Wind Farm Planning Assistant

An AI-powered multi-agent system with specialized agents that work together to analyze terrain, design optimal turbine layouts, and simulate wind farm performance using Amazon Bedrock and advanced wind modeling tools.

Table of Contents

• Features
• Agents
• Architecture
• Project Structure
• Deployment
• Additional Resources
• Acknowledgements

Features

• Intelligent Terrain Analysis: Automated geographic constraint analysis with buildable land identification
• Optimal Layout Design: AI-driven turbine placement considering spacing and terrain requirements
• Advanced Wake Modeling: PyWake-powered energy production simulations
• Executive Reporting: Automated report generation with visualizations and recommendations
• Multi-Agent Orchestration: Coordinated workflow management across specialized agents
• Deployment in Amazon Bedrock AgentCore: Deployment of agents in AgentCore Runtime and MCP server in AgentCore Gateway
• Front-end webapp: Chat interface with interactive maps and charts and thought tracing

Agents

• Planning Assistant: Orchestrates the complete workflow
• Terrain Agent: Analyzes geographic constraints and provides GeoJSON of buildable land
• Layout Agent: Places turbines while considering terrain and spacing requirements
• Simulation Agent: Performs wake modeling and energy production calculations
• Report Agent: Generates executive summaries and visualizations

Example Workflow

1. Terrain Analysis

   Analyze terrain at 35.067482, -101.395466 with turbine model "IEA_Reference_3.4MW_130. Use a setback from roads of 1.2 * turbine height."
   

2. Layout Design

   Design a 30MW wind farm at location lat:35.067482, lon:-101.395466 using IEA_Reference_3.4MW_130 turbines.
   

3. Energy Simulation

   Run wake simulation for the layout with wind conditions 
   at coordinates 35.067482, -101.395466
   

4. Report Generation

   Generate executive report with charts and recommendations
   

5. Full Workflow

   Run through the full workflow required to design a 30MW wind farm at location lat:35.067482, lon:-101.395466 using IEA_Reference_3.4MW_130 turbines and output a report with all findings and simulation results.
   

Architecture

Web Application

The web application is implemented with Python FastAPI and React frontend, deployed separately from the agents and tools.

Note: This is a sample frontend demonstrating agent interactions. It does not implement project management, authentication, or authorization features required for production use. We recommend to implement your own user experience with appropriate authentication mechanisms and integrations.

Key Technologies

• Amazon Bedrock: Claude Sonnet 4 for AI reasoning
• Amazon Bedrock AgentCore: Secure scalable runtime for agent deployment
• Amazon S3 Vectors: Knowledge base using object storage
• Strands Agents: Agent development framework
• MCP (Model Context Protocol): Tool integration
• PyWake: Advanced wake modeling and yield simulation
• Folium/GeoPandas: Geo-spatial visualization

Project Structure

root/
├── agents/                      # AI agent implementations
│   ├── prompts/                 # System prompts used by the agents
│   ├── tools/                   # Agent-specific tools
│   ├── terrain_agent.py
│   ├── layout_agent.py
│   ├── simulation_agent.py
│   └── reporting_agent.py
├── mcp_tools/                   # MCP server for tool integration
└── web_app/                     # Web application interface

Deployment

The project uses AWS SAM (Serverless Application Model) for infrastructure deployment.

Prerequisites

• AWS Account with appropriate permissions
• AWS CLI configured
• AWS SAM CLI installed
• Docker for building container images

Security Recommendations

⚠️ Important: This sample does not implement authentication, authorization, or LLM guardrails. Implement these according to your organization's requirements if you intend to develop further for production use.

Recommended security measures:

• Amazon Bedrock Guardrails: Protect against LLM-based threats
• Amazon Cognito: User management and authorization
• VPC Configuration: Network isolation for sensitive workloads
• IAM Least Privilege: Minimal required permissions
• API Rate Limiting: Prevent abuse and control costs

Environment Variables

Create a .env file with the following required variables:

• AWS_REGION: Target AWS region (this sample supports us-west-2 region)
• AWS_ACCOUNT_ID: Your AWS account ID
• AWS_ROLE_NAME: IAM role for GitHub Actions OIDC
• OVERPASS_API_URL: Overpass API instance Url - see docs at: OpenStreetMap Wiki
• NREL_API_BASE_URL: NREL API base URL - see docs at: NREL Developer Docs
• NREL_API_KEY: NREL API key - you can obtain one at: NREL Developer Portal
• NREL_API_EMAIL: NREL API email you used to register for the API key
• [OPTIONAL]KNOWLEDGE_BASE_ID: optional knowledge base ID to retrieve setbacks used by terrain agent; if no valid ID is given, the default setbacks given in the agent's system prompt are used.

