Traffic congestion remains a significant challenge in urban management, with traditional fixed-cycle traffic signal systems struggling to adapt to dynamic traffic conditions. This paper proposes an adaptive traffic signal control method based on a Graph Neural Network (GNN) and a dynamic entropy-constrained Soft Actor–Critic (DESAC) algorithm. The approach first extracts both global and local features of the traffic network using GNN and then utilizes the DESAC algorithm to optimize traffic signal control at both single and multi-intersection levels. Finally, a simulation environment is established on the CityFlow platform to evaluate the proposed method’s performance through experiments involving single and twelve intersection scenarios. Simulation results on the CityFlow platform demonstrate that G-DESAC significantly improves traffic flow, reduces delays and queue lengths, and enhances intersection capacity compared to other algorithms. In single intersection scenarios, G-DESAC achieves a higher reward, reduced total delay time, minimized queue lengths, and improved throughput. In multi-intersection scenarios, G-DESAC maintains high rewards with stable and efficient optimization, outperforming DQN, SAC, Max-Pressure, and DDPG. This research highlights the potential of deep reinforcement learning (DRL) in urban traffic management and positions G-DESAC as a robust solution for practical traffic signal control applications, offering substantial improvements in traffic efficiency and congestion mitigation.

To solve the decision problem of train stopping schemes, this paper introduces the static game into the optimal configuration of stopping time to realize the rational decision of train operation. First, a train energy consumption model is constructed with the lowest energy consumption of train operation as the optimization objective. In addition, a Mustang optimization algorithm based on cubic chaos mapping, the population hierarchy mechanism, the golden sine strategy, and the Levy flight strategy was designed for solving the problem of it being easy for the traditional population intelligence algorithm to fall into a local optimum when solving complex problems. Lastly, simulation experiments were conducted to compare the designed algorithm with PSO, GA, WOA, GWO, and other cutting-edge optimization algorithms in cross-sectional simulations, and the results show that the algorithm had excellent global optimization finding and convergence capabilities. The simulation results show that the research in this paper can provide effective decisions for the dwell time of trains at multiple stations, and promote the intelligent operation of the train system.

The core idea of prediction-based anomaly detection is to identify anomalies by constructing a prediction model and comparing predicted and observed values. However, most existing traffic flow prediction models primarily focus on spatio-temporal features, neglecting comprehensive frequency-domain feature learning. Additionally, anomaly detection accuracy is often limited by insufficient prediction error analysis. To address this limitation, this paper proposes a prediction-based anomaly detection method for traffic flow data with multi-domain feature extraction. The prediction model is built as follows: first, Bidirectional Long Short-Term Memory network (Bi-LSTM) and a Graph Attention Network (GAT) extract temporal and spatial features, respectively. Then, Fast Fourier Transform (FFT) converts time-domain signals into the frequency domain, where Transformer learns magnitude and phase features. Finally, a prediction model is constructed using the extracted time-domain and frequency-domain features. For error analysis, this paper innovatively applies Chebyshev’s inequality to determine the error threshold, identifying anomalies based on whether errors exceed this threshold. Experimental results show that integrating multi-domain features can more comprehensively capture data characteristics and improve model prediction accuracy. In the anomaly detection experiment, it was verified that constructing a high-accuracy prediction model and conducting reasonable error analysis can effectively enable anomaly detection in the data.

This paper proposes a flow battery-lithium-ion battery hybrid energy storage system (HESS) bi-level optimization planning method to address flexibility supply-demand balance challenges in regional power grids with high renewable penetration at 220 kV and above voltage levels. The method establishes a planning-operation coordination framework: Upper-level planning minimizes total lifecycle investment and operation-maintenance costs; Lower-level operation incorporates multiple constraints including flexibility gap penalties, voltage fluctuations, and line losses, overcoming single-timescale limitations. The approach enhances global search capability through the Improved Weighted Average Algorithm (IWAA) and optimizes power allocation accuracy using adaptive Variational Mode Decomposition (VMD). Validation using grid data from Southwest China demonstrates significant improvements across five comparative schemes. Results show substantial reductions in total investment costs, penalty costs, voltage fluctuations, and line losses compared to benchmark solutions, enhancing grid power supply stability and verifying the effectiveness of the model and algorithm.

With the continued electrification and digitalization of urban logistics, electric freight routing increasingly requires the coordinated consideration of customer time windows, vehicle capacity, limited battery range, and en-route charging. This study formulates an electric vehicle routing problem with time windows (EVRPTW) for smart-city electric freight and develops a multi-strategy improved ant colony optimization algorithm (IACO). The proposed model integrates customer service, route continuity, time windows, vehicle capacity, battery-energy propagation, and en-route charging. IACO combines a route–charging-state representation with feasibility-guided sweep-insertion initialization, max–min pheromone control, multi-representative guidance, reachable charging-station insertion, greedy feasibility repair, and 2-opt local search, forming a multi-stage search process that integrates global exploration, feasibility restoration, and local intensification. Computational experiments on an R-C benchmark scenario with 51 customers and 9 charging stations compare IACO with ACO, GA, TS, LNS, SA, PSO, and WOA over 100 independent runs under a common 300-iteration limit. Under the current experimental protocol, IACO records a representative generalized cost of 570.88, with reductions of 7.75–44.65% relative to the seven comparison methods, while its median CPU time is 28.42 s. These results demonstrate a clear solution-quality–computation trade-off and indicate the potential of IACO for plan-level electric freight routing and en-route charging coordination.

To address the coordinated operation of renewable generation, load, and storage in zero-carbon parks, this paper proposes a source–load–storage (SLS) coordinated multi-resource optimization method. First, a parallel forecasting model combining partial least squares regression (PLSR) and ModernTCN, denoted PLSR-Modern TCN, is developed. PLSR extracts an eight-dimensional latent representation from each lagged-load window, while ModernTCN independently captures nonlinear temporal dependencies from the original one-channel sequence. The two representations are aligned by sample index, concatenated, and mapped to the one-day-ahead forecasting horizon. The model achieves an R² of 0.987, outperforming traditional TCN, convolutional neural network (CNN), random forest, and linear regression models. Based on the forecasts, a multi-objective SLS optimization model is established by considering time-of-use electricity prices, supply–demand balance, renewable curtailment, operation cost, carbon emissions, energy storage operation, PCC voltage, and equivalent harmonic power. The entropy weight method determines the objective weights, and an improved genetic algorithm (IGA) solves the optimization model. Simulation results show that IGA achieves the lowest mean comprehensive fitness and the smallest repeated-run variation among the compared algorithms. In the illustrative scheduling result, IGA provides a modest energy-related operating-cost reduction of approximately 0.11–0.13% and a carbon-emission reduction of approximately 2.2–4.5%.