• No results found

A PPLICATION E XAMPLE

The proposed tool has been tested on a practical distribution network in New South Wales, Australia. Results will be presented in the following subsections.

3.4.1 Test Network

The test distribution network consists of an 80 km 11 kV backbone with 18 spurs of

different lengths in the range of 5 to 30 km. This rural network serves several sparsely populated areas. Three series voltage regulators are installed along the main feeder to control the voltage at remote ends. A simplified schematic diagram of the test network is shown in Fig. 3.9. The transformer symbols shown inside the PV cluster indicate the downstream LV feeder contains solar PV units. LV feeders are modeled using four-wire configuration and neutral wire is considered as solidly grounded at the customer service drop. Voltage and power flow profiles for the investigation are generated by performing load flow analysis of the test network.

Fig. 3.9. Test distribution network with locations of PV clusters.

3.4.2 Dimensionality Reduction of Results Database using SAX

For a 24 hours voltage profile with 5-min resolution data, the length of the time series is 288. Using SAX, the voltage profile time series can be reduced to a lower dimension, w, where w is the size of the SAX word. The compression ratio achieved using SAX can be defined as,

w

n ratio n

compressio (3.9)

where, n is the length of original data and w is the SAX word size. With a 20 word SAX representation, the compression ratio achieved for 1 day’s voltage profile data is 14.4.

Similarly, line flow data can be converted to SAX word and considerable reduction in data dimensionality can be achieved. Reduced data dimension will incur less computational effort in clustering and pattern recognition tasks to be performed on the data collected by smart meters.

3.4.3 Identification of Network Location Affected by Solar PV Clusters

Voltage profile time series data was clustered into 4 groups using k-means algorithm and the results are presented in Fig. 3.10. Clustering results of the original raw data are shown in Fig. 3.10(a) and that from SAX representations are shown in Fig. 3.10(b). It is observed in Fig. 3.10(a) that all of the four clusters created from the original data include voltage rise behaviour, and cluster 1 contains a mix (low and high) of voltage rise patterns. However, Fig. 3.10(b) shows that based on SAX representation of the data, the profiles with significant voltage rise are now grouped into cluster 4 only. The node locations corresponding to these voltage profiles can be identified using the cluster indices and further analysis of voltage rise now can be performed.

Similarly, daylong power flow profiles from the system nodes have been clustered using the original data as shown in Fig. 3.11(a) and using the approximated time series data shown in Fig. 3.11(b). Cluster 3 in Fig. 3.11(a) shows that the reverse power flow profiles due to solar PV have been clustered together with the ones that have not been affected by PV operation. On the other hand, when clustered using time series data obtained from SAX approximation, Fig. 3.11(b) shows that all the time series dataset with reverse power flow profiles at midday have been grouped into cluster 2 only.

 

Fig. 3.10. Clustering of voltage profiles for identification of network locations showing voltage rise: (a) using original time series data (b) using SAX approximation.

0 10 18

0.95 1 1.05 1.1

0 10 18

0.95 1 1.05 1.1

0 10 18

0.95 1 1.05 1.1

0 10 18

0.95 1 1.05 1.1

0 10 18

0.95 1 1.05 1.1

0 10 18

0.95 1 1.05 1.1

0 10 18

0.95 1 1.05 1.1

0 10 18

0.95 1 1.05 1.1

(b) (a)

Clustering on SAX Representation of Time Series Data

Cluster 1 Cluster 2 Cluster 3 Cluster 4

Cluster 1 Cluster 2 Cluster 3 Cluster 4

Clustering on Original Time Series Data

Fig. 3.11. Clustering of power flow profiles for identification of network locations showing reverse power flow: (a) using original time series data (b) using SAX approximation.

The main advantage of applying SAX was found in reduction of computational time.

A comparison was made on average runtime of k-means routine for 10 consecutive runs with original and SAX data. Reduction in elapsed time was about 16 times and 10 times for voltage and power flow profiles, respectively.

3.4.4 Detection of Abnormal Voltage Profile

Consecutive 14 days voltage profile time series data of one of the network buses affected by solar PV were analysed as 14 time series patterns. The original time series data of each day’s voltage profile was converted into SAX words. The distances of each of the SAX time series from each other were calculated to determine the time series that possesses the largest distance from its nearest non-self match. For simulation tests, cloud passing effects have been embedded (by incorporating sharp reductions in the PV output) in the time series data for two of the 14 days considered. The levels of reductions in the PV output are selected based on historical data. Detection of the voltage profile anomaly due to distortion by cloud passing effect was performed successfully, as shown in Fig. 3.12 below.

