Table of Contents

• Introduction

• Requirements

• Input_data

• Training_prediction

• Results

• Update

Introduction

This notebook presents an example of how to use rhced module to disaggregate heating and cooling electricity consumption from the net electricity consumption. Based on the sample data obtained from a smart thermostat and power meter, we will give a realistic example of the use of this model for users.

Requirements

There are several packages that needs to be installed. Please see this page for the required packages. In addition, change the directory and add path to use the modules.

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import pickle
import os
import re
import pathlib
import sys

%matplotlib inline

Change working directory and insert path.

init_wd=os.getcwd() #get current working directory
print(f'Current working directory is {init_wd}.') # The default is ~/rhced/notebook
# Add ~/rhced directory to the system path so that the module `rhced` can be loaded.
path=pathlib.Path(init_wd).parents[0].__str__()  
if not path in sys.path: 
    sys.path.insert(1, path)
os.chdir(path)
print(f'directory {path} is added to system path.')

Current working directory is c:\Users\ham9\rhced\notebook.
directory c:\Users\ham9\rhced is added to system path.

Loading modules. Warnings related to theano can be nelgected. If there are errors to import modules, please create a separate environment by using conda or virtualenv and install pymc3 > 3.11 again.

from rhced.data_utils import * # for data pre-processing
from rhced.training import *   # training modules
from rhced.prediction import * # prediction modules

WARNING (theano.tensor.blas): Using NumPy C-API based implementation for BLAS functions.

Input_data

Input data consists of three parts: (1) meta data, (2) thermostat data, and (3) meter data.

(1) Meta data

Each housing unit is classified two hierarchy: bldg and unitcode. bldg indicates the unique identifier of the building or site. unitcode is the unique identifier of the housing unit in a certain bldg. You need to specify the start_date and end_date to get access to the specific data. bldg, unitcode, start_date, and end_date are used for a unique identifier of (2) thermostat data and (3) meter data. So, please follow the same format given below.

In addition, max_values shows the maximum value of net, heatpump heating, heatpump heating with defrost control, heatpump_cooling, and auxiliary heating in Watt. During the calculation, the results are strictly restricted by the these maximum numbers, so please put numbers with a certain amount of margin. 120-150% can be a good choice.

output_path is used for storing outputs. Also, time interval is the model's time interval. We use 15-min interval due to defrost control. See section 2.2 in the paper.

bldg="sample_bldg" # building identifier. Use location identifier for single family detached house.
unitcode="sample" # housing unit identifier. 

start_date="2016-01-19"
end_date="2016-01-25"

max_values={'net_max':10000,'heat_max':2000,'cool_max':2000,'aux_max':6000,'df_max':4000} # unit of [Watt]

# define output_path. In this directory, we will create output_path/bldg/unitcode/ and store the outputs.
output_path=pathlib.Path().joinpath("outputs",bldg,unitcode)
time_interval=15 # due to defrost control, we use 15-min interval data (see Section 2.2 in the paper)

(2) Thermostat data

Once you put all meta data information, what you need to do is just put the data for training in the specified data directory.

Then, unit_prediction function in the below read data from the directory.
For example, in this example, the meta data is bldg:sample_bldg, unitcode:sample, start_date:2016-01-19, end_date: 2016-01-25. In this case, the thermostat_data's directory is ~/rhced/data/sample_bldg/sample/raw_data/thermostat_data_sample_2016-01-19_2016-01-25.csv. The meter_data's directory is ~/rhced/data/sample_bldg/sample/raw_data/meter_data_sample_2016-01-19_2016-01-25.csv.

To example data format, we import thermostat_data and meter_data.

# import thermostat_data and meter_data
thermostat_data_path=pathlib.Path().joinpath("data",bldg,unitcode,"raw_data","thermostat_data_sample_2016-01-19_2016-01-25.csv")
meter_data_path=pathlib.Path().joinpath("data",bldg,unitcode,"raw_data","meter_data_sample_2016-01-19_2016-01-25.csv")

df_thermostat=pd.read_csv(thermostat_data_path.__str__())
df_meter=pd.read_csv(meter_data_path.__str__())

thermostat_data has 7 columns.

