Deep Lossless Compression Algorithm Based on Arithmetic Coding for Power Data.

Area of Science:

• Computer Science
• Electrical Engineering
• Data Science

Background:

• Classical lossless compression relies on manual encoding and quantification.
• Deep learning offers data-driven approaches with superior performance in specific domains.
• Neural networks can automatically learn features and adapt to data distributions.

Purpose of the Study:

• To propose an efficient deep lossless compression algorithm for minute-level power data.
• To compare the effectiveness of Bi-directional Long Short-Term Memory (Bi-LSTM) and Transformers for this task.
• To leverage arithmetic coding for quantifying neural network outputs.

Main Methods:

• Developed a deep learning model utilizing Bi-LSTM and Transformers.
• Trained and evaluated models on minute-level power data.
• Employed arithmetic coding to quantify the neural network's output probability distribution.
• Assessed performance based on compression ratio (CR).

Main Results:

• The deep learning approach automatically extracts relevant features from power data.
• The model adapts to the probability distribution quantification.
• Achieved an average compression ratio (CR) of 4.06 for minute-level power data.
• Outperformed classical entropy coding methods in terms of compression ratio.

Conclusions:

• The proposed deep lossless compression algorithm is efficient for minute-level power data.
• Data-driven deep learning models, specifically Bi-LSTM and Transformers, show promise in lossless compression.
• Automatic feature extraction and adaptive quantification enhance compression performance.