Skip to main content

​
Getting Started with Layer 2 Forecasting

Get energy predictions running in just a few minutes with Qubit’s AI-powered forecasting models. This guide walks you through installation, setup, and generating your first forecasts.

​
Installation

1

Install the Package

pip install qubit-energy-forecasting

# Or with specific ML backends
pip install qubit-energy-forecasting[all]  # All models
pip install qubit-energy-forecasting[deep]  # TensorFlow/Keras
pip install qubit-energy-forecasting[prophet]  # Facebook Prophet
2

Verify Installation

import qubit.forecasting
print(f"Qubit Forecasting v{qubit.forecasting.__version__}")

# Check available models
from qubit.forecasting import SolarForecaster, DemandForecaster
print("✅ Installation successful!")

​
Quick Examples

​
Solar Generation Forecast

Generate a 24-hour solar generation forecast: 
from qubit.forecasting import SolarForecaster
import pandas as pd
import numpy as np

# 1. Create sample historical data
dates = pd.date_range('2024-01-01', periods=720, freq='1H')
np.random.seed(42)

# Simulate solar generation pattern
hours = dates.hour
seasonal_factor = 0.8 + 0.4 * np.sin(2 * np.pi * dates.dayofyear / 365)
daily_pattern = np.where(
    (hours >= 6) & (hours <= 18),
    1000 * np.exp(-((hours - 12) ** 2) / 18),
    0
)

historical_data = pd.DataFrame({
    'irradiance': daily_pattern * seasonal_factor + np.random.normal(0, 50, len(dates)),
    'temperature': 25 + np.random.normal(0, 5, len(dates)),
    'cloud_cover': np.random.beta(2, 5, len(dates)),
    'generation': daily_pattern * seasonal_factor * 0.18 + np.random.normal(0, 10, len(dates))
}, index=dates)

historical_data = historical_data.clip(lower=0)

# 2. Initialize and train forecaster
forecaster = SolarForecaster(
    system_capacity_kw=1000,
    panel_type="polycrystalline",
    tilt_angle=35
)

# Prepare features and target
X = historical_data[['irradiance', 'temperature', 'cloud_cover']]
y = historical_data['generation']

# Train the model
forecaster.fit(X, y)
print("✅ Model trained successfully!")

# 3. Generate forecast
# Create future weather forecast (normally from weather API)
future_weather = pd.DataFrame({
    'irradiance': [800, 850, 900, 950, 1000, 950, 900, 800, 600, 300, 0, 0,
                   0, 0, 0, 0, 0, 0, 200, 600, 850, 950, 900, 800],
    'temperature': [20, 22, 25, 27, 30, 32, 30, 28, 25, 22, 18, 15,
                   12, 10, 10, 12, 15, 18, 20, 22, 25, 27, 25, 22],
    'cloud_cover': [0.2] * 24
}, index=pd.date_range(dates[-1] + pd.Timedelta(hours=1), periods=24, freq='1H'))

forecast = forecaster.predict(
    X.tail(24),  # Last 24 hours for context
    horizon="24h",
    weather_forecast=future_weather,
    confidence_levels=[0.80, 0.95],
    asset_id="solar_farm_001"
)

print(f"🌞 24-hour Solar Forecast Generated!")
print(f"Peak generation: {forecast.peak_value:.2f} kW at {forecast.peak_time.strftime('%H:%M')}")
print(f"Total energy: {forecast.total:.2f} kWh")
print(f"Confidence: 80% interval [{forecast.confidence_intervals[0.8][0].max():.0f}, {forecast.confidence_intervals[0.8][1].max():.0f}] kW")

​
Demand Forecast

Generate electricity demand predictions: 
from qubit.forecasting import DemandForecaster
import pandas as pd
import numpy as np

# 1. Create sample load data
dates = pd.date_range('2024-01-01', periods=720, freq='1H')
np.random.seed(42)

# Simulate demand patterns
hours = dates.hour
weekday_pattern = 50 + 30 * (
    ((hours >= 7) & (hours <= 9)) |  # Morning peak
    ((hours >= 17) & (hours <= 21))  # Evening peak
).astype(int)

weekend_pattern = 40 + 10 * np.sin(2 * np.pi * hours / 24)
is_weekend = dates.dayofweek >= 5

base_load = np.where(is_weekend, weekend_pattern, weekday_pattern)
temperature_effect = 2 * (25 - np.abs(dates.hour - 12)) # Temperature effect

historical_load = pd.DataFrame({
    'temperature': 20 + 10 * np.sin(2 * np.pi * dates.dayofyear / 365) + np.random.normal(0, 2, len(dates)),
    'load': base_load + temperature_effect + np.random.normal(0, 5, len(dates))
}, index=dates)

historical_load['load'] = historical_load['load'].clip(lower=0)

# 2. Initialize and train demand forecaster
demand_forecaster = DemandForecaster(
    customer_type="mixed",
    include_weather=True,
    include_calendar=True
)

# Prepare data
X_demand = historical_load[['temperature']]
y_demand = historical_load['load']

# Train model
demand_forecaster.fit(X_demand, y_demand)
print("✅ Demand model trained successfully!")

# 3. Generate demand forecast
future_temp = pd.DataFrame({
    'temperature': 22 + 3 * np.sin(2 * np.pi * np.arange(24) / 24) + np.random.normal(0, 1, 24)
}, index=pd.date_range(dates[-1] + pd.Timedelta(hours=1), periods=24, freq='1H'))

demand_forecast = demand_forecaster.predict(
    X_demand.tail(24),
    horizon="24h",
    temperature_forecast=future_temp,
    asset_id="load_mixed_001"
)

print(f"⚡ 24-hour Demand Forecast Generated!")
print(f"Peak demand: {demand_forecast.peak_value:.2f} kW at {demand_forecast.peak_time.strftime('%H:%M')}")
print(f"Average demand: {demand_forecast.point_forecast.mean():.2f} kW")