Column	Unit	Type	Description
timestamp		str	Timestamp of the data point. The format should be %Y-%m-%d %H:%M:%S (e.g., 2016-01-02 13:15:00)
unitcode		str	Housing unit identifier. Any character is fine.
operation		str	Operation of HC system. See Section 3.3 in the paper.
T_in	°C	numeric	Indoor air temperature
rh_in	-	numeric	Indoor air relative humidity in 0-1 scale.
T_out	°C	numeric	Outdoor air temperature
rh_out	-	numeric	Outdoor air relative humidity in 0-1 scale.

df_thermostat.head(3)

	timestamp	unitcode	operation	T_in	rh_in	T_out	rh_out
0	2016-01-19 00:00:00	sample	idle	21.111111	0.34	0.527778	0.90
1	2016-01-19 00:05:00	sample	idle	21.111111	0.34	0.444444	0.91
2	2016-01-19 00:10:00	sample	idle	21.111111	0.34	0.388889	0.91

meter_data require 3 columns (timestamp, unitcode, and net). But, for the validation purpose, you can have HC data, too.

Column	Unit	Type	Description
timestamp		str	Timestamp of the data point. The format should be %Y-%m-%d %H:%M:%S (e.g., 2016-01-02 13:15:00)
unitcode		str	Housing unit identifier. Any character is fine.
net	W	numeric	Average power of net electricity during sampling interval in Watt.
ahu	W	numeric	Average power of air handler electricity during sampling interval in Watt (only necessary for validation purpose).
heatpump	W	numeric	Average power of heat pump electricity during sampling interval in Watt (only necessary for validation purpose).
hvac	W	numeric	Average power of air HC system (i.e., ahu+heatpump) during sampling interval in Watt (only necessary for validation purpose).

df_meter.head(3)

	timestamp	unitcode	ahu	heatpump	hvac	net
0	2016-01-19 00:00:00	sample	4.210000	34.070000	38.280000	50.843333
1	2016-01-19 00:05:00	sample	4.180000	34.093333	38.273333	50.973333
2	2016-01-19 00:10:00	sample	4.203333	34.426667	38.630000	51.023333

Training_prediction

The disaggregation model is unsupervised learning, so it is not necessarily to distinguish training/prediction. However, once the model is trained, it is not necessary to train or update model when there is new data. We just want to know the result. Therefore, training indicates the process of model update with new data, and prediction is the process of disaggregation with data and trained model parameters.

Both training and prediction are done by using unit_prediction function. The function gets all the meta data information (unitcode, bldg, start_date, end_date, max_values, output_path, and time_interval).

Other function arguments control the model update.

• new_traiing: Whether it is new training. If it is the first training of the specific housing unit, please put True. Otherwise, it should be False|
• force_update: When new data is available, the model check if the update is necessary. If new data doesn't include new HC operation, it is not updated. But, for the training purpose, you can force model update by putting True for this variable.|

The remaining arguments control model inference. n_samples is the number of samples to approximate the posterior distribution. n_training is the number of steps of ADVI learning. It should be large enough to guarantee the stochastic optimization of ADVI convergence. n_inference indicates the number of repeated ADVI. ADVI uses stochastic optimization, so the inference result may be local optima even if the convergence is observed with large n_training. To prevent this, you can repeat the inference with different initial values. This argument controls how many inference you want to repeat.

df,outputs=unit_prediction(unitcode=unitcode,bldg=bldg,start_date=start_date,end_date=end_date,max_values=max_values,time_interval=time_interval,output_path=output_path,new_training=True,force_update=True,n_samples=10000,n_training=60000,n_inference=5)

C:\Users\ham9\anaconda3\envs\rhced\lib\site-packages\pandas\core\groupby\generic.py:303: FutureWarning: Dropping invalid columns in SeriesGroupBy.agg is deprecated. In a future version, a TypeError will be raised. Before calling .agg, select only columns which should be valid for the aggregating function.
  results[key] = self.aggregate(func)

Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
New training, so new training and export the trained results.
New training

C:\Users\ham9\anaconda3\envs\rhced\lib\site-packages\theano\gpuarray\dnn.py:192: UserWarning: Your cuDNN version is more recent than Theano. If you encounter problems, try updating Theano or downgrading cuDNN to a version >= v5 and <= v7.
  warnings.warn(
WARNING (theano.tensor.opt): Optimization Warning: The Op erfcx does not provide a C implementation. As well as being potentially slow, this also disables loop fusion.

