In [2]:
from __future__ import print_function, division
import matplotlib
matplotlib.use('nbagg') # interactive plots in iPython. New in matplotlib v1.4
# %matplotlib inline
In [3]:
import matplotlib.pyplot as plt
from nilmtk import DataSet, MeterGroup
import pandas as pd
import numpy as np
from time import time
In [4]:
from pybrain.supervised import RPropMinusTrainer
from pybrain.datasets import SequentialDataSet
from pybrain.structure import RecurrentNetwork, FullConnection, FeedForwardNetwork
from pybrain.structure.modules import LSTMLayer, BiasUnit, LinearLayer, TanhLayer, SigmoidLayer
In [5]:
CONFIG = dict(
EPOCHS_PER_CYCLE = 5,
CYCLES = 6,
HIDDEN_LAYERS = [],
PEEPHOLES = True,
TRAINERCLASS = RPropMinusTrainer,
# instead, you may also try
# TRAINERCLASS = BackpropTrainer(net, dataset=trndata, verbose=True,
# momentum=0.9, learningrate=0.00001)
INPUTS = [], #, 'hour of day (int)', 'outside temperature', 'is business day (-1, 1)'
EXPERIMENT_NUMBER = 24
)
In [6]:
# Load dataset
dataset = DataSet('/data/mine/vadeec/merged/ukdale.h5')
dataset.set_window("2014-01-01", "2014-01-07")
elec = dataset.buildings[1].elec
In [7]:
# Select top-5 meters identified in UK-DALE paper
# APPLIANCES = ['kettle', 'dish washer', 'HTPC', 'washer dryer', 'fridge freezer']
APPLIANCES = ['kettle', 'toaster']
selected_meters = [elec[appliance] for appliance in APPLIANCES]
selected_meters.append(elec.mains())
selected = MeterGroup(selected_meters)
In [8]:
df = selected.dataframe_of_meters()
In [9]:
# Use human-readable column names
df.columns = selected.get_labels(df.columns)
In [10]:
mains = (df['Toaster'] + df['Kettle']).fillna(0).diff().dropna()
appliances = df['Toaster'].fillna(0).diff().dropna()
del df
In [11]:
# Constrain outputs to [-1,1] because we're using TanH
maximum = appliances.abs().max()
appliances /= maximum
mains_same_scale_as_appliances = mains / maximum
# standardise input
mains = (mains - mains.mean()) / mains.std()
In [12]:
ax = mains.plot()
ax = appliances.plot(ax=ax)
plt.show()
In [13]:
# Build PyBrain dataset
N_OUTPUTS = 1
N_INPUTS = 1
N = len(mains)
ds = SequentialDataSet(N_INPUTS, N_OUTPUTS)
ds.newSequence()
ds.setField('input', pd.DataFrame(mains).values)
ds.setField('target', pd.DataFrame(appliances).values)
In [14]:
ds.getSequence(0)
Out[14]:
In [19]:
# Build network
net = RecurrentNetwork()
def lstm_layer_name(i):
return 'LSTM{:d}'.format(i)
# Add modules
net.addInputModule(LinearLayer(dim=ds.indim, name='in'))
net.addOutputModule(TanhLayer(dim=ds.outdim, name='out'))
net.addModule(TanhLayer(10, name='tanh_input'))
# Bias
bias = BiasUnit()
net.addModule(bias)
#c_output_bias = FullConnection(bias, net['out'], name='c_output_bias')
#c_output_bias._setParameters(np.zeros(1))
#net.addConnection(c_output_bias)
c_tanh_input_bias = FullConnection(bias, net['tanh_input'], name='c_tanh_input_bias')
c_tanh_input_bias._params = np.zeros(c_tanh_input_bias.paramdim)
net.addConnection(c_tanh_input_bias)
forwards_connection = FullConnection(net['in'], net['tanh_input'], name='c_in_to_tanh')
forwards_connection._params = np.random.uniform(-0.2, 0.2, size=forwards_connection.paramdim)
net.addConnection(forwards_connection)
connect_to_out = FullConnection(net['tanh_input'], net['out'], name='c_tanh_to_out')
connect_to_out._params = np.random.uniform(-0.2, 0.2, size=connect_to_out.paramdim)
net.addConnection(connect_to_out)
net.sortModules()
print(net)
In [20]:
# define a training method
trainer = CONFIG['TRAINERCLASS'](net, dataset=ds, verbose=True)
In [21]:
# carry out the training
net.reset()
# train_errors = []
t0 = time()
EPOCHS = CONFIG['EPOCHS_PER_CYCLE'] * CONFIG['CYCLES']
# trainer.trainUntilConvergence(maxEpochs=EPOCHS, verbose=True)
# start_time = time()
print("Starting training with", EPOCHS, "epochs...")
for i in xrange(CONFIG['CYCLES']):
trainer.trainEpochs(CONFIG['EPOCHS_PER_CYCLE'])
# train_errors.append(trainer.testOnData())
# epoch = (i+1) * CONFIG['EPOCHS_PER_CYCLE']
# seconds_elapsed = time() - start_time
# seconds_per_epoch = seconds_elapsed / epoch
# seconds_remaining = (EPOCHS - epoch) * seconds_per_epoch
# td_elapsed = timedelta(seconds=seconds_elapsed)
# td_elapsed_str = str(td_elapsed).split('.')[0]
# eta = (datetime.now() + timedelta(seconds=seconds_remaining)).time()
# eta = eta.strftime("%H:%M:%S")
# print("\r epoch = {}/{} error = {} elapsed = {} ETA = {}"
# .format(epoch, EPOCHS, train_errors[-1], td_elapsed_str, eta),
# end="")
# stdout.flush()
print("Finished training. total seconds =", time() - t0)
In [22]:
# Disaggregate!
START = "2014-01-01"
END = "2014-01-03"
print("Starting disaggregation...")
net.reset()
estimates = pd.Series(index=appliances[START:END].index)
for date, mains_value in mains[START:END].iteritems():
estimates[date] = net.activate(mains_value)
In [23]:
estimates.plot()
plt.show()
In [24]:
mains[START:END].plot()
plt.show()
In [25]:
appliances[START:END].plot()
plt.show()
In [26]:
ax = estimates[START:END].cumsum().plot(label='estimates')
ax = mains_same_scale_as_appliances[START:END].cumsum().plot(ax=ax, label='aggregate')
ax = appliances[START:END].cumsum().plot(ax=ax)
plt.legend()
plt.show()
In [27]:
estimates.cumsum().to_hdf('neuronilm_estimates_{:03d}.hdf'.format(CONFIG['EXPERIMENT_NUMBER']), 'df')
In [ ]: