1.

Introduction

The purpose of this technical report is to document the process and

results of integrating offshore wind generation into an onshore power
system. This work uses PowerFactory software and Python integration,
along with the application of Genetic Algorithms (GA) to optimize
system expansion and wind power penetration. The IEEE 39-Bus
system was used as the initial base model for simulations.

2. Initial System Testing with PowerFactory

The first phase involved validating the existing 39-Bus System data
used for initial testing, particularly focusing on confirming the system
topology (bus and line locations) based on a provided map. This
ensured the correct representation of the system in PowerFactory,
which was essential before proceeding with power flow and optimal
power flow analyses. The single line diagram for the 39-Bus System is
presented in Figure 1:

Figure 1 - 39-Bus System on PowerFactory.

Once the system data was verified, the next step was to run Power
Flow (PF) and Optimal Power Flow (OPF) analyses. The goal of these
tests was to understand how the system behaved under normal
operating conditions and under optimized scenarios, e.g., where the
system’s operational costs could be minimized.
The PF analysis was conducted to ensure that the system could
balance generation and load while maintaining voltage stability
across the network. This was a baseline check to ensure no significant
issues such as voltage collapse or line overloads under normal
conditions.

The OPF analysis aimed to optimize the operation of the system while
meeting demand. However, the initial results showed an unexpected
behavior where the excess generation was being directed entirely to
the reference bus (slack bus), meaning the other generators in the
system were not being properly dispatched. The system was failing to
adjust generation outputs optimally based on the system’s demand
and cost structure. The generation results were fixed at the set values
for PF base case analysis.

To address this issue, generator control settings and operational

constraints were activated manually, ensuring more realistic system
behavior during OPF simulations. This allowed the OPF algorithm to
adjust the generation output of each generator dynamically, rather
than defaulting on all excess generation to the reference bus.

After activating the necessary constraints and controls, the OPF was
able to properly dispatch generation across the network. Each
generator contributed based on its assigned cost, while ensuring that
voltage profiles remained within acceptable ranges and lines were not
overloaded.

The revised OPF was evaluated under various objective functions and
load scenarios to evaluate the system's responses. The possible
objective functions of PowerFactory are minimization of losses, costs,
load shedding or control variable deviations or maximization of
reactive power reserve. The main goal of this project is to minimize
the operational costs of the system, once the offshore wind parks are
connected to it, so the selected object function is the minimization of
costs.

As the primary objective was to minimize the total cost of generation

while satisfying demand, each generator was assigned a random cost,
and the OPF worked to minimize the total operational cost by
prioritizing cheaper units.

The system behaved as expected once the generator controls and

constraints were activated. The generators responded to variations in
load and system conditions, adjusting their outputs based on the cost
functions and the operational constraints.
In scenarios where load was reduced, the OPF reduced generation
accordingly, prioritizing lower-cost generators. However, in some
cases, the OPF struggled to find a feasible solution when there were
tight voltage limits. This was resolved by slightly relaxing the voltage
limits or simply disregarding it at all, which allowed the OPF to find a
solution.

Key Findings:

 Generator Controls: Activating generator controls and constraints

is essential for the OPF to function properly and dispatch
generators based on their cost and capacity.
 Voltage Limits: Tight voltage constraints can impact the feasibility
of OPF solutions, particularly under high wind generation or load
variation scenarios. Flexibility in voltage limits is sometimes
necessary to obtain a feasible solution.

3. Python-PowerFactory Integration

The integration of Python with PowerFactory was a critical component

in automating the control, modification, and analysis of the power
system model. This integration aimed to facilitate automated
processes, allowing for flexibility in adjusting system parameters and
running simulations such as Power Flow and Optimal Power Flow
through Python scripts. This integration also enabled efficient data
handling by exporting results to Excel and other formats for further
analysis.

The first stage of Python-PowerFactory integration involved setting up

basic communication between Python and PowerFactory. The
objective was to enable Python scripts to access system parameters
(e.g., line and generator data) and modify them (e.g., change line
impedances or adjust generator outputs). The script was also able to
run PF and OPF simulations and retrieve the results for additional
analysis.

The first challenge was identifying the correct attribute names within
PowerFactory’s object model to ensure that Python could access and
modify the desired parameters. PowerFactory uses specific naming
conventions for attributes such as line impedance, length, and
loading. These had to be correctly mapped to Python variables.
The integration was successful in running PF and OPF with adjustable
parameters, essential for future work on system optimization. One of
the significant advantages of Python-PowerFactory integration was
the ability to automate repeated OPF runs with different
configurations, which will be essential for optimizing the system as
wind power plants are integrated.

The ability to export results into Excel streamlined the post-simulation

analysis process, enabling further processing, visualization, and
reporting.

Main Achievements:

 Full Automation of PF and OPF: Python successfully automated the

entire PF and OPF processes, from modifying parameters to
running simulations and exporting results.
 Dynamic System Adjustments: The integration allowed for on-the-
fly changes to the system setup, making it easy to evaluate
different wind integration scenarios and system expansions.
 Cost and System Optimization: The automated OPF process
enabled the analysis of generation costs and system behavior
under various configurations, laying the groundwork for further
optimization studies.

4. Wind Generation Integration

The integration of offshore wind generation into the existing onshore

power system was one of the primary goals of the project. This
involved connecting to the 39-Bus System new wind generation
sources through an offshore transmission system (represented by
orange lines) and Points of Common Coupling (PCCs) (represented by
the red circles), as shown in Figure 2:
 Figure 2 - 39-Bus System and the connection of wind power plants.

The aim was to study the effects of adding wind power, optimize its
dispatch, and ensure system stability, all while minimizing operational
costs. Initially, the study is focused on analyzing the placement of the
PCCs. Two PCCs were chosen to function as the main connection
points for the offshore wind generation. These points were selected
based on their proximity to the offshore wind farms and the onshore
system buses located on the coast. The two PCCs were nominated
buses 40 and 41, and were interconnected by a single line to enhance
flexibility and allow for better distribution of power generated by the
wind farms. Wind generation was modeled by adding two new
generators (one at each PCC) with an initial capacity of 500 MW each.
These wind generators were assumed to have a maximum capacity
for injecting power into the grid.

The wind generation was assumed to penetrate the system at zero

cost, meaning the system would prioritize dispatching wind energy
over more expensive conventional generation units. Other generators
in the system were assigned costs based on their distance from the
PCCs, where generators closest to the wind generators were assigned
higher costs. This hierarchical cost structure allowed the system to
operate efficiently, prioritizing clean wind energy and minimizing
overall generation costs.

The transmission lines between the PCCs and the existing busbars
were modeled as candidate lines, meaning they were considered as
possible but not guaranteed additions to the system. The Genetic
Algorithm later will evaluate whether these lines should be built and,
if so, which ones should be prioritized based on cost and system
performance.

Once the wind generation and transmission lines were integrated, a

series of tests were conducted to evaluate the system’s behavior
under different wind penetration scenarios. The tests focused on
ensuring that the system could manage the additional power flow
from the wind generators without violating operational limits such as
line capacity or voltage constraints.

OPF was used to determine how the wind generation would be

dispatched under optimal conditions, while minimizing system costs.
Since wind generation had zero costs, the OPF ensured that wind
power was always dispatched first, followed by conventional
generators.

In some cases, the system struggled to maintain voltage within

acceptable limits, especially when wind generation was high. This was
particularly problematic in areas close to the PCCs, where large
amounts of power were being injected into the grid.

A detailed classification of lines was developed to differentiate

between original onshore system lines and wind power connection
lines. To separate these lines visually and functionally in the system, a
new grid was created using PowerFactory’s tool that allowed copying
the bus image. This ensured that the same buses could be
represented on different grids without altering the system’s core data.
This setup helped detailed analysis of how wind generation impacted
specific sections of the network. The new grid was called Wind Power
and is represented in Figure 3:
 Figure 3 - Wind Power grid in PowerFactory.

