Power Distribution Network (Transactive)

From Mindworks
Jump to navigation Jump to search
Transactive Energy
Sponsors Avista
Team Name Team Transactive
Duration Fall 2019 - Spring 2020
Client University of Idaho
Lead Hussain Beleed
Instructor Lead Yacine Chakhchoukh
Mentors
  • Brian Johnson
  • Herbert Hess
Grad Students
  • Andrew Miles
  • Matthew Holmen
Team Members
  • Karstin Neill
  • Christine Page
  • James Todd
  • Nikki Tran

The goal of this project is to implement a small-scale simulated model of the power distribution network representing the impact of reverse power flows and analyze and discuss their impact on the stability and operation of the distribution system and on pricing.


Problem Definition[edit | edit source]

Transactive Energy Diagram.png

A first step toward implementing electricity markets at the distribution system level and incorporating prosumers as providers within the public utility system requires a real-time market auction system. The above illustrates an example of the process flow for a transactive energy system.

A previous senior design team developed a distribution system model in OpenDSS, and implemented planning level studies for analyzing transactions. The next step is to build an updated model with real-time communication between the small-scale model and an auctioning system for data analyzing purposes. This way, real-time data can be obtained by the use of the RTDS. A pricing system will also be created based on the results of the small-scale model.

A modular integrated real-time auction system will need to be created so that users can buy and sell power between themselves and the current power producer, as well as allow for transaction prioritization on sponsor defined variables. The designed auction must be secure, maintainable, modular, and well-documented so that features can be added, dropped, or changed at a later date.

Background[edit | edit source]

In the modern power grid, consumers are transforming into prosumers where electric power end-users could also generate their own energy from renewable resources such rooftop photovoltaic panels (PVs) or have stored energy in batteries (electric vehicle) and inject it back into the grid to sell the excess energy. This results in reverse power flows that could have an impact on the stability and the economics of the operation of the power distribution and transmission systems. The power distribution system could benefit from this local distributed generation if the adequate controls and sensing are developed.

Deliverables[edit | edit source]

IEEE 34 bus system.png
  • Working 34-bus RSCAD model, and ATP IEEE 34-bus model for validation as shown above.
  • Real-time pricing system, with implemented locational marginal pricing (LMPs).
  • Visualizations of both of both the pricing system and RSCAD/ATP models.
  • Implementation of PV into RSCAD/ATP models
  • Documentation about the program and model's inner workings, as well as the team's process in creating both.
  • A description of difficulties faced while creating the above.

Specifications[edit | edit source]

Functional Requirements
The final product should be able to accept real-time bid input into the system, implement a changing locational marginal pricing (LMP), and conduct transactions without error. This should be able to be accomplished by using the continuous data being collected by an SEL RTAC (real-time automation controller) that is going to distribute data to the bid system.
Reliability Requirements
The system should process bids and transactions without error. It should be designed to operate for a defined amount of time without maintenance.

Model Requirements
The model should be built in RSCAD off of the 34-bus model. The model must be able to accept input from the auctioning system in real-time. It must be built to handle expected and unexpected line outages, as well as have constraints such as voltage regulators, low voltages, and line conductors.
The model that will be used to verify the RSCAD model will be built in ATP. This will allow for the data to show statically if the RSCAD model is correct.

Software Requirements
The software should be designed to function constantly without restart. The software should be designed such that a single transaction failure should not cause system failure. The auctioning system should be able to take in real-time bids. The software should be able to relay transaction information to the model, and from the model to the auctioning system.

Design Considerations[edit | edit source]

  • Transferring a small-scale model from OpenDSS to a real-time automated program RSCAD
  • Develop full communication between small-scale model and an auctioning system for data analysis
  • Real-time communication between the RSCAD system and the developed auction system

Project Learning[edit | edit source]

EE side[edit | edit source]

RSCARD/ATP Models:

Early Project:

RSCAD model transactive.PNG
ATP model transactive.png

Late Project:

RSCAD model2 transactive.PNG
RSCAD model3 transactive.PNG
ATP model2 transactive.PNG



The EE side reviewed the OpenDSS 13-bus model and learned OpenDSS by attending a conference. This allowed the EE side to start working on the RSCAD and ATP 34-bus models. We then went through learning how to model different kinds of loads as well as trying to master three different modeling programs. While working through these modeling programs, we figured out the weird quirks of the programs (such as OpenDSS's inability to do controls, how ATP can't do three-phase loads so a splitter needs to be used to be able to do so, and how RSCAD can use real-time simulations using the RTDS to be able to control the system without stalling the simulation). We also are looking into which spot and distribution loads will be replaced with solar panels (solar power) to be able to emulate AVISTA's 34-bus system better.