# Peak metrics
peak_metrics = demand_forecaster.calculate_peak_metrics(demand_forecast)
print(f"Load factor: {peak_metrics['load_factor']:.2%}")

​
Integration with Layer 1

Layer 2 is designed to work seamlessly with Layer 1 data: 
from qubit.connectors.mqtt import MQTTConnector
from qubit.forecasting import SolarForecaster
from qubit.adapters import AdapterPipeline

# 1. Set up Layer 1 data pipeline
connector = MQTTConnector({
    "broker": "mqtt://your-broker.com:1883",
    "topics": ["solar/+/telemetry"]
})

adapter_pipeline = AdapterPipeline([
    ("units", UnitConverter()),
    ("timezone", TimezoneAdapter()),
    ("validation", SchemaValidator("timeseries"))
])

# 2. Initialize forecaster
forecaster = SolarForecaster(system_capacity_kw=1000)

# 3. Process real-time data and forecast
@connector.on_message
async def process_and_forecast(mqtt_message):
    # Layer 1: Normalize data
    timeseries = adapter_pipeline.process(mqtt_message)
    
    # Layer 2: Generate forecast when enough data
    if len(historical_buffer) > 168:  # 1 week of data
        forecast = forecaster.predict(
            historical_buffer.tail(24),
            horizon="24h"
        )
        
        # Send forecast to Layer 3 or storage
        await send_to_layer3(forecast.to_timeseries())

await connector.start()

​
Configuration

​
Environment Variables

# Optional configuration
export QUBIT_FORECASTING_CACHE_DIR="/tmp/qubit_models"
export QUBIT_FORECASTING_LOG_LEVEL="INFO"
export QUBIT_FORECASTING_PARALLEL_JOBS="4"

​
Configuration File

forecasting:
  default_horizon: "24h"
  default_resolution: "1h"
  cache_models: true
  
models:
  solar:
    default_capacity: 1000  # kW
    include_weather: true
    confidence_levels: [0.8, 0.95]
    
  demand:
    customer_segments: ["residential", "commercial", "industrial"]
    include_calendar: true
    
performance:
  batch_size: 32
  n_jobs: -1
  memory_limit: "4GB"

​
Docker Deployment

Run forecasting models in containers: 
FROM python:3.9-slim

# Install forecasting package
RUN pip install qubit-energy-forecasting[all]

# Copy your forecasting script
COPY forecast_app.py /app/
WORKDIR /app

# Run the forecasting service
CMD ["python", "forecast_app.py"]

​
API Server

Deploy forecasting as a REST API: 
from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from qubit.forecasting import SolarForecaster, DemandForecaster
import pandas as pd

app = FastAPI(title="Qubit Energy Forecasting API")

# Initialize models
solar_forecaster = SolarForecaster(system_capacity_kw=1000)
demand_forecaster = DemandForecaster(customer_type="mixed")

class ForecastRequest(BaseModel):
    asset_id: str
    model_type: str  # "solar" or "demand"
    horizon: str = "24h"
    historical_data: dict
    weather_data: dict = None

@app.post("/forecast")
async def generate_forecast(request: ForecastRequest):
    try:
        if request.model_type == "solar":
            forecaster = solar_forecaster
        elif request.model_type == "demand":
            forecaster = demand_forecaster
        else:
            raise HTTPException(400, "Invalid model_type")
        
        # Convert data to DataFrame
        X = pd.DataFrame(request.historical_data)
        
        # Generate forecast
        forecast = forecaster.predict(
            X,
            horizon=request.horizon,
            asset_id=request.asset_id
        )
        
        return {
            "asset_id": forecast.asset_id,
            "forecast": forecast.point_forecast.tolist(),
            "timestamps": forecast.timestamps.isoformat().tolist(),
            "peak_value": forecast.peak_value,
            "total": forecast.total
        }
    
    except Exception as e:
        raise HTTPException(500, str(e))

if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=8080)

​
Troubleshooting

# Missing dependencies
pip install --upgrade qubit-energy-forecasting[all]

# Python version issues
python --version  # Requires 3.9+

# Virtual environment
python -m venv venv
source venv/bin/activate  # Linux/Mac
venv\Scripts\activate     # Windows
pip install qubit-energy-forecasting

# Reduce batch size
forecaster = XGBoostForecaster(
    batch_size=16,  # Reduce from default 32
    n_jobs=2        # Limit parallel jobs
)

# Clear model cache
forecaster.clear_cache()

# Use lighter models
forecaster = ARIMAForecaster()  # Instead of LSTM

# Enable model caching
forecaster.cache_model = True

# Pre-compute features
features = feature_engineer.extract(data)
forecaster.fit(features, target, cache_features=True)

# Use ensemble weights
ensemble.optimize_weights(X_train, y_train, method='fast')

# Validate input data
from qubit.forecasting.validation import validate_timeseries

errors = validate_timeseries(data)
if errors:
    print("Data validation errors:", errors)

# Handle missing values
data_cleaned = data.fillna(method='ffill').fillna(method='bfill')

# Remove outliers
from qubit.forecasting.preprocessing import remove_outliers
data_clean = remove_outliers(data, method='iqr', threshold=3)

​
Next Steps

Explore Models

Learn about different forecasting algorithms and when to use them

Production Deployment

Deploy forecasting models at scale with Kubernetes and monitoring

GitHub Examples

Browse complete example applications and use cases

Layer 3 Integration

Use forecasts for energy system optimization
You’re now ready to generate production-quality energy forecasts! Layer 2 provides the predictions that power intelligent energy optimization in Layer 3.