In order to solve the problems of difficulty in finding target points, sparse rewards, and slow convergence when using deep reinforcement learning algorithms for path planning of agricultural robots, a path-planning method based on multi-target point navigation integrated improved deep Q-network algorithm (MPN-DQN) was proposed. The laser simultaneous localization and mapping (SLAM) was used to scan the global environment to construct a prior map and divide the walking row and crop row areas, and the map boundary was expanded and fitted to form a forward bow-shaped operation corridor. The middle target point was used to segment the global environment, and the complex environment was divided into a multi-stage short-range navigation environment to simplify the target point search process. The deep Q-network algorithm was improved from three aspects: action space, exploration strategy and reward function to improve the reward sparsity problem, accelerate the convergence speed of the algorithm, and improve the navigation success rate. Experimental results showed that the total number of collisions of agricultural robots equipped with the MPN-DQN algorithm was 1, the average navigation time was 104.27 s, the average navigation distance was 16.58 m, and the average navigation success rate was 95%.

To achieve fault type judgment, fault line localization, and fault distance judgment in the grid, a novel hybrid network model integrating fast Fourier transform (FFT), convolutional neural networks (CNN), and graph convