octoanalytics 0.2.0

Quantitative analysis for power markets

octoanalytics is a Python package by Octopus Energy providing tools for quantitative analysis and risk calculation on energy data. It facilitates analyzing energy consumption time series, incorporating temperature data, forecasting consumption, retrieving market prices, and computing risk premiums.

---

Key Features

• Smoothed Temperature Retrieval: Fetches hourly smoothed temperature data for major French cities and computes a national average.
• Energy Consumption Forecasting: Gradient Boosting model based on time features and temperature.
• Interactive Plotting: Visualize forecasts vs actual consumption with MAPE annotation.
• Spot and Forward Price Data: Functions to query EPEX spot prices and EEX forward prices from Databricks.
• Risk Premium Calculation: Computes risk premiums from forward price curves and forecast errors.
• Data Preprocessing: Automatic feature engineering, imputation, and scaling.

---

Installation

To install octoanalytics, use:

pip install octoanalytics

Dependencies such as pandas, numpy, scikit-learn, holidays, plotly, tqdm, and tentaclio will be installed automatically.

---

Usage

Importing the package

from octoanalytics import eval_forecast, plot_forecast, calculate_mape, get_temp_smoothed_fr, get_spot_price_fr, get_forward_price_fr, get_pfc_fr, calculate_prem_risk_vol

Data format for forecasting

The input data should be a DataFrame with at least:

• A datetime column (default named 'datetime')
• A consumption column (default named 'consumption')

Example:

import pandas as pd

data = pd.DataFrame({
    'datetime': ['2025-01-01 00:00', '2025-01-01 01:00', '2025-01-01 02:00'],
    'consumption': [120.5, 115.3, 113.7]
})

data['datetime'] = pd.to_datetime(data['datetime'])

Forecasting energy consumption

Use the eval_forecast function to train and predict consumption:

forecast_df = eval_forecast(data)
print(forecast_df.head())

This returns the test set with a 'forecast' column containing predicted values.

Plotting forecasts

Visualize actual vs predicted consumption with:

plot_forecast(data)

Calculate MAPE (Mean Absolute Percentage Error)

mape_value = calculate_mape(data)
print(f"MAPE: {mape_value:.2f}%")

Retrieve smoothed temperature data for France

temp_df = get_temp_smoothed_fr('2025-01-01', '2025-01-31')
print(temp_df.head())

Retrieve spot prices for French electricity market

Requires a Databricks token:

spot_prices = get_spot_price_fr(token='your_token_here', start_date='2025-01-01', end_date='2025-01-31')
print(spot_prices.head())

Retrieve forward prices and Price Forward Curve (PFC)

forward_prices = get_forward_price_fr(token='your_token_here', cal_year=2026)
pfc = get_pfc_fr(token='your_token_here', cal_year=2026)

Calculate premium risk volatility

premium = calculate_prem_risk_vol(token='your_token_here', input_df=data, datetime_col='datetime', target_col='consumption', plot_chart=True, quantile=50)
print(f"Risk premium at 50th percentile: {premium}")

---

Function Descriptions

eval_forecast(df, datetime_col='datetime', target_col='consumption')

Trains a Gradient Boosting model using time features and smoothed temperature data to forecast energy consumption. Splits data into train/test sets and returns test data with forecasts.

---

plot_forecast(df, datetime_col='datetime', target_col='consumption')

Plots interactive time series comparing actual consumption with forecasts, showing MAPE on the plot.

---

calculate_mape(df, datetime_col='datetime', target_col='consumption')

Returns the MAPE between actual and predicted consumption using the forecasting model.

---

get_temp_smoothed_fr(start_date, end_date)

Fetches hourly smoothed temperatures averaged over multiple major French cities.

---

get_spot_price_fr(token, start_date, end_date)

Retrieves hourly spot prices for the French electricity market (EPEX) from Databricks.

---

get_forward_price_fr(token, cal_year)

Fetches annual forward prices for French electricity (EEX) for a specified calendar year.

---

get_pfc_fr(token, cal_year)

Retrieves and resamples hourly Price Forward Curve data for French electricity (EEX) for a specified calendar year.

---

calculate_prem_risk_vol(token, input_df, datetime_col, target_col, plot_chart=False, quantile=50)

Calculates a risk premium based on forecast errors and forward price curves. Returns the premium value for the requested quantile and optionally plots the distribution.

---

Author

• Jean Bertin
• Email: [e-mail]
• Status: In development (planning)

---

License

MIT License — see LICENSE file for details.

---

Contributions

Contributions are welcome! Please open issues or pull requests on GitHub for suggestions or bug reports.