Intelligent analysis algorithm for power engineering data based on improved BiLSTM.

Area of Science:

• Electrical Engineering
• Artificial Intelligence
• Data Science

Background:

• Traditional Bidirectional Long Short-Term Memory (BiLSTM) models struggle with long time series and the specific characteristics of power engineering data.
• Existing models show insufficient accuracy in predicting issues within specialized power system applications.

Purpose of the Study:

• To enhance BiLSTM's capability in processing long time series data relevant to power engineering.
• To develop an intelligent analysis algorithm that integrates multi-dimensional features and adapts to power data characteristics.
• To improve the accuracy and robustness of power load forecasting and equipment fault diagnosis.

Main Methods:

• Integration of an attention mechanism into the BiLSTM architecture to improve long time series capture.
• Development of an intelligent analysis algorithm incorporating multi-dimensional feature fusion and a multi-head self-attention mechanism.
• Optimization of the model specifically for power engineering data, enhancing noise robustness by 8%.

Main Results:

• In load forecasting, the improved BiLSTM achieved mean square errors of 0.02 (summer) and 0.025 (winter), with R² values of 0.985 and 0.982, respectively.
• For equipment fault diagnosis, the enhanced BiLSTM demonstrated significantly higher accuracy compared to models like Gated Recurrent Unit (GRU) across current, voltage, temperature, and pressure parameters.
• The model exhibits faster convergence speed despite a slight increase in training time and memory usage.

Conclusions:

• The improved BiLSTM algorithm, incorporating self-attention and multi-dimensional feature fusion, significantly enhances accuracy and robustness in power engineering data analysis.
• The model shows superior adaptability for complex temporal patterns and multi-source data modeling in power systems.
• This approach offers a more effective solution for critical tasks like power load forecasting and equipment fault diagnosis.