Towards a Data-Driven Approach to Ground-Fault Location
Mauro Caporuscio1, Pieternella Cijvat2, Hans Ottosson3
Core research areas: Computer Science (Data-driven technology), Electrical Engineering (Power Grids)
Summary of activities and results of the seed project.
The SEED project conducted by Linnaeus University and Hughes Power System aimed at investigating state- of-the-art in ground-fault location methods at the distribution level, as well as at performing an initial feasibility study about the possibility of employing simulation-based data-driven solutions. This was concretely translated into the following goals:
• State-of-the-art review and analysis of existing fault location methods used at the distribution level nowadays; Their respective limits, drawbacks and advantages.
• Understand specificity of distribution power system in comparison to transmission system.
• Develop a distribution power system model with the use of a suitable EMTP software, such as Matlab/Simulink. Identify the degree of accuracy of the model required for fault transient simulation, simplification and relevant assumption.
• Use the distribution system model to generate big data of faulted transient signals in time domain.
In addition, we aimed at (1) identifying and specifying the industrial needs for the future, and (2) planning the development and commercialization of the envisioned approach.
The state-of-the-art analysis has been carried out as well as the related report completion. A 11-bus, 10 km long radial Distribution Power System and its component has been modeled in Matlab/Simulink (see Figure 1). This system contains five branches in order to test how the method reacts to the multiple possible location problem. It also has a distributed generator represented by a 20 MVA synchronous generator to fit the recent increase of distributed generation within modern power system.
Concerning faulted transient signals, substation feeder voltage has been elected as the faulted data according to the state-of-the art analysis results. To run a large amount of simulation, a MatLab «data generation»
script has been coded. It enables to run any fault scenario at any of the 38 predefined locations in the system (every 500 m). The possible fault scenarios depend on system parameter such as distributed generation penetration, load value, inception angle, fault impedance and faulted phase. A total of 6384 simulations (168 fault scenarios at 38 locations) have been carried out. This number can easily be increased if needed to meet the secondary objective.
The review of the signal processing theory and its implementation in Matlab has been partially completed.
The technique chosen for this application is the wavelet transform: most used for high frequency transient signal since it allows both time and frequency analysis of the signals. Finally, study of supervised machine
1 mauro.caporuscio@lnu.se - Linnaeus University (Dep. of Computer Science and Media Technology)
2 ellie.cijvat@lnu.se - Linnaeus University (Dep. of Physics and Electrical Engineering)
3 hans.ottosson@hughespowersystem.com - HUGHES Power System Figure 1: 11-bus Distribution Power System
learning theory has been started recently. However, the fault location algorithm hasn’t been developed yet, and hence, nor the results and their analysis.
Results
Two examples of faulted transient voltages at the substation system are shown in Figure 2. For instance, one can see the impact of the inception angle on the electromagnetic transient.
Figure 2: Ground fault of 0.01 Ohm on phase A at 2500 m on path 2
Voltage transient amplitudes are greater when the inception angle approaches 90°, in other word when faulted phase voltage is close to its maximum.
As signal processing result, Figure 3 shows the discrete wavelet transform of a generated transient voltage.
Wavelet transform decomposition enables to extract relevant feature of the signal, that can be used to train a machine learning model.
Figure 3: Level 5 discrete wavelet transform
Challenges, deviations, and future work
According to the Project Proposal, expected results were:
O1.1 A detailed reasoning about existing solutions for the ground-fault location problem
R
O1.2 A simulation study proving the feasibility of the envisioned approach
R
O2.1 A Requirements specification for future Smart Electronics products
Q
O2.2 A Project proposal, including implementation and commercialization plans, to be
submitted for external funding to Vinnova’s Smarter Electronic Systems 2020 call
Q
While O1.1 and O1.2 have been fulfilled, we have not been able to achieve O2.1 and O2.2. Challenges encountered in the project, and the related deviations, are mostly due to the sudden spreading of COVID- 19, which heavily affected the project development and impacted on the planned activities.
However, even though the SEED project is formally over, we intend to address the unfulfilled objectives, and apply for external funding as soon as possible.
Budget
Main costs are reported in the following table.
Participant Item SEED2020 Co-financing TOTAL[SEK]
LNU
Student Internship:
Antoine Dupuis 50.000 50.000
R&D time - (LNU) 50.000 50.000
HPS R&D time (125
hours, 800 SEK/h) 50.000 (HPS) 50.000 100.000
Travels and Meetings organization
0 - 0
Total 100.000 100.000 200.000
It is worth noticing that due to the COVID-19, we did not have any cost related to Travel and Meetings organization.
