First Name: 
Konstantinos
Last Name: 
Theodorakos

Datasets & Analysis

Model 2.1.4

 

Clustered-meter average monthly ratios

 

 

Monthly (kWh) consumption clustering:

 

1. Twelve clustering models (one per month of signup): Spectral Clustering with nearest neighbors.

2. Clustering using (daily kWh) consumption extracted features. Cluster count set to 2-4: final count is decided by maximum silhouette score of manhattan distances.

3. Meters not in the training clustering data, are classified to a specific cluster using a Gaussian Process Classifier (GPC) on the same dataset (12 classifiers total).

1 Views

Model 2.1.3

 

Clustered-meter average monthly ratios

 

 

Monthly (kWh) consumption clustering:

 

1. Twelve clustering models (one per month of signup): Spectral Clustering with nearest neighbors.

2. Clustering using (daily kWh) consumption extracted features. Cluster count set to 2-4: final count is decided by maximum silhouette score of manhattan distances.

3. Meters not in the training clustering data, are classified to a specific cluster using a Gaussian Process Classifier (GPC) on the same dataset (12 classifiers total).

1 Views

Model 2.1.1

 

Clustered-meter average monthly ratios

 

 

Monthly (kWh) consumption clustering:

 

1. Twelve clustering models (one per month of signup): Spectral Clustering with nearest neighbors.

2. Clustering using (daily kWh) consumption extracted features. Cluster count set to 2-4: final count is decided by maximum silhouette score of manhattan distances.

3. Meters not in the training clustering data, are classified to a specific cluster using a Gaussian Process Classifier (GPC) on the same dataset (12 classifiers total).

2 Views

 

 

Model 2.15

 

Clustered-meter average monthly ratios

 

 

Monthly (kWh) consumption clustering:

 

1. Twelve clustering models (one per month of signup): Uniform Manifold Approximation and Projection (UMAP) sub-space euclidean Gaussian Mixture Models (GMM) clustering.

2. Clustering using (daily kWh) consumption extracted features. Cluster count set to 4.

 

Forecasting for a meter: Using the mean monthly consumption + in-cluster month-to-month ratios.

 

Clustering

 

1 Views

Model 1.6

 

Linear Boosting (XGBoost GBLinear) + Clustered-meter average monthly ratios

 

 Monthly (kWh) consumption clustering:

 

1. Twelve clustering models (one per month of signup): Spectral Clustering

2. Clustering using (daily kWh) consumption extracted features. Cluster count set to 4 (empirically better than the 5 or 6 suggested by the elbow distortion method).

 

Forecasting for a meter: Using the mean monthly consumption linear prediction + in-cluster month-to-month ratios.

 

Clustering

 

1 Views

Model 2.12

 

Clustered-meter average monthly ratios

 

 

Monthly (kWh) consumption clustering:

 

1. Twelve clustering models (one per month of signup): Sub-space euclidean Gaussian Mixture Models (GMM) clustering.

2. Clustering using (daily kWh) consumption extracted features. Cluster count set to 2-6: Adjusted (with penalty term for cluster count un-evenness) silhouette index choses the cluster count.

 

Forecasting for a meter: Using the mean monthly consumption + in-cluster month-to-month ratios.

 

2 Views

 

Model 1.5.1

 

Dropout Additive Regression Trees (DART) + Clustered-meter average monthly ratios

 

 Monthly (kWh) consumption clustering:

 

1. Twelve clustering models (one per month of signup): Spectral Clustering

2. Clustering using (daily kWh) consumption extracted features. Cluster count set to 4 (empirically better than the 5 or 6 suggested by the elbow distortion method).

 

Forecasting for a meter: Using the mean monthly consumption + in-cluster month-to-month ratios.

 

Clustering

 

1 Views

Model 1.5

 

Dropout Additive Regression Trees (DART) + Clustered-meter average monthly ratios

 

 Monthly (kWh) consumption clustering:

 

1. Twelve clustering models (one per month of signup): Spectral Clustering

2. Clustering using (daily kWh) consumption extracted features. Cluster count set to 4 (empirically better than the 5 or 6 suggested by the elbow distortion method).

 

Forecasting for a meter: Using the mean monthly consumption + in-cluster month-to-month ratios.

