MetaEnergy
MetaEnergy

MetaCast - Natural Gas Demand Forecasting

Machine-learning forecasts for gas demand, load planning, and procurement decisions

AI, Forecasting, Machine Learning, Gas, Optimization

Overview

MetaCast forecasts gas consumption across hourly, day-ahead, seasonal, and planning horizons. It combines historical consumption, weather data, calendar effects, and optional economic drivers to produce forecasts with confidence intervals.

Key Features

  • Multi-model Forecasting
  • Weather & Calendar Features
  • Confidence Intervals
  • API Delivery & Retraining

Challenge

Gas utilities need reliable demand forecasts to plan procurement, storage, nominations, and network operations. Manual forecasting is slow to update, difficult to validate, and vulnerable to weather-driven seasonality and unusual consumption patterns.

Solution

MetaEnergy developed an ML forecasting engine using ensemble methods such as gradient-boosted trees, recurrent models, and statistical baselines. The platform automates data preparation, feature engineering, model comparison, retraining, and API delivery to operational systems.

Impact & Results

  • Supports 10+ years of historical consumption data and forecasts for 100+ delivery points. In validated deployments, the approach can improve day-ahead and seasonal forecast quality while identifying measurable procurement and imbalance optimization opportunities.

Tech Stack

  • Python
  • TensorFlow
  • Scikit-learn
  • PostgreSQL
  • FastAPI
  • Docker

System Architecture

ML-powered demand forecasting pipeline - from weather data collection to ensemble predictions

  • metaenergy.ge · demand forecasting platform v2.0
  • Weather Data Collection
  • Data Processing Layer
  • Weather Data Collector
  • Scheduled polling service
  • Unified query interface
  • Norwegian Met Institute
  • Commercial forecast API
  • Historical · 10yr archive
  • Batch script · Scheduler
  • Orchestrator Service
  • Task routing · Concurrency
  • Mutex · Deadlock prevention
  • Station-level partitioning
  • Data Cleaning & Merging
  • Outlier removal · Imputation
  • Rolloff temp · Holidays · Lags
  • Per-group routing · Retry
  • MAPE-weighted · Meta-learners · Bayesian
  • BaseForecaster (ABC)
  • Unified interface · Fit / Predict
  • OLS · SARIMAX · Holt-Winters
  • XGBoost · RF · Prophet
  • LSTM · GRU · DeepAR · TFT · TCN
  • REST API · JSON payload
  • Network share · Versioned
  • Consumer endpoint · Caching

Forecasting, Analytics & Decision Support

Forecasts and operational analytics for planning, procurement, balancing, and network decisions

MetaEnergy builds forecasting and analytics workflows that turn historical consumption, weather signals, calendar effects, and operational data into repeatable planning outputs. The value is not only a model - it is a workflow that operators can monitor, compare, audit, and improve.

What we deliver

  • Demand forecasting
  • Weather and calendar feature modeling
  • Confidence intervals
  • KPI dashboards
  • Anomaly detection
  • Forecast APIs and scheduled retraining

Use this service when

  • Forecasting is still spreadsheet-based or hard to validate
  • Procurement, storage, nominations, or balancing depend on better demand estimates
  • You need day-ahead, seasonal, or long-range planning forecasts
  • You need model comparison, backtesting, or forecast auditability

Related posts