100.00% [60000/60000 01:38<00:00 Average Loss = -967.16]

Finished [100%]: Average Loss = -967.17

New training

100.00% [60000/60000 01:56<00:00 Average Loss = -967.25]

Finished [100%]: Average Loss = -967.26

New training

100.00% [60000/60000 01:56<00:00 Average Loss = -965.62]

Finished [100%]: Average Loss = -965.67

New training

100.00% [60000/60000 01:48<00:00 Average Loss = -961.85]

Finished [100%]: Average Loss = -961.82

New training

100.00% [60000/60000 01:49<00:00 Average Loss = -965.81]

Finished [100%]: Average Loss = -965.84

During this training, heat1 hours: 28.75,             heat2 hours: 0.0,             df1 hours: 83.5,             df2 hours: 0.0,             cool1 hours: 0.0,             cool2 hours: 0.0,             aux1 hours: 4.75,             fan1 hours: 0.0
So far, heat1 hours: 28.75,             heat2 hours: 0.0,             df1 hours: 83.5,             df2 hours: 0.0,             cool1 hours: 0.0,             cool2 hours: 0.0,             aux1 hours: 4.75,             fan1 hours: 0.0
feather file is created at outputs\sample_bldg\sample\df_2016-01-19_2016-01-25.feather
lower: 66.07166265891901
median: 141.42626536827805
upper: 218.6337332194337
measurement: 126.32639104166667

# #load data without training.
# import pyarrow.feather as feather
# df_values_path=output_path.joinpath("df_2016-01-19_2016-01-25.feather")
# df=feather.read_feather(df_values_path.__str__())

Results

The results of unit_prediction function are given in two formats. outputs is a dictionary that gives all the posterior distribution of parameter and variables. df is a summary of outputs in a Pandas dataframe format. df gives all the HC signals and 2.5 (lower), 50 (mid), and 97.5 (upper) percentiles of the posterior distribution of each operation's power.

Since P_{HC_operation} is given in power (Watt), the median of predicted HC kWh value of the data period can be calculated by np.nanmean(df['P_hc_mid'])/1000*24*n_days. The mean of P_hc_mid is divided by 1000 to make kW, and then it is multiplied the number of hours during the data period to calculate kWh.

df.head(3)

	timestamp	i_hc	i_heat1_all	i_cool1	i_aux1	i_df1	i_heat1	i_fan1	P_hc_lower	P_hc_mid	...	P_aux1_mid	P_aux1_upper	P_aux1_uq	P_aux1_lq	P_fan1_lower	P_fan1_mid	P_fan1_upper	P_fan1_uq	P_fan1_lq	hc
0	2016-01-19 00:00:00	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	38.394444
1	2016-01-19 00:15:00	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	38.722222
2	2016-01-19 00:30:00	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	...	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	38.444444

3 rows × 39 columns

In addition, you can visualize timeseries prediction results.

# 
x_time=pd.to_datetime(df['timestamp'])
fig, ax = plt.subplots(nrows=1, ncols=1,  figsize=(12,5))
ax.plot(x_time,df['hc'],'-',linewidth=0.8,color="red",label="Observation")
ax.fill_between(x=x_time,y1=df['P_hc_lower'],y2=df['P_hc_upper'],alpha=0.3,color="skyblue",label="95% uncertainty")
ax.plot(x_time,df['P_hc_mid'],'--',linewidth=0.5,label='Median')
ax.set(ylabel="Power [W]",xlabel="Time")
ax.legend(fontsize=11,loc='upper right')

<matplotlib.legend.Legend at 0x19eed06f610>

Update

As shown in Figure 11 in the paper, the model automatically updates model parameters when new HC operation is observed. In the below code, the model predicts for the several weeks of data and updates whenever necessary. The results are not explicitly visualized in this notebook, but the purpose of below code is to give how to do it.

start_date_="2016-01-26"
end_date_="2016-02-01"

start_date_vector=pd.date_range(start_date_, periods=24, freq='7D')
end_date_vector=pd.date_range(end_date_, periods=24, freq='7D')

for i_start,i_end in zip(start_date_vector,end_date_vector):
    start_date=i_start.strftime("%Y-%m-%d")
    end_date=i_end.strftime("%Y-%m-%d")
    df,outputs=unit_prediction(unitcode=unitcode,bldg=bldg,start_date=start_date,end_date=end_date,output_path=output_path,new_training=False,force_update=False,time_interval=time_interval,n_samples=10000,n_training=60000,n_inference=3)

last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-01-19_2016-01-25.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-01-19_2016-01-25.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
Posterior prediction only
feather file is created at outputs\sample_bldg\sample\df_2016-01-26_2016-02-01.feather
lower: 79.57744901520313
median: 164.55798660984797
upper: 250.53922114286632
measurement: 153.6213508333333
last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-01-19_2016-01-25.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-01-19_2016-01-25.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
Posterior prediction only
feather file is created at outputs\sample_bldg\sample\df_2016-02-02_2016-02-08.feather
lower: 51.252438625257106
median: 93.43201568913724
upper: 135.71121032757208
measurement: 86.84550439716313
last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-01-19_2016-01-25.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-01-19_2016-01-25.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
Posterior prediction only
feather file is created at outputs\sample_bldg\sample\df_2016-02-09_2016-02-15.feather
lower: 11.183752803070206
median: 18.352229139735286
upper: 26.38410375088394
measurement: 18.630678398398402
last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-01-19_2016-01-25.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-01-19_2016-01-25.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
Posterior prediction only
feather file is created at outputs\sample_bldg\sample\df_2016-02-16_2016-02-22.feather
lower: 6.069328585578284
median: 7.2253006444637435
upper: 8.263450986238858
measurement: 7.530834722222222
last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-01-19_2016-01-25.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-01-19_2016-01-25.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
Posterior prediction only
feather file is created at outputs\sample_bldg\sample\df_2016-02-23_2016-02-29.feather
lower: 26.484950740359842
median: 49.041526372475396
upper: 74.1370234923687
measurement: 43.57632047832586
last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-01-19_2016-01-25.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-01-19_2016-01-25.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
Posterior prediction only
feather file is created at outputs\sample_bldg\sample\df_2016-03-01_2016-03-07.feather
lower: 34.95658069341421
median: 59.43217606979067
upper: 87.84865850437752
measurement: 56.67537797101449
last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-01-19_2016-01-25.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-01-19_2016-01-25.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
Posterior prediction only
feather file is created at outputs\sample_bldg\sample\df_2016-03-08_2016-03-14.feather
lower: 26.023287597074972
median: 41.9482586450786
upper: 59.00726677415236
measurement: 39.80296416666667
last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-01-19_2016-01-25.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-01-19_2016-01-25.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
Posterior prediction only
feather file is created at outputs\sample_bldg\sample\df_2016-03-15_2016-03-21.feather
lower: 2.2716298450050614
median: 3.275182971186135
upper: 4.266343971452444
measurement: 3.5629765517241383
last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-01-19_2016-01-25.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-01-19_2016-01-25.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
Posterior prediction only
feather file is created at outputs\sample_bldg\sample\df_2016-03-22_2016-03-28.feather
lower: 13.966181836722189
median: 26.72747830239086
upper: 39.722454785412374
measurement: 23.747451137254906
last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-01-19_2016-01-25.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-01-19_2016-01-25.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
Model update and then posterior prediction.

100.00% [60000/60000 01:29<00:00 Average Loss = -1,385.7]

Finished [100%]: Average Loss = -1,385.5

100.00% [60000/60000 01:28<00:00 Average Loss = -1,387.1]

Finished [100%]: Average Loss = -1,387.1

100.00% [60000/60000 01:18<00:00 Average Loss = -1,385.4]

Finished [100%]: Average Loss = -1,385.3

During this training, heat1 hours: 3.75,             heat2 hours: 0.0,             df1 hours: 5.0,             df2 hours: 0.0,             cool1 hours: 11.25,             cool2 hours: 0.0,             aux1 hours: 0.0,             fan1 hours: 0.0
So far, heat1 hours: 32.5,             heat2 hours: 0.0,             df1 hours: 88.5,             df2 hours: 0.0,             cool1 hours: 11.25,             cool2 hours: 0.0,             aux1 hours: 4.75,             fan1 hours: 0.0

c:\Users\ham9\rhced\rhced\training.py:10: RuntimeWarning: overflow encountered in exp
  return 1./(1+np.exp(-500*(-x+x0)))

feather file is created at outputs\sample_bldg\sample\df_2016-03-29_2016-04-04.feather
lower: 12.152008999754775
median: 16.336477481345973
upper: 24.858806383064813
measurement: 16.26179611111111
last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-03-29_2016-04-04.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-03-29_2016-04-04.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
Posterior prediction only
feather file is created at outputs\sample_bldg\sample\df_2016-04-05_2016-04-11.feather
lower: 3.8905399651109605
median: 4.901591150209426
upper: 5.812001289461128
measurement: 3.975558055555555
last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-03-29_2016-04-04.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-03-29_2016-04-04.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
Posterior prediction only
feather file is created at outputs\sample_bldg\sample\df_2016-04-12_2016-04-18.feather
lower: 14.500644279777962
median: 18.162839254382284
upper: 21.443156278224386
measurement: 15.469370555555555
last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-03-29_2016-04-04.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-03-29_2016-04-04.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
Posterior prediction only
feather file is created at outputs\sample_bldg\sample\df_2016-04-19_2016-04-25.feather
lower: 24.476746908084923
median: 30.458194760886343
upper: 35.73667401626338
measurement: 29.193363643643643
last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-03-29_2016-04-04.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-03-29_2016-04-04.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.