Fig. 3.12. Detection of unusual voltage behaviour.

5 10 15 20

-0.1 -0.05 0 0.05

5 10 15 20

-0.1 -0.05 0 0.05

5 10 15 20

-0.1 -0.05 0 0.05

5 10 15 20

-0.1 -0.05 0 0.05

5 10 15 20

-0.1 -0.05 0 0.05

5 10 15 20

-0.1 -0.05 0 0.05

(b)

Cluster 2 Cluster 1

Cluster 2

Cluster 3 (a) Cluster 1

Clustering on SAX Representation of Time Series Data Clustering on Original Time Series Data

Cluster 3

0 500 1000 1500 2000 2500 3000 3500

0.9 0.95 1 1.05 1.1

Time Series Data Points

Voltage [pu]

Day1 Day2 Day 5 Day13Day14

Unsual Voltage Profiles (distorted due to cloud passing effect)

3.4.5 Detailed Analysis of Identified Dataset Related to Solar PV Impacts

Once the patterns and anomalies of interest for PV impact assessment have been obtained from a large set of time series database, more detailed analysis can be carried out. The application of such detailed analysis may include determination of better network operation strategy with high PV penetration, or, design of corrective actions to mitigate PV impacts, etc. The type and methodology of the next stage of analysis will depend on its application. Load flow solution will be commonly required to perform such detailed analysis. This thesis will only show an example related to the PV impact on upstream voltage regulator operation.

Clustering of voltage profiles of the test system has identified that the LV feeders downstream to the third voltage regulator is experiencing voltage rise issues. Based on this information, it is important to investigate how the load centre voltage of this regulator is being affected by the voltage rise in the downstream LV feeders and how is this affecting the tap changer operation. The voltage profile at the load centre of this regulator was studied and presented in Fig. 3.13(a) which shows that the voltage is marginally equal to the upper limit set in the controller. This makes an intuitive suggestion that tap operation may have been performed by the regulator to keep the voltage within upper limit and this is verified by the tap operation profile shown in Fig.

3.13 (b). With the load and PV generation data corresponding to this scenario available at hand, load flow analysis was performed to investigate how the tap operation would be without solar PV. The load center voltage without PV is shown in Fig. 3.13(a) that shows the voltage is well below the threshold and no tap operation would have been performed without PV.

Fig. 3.13. Impact of midday voltage rise on tap changer operation: (a) Voltage profile at the load center of the third voltage regulator (b) Tap operation corresponding to the voltage profile in Fig. 3.13(a) (c) Fluctuations introduced by passing clouds in the load center voltage (d) Tap operation corresponding to the voltage profile in Fig. 3.13(c).

Passing clouds introduce fluctuations in the distribution feeder voltage with solar PV and may also produce fluctuations in the tap changer operation. The load and PV generation data corresponding to the most anomalous day identified in Fig. 3.12 by the anomaly detection technique were used in the load flow analysis. Tap operation of the third voltage regulator with this data was examined. Passing cloud induced fluctuations were observable from the load center voltage, as shown in Fig. 3.13(c), and fluctuation in tap operations were observed at the same time periods, as shown in Fig. 3.13(d). It is to be noted that the voltage fluctuation at about 15:00 hours did not impact the regulator operation as voltage rise at this moment was not so high as to cause the tap changer to operate.

Since archived data from real-time measurements is used in the PV impact analysis, more accurate results should be obtained when compared to those using assumption based synthetic data. This will help utility planning engineers to understand the PV impacts on the distribution systems better. The proposed approach can be used to track and visualise the identified PV impacts of interest, and also identify unique signatures within the utility system. This can be achieved using the three-phase load flow analysis, capable of dealing with network asymmetry, load unbalance, single-phase solar PV integration, and their impact on upstream network, described in section 3.3.5. Using the unique signatures of the identified PV impacts, the proposed approach can assist in providing distribution system infrastructure security via early warning systems of solar PV impacts in real-time.

4 8 12 16 20 24

0.96 0.98 1 1.02 1.04

Time [hour]

Load Cen. Volt [pu]

4 8 12 16 20 24

-6 -4 -2 0 2

Time [hour]

Tap Numer

4 8 12 16 20 24

0.96 0.98 1 1.02 1.04

Time [hour]

Load Cen. Volt. [pu]

4 8 12 16 20 24

-6 -4 -2 0 2

Time [hour]

Tap Numer

Tap operation due to voltage rise at midday

Fluctuations caused by passing clouds

Fluctuations in Tap operation

(a)

(c) (d)

(b) Upper Threshold

Voltage without PV