Gaussian transactive.png
Resistor transactive.png

In the updated ATP model, three lookup tables have been added, one for each phase, to model an irradiance curve in the form of a Gaussian distribution, pictured above. Each phase of the irradiance curve is added up to total 100kW, which is the size of the PV load. This is then converted into a DC current, which is found by dividing the power by the maximum DC voltage of 600V. Using a shunt capacitor and a series resistor pictured above, the DC current is injected into a bus of the system.

SAM transactive.PNG
Season-adjusted SAM.png

Using SAM (System Advisory Model), which is a program created by NREL that pings the nearest weather station for data at a given location that you enter in, it gives you irradiance information in the form of 8760 data. The data collected for Spokane, chosen to represent Avista's service area is shown above. This data will be normalized to represent four days that for each seasonal load for Spokane: spring, summer, fall, and winter.

CS side[edit | edit source]

Messy graph.png

The CS side of the project first began work on the system flow and pricing tool portion of the project by visualizing the problem as a variant of the famous Knapsack problem. While working to develop this tool, several steps were made. The team came up with a comparison of various implementations of the Knapsack Problem, including a brute force, dynamic programming, and evolutionary computation method. The group found that the dynamic method was by far the fastest implementation with a very fast run time, whereas the evolutionary method had an execution time on the same order of magnitude, and used less memory--so it is perhaps more scalable for larger systems. A large portion of time was also spent developing a parser for the output OpenDSS consumer and bus information files. This implementation was done in Python, originally in a Pythonic-list-comprehension method and later transferred into a regular expression implementation method to allow for a more moddable workflow. The parser created graphs using NetworkX, a Python library for graph theory, and originally separated the phases of the network into three separate graphs. Per a suggestion from our client, this idea was discarded in favor of a single multigraph due to it being a closer representation of the overall system. This too caused issues due to the fact that the NetworkX library does not have a method to print multigraphs with separate lines. A solution was attempted to be created by hand by changing the ordering within the nodes libraries dictionaries, by adding the nodes in certain orders, and by exporting the graphs to .DOT files to be graphed in a different Python library called PyDot.

At this point, the CS side transitioned from a Knapsack Problem visualization of the network to a Min Cut/Max Flow visualization, as it was suggested to be a more applicable solution type for the goals required. This implementation also required a great deal of change, because the normal solution to the multi-source multi-sink Min Cut/Max Flow problem creates infinite capacity lines between a single imaginary source and sink and the various real ones to transfer a difficult problem to a more tractable one. This, however, does not match the reality of the situation, because this would consider the problem constraints satisfied as long as the correct overall amount of power was transferred to any subset of the consumers, instead of the right amount to each. Therefore, it was decided that the client would be approaching the problem as though it were a multi commodity network flow problem.

The CS side of the team then pursued visualization of the problem. The graph theory version of the graph read in from the 34 bus system is shown above. The parser that was previously used was rebuilt to handle output from the EE models, including a node filtering mechanism.

Final Design[edit | edit source]

Final RSCAD model.png
IEEE 34 bus system ATP.png

The above shows the final implementations of the RSCAD model and ATP models.


Final PV Implementation.png

In order to implement a PV system in the ATP model, it must be implemented as a single-phase DC current. Using the seasonal irradiance curves for each phase in a lookup table in ATP, they are added up to calculate the total power of each PV system. We chose a PV load of 100 kW to match what last year’s group had used as one of their PV system loads. The total power is divided by the DC voltage, which is assumed to be 600 V and split into a positive and negative DC current. The positive DC current is then connected with a 1 mOhm series resistor and a 20 mF shunt capacitor. The PV system in ATP is shown above.