Manual Deployment

Replace the environment variable values and execute the following commands:

# Set environment variables
export AWS_REGION=us-west-2
export AWS_ACCOUNT_ID=<your-account-id>
export OVERPASS_API_URL=<overpass-api-url>
export NREL_API_BASE_URL=<base-url>
export NREL_API_KEY=<your-api-key>
export NREL_API_EMAIL=<your-email>

# Login to ECR
aws ecr get-login-password --region $AWS_REGION | docker login --username AWS --password-stdin $AWS_ACCOUNT_ID.dkr.ecr.$AWS_REGION.amazonaws.com

# Create ECR repositories
aws ecr create-repository --repository-name wind-dev-agent --region $AWS_REGION || true
aws ecr create-repository --repository-name wind-tools-lambda-image --region $AWS_REGION || true
aws ecr create-repository --repository-name wind-farm-app --region $AWS_REGION || true

# Build and push container images
docker build -t $AWS_ACCOUNT_ID.dkr.ecr.$AWS_REGION.amazonaws.com/wind-dev-agent:latest ./agents --platform linux/arm64 --push
docker build -t $AWS_ACCOUNT_ID.dkr.ecr.$AWS_REGION.amazonaws.com/wind-tools-lambda-image:latest ./agents --platform linux/arm64 --provenance=false --push
docker build -t $AWS_ACCOUNT_ID.dkr.ecr.$AWS_REGION.amazonaws.com/wind-farm-app:latest ./web_app --platform linux/amd64 --push

# Deploy with SAM
sam deploy --parameter-overrides \
  ParameterKey=AgentImageUri,ParameterValue=<account-id>.dkr.ecr.<region>.amazonaws.com/wind-dev-agent:latest \
  ParameterKey=ToolImageUri,ParameterValue=<account-id>.dkr.ecr.<region>.amazonaws.com/wind-tools-lambda-image:latest \
  ParameterKey=AppImageUri,ParameterValue=<account-id>.dkr.ecr.<region>.amazonaws.com/wind-farm-app:latest \
  ParameterKey=OverpassApiUrl,ParameterValue=$OVERPASS_API_URL \
  ParameterKey=NrelApiBaseUri,ParameterValue=$NREL_API_BASE_URL \
  ParameterKey=NrelApiKey,ParameterValue=$NREL_API_KEY \
  ParameterKey=NrelApiEmail,ParameterValue=$NREL_API_EMAIL \
  --capabilities CAPABILITY_IAM

Cleanup

To remove all resources and avoid charges:

# Delete the CloudFormation stack
aws cloudformation delete-stack --stack-name wind-farm-planning

# Remove ECR images
aws ecr batch-delete-image --repository-name wind-dev-agent --image-ids imageTag=latest
aws ecr batch-delete-image --repository-name wind-tools-lambda-image --image-ids imageTag=latest
aws ecr batch-delete-image --repository-name wind-farm-app --image-ids imageTag=latest

Troubleshooting

Common Issues:

• ECR Login Failed: Ensure AWS CLI is configured with correct region
• Docker Build Failed: Check Docker daemon is running and platform matches target
• SAM Deploy Failed: Verify all environment variables are set correctly
• API Errors: Check NREL API key is valid and has sufficient quota

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

Additional Resources

• AWS Bedrock - AI model capabilities
• AWS Bedrock AgentCore - Agent deployment platform
• Strands SDK - Multi-agent framework
• PyWake - Wake modeling techniques

Data Sources:

• NREL Wind Toolkit - Wind resource data
• NREL Turbine Models - Turbine specifications
• Overpass API - OpenStreetMap features
• USGS - Geospatial data

Acknowledgments

Special thanks to NREL, OpenStreetMap contributors, and the open-source community for providing the data and tools that make this project possible.