 

Clustering

 

1 Views

 

Model 2.12

 

Clustered-meter average monthly ratios

 

 

Monthly (kWh) consumption clustering:

 

1. Twelve clustering models (one per month of signup): Spectral Clustering

2. Clustering using (daily kWh) consumption extracted features. Cluster count set to 4 (empirically better than the 5 or 6 suggested by the elbow distortion method).

 

Forecasting for a meter: Using the mean monthly consumption + in-cluster month-to-month ratios.

 

Clustering

 

1 Views

 

Model 2.11

 

Clustered-meter average monthly ratios

 

 

Monthly (kWh) consumption clustering:

 

1. Twelve clustering models (one per month of signup): Spectral Clustering

2. Clustering using (daily kWh) consumption extracted features. Cluster count set to 4 (empirically better than the 5 or 6 suggested by the elbow distortion method).

 

Forecasting for a meter: Using the mean monthly consumption + in-cluster month-to-month ratios.

 

Clustering

 

1 Views

 

Model 2.10

 

Clustered-meter average monthly ratios

 

 

Monthly (kWh) consumption clustering:

 

1. Twelve clustering models (one per month of signup): Spectral Clustering

2. Clustering using (daily kWh) consumption extracted features. Cluster count set to 4 (empirically better than the 5 or 6 suggested by the elbow distortion method).

 

Forecasting for a meter: Using the mean monthly consumption + in-cluster month-to-month ratios.

 

Clustering

 

1 Views

Model 2.9

 

Clustered-meter average monthly ratios

 

 

Monthly (kWh) consumption clustering:

 

1. Twelve clustering models (one per month of signup): Spectral Clustering

2. Clustering using (daily kWh) consumption extracted features. Cluster count set to 4 (empirically better than the 5 or 6 suggested by the elbow distortion method).

 

Forecasting for a meter: Using the mean monthly consumption + in-cluster month-to-month ratios.

 

Clustering

 

1 Views

 

Model 2.8

 

Clustered-meter average monthly ratios

 

 

Monthly (kWh) consumption clustering:

 

1. Twelve clustering models (one per month of signup): Spectral Clustering

2. Clustering using (daily kWh) consumption extracted features. Cluster count set to 4 (empirically better than the 5 or 6 suggested by the elbow distortion method, given 2-20 clusters range).

 

Features (170) from weekend/weekday and full series:

- Statistical: median, variance, quantiles, ...

1 Views

 

Model 2.7

 

Clustered-meter average monthly ratios

 

 

Monthly (kWh) consumption clustering:

 

1. Twelve clustering models (one per month of signup): Spectral Clustering

2. Clustering using (daily kWh) consumption extracted features. Cluster count set to 4 (empirically better than the 5 or 6 suggested by the elbow distortion method).

 

Features (170) from weekend/weekday and full series:

- Statistical: median, variance, quantiles, ...

- Time-series: autocorrelations, trends, seasonalities, ...

1 Views

Model 2.6

 

Clustered-meter average monthly ratios

 

 

Monthly (kWh) consumption clustering:

 

1. Twelve clustering models (one per month of signup): Spectral Clustering.

2. Clustering using (daily kWh) consumption extracted features. Cluster count set to 4 (empirically better than the 5 or 6 suggested by the elbow distortion method).

 

Features (170) from weekend/weekday and full series:

- Statistical: median, variance, quantiles, ...

- Time-series: autocorrelations, trends, seasonalities, ...

 

1 Views

Model 2.5

 

Clustered-meter average monthly ratios

 

 

Monthly (kWh) consumption clustering:

 

1. Twelve clustering models (one per month of signup): K-medoids Clustering

2. Clustering using (daily kWh) consumption extracted features. Cluster count set to 4.

 

Features (170) from weekend/weekday and full series:

- Statistical: median, variance, quantiles, ...

- Time-series: autocorrelations, trends, seasonalities, ...

 

1 Views

Model 2.4

 

Clustered-meter average monthly ratios

 

 

Monthly (kWh) consumption clustering:

 

1. Twelve clustering models (one per month of signup): Spectral Clustering

2. Clustering using (daily kWh) consumption extracted features. Cluster count set to 3-6: final count is decided by maximum silhouette score of euclidean distances.

 

Additional Preprocessing:

- Missing daily consumption values filled-in with (full) time-series mean (per meter).