5. Optimization using Genetic Algorithm (GA)

The application of a Genetic Algorithm was a key approach for

optimizing the integration of offshore wind generation into the
onshore system. The GA was used to determine the optimal
configuration of new transmission lines, minimizing construction costs
while maximizing the efficiency and reliability of power flow within the
network. The primary goal was to evaluate and select the most
beneficial candidate lines to connect the wind farms to the system
and to manage increased wind generation, using both cost and
operational criteria.

The DEAP (Distributed Evolutionary Algorithms in Python) library was

chosen to implement the GA due to its flexibility, ease of use, and
compatibility with multiprocessing tools. DEAP is well-suited for
evolutionary algorithms, including Genetic Algorithms, Genetic
Programming, and Evolution Strategies. It supports rapid prototyping
and allows detailed control over algorithmic behavior, making it ideal
for this type of optimization problem.

DEAP’s framework allows easy adjustment of evolutionary

parameters, such as crossover and mutation rates, selection
methods, and objective functions. It also supports parallel
computation, which is important for handling complex optimization
problems like this, though challenges with software licenses limited
the use of parallelization (as discussed later).
The GA worked with binary encoding, where each gene represented
the decision to build or not build a candidate transmission line. For a
system with ten candidate lines, an individual might look like this [1,
0, 1, 0, 0, 1, 1, 0, 0, 1]. This indicates that five of the ten candidate
lines are selected to be built in this solution. Initially only the
candidate lines on the Wind Power grid were considered, so the
individual was built with six genes. Other key components of the GA
will be discussed in detail next.

Selection, crossover, and mutation methods

To generate new solutions (offspring), the GA uses crossover and

mutation operations. These processes introduce diversity into the
population and help the algorithm explore a wider solution space.

The crossover method chosen was a two-point crossover (cxTwoPoint),

in which two parent chromosomes are selected, and two crossover
points are chosen. The genes between these two points are swapped
between the parents to create two new offspring.

The mutation method used was bit flipping (mutFlipBit), which flips a
gene from ‘0’ to ‘1’ or from ‘1’ to ‘0’ with a certain probability. This
introduces new genes into the population and prevents the GA from
becoming stuck in local optima.

The selection method used was the tournament (selTournament). A

group of individuals is randomly selected, and the best individual from
this group is chosen to reproduce. This encourages competition and
ensures that stronger individuals have a higher chance of passing on
their genes.

Population and Generation Management

The size of the population (the number of individual solutions) and the
number of generations (iterations of the GA) were crucial parameters.
These were adjusted based on the complexity of the problem,
particularly the number of candidate transmission lines (genes in
each chromosome).

Objective Function

The GA required an objective function (also known as a fitness

function) to evaluate the quality of each solution. The objective
function was initially considered multi-objective, evaluating both the
cost of building transmission lines and the operational performance of
the system. However, some trade-offs and adjustments were made
during the testing process.
Initial Multi-Objective Function: Line Construction Costs and
Power Flow

Initially, a multi-objective function was implemented to

simultaneously consider both line construction costs and power flow
performance – the GA evaluated how well the selected lines managed
the power flow, ensuring that lines operated close to their optimal
loading without overloading:

Objective funciton ( OF )=line cost ( LC ) + power flow perfomance( PFP)

The first part of the objective function focused on minimizing the

construction costs of the selected transmission lines. The formula for
line costs was:

LC=(V line /100)∗length

Where:

 V line represents the line's voltage rating.

 length represents the physical length of the line (in kilometers).

This objective function aimed to minimize the total cost of the built
transmission lines, while ensuring that enough capacity was added to
accommodate the wind generation.

The second part of the objective function focuses on analyzing the

power flow performance. It is formulated as:

1
NB
| i |
V i−V nom 1
NL max
fj
PFP= ∑ + ∑
NL j=1 |f j−f max
j |
NB i=1 V nomi

Where:

 NB is the number of the system’s buses.