RSCAD PV 2.png

In RSCAD, the modeling program as a solar panel block that can be added. For the PV solar panel model, the insolation level and temperature affect the output DC voltage and current, which are then converted to AC through a three-phase DC/AC voltage source converter. In this case, the insolation level is determined by a table of values pulled from the SAM data for one day in either spring, summer, fall, or winter. The PV system in RSCAD is shown above.

RSCAD PV 3.png

The DC/AC converter is used to convert the single-phase DC currents into three-phase AC currents. Inside the PV controls, Park’s Transformation is used to reduce the three AC quantities to two dc quantities rotating at the same speed about a fixed axis. It is used to reduce complexity for the control system with the development of three-phase converters. The Park’s Transformation implementation is shown above.

ATP PV Implementation.png

The above shows how the PV system was implemented in ATP. Using a lookup table for each phase, the seasonal irradiance curves were modeled for Spokane. The three phases are added together to get the total power of each PV system. The total power is divided by twice the amount of the DC voltage of 600V. By dividing the total power into two parts, a positive DC current and a negative DC current are created. The positive DC current was used as a current injection connected to a resistor and shunt capacitor for each phase.

Season-adjusted SAM.png

Due to cloud coverage, rainy days, or objects obscuring weather data measurements, some hours of the day are producing 0 kW, even if it is in the middle of the day. To make up for this, four representative days were created for each season to test out model. The irradiance curves of each season are shown above.

In the current implementation of the Python program, the parser attempts to open and read from an ATP .LIS file. Our first implementation of this functionality used a more rigid, hard-coded version of the file parsing, but after reworking this several times due to a changing file structure, we switched to a regular expression approach to file parsing. This design allows the software to take in any number of different variables that may be present in the data, provided consistent formatting is used. By the end of the first section of our program, the data is stored as a list of dictionaries, which is efficient and easy to access.

Shell layout.png

To visualize the data gathered by our parser, we developed a mathematical graph theory representation of the buses and their connections. We are now using the NetworkX shell layout, which attempts to lay the nodes equidistant on a circle’s circumference with as little overlap between edges and nodes as possible . Filtering

Filtered shell.png

These filters limit what is shown by excluding or including subsets of the information denoted by bus names. This allows for users to “zoom in” on a small subset of the power grid buses, or to remove unnecessary buses without altering the actual data set so text on the nodes and edges to be larger without causing overlap issues.

Data set comparison.png

In developing this project, we wanted to create a set of tools that we could use to work with the power grid data. One such tool that we thought would be useful would be the ability to compare two power grids. This could happen when testing theoretical data, such as the result of seeing what would happen if a change in a physical power grid were made, or comparing data at different time stamps on the same grid.

Future Work[edit | edit source]

The next steps for the project will involve validating the ATP and RSCAD models by using the OpenDSS 34-Bus model’s data as well as running the RSCAD model with the RTDS to match the results of the ATP model. Then, communication between the RSCAD model and an automation controller needs to be established with real-time communication between the RSCAD model and an auctioning system for data analyzing purposes. This way, real-time data can be obtained by the use of the RTDS.

One of the next steps in the program is coordinate-based graphing. If coordinates of nodes could be specified, then it would be easy to switch between different visualizations (such as switching between graph theory, real-world coordinates, and an electrical engineering format). This would make communication between teams composed of people with different backgrounds to all be able to understand the information in the format with which they are most familiar.

A useful change that was beyond the scope of this class would be to implement images that can be scrolled into dynamically. This would drastically reduce the number of issues our graph visualization has with readability without compromising on what is visible in the graph as a whole.

The current comparison of the two node sets in the comparison function tests simple equality (if a bus exists in both sets and if its variables such as power and voltage are the same). A feature that would be relatively simple to implement would be to test whether values associated with buses are within a certain distance of each other (have similar powers or voltages). Because we did not know the specifics of real-world numbers, or what would be most useful to compare, we did not implement this behavior. However, it would be straightforward for this to be developed in the future.

Please refer to the project paper for further ideas on the project's direction in following semesters.