 

1 Views

 

 

Model 2.3

 

Clustered-meter average monthly ratios

 

 

Monthly (kWh) consumption clustering:

 

1. Twelve clustering models (one per month of signup): Agglomerative Clustering

2. Clustering using (daily kWh) consumption extracted features. Cluster count set to 5-12: final count is decided by maximum silhouette score of euclidean distances.

 

Features (170) from weekend/weekday and full series:

- Statistical: median, variance, quantiles, ...

1 Views

Smart Meters IEEE contest

 

 

Model 2.2

 

Clustered-meter average monthly ratios

 

 

Monthly (kWh) consumption clustering:

 

1. Twelve clustering models (one per month of signup): Spectral Clustering with nearest neighbors.

2. Clustering using (daily kWh) consumption extracted features. Cluster count set to 2-23: final count is decided by maximum silhouette score of manhattan distances.

 

Features (170) from weekend/weekday and full series:

- Statistical: median, variance, quantiles, ...

1 Views

Model 2.1

 

Clustered-meter average monthly ratios

 

 

Monthly (kWh) consumption clustering:

 

1. Twelve clustering models (one per month of signup): Spectral Clustering with nearest neighbors.

2. Clustering using (daily kWh) consumption extracted features. Cluster count set to 2-4: final count is decided by maximum silhouette score of manhattan distances.

3. Meters not in the training clustering data, are classified to a specific cluster using a Gaussian Process Classifier (GPC) on the same dataset (12 classifiers total).

1 Views

Model 2

 

Clustered-meter average monthly ratios

 

 

Monthly (kWh) consumption clustering:

 

1. Twelve clustering models (one per month of signup): Spectral Clustering with nearest neighbors.

2. Clustering using (daily kWh) consumption extracted features. Cluster count set to two.

3. Meters not in the training clustering data, are classified to a specific cluster using a Gaussian Process Classifier (GPC) on the same dataset (12 classifiers total).

 

Features from weekend/weekday and full series:

1 Views

Model 1.4

Gradient Boosting Trees using GBtree (Python xgboost library)

 

Monthly (kWh) consumption forecasting:

1. Twelve regression models (one per month).

2. Regression using (daily kWh) consumption extracted features.

 

From weekend/weekday and full series:

- Statistical: median, variance, quantiles, ...

- Time-series: autocorrelations, trends, seasonalities, ..

 

Regression model training: Nested cross-validation.

- Outer k-fold CV loop splits held-out test data.

1 Views

Model 1.3

 

Gradient Linear Booster using Dart (Python xgboost library)

 

Monthly (kWh) consumption forecasting:

1. Twelve regression models (one per month).

2. Regression using (daily kWh) consumption extracted features.

 

From weekend/weekday and full series:

- Statistical: median, variance, quantiles, ...

- Time-series: autocorrelations, trends, seasonalities, ..

 

Regression model training: Nested cross-validation.

- Outer k-fold CV loop splits held-out test data.

1 Views

Model 1.2

 

Gradient Linear Booster (Python xgboost library)

 

Monthly (kWh) consumption forecasting?

1. Twelve regression models (one per month).

2. Regression using (daily kWh) consumption extracted features.

 

From weekend/weekday and full series:

 

Statistical: median, variance, quantiles, ...

Time-series: autocorrelations, trends, seasonalities, ..

 

Regression model training: 60% (train),20% (validation), 20% (test).

 

Change vs Model 1: Linear interpolation on the y labels.

4 Views

Model 1.1

 

Gradient Boosted Trees (Python xgboost library)

 

Monthly (kWh) consumption forecasting?

1. Twelveregression models (one per month).

2. Regressionusing (daily kWh) consumption extracted features.F

 

rom weekend/weekday and full series:

Statistical: median, variance, quantiles, ...

Time-series: autocorrelations, trends, seasonalities, ..

 

Regression model training: 60% (train),20% (validation), 20% (test).

 

Change vs Model 1: Linear interpolation on the y labels.

5 Views

Long-term (up to 11 months ahead) forecasting using regression Ensemble Trees. 

For monthly (kWh) consumption forecasting, we trained twelve regression models (one per month), using (daily kWh) consumption extracted features. From weekend/weekday and full series (170 features total):

  • Statistical: median, variance, quantiles, ...
  • Time-series: autocorrelations, trends, seasonalities, ...

Data separation:

  • General regression learning workflow: 60% (train), 20% (validation), 20% (test).

Model types:

3 Views