 NL is the number of system’s lines.
 V i is the actual value of voltage level on the i bus.
nom
 V i is the nominal value of voltage level on the i bus.
 f j is the active power flow on the j line.
max
 fj is the maximum active power flow on the j line.

However, during testing, it was observed that the power flow portion
of the objective function tended to dominate, generating
disproportionately high values compared to the cost portion. This
made it difficult to balance the two objectives, as the GA heavily
favored solutions with good power flow performance, even if they
were more expensive.
After calibration tests, the decision was made to simplify the objective
function by focusing primarily on construction costs and removing the
power flow portion from the objective function. This allowed the GA to
prioritize minimizing costs while ensuring that the system could still
manage the required wind generation.

Penalty functions to eliminate unfeasible solutions

In addition to the objective function, penalty functions were used to

discard infeasible solutions. Penalties were applied when certain
conditions were not met, ensuring that only valid configurations
progressed in the GA. Key penalties included:

 No candidate line connected: If no candidate lines were built, the

solution was discarded.
 No connection to the onshore system: If the PCCs were not
properly connected to the onshore grid, the solution was penalized.
 No feasible OPF solution: If the OPF could not find a feasible
solution (due to voltage limits, overloaded lines, etc.), the
individual was penalized or discarded.
 Wind park dispatch not feasible: If the OPF failed to fully dispatch
the capacity of the wind farms, the solution was penalized.

Parallelization Attempts

To speed up the GA process, parallelization was attempted using

DEAP’s parallel processing capabilities, particularly in the evaluation
of the OPF solutions. The idea was to run multiple OPF simulations
simultaneously across different CPU cores.

Despite DEAP's ability to parallelize the process, issues arose due to

the licensing restrictions of PowerFactory. Only one instance of
PowerFactory could be run on the same computer, preventing the use
of multiple licenses for parallel OPF evaluations. This limited the
ability to fully exploit the parallelization benefits.

Results from Initial GA Runs

The use of a Genetic Algorithm to optimize the integration of wind

generation showed significant potential. The GA efficiently selected
transmission line configurations that minimized construction costs
while ensuring the system could manage the additional wind
generation.

6. System Expansion

The system expansion phase focused on accommodating the

increased penetration of offshore wind generation into the existing
power grid. As more wind power was introduced, the grid's capacity
needed to be expanded to avoid overloading existing lines and to
ensure that the power from wind farms could be transmitted
efficiently. This required the addition of new transmission lines to the
network, strategically placed to manage the increased load while
keeping costs manageable.

The system expansion strategy involved identifying and creating new

candidate lines in the original 39-Bus System. These candidate lines
were chosen based on their potential to relieve congested lines and to
better integrate the power coming from offshore wind generation. Key
considerations in identifying candidate lines included proximity to the
wind generation and existing line usage.

The new candidate lines were placed close to the points where the
wind generation was injected into the grid (i.e., near the PCCs). This
helped distribute the wind power more evenly across the system and
reduced the load on the original lines. Also, lines that were already
operating at 90% or more of their capacity were prioritized for
expansion. Additional lines were created near these high-usage lines
to help offload some of the power and prevent overloading.

To facilitate the system expansion, new transmission lines were

introduced into the PowerFactory model. A new grid model (referred
to as the “expand” grid) was created to represent the expansion lines
and is shown in Figure 4. This grid was designed to separate the
original system’s lines from the newly created candidate lines. The
new grid also allowed for visual separation, making it easier to
identify which lines were part of the original network and which lines
were proposed for expansion.
 Figure 4 - Expand grid on PowerFactory.

A new classification area was introduced to differentiate between the

original system's lines and the candidate lines for both wind
integration and system expansion. This allowed for more precise
analysis of how the new lines interacted with the existing grid.

One of the primary challenges in system expansion was ensuring that

the new lines were economically justified. To achieve this, the
expansion lines were subject to a penalty structure within the
optimization algorithm, ensuring that they were only built if
necessary.

