Logistics distribution transportation vehicle route decision-making based on artificial neural network algorithm

Abstract

This paper proposes a logistics distribution route optimization model that integrates long short-term memory (LSTM) and feedforward neural networks (FNN). LSTM captures dynamic time-series features (e.g., traffic, weather), while FNN handles static features (e.g., customer location, historical demand). Their outputs are fused via a fully connected layer and passed through Softmax or Sigmoid to generate the optimal route. Experiments show the model significantly reduces delivery time and cost while improving customer satisfaction. For example, in Task 1—delivering 10 kg of food in Downtown Area A—it achieves an average delivery time of 29 minutes, a cost of 34 yuan, and a satisfaction rating of 4.9, demonstrating its effectiveness as a novel solution for route optimization.