Neural network-driven user behavior forecasting and personalized recommendation in power marketing

Area of Science:

• Artificial Intelligence
• Power Systems Engineering
• Data Science

Background:

• Smart grids and complex power marketing necessitate accurate user behavior prediction and personalized recommendations.
• Existing methods struggle with data sparsity, cold-start issues, fixed strategies, and adapting to dynamic user behavior, impacting recommendation accuracy and customer experience.

Purpose of the Study:

• To propose a novel neural network model for enhanced user behavior prediction and personalized recommendation in power marketing.
• To address limitations of current methods, including data sparsity and inability to adapt to dynamic user behavior.

Main Methods:

• Utilized Graph Convolutional Networks (GCN) to model user-product interactions.
• Employed Deep Deterministic Policy Gradient (DDPG) for dynamic recommendation strategy optimization.
• Integrated Multi-Layer Perceptron (MLP) for user behavior prediction.

Main Results:

• The proposed model demonstrated significant improvements in recommendation precision, recall, and AUC compared to traditional methods.
• Achieved an average improvement of 5.4% in Precision, 7.7% in Recall, and 3.3% in AUC over state-of-the-art methods.
• GCN, DDPG, and MLP integration enhanced handling of multi-dimensional user behaviors and real-time feedback adaptation.

Conclusions:

• The neural network model offers a more accurate, personalized, and dynamic approach to user behavior prediction and recommendation in power marketing.
• The model enhances customer experience and improves overall business efficiency in the power marketing sector.