Skip to main content
📝The Challenge
Inaccurate ETAs are a primary source of operational friction in the maritime industry, leading to increased costs, scheduling conflicts, and inefficient port operations. This module addresses this by moving beyond simple calculations to a data-driven prediction model that accounts for real-world, dynamic variables.

​
Data Flow & Processing

The ETA prediction model relies on a robust, four-stage data pipeline to ensure accuracy and reliability, transforming raw inputs into actionable insights.

​
Core Algorithms & Calculations

The ETA prediction model is built on a comprehensive data pipeline that transforms raw, diverse inputs into a refined, reliable forecast. This process involves five key stages, from initial data collection to the final continuous calculation.

​
1. Data Ingestion: Weather & Route Polling

The process begins by polling external APIs for essential voyage context. The system is designed to periodically query Navtor for detailed routing information and Stormglass for real-time and forecasted weather conditions along the vessel’s planned route. Algorithm Logic:
  • Establish secure connections to Navtor and Stormglass APIs.
  • Request route data for a specific vessel and voyage.
  • Request weather parameters (wind speed, wave height, currents) for the specific geographical points along the route.
  • Store the raw, unstructured JSON responses for processing.

​
2. Data Extraction: Email & Noon Report Parsing

A significant portion of operational data arrives in unstructured formats, such as noon report emails. The system uses a sophisticated parsing engine to extract critical information from this text. Algorithm Logic:
  • Monitor a designated inbox for incoming noon report emails.
  • Use regular expressions (regex) and keyword matching to identify and isolate key data points (e.g., “SOG:”, “Remaining Dist:”, “ETA:”).
  • Extract values for vessel speed, remaining distance, fuel consumption, and the reported ETA.
  • Temporarily store this extracted, key-value data for the next stage.
import re

def parse_noon_report_email(email_body):
    # Use regex to find key-value pairs in the email text
    # Note: These are simplified patterns for illustration
    sog_pattern = re.compile(r"SOG:\s*([\d\.]+)\s*knots")
    eta_pattern = re.compile(r"ETA:\s*(\d{4}-\d{2}-\d{2}\s*\d{2}:\d{2})")
    
    sog_match = sog_pattern.search(email_body)
    eta_match = eta_pattern.search(email_body)
    
    extracted_data = {
        "SOG": sog_match.group(1) if sog_match else None,
        "ETA": eta_match.group(1) if eta_match else None,
        # ... other extracted fields
    }
    
    return extracted_data
Alternative Method: GPT-Based Parsing For more complex or less structured reports, a Large Language Model (LLM) is be used for more robust and flexible data extraction. 
import openai

def parse_with_gpt(email_body):
    # Prepare a prompt that instructs the model to extract key information
    prompt = f"""
    Extract the following entities from this noon report:
    - Speed Over Ground (SOG) in knots
    - Estimated Time of Arrival (ETA) in YYYY-MM-DD HH:MM format
    
    Report: "{email_body}"
    
    Return the result as a JSON object.
    """
    
    # Call the OpenAI API (or any other LLM provider)
    response = openai.Completion.create(
      engine="text-davinci-003",
      prompt=prompt,
      max_tokens=100
    )
    
    # The model's response will be a JSON string that can be parsed
    extracted_data_json = response.choices[0].text
    return json.loads(extracted_data_json)

​
3. Data Standardization & Enrichment

Once data is extracted, it must be converted into a standardized, structured format. This stage involves cleaning the data, converting units, and enriching it with information from other sources. Code Implementation: 
def standardize_report_data(extracted_data):
    # Convert speed from knots to a standard float
    standardized_speed = float(extracted_data.get("SOG"))
    
    # Standardize date and time formats to UTC
    reported_eta_str = extracted_data.get("ETA")
    standardized_eta = convert_to_utc(reported_eta_str)
    
    # Enrich with data from other sources
    vessel_dwt = get_vessel_particulars(vessel_id)
    
    # Create a clean, structured data object
    structured_report = {
        "speed_knots": standardized_speed,
        "eta_utc": standardized_eta,
        "vessel_dwt": vessel_dwt,
        # ... other fields
    }
    return structured_report

​
4. Data Validation: Time-Based Correction

A key source of ETA error stems from the timing of noon report submissions. Reports filed after midday can be incorrectly timestamped to the following day. The system applies a specific logical check to correct this. Algorithm Logic: The algorithm checks the timestamp of each incoming noon report. If the report’s time is after 12:00 PM (noon), but the associated date has been advanced to the next day, the algorithm corrects the date back to the actual day of submission. Code Implementation: 
def correct_noon_report_date(report):
    # Check if the report time is post-meridian (after 12:00 PM)
    is_after_noon = report.time > '12:00:00'
    
    # Check if the date has been incorrectly advanced
    date_is_advanced = report.date > actual_submission_date
    
    if is_after_noon and date_is_advanced:
        # Revert the date to the correct day
        report.date = actual_submission_date
        
    return report

​
6. Voyage Status & Geolocation Processing

After the primary ETA is calculated, the data is further enriched with voyage status and geolocation information to provide a complete operational picture on the dashboard. Geolocation Mapping: This function maps the vessel’s current location to a standardized geographical region for easier tracking and filtering. Code Implementation: 
def map_location(latitude, longitude):
    # This function would contain logic to map coordinates to defined regions
    # Example:
    if 30.0 < latitude < 60.0 and -30.0 < longitude < 0.0:
        return "North Atlantic"
    elif 25.0 < latitude < 45.0 and 35.0 < longitude < 65.0:
        return "Arabian Sea"
    else:
        return "Unknown Region"
Voyage Status Creation: The system generates a dynamic, human-readable status for each voyage based on its current operational data. Code Implementation: 
def create_voyage_status(voyage_data):
    # This function would create a status string based on vessel activity
    # Example:
    if voyage_data.get('speed_knots') > 1:
        status = f"En route to {voyage_data.get('destination_port')}"
    else:
        status = f"Alongside at {voyage_data.get('current_port')}"
    
    return status
Final Data Processing: This function orchestrates the final data processing steps, bringing together all the calculated and enriched data points into a final, dashboard-ready object. Code Implementation: 
def process_eta_data(voyage_id):
    # This function would be the main orchestrator for a single voyage
    # 1. Fetch the latest validated report data
    report_data = get_validated_report(voyage_id)
    
    # 2. Calculate the live ETA
    live_eta = calculate_live_eta(report_data)
    
    # 3. Map the location
    geo_location = map_location(report_data.get('lat'), report_data.get('lon'))
    
    # 4. Create a human-readable status
    voyage_status = create_voyage_status(report_data)
    
    # 5. Assemble the final data object for the dashboard
    dashboard_payload = {
        "voyage_id": voyage_id,
        "live_eta_utc": live_eta,
        "location_region": geo_location,
        "current_status": voyage_status,
        "last_updated": get_current_utc_time()
    }
    
    return dashboard_payload