def initialize(self, data_vec):
extra_amount = 0 #Create the initial clusters from the 1st bit of data
while True:
current_time = 0
initial_data = data_vec[current_time : current_time + self.settings.init_size - extra_amount]
initial_delta_power = [np.round(initial_data[i + 1] - initial_data[i], 2) for i in range(len(initial_data) - 1)]
initial_events = [i for i in range(len(initial_delta_power)) if (initial_delta_power[i] > self.settings.T_Positive or initial_delta_power[i] < self.settings.T_Negative)]
try:
trial_clusters = gsp.refined_clustering_block(initial_events, initial_delta_power, self.settings.sigma, self.settings.ri)
break
except np.linalg.LinAlgError:
extra_amount += 1
print("\tSVD didn't converge, retrying with {} less data".format(extra_amount))
continue
self.trial_clusters, self.pairs = gsp.pair_clusters_appliance_wise(self.trial_clusters, data_vec, initial_delta_power, self.settings.instancelimit)
return (initial_delta_power, initial_events,self.settings.init_size - extra_amount)
beta = 0.5 # this defines the minimum number of times an appliance is set ON in considered time duration instancelimit = 3 #%% main_val = df.values # get only readings main_ind = df.index # get only timestamp data_vec = main_val signature_database = "signature_database_labelled.csv" #the signatures were extracted of power analysis from April 28th to 30th threshold = 2000 # threshold of DTW algorithm used for appliance power signature matching delta_p = [round(data_vec[i+1] - data_vec[i], 2) for i in range(0, len(data_vec) - 1)] event = [i for i in range(0, len(delta_p)) if (delta_p[i] > T_Positive or delta_p[i] < T_Negative) ] # initial and refined clustering block of Figure 1 in the paper clusters = gsp.refined_clustering_block(event, delta_p, sigma, ri) # Feature matching block of Figure 1 in the paper finalclusters, pairs = gsp.pair_clusters_appliance_wise(clusters, data_vec, delta_p, instancelimit) appliance_pairs = gsp.feature_matching_module(pairs, delta_p, finalclusters, alpha, beta) # create appliance wise disaggregated series power_series, appliance_signatures = gsp.generate_appliance_powerseries(appliance_pairs, delta_p) # label the disaggregated appliance clusters by comparing with signature DB labeled_appliances = gsp.label_appliances(appliance_signatures, signature_database, threshold) # Attach timestamps to generated series power_timeseries = gsp.create_appliance_timeseries(power_series, main_ind) # create pandas dataframe of all series
delta_p = [(data_vec[i + 1] - data_vec[i]).astype(int)
for i in range(0,
len(data_vec) - 1)]
event = [
i for i in range(0, len(delta_p))
if (delta_p[i] > settings.T_Positive or delta_p[i] < settings.T_Negative)
]
np.savetxt('d_events.txt',
np.array(event, dtype=np.dtype(np.int32)).astype(int),
fmt='%i')
print('Events is {}'.format(len(event)))
# initial and refined clustering block of Figure 1 in the paper
print('Clusters with {} {} {} {}'.format(len(event), len(delta_p),
settings.sigma, settings.ri))
clusters = gsp.refined_clustering_block(event, delta_p, settings.sigma,
settings.ri)
print('Expected {} got {}'.format(len(event), sum([len(c) for c in clusters])))
print('Got {} clusters'.format(len(clusters)))
# Feature matching block of Figure 1 in the paper
finalclusters, pairs = gsp.pair_clusters_appliance_wise(
clusters, data_vec, delta_p, settings.instancelimit)
print('Found {} pairs'.format(pairs))
print('Got {} clusters and {} final clusters'.format(len(clusters),
len(finalclusters)))
appliance_pairs = gsp.feature_matching_module(pairs, delta_p, finalclusters,
settings.alpha, settings.beta)
# create appliance wise disaggregated series
power_series, appliance_signatures = gsp.generate_appliance_powerseries(
appliance_pairs, delta_p)