('Chiller load /%', 'kw/Ton', c='Returning Water Temp /F',

                       colormap='coolwarm', norm=colors.Normalize(50, 90), ax=axes[6])

mean, std = scatterdf.mean().values, scatterdf.std().values

axes[6].set_title('Average kw/Ton %.2f+/-%.2f, Chiller load %.2f+/-%.2f, Ret. temp %.2f+/-%.2f)' % \

                  (mean[1], std[1], mean[0], std[0], mean[2], std[2]))

axes[6].errorbar(mean[0], mean[1], xerr=std[0], yerr=std[1])

axes[6].set_xlim(0, 100)

axes[6].set_ylim(0, 1)





for ax in axes:

    ax.set_facecolor('honeydew')

    ax.grid(True, 'both')

plt.tight_layout()

```