c:\Users\ham9\rhced\rhced\training.py:10: RuntimeWarning: overflow encountered in exp
  return 1./(1+np.exp(-500*(-x+x0)))

Posterior prediction only
feather file is created at outputs\sample_bldg\sample\df_2016-04-26_2016-05-02.feather
lower: 50.751056135633576
median: 63.191676463177295
upper: 74.1186867430763
measurement: 60.60450805555556
last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-03-29_2016-04-04.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-03-29_2016-04-04.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
Posterior prediction only
feather file is created at outputs\sample_bldg\sample\df_2016-05-03_2016-05-09.feather
lower: 68.57043003619972
median: 85.25363648262419
upper: 99.81014898614974
measurement: 85.35886694444447
last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-03-29_2016-04-04.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-03-29_2016-04-04.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
Posterior prediction only
feather file is created at outputs\sample_bldg\sample\df_2016-05-10_2016-05-16.feather
lower: 81.80365306180686
median: 101.751981870556
upper: 119.17203864249615
measurement: 102.33163166666665
last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-03-29_2016-04-04.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-03-29_2016-04-04.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
Posterior prediction only
feather file is created at outputs\sample_bldg\sample\df_2016-05-17_2016-05-23.feather
lower: 106.72048565592148
median: 133.20477280280275
upper: 156.07021080206368
measurement: 139.0038266865969
last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-03-29_2016-04-04.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-03-29_2016-04-04.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
Posterior prediction only
feather file is created at outputs\sample_bldg\sample\df_2016-05-24_2016-05-30.feather
lower: 68.42051528903936
median: 85.26500335554991
upper: 99.92968525553678
measurement: 85.99558833333336
last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-03-29_2016-04-04.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-03-29_2016-04-04.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
Posterior prediction only
feather file is created at outputs\sample_bldg\sample\df_2016-05-31_2016-06-06.feather
lower: 78.72420680513137
median: 98.33230469778967
upper: 115.33069342115549
measurement: 99.810345
last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-03-29_2016-04-04.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-03-29_2016-04-04.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
Posterior prediction only
feather file is created at outputs\sample_bldg\sample\df_2016-06-07_2016-06-13.feather
lower: 105.6117778952692
median: 132.15556493468338
upper: 155.03272974976903
measurement: 138.68138805555557
last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-03-29_2016-04-04.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-03-29_2016-04-04.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
Posterior prediction only
feather file is created at outputs\sample_bldg\sample\df_2016-06-14_2016-06-20.feather
lower: 107.42715463265027
median: 134.23102321755346
upper: 157.47190322754648
measurement: 141.00933805555556
last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-03-29_2016-04-04.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-03-29_2016-04-04.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
Posterior prediction only
feather file is created at outputs\sample_bldg\sample\df_2016-06-21_2016-06-27.feather
lower: 105.64300991610396
median: 132.2016487147577
upper: 155.03151398975072
measurement: 141.42071611111112
last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-03-29_2016-04-04.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-03-29_2016-04-04.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
Posterior prediction only
feather file is created at outputs\sample_bldg\sample\df_2016-06-28_2016-07-04.feather
lower: 102.74847581175996
median: 128.75022688243598
upper: 151.09059619591432
measurement: 137.19567083333334
last_prior_values_path is outputs\sample_bldg\sample\prior_values_2016-03-29_2016-04-04.pickle
last_input_values_path is outputs\sample_bldg\sample\input_values_2016-03-29_2016-04-04.pickle
Model time_interval from previous training output is 15.0 minutes. If not please double check.
Drop NA data to create training data.
scale const is {'T_out_max': 40.0, 'T_out_min': -20.0, 'T_out_upper': 1.0, 'T_out_lower': -1.0, 'T_in_max': 30.0, 'T_in_min': 10.0, 'T_in_upper': 1.0, 'T_in_lower': -1.0, 'wb_in_max': 30.0, 'wb_in_min': 0.0, 'wb_in_upper': 1.0, 'wb_in_lower': -1.0, 'net_max': 10000, 'net_min': 0.0, 'heat_max': 2000, 'cool_max': 2000, 'df_max': 4000, 'aux_max': 6000}.
Posterior prediction only
feather file is created at outputs\sample_bldg\sample\df_2016-07-05_2016-07-11.feather
lower: 111.70222700533552
median: 139.1130532628102
upper: 162.84048776702605
measurement: 145.31479979674796