app.py

# app.py

import pandas as pd
import matplotlib.pyplot as plt
from prophet import Prophet
import gradio as gr
import pytz

# Step 1: Download the dataset from Open Power System Data
df = pd.read_csv("https://data.open-power-system-data.org/time_series/2020-10-06/time_series_60min_singleindex.csv", parse_dates=["utc_timestamp"])

# Step 2: Convert UTC to Europe/Berlin timezone
df['utc_timestamp'] = pd.to_datetime(df['utc_timestamp']).dt.tz_convert('Europe/Berlin')

# Step 3: Prepare the data for Prophet
df_prophet = df[["utc_timestamp", "DE_load_actual_entsoe_transparency"]].copy()
df_prophet = df_prophet.rename(columns={"utc_timestamp": "ds", "DE_load_actual_entsoe_transparency": "y"})
df_prophet["ds"] = df_prophet["ds"].dt.tz_localize(None)  # Remove timezone
df_prophet = df_prophet.dropna()  # Drop missing values

# Optional: Filter data for one year (2020)
df_prophet = df_prophet[df_prophet['ds'].dt.year == 2020]

# Step 4: Train the Prophet model
model = Prophet()
model.fit(df_prophet)

# Step 5: Create future DataFrame and forecast
def forecast_plot(hours):
    future = model.make_future_dataframe(periods=hours, freq='H')
    forecast = model.predict(future)
    forecast['ds_local'] = pd.to_datetime(forecast['ds']).dt.tz_localize('UTC').dt.tz_convert('Europe/Berlin')

    # Calculate the highlights from the forecast
    peak_demand = forecast['yhat'].max()
    peak_time = forecast['ds_local'][forecast['yhat'].idxmax()]
    lower_bound = forecast['yhat_lower'].min()
    upper_bound = forecast['yhat_upper'].max()

    # Plot forecasted energy load
    plt.figure(figsize=(12,6))
    plt.plot(forecast['ds_local'], forecast['yhat'], label='Forecast', color='#FF6347', linewidth=2)  # Orange-ish color
    plt.fill_between(forecast['ds_local'], forecast['yhat_lower'], forecast['yhat_upper'], alpha=0.3, color='#FF6347')
    plt.xlabel("Local Time (Europe/Berlin)", fontsize=12, fontweight='bold')
    plt.ylabel("Energy Load (MW)", fontsize=12, fontweight='bold')
    plt.title("⚡ German Energy Load Forecast (Zoomed into Future)", fontsize=16, fontweight='bold')
    plt.legend()
    plt.grid(True, linestyle='--', alpha=0.6)
    plt.xticks(rotation=45)
    plt.tight_layout()
    plt.show()  # Show plot immediately

    # Prepare the explanation text with Markdown style formatting
    explanation = f"""
    ## 📊 **Forecast Highlights**
    - **Peak Energy Demand**: {peak_demand:.2f} MW at **{peak_time.strftime('%Y-%m-%d %H:%M:%S')}**.
    - **Confidence Interval**: The forecasted energy load is expected to be between 
      **{lower_bound:.2f} MW** and **{upper_bound:.2f} MW**.
    - The trend shows that the energy demand is expected to [increase/decrease], 
      with a potential peak at the specified time.
    \n\n
    ## 🔍 **Key Observations**:
    - The forecast suggests that **energy demand** will **reach its highest point** at **{peak_time.strftime('%H:%M')}**.
    - This peak could be a critical time for grid management to prevent overloading.
    - **Confidence interval** gives us a range, helping to understand the possible **fluctuations in demand**.
    """

    # Return both the plot and explanation
    return plt, explanation

# Step 6: Create Gradio interface
gr.Interface(
    fn=forecast_plot,
    inputs=gr.Slider(24, 168, step=24, label="Forecast Hours (1 to 7 days)"),
    outputs=[gr.Plot(), gr.Textbox()],
    title="⚡ Smart Energy Load Forecasting (Germany)",
    description="Predict the energy demand using Facebook Prophet. Select the number of hours to forecast (1-7 days).",
    theme="huggingface",  # You can add more themes if desired
).launch()