Enterprise-style forecasting and operational intelligence platform for industrial energy analytics, predictive operations, and operational decision support.
This project simulates an industrial operation and builds a forecasting workflow to predict energy consumption, estimate operational cost, analyze demand behavior, and expose forecasting logic through a FastAPI service.
Industrial operations depend on accurate forecasting to control energy cost, plan capacity, reduce operational risk, and anticipate high-load periods.
This project demonstrates how time series forecasting and operational intelligence can answer practical business questions:
- How much energy will the operation consume?
- What will the expected operating cost be?
- Which operational factors drive energy demand?
- When do high-load periods occur?
- How can forecasting support planning and decision-making?
Industrial facilities often face fluctuating energy demand caused by production throughput changes, ambient temperature variation, maintenance events, operational anomalies, shift schedules, and daily/weekly seasonality.
Without forecasting, teams react after cost increases or overloads occur.
This platform provides a forecasting and analytics layer that helps operations teams move from reactive monitoring to predictive planning.
The platform includes:
- Synthetic industrial time series data generation
- Temporal exploratory data analysis
- Lag-based feature engineering
- Rolling window feature engineering
- Temporal train/test split
- Random Forest forecasting model
- Operational KPI analysis
- Feature importance analysis
- FastAPI forecasting endpoint
- Professional documentation and visual reporting
The model tracks future industrial energy consumption using temporal features and operational drivers.
The model identifies production throughput, ambient temperature, anomalies, and maintenance events as key operational drivers.
FastAPI provides an interactive enterprise-style API interface for forecasting requests.
Operational Data
|
v
Data Generation / Ingestion
|
v
Temporal EDA
|
v
Feature Engineering
- lag features
- rolling windows
- calendar features
- operational events
|
v
Temporal Train/Test Split
|
v
Forecasting Model
|
v
Evaluation + KPI Layer
|
v
FastAPI Forecasting Service
|
v
Operational Intelligence Outputs
operational-intelligence-forecasting/ ├── app/ │ ├── __init__.py │ └── main.py ├── data/ │ ├── raw/ │ └── processed/ ├── docs/ │ ├── architecture.md │ ├── business_context.md │ └── model_card.md ├── models/ ├── notebooks/ │ └── 01_temporal_eda.ipynb ├── reports/ │ ├── figures/ │ └── screenshots/ │ ├── actual_vs_predicted.png │ ├── feature_importance.png │ └── api_swagger.png ├── src/ │ ├── forecasting/ │ ├── visualization/ │ ├── generate_operational_data.py │ └── inspect_data.py ├── tests/ ├── Dockerfile ├── requirements.txt ├── README.md └── .gitignore
The project uses a simulated industrial operations dataset with hourly records.
| Variable | Description |
|---|---|
| timestamp | hourly timestamp |
| energy_consumption_kwh | target variable |
| production_throughput_units | production output |
| ambient_temperature_c | environmental temperature |
| maintenance_event | planned maintenance indicator |
| anomaly_event | abnormal operating condition |
| energy_price_usd_per_kwh | hourly energy price |
| estimated_energy_cost_usd | estimated energy cost |
The dataset was designed to reflect realistic operational behavior including daily cycles, weekly behavior, seasonal effects, maintenance periods, and anomaly events.
- hour
- day_of_week
- month
- is_weekend
- production_throughput_units
- ambient_temperature_c
- maintenance_event
- anomaly_event
| Feature | Meaning |
|---|---|
| lag_1 | energy consumption 1 hour ago |
| lag_24 | energy consumption at the same hour yesterday |
| lag_168 | energy consumption at the same hour one week ago |
| Feature | Meaning |
|---|---|
| rolling_mean_24 | average energy consumption over the previous 24 hours |
| rolling_std_24 | variability of energy consumption over the previous 24 hours |
The forecasting model uses a Random Forest Regressor trained with temporal features and operational variables.
The dataset is split using a chronological train/test split to avoid temporal leakage.
Past data -> training set Future data -> test set
No random shuffle is used because time series forecasting must preserve temporal order.
| Metric | Result |
|---|---|
| MAE | ~29 kWh |
| RMSE | ~36 kWh |
| R² | ~0.84 |
| Average Percentage Error | ~4.1% |
The model predicts future industrial energy consumption with an average error of approximately 4.1%, making it useful for operational planning, cost estimation, and decision support.
The platform estimates and analyzes:
- Average actual energy consumption
- Average predicted energy consumption
- Absolute forecast error
- Percentage forecast error
- Total actual energy cost
- Total predicted energy cost
- Cost forecast error
- Peak actual energy demand
- Peak predicted energy demand
| KPI | Result |
|---|---|
| Average Actual Consumption | ~718 kWh |
| Average Predicted Consumption | ~719 kWh |
| Forecast Error | ~4.1% |
| Cost Forecast Error | ~$291 USD |
| Peak Actual Demand | ~1235 kWh |
| Peak Predicted Demand | ~1145 kWh |
The project includes a FastAPI service for operational forecasting.
Run locally:
uvicorn app.main:app --reload
Open:
http://127.0.0.1:8000/docs
Available endpoints:
| Endpoint | Method | Description |
|---|---|---|
| / | GET | API root |
| /health | GET | health check |
| /forecast | POST | forecast energy consumption and cost |
Example request:
{
"production_throughput_units": 120,
"ambient_temperature_c": 31,
"maintenance_event": 0,
"anomaly_event": 0
}
Example response:
{
"predicted_energy_kwh": 801.0,
"predicted_cost_usd": 104.13,
"operational_status": "normal"
}
Build image:
docker build -t operational-intelligence-forecasting .
Run container:
docker run -p 8000:8000 operational-intelligence-forecasting
Then open:
http://127.0.0.1:8000/docs
- Python data science workflow
- Time series forecasting
- Temporal feature engineering
- Lag features
- Rolling window features
- Autocorrelation analysis
- Seasonality analysis
- Temporal leakage prevention
- Machine learning regression
- Model evaluation
- Operational KPI design
- FastAPI deployment
- Git/GitHub workflow
- Enterprise-style project documentation
This platform shows how forecasting can support industrial operations by helping teams:
- Anticipate energy demand
- Estimate operating costs
- Identify high-load periods
- Understand operational drivers
- Support maintenance planning
- Improve decision-making
- Reduce reactive operations
This project is designed as a portfolio-grade demonstration of applied data science for industrial operations.
It combines machine learning, time series forecasting, operational analytics, API deployment, and business storytelling in a single end-to-end project.
Relevant roles:
- Data Analyst
- Data Scientist Jr
- Machine Learning Engineer Jr
- Operations Analyst
- Supply Chain Analyst
- Industrial Analytics Analyst
- Energy Analytics Analyst
- Operational Intelligence Analyst
- Connect FastAPI endpoint to trained model artifact
- Add Streamlit dashboard
- Add Prophet baseline model
- Add XGBoost or LightGBM forecasting model
- Add scenario simulation
- Add alerting for high-load predictions
- Add cloud deployment
- Add automated model retraining pipeline
Portfolio project focused on Industrial Analytics, Forecasting, Operational Intelligence, and Predictive Operations.