NNISGGA: A Lightweight Method for Generating Large-Scale Road Network Maintenance Plans

The performance of road deteriorates over time due to various factors, significantly impacting traffic safety and driving comfort. To maintain the quality of road services, proper pavement maintenance and rehabilitation (M&R) is essential for improving road performance. Existing pavement M&R decision algorithms often suffer from issues like dimensional explosion and local optima, rendering them unsuitable for large-scale road networks. To address these challenges, this study proposes a lightweight method for generating large-scale road network maintenance plans (NNISGGA). First, two deep neural networks are employed to calculate the optimization objective and generate the probability of implementing each M&R action on every road segment, respectively. Subsequently, M&R actions with low probabilities on each segment are eliminated, and the highest-probability action for each segment is selected to form a high-quality solution. Finally, an initial solution-guided genetic algorithm (ISGGA) is established for maintenance plans generation and optimization, which can initialize the population by mutating high-quality solutions and generating random solutions. Case studies were conducted to analyze this method, yielding the following results: Compared to several baseline algorithms, the NNISGGA algorithm generated maintenance plans with the highest average effectiveness-cost ratio (E/C), demonstrating that this algorithm-generated plans are closer to the global optimum. The NNISGGA algorithm reduces the number of combinations required for optimized M&R decisions and decreases the average runtime of optimization. Furthermore, compared to currently popular genetic algorithm (GA), the NNISGGA algorithm converges faster during operation. Therefore, this method can generate reasonable large-scale maintenance plans for transportation management departments. © 2026 IEEE.