Conclusion[edit | edit source]

This semester, we successfully:

  • Completed the IEEE 34-Bus model in both RSCAD and ATPdraw with PV implementation
  • Developed toolset to parse, organize, and analyze grid data in an extensible structure
  • Helped Avista explore the physical and economic feasibility of participating in a transactive energy network for their business mode

Validation[edit | edit source]

Transactive Validation Plan 1.jpg
Transactive Validation Plan 2.jpg

While this plan was drawn up to meet project requirements, the team opted to use the EE standard for project validation. While the above are example tests, the project was actually implemented under a far greater number of small tests, in which a successful compilation was required before moving to the following one. The GOOSE protocol implementation was not completed due to the shutdown of campus due to COVID-19.

Schedule[edit | edit source]

Fall Semester:[edit | edit source]

October

  • 10/08 - Project Schedule (All)
  • 10/15 - Snapshot Day #1 (All)
  • 10/16 - Logbook and Portfolio Review (All) - Hussain will be gone at conference
  • 10/24 - Design Validation Plan (All)
  • 10/25 - Literature Review (All)
  • 10/31 - Project Value Proposition (All)
  • 10/31 - Code Setup and Graphs with consumers (CS)
  • 10/31 - ATP model draft (EE)
  • 10/31 - RSCAD model draft (EE)

November

  • 11/07 - Draft Wikipage (All) (1-2 weeks)
  • 11/14 - Logbook and Portfolio Review (All)
  • 11/22 - Concept Design Review (All)
  • 11/22 - Team Member Citizenship (All)
  • 11/22 - Test and compare ATP/RSCAD models for validation (EE)
  • 11/24 - Add Producers/Sinks into graph (CS)

December

  • 12/06 - Snapshot Day #2 (All)
  • 12/13 - Portfolio, Wikipage, Logbook Review (All)

Spring Semester:[edit | edit source]

Spring Semester: January

  • 01/24 - ATP controls draft (EE)
  • 01/24 - RSCAD controls draft (EE)
  • 01/24 - Have working min cut/max flow solution allowing for line removal (CS)
  • 01/28 - Design Expo Registration (All)
  • 01/28 - Test design--many transactions, conflicting transactions, stress testing (CS)

February

  • 02/10 - File Reading (CS)
  • 02/21 - Engineering Release Review (All)
  • 02/21 - Logbook Review (All)
  • 02/28 - ATP testing (EE)

March

  • 03/09 - Algorithm Implemented (CS)
  • 03/10 - Snapshot Day #3 (All)
  • 03/10 - Portfolio, Wikipage, Team Member Citizenship (All)
  • 03/10 - PV Implementation in RSCAD & ATP (EE)
  • 03/30 - CAL ISO data integrated (CS)
  • 03/27 - RSCAD auctioning system testing (All)

April

  • 04/06 - Graphical Output (CS)
  • 04/17 - Logbook Review (All)
  • 04/17 - Real-time data collection (All)
  • 04/24 - Demonstration testing (All)
  • 04/27-30 - EXPO preparation (All)

May

  • 05/01 - EXPO
  • 05/08 - Final Course Deliverables Due (Final hardware/software, Design report/project portfolio, Wikipage, Project Poster, Work Area Check-Out, Electronic Archive/File Management)

Team Members[edit | edit source]

Karstin Neill.jpg

Name: Karstin Neill
Major: Computer Science
Hometown: Lewistown, MT
Responsibilities: Secondary Team Documentation Lead
Email: neil1648@vandals.uidaho.edu


Christine Page.png

Name: Christine Page
Major: Electrical Engineering
Hometown: Lake Stevens, WA
Responsibilities: Team Lead
Email: page3114@vandals.uidaho.edu



James Todd.jpg

Name: James Todd
Major: Computer Science/Mathematics
Hometown: Moscow, ID
Responsibilities: Wiki Master, Secondary Team Budget Lead
Email: todd1391@vandals.uidaho.edu



Nikki Tran.jpg

Name: Nikki Tran
Major: Electrical Engineering
Hometown: Orofino, ID
Responsibilities: Team Budget Lead, Team Documentation Lead, Secondary Team Lead
Email: tran6924@vandals.uidaho.edu



Additional Documentation[edit | edit source]

Project Schedule
Meeting Minutes
Snapshot 1
Snapshot 2
Design Review
Meeting Agendas
Budget
Value Proposition
Product Requirements
Project Proposal
Design Validation Plan