Skip to content

Topology Optimizer#

This package implements a topology optimizer for electrical transmission grids using a two-stage approach. It employs multi-objective optimization to determine substation reconfigurations that reduce line overloads and improve grid stability through reconfiguration of branches in a substation + opening a busbar coupler and branch disconnections.

Overview#

The topology optimizer consists of two main optimization stages working in a coordinated pipeline:

1. DC Stage (High-Performance Screening)#

The DC optimizer performs rapid exploration of the topology space using an accelerated DC loadflow solver with GPU support. It uses a discrete Map-Elites genetic algorithm to maintain a diverse repertoire of solutions across multiple objectives:

  • Fitness metrics: Line overload energy, switching distance, number of split substations
  • Search algorithm: Quality-diversity optimization maintaining solutions across descriptor dimensions
  • Performance: GPU-accelerated computation using JAX for massive parallel evaluation
  • Output: Pareto-optimal topologies sent to the AC stage for validation

Entry point: initialize_optimization

2. AC Stage (Validation and Refinement)#

The AC optimizer validates promising DC solutions using full AC loadflow calculations and applies further evolutionary refinements:

  • Validation: Full AC power flow analysis with N-1 contingency checking. Use optimization_loop
  • Evolution operators: Pull (from DC), reconnection, coupler closing: Use evolution_try
  • Early stopping: Rejection based on worst-case overload comparison
  • Selection strategy: Filtering method using median, dominator, and discriminator filters

Key Data Structures#

Topology Representation#

  • Actions: List of substation switching indices from the ActionSet
  • Disconnections: Branch outage specifications for N-1 analysis
  • PST Setpoints: Phase-shifting transformer positions
  • Metrics: Multi-objective fitness values and constraint violations

Strategy Collections#

  • DC Repertoire: Map-Elites grid maintaining diversity across descriptor dimensions
  • AC Database: SQLite storage for validated topologies with loadflow references
  • Message Protocols: Standardized formats for inter-optimizer communication

Prerequisites#

The optimization process requires preprocessed grid data from the Importer package:

  1. Static Information: Grid electrical parameters and topology
  2. Action Set: Enumerated switching possibilities
  3. N-1 Definition: Contingency analysis specifications

Running an Optimization#

Method 1: Kafka-based Distributed Execution#

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
# Start Kafka infrastructure
cd dev-deployment
docker-compose up

# Launch DC optimizer worker
cd packages/topology_optimizer_pkg
python -m topology_optimizer.dc.worker.worker \
    --processed_gridfile_folder=/path/to/preprocessed/data

# Launch AC optimizer worker  
python -m topology_optimizer.ac.worker \
    --processed_gridfile_folder=/path/to/preprocessed/data \
    --loadflow_result_folder=/path/to/results

Method 2: Direct Python Integration#

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
from topology_optimizer.dc.worker.optimizer import initialize_optimization, run_epoch
from topology_optimizer.interfaces.messages.dc_params import DCOptimizerParameters

# Configure optimization parameters
params = DCOptimizerParameters(
    ga_config=BatchedMEParameters(runtime_seconds=300),
    loadflow_solver_config=LoadflowSolverParameters(max_num_splits=4)
)

# Initialize and run optimization
optimizer_data, stats, initial_strategy = initialize_optimization(
    params=params,
    optimization_id="my_optimization", 
    static_information_files=["grid_data.pkl"]
)

Advanced Topics#