The construction costs of expansion lines were intentionally set higher

than those of the wind connection lines. This created a cost penalty
that discouraged the use of expansion lines unless the wind
connection lines alone could not manage the load.

Once the expansion lines were introduced, the system was subjected
to stress tests by gradually increasing the amount of wind generation
to assess how the system managed higher levels of renewable
penetration.

The lengths and capacities of the candidate lines were diversified to

reflect different construction costs. This created a more granular cost
structure, allowing the GA to balance the use of wind connection lines
and system expansion lines more effectively.

While the new expansion lines provided additional capacity, the

primary goal remained to maximize the use of wind connection lines
before resorting to system expansion. To achieve this, a stepwise wind
connection strategy was adopted.

Wind connection lines with different capacities were introduced in

steps, ensuring that smaller lines were used first, followed by larger
lines only if necessary. These lines had reduced construction costs
and were intended to create levels of wind connection that would be
used before any system expansion lines. The new Wind Power grid
topology is shown in

.
 Figure 5 - Modified Wind Power grid on PowerFactory.

7. Results

The following results were obtained applying the Genetic Algorithm to

the 39-Bus System, considering the modified Wind Power grid and the
Expand grid. Since each connection and expansion line corresponds
to a gene on the individual, this problem has seventeen genes to
decide. This is a big problem with over 100,000 combinations of
genes, so to solve it the following GA parameters were set:

 Population size: 160

 Number of generations: 100
 Crossover probability: 5%
 Mutation probability: 20%
 Elitism is considered

The maximum power generated in each offshore wind power plant is

500 MW.

The construction costs for each line are shown in Table 1.

Line Grid Length (km) Cost ($)

Line 16 - 41 Wind Power 50 172.5
Line 16 - 41 (1) Wind Power 50 172.5
Line 24 - 41 Wind Power 40 138
Line 24 - 41 (1) Wind Power 40 138
Line 26 - 40 Wind Power 80 276
Line 26 - 40 (1) Wind Power 80 276
Line 27 - 41 Wind Power 60 207
Line 27 - 41 (1) Wind Power 60 207
Line 28 - 40 Wind Power 70 241.5
Line 28 - 40 (1) Wind Power 70 241.5
Line 40 - 41 Wind Power 90 310.5
Line 40 - 41 (1) Wind Power 90 310.5
Line 2-3 Expand 110 379.5
Line 26 - 24 Expand 138.8625 479.0756
Line 26 - 27 Expand 100 345
Line 28 - 24 Expand 188.0595 648.8053
Line 8-9 Expand 150 517.5

The value of the best fitness function for each generation is presented
in the graph in Figure 6. The final cost of investment for the
constructed lines is $759.

Figure 6 - Fitness values over generations.

Figure 7 shows the original onshore system results.

 Figure 7 - Results for the 39-Bus System.

Figure 8 shows the Wind Power grid, and is possible to note that the
lines selected by the GA are:

 Line 28 - 40
 Line 28 - 40 (1)
 Line 24 - 41
 Line 24 – 41 (1)
 Figure 8 - Results for connection lines.

For this case it was not necessary to build any expansion lines.

8. Future Work

After the initial phase of wind integration testing, the next steps were
identified for further analysis and optimization:

 Optimization of PCC Locations: The next phase would involve

treating the location of the PCCs as an optimization variable. This
would allow further refinement of where wind generation is
injected into the system for optimal performance.
 Transmission Technology Evaluation: Both HVAC (High Voltage
Alternating Current) and HVDC (High Voltage Direct Current)
technologies were considered for the connection lines between
offshore wind farms and the PCCs. Future analysis will evaluate
which technology provided better performance in terms of cost,
efficiency, and stability.
 Increased Wind Generation Penetration: Further tests would be
conducted by increasing the capacity of the wind generators to
evaluate how the system behaved under higher wind generation
levels. This would stress the system further and provide insights
into how much wind power could realistically be integrated.