Gas Price Prediction Using LSTM

Description

This project predicts natural gas futures prices using an LSTM (Long Short-Term Memory) neural network.
The model is trained on historical gas price and trade volume data fetched from Tinkoff Invest API.
It forecasts future gas prices for the next 10 hours and visualizes the predictions alongside actual prices for the last 30 days.
Additionally, the app calculates daily trading profits based on recent operations.

Example:

---

Table of Contents

• Features
• Requirements
• Installation
• Usage
• Data
• Model
• Results
• Contributing

---

Features

• Fetches real-time gas price data from Tinkoff Invest API.
• Preprocesses data to handle missing values.
• Builds and trains an LSTM model for time series prediction.
• Predicts gas prices for the next 10 hours.
• Visualizes model predictions and forecasts:
  • 30-day historical data
  • 2-day test predictions
  • 10-hour forecast
• Calculates daily trading profit and percentage based on recent gas trades.

---

Requirements

• Python 3.8 - 3.11
• tinkoff-investments
• numpy
• pandas
• scikit-learn
• matplotlib
• tensorflow
• python-dotenv
• tkinter

---

Installation

1. Clone the repository:

git clone https://github.com/flild/AIMarketPrediction.git
cd AIMarketPrediction

2. Install the required libraries:

pip install -r requirements.txt

3. Create a .env file in the project directory and enter your Tinkoff API token and account ID:

TINKOF_API="your_token"
TIN_ACC_ID="your_account_id"

4. Find your account ID using the script:

python ScriptForAccountId.py

---

Usage

1. Run the application:

python app.py

2. App Features:
   • "Просчитать график" – fetches data, trains the model, predicts prices, and displays the 30-day chart with predictions.
   • "Посчитать сделки" – calculates today's profit from gas trades in rubles and percentage.

---

Data

• Source: Tinkoff Invest API
• Attributes:
  • Gas_Price: Closing price of natural gas futures.
  • Trade_Volume: Volume of trades.

---

Model

• Architecture: Sequential LSTM model with two LSTM layers and dropout regularization.
• Loss Function: Mean Squared Error (MSE).
• Optimizer: Adam optimizer.
• Training: Trained for 30 epochs with a batch size of 32.
• Prediction:
  • Predicts prices for the last 2 days (test data).
  • Forecasts gas prices for the next 10 hours.

---

Results

• Visualizations include:

  • Actual gas prices for the last 30 days.
  • 2-day test predictions compared with actual prices.
  • Forecasted gas prices for the next 10 hours.

• Profit Calculation:

  • Displays the daily trading profit in rubles and percentage.

• Model Performance:

  • Mean Squared Error (MSE) is printed after evaluation.

---

Contributing

Contributions are welcome!
Please open an issue or submit a pull request.

---

Future Improvements

• Extend forecast beyond 10 hours.
• Add support for other commodities.
• Optimize the model for better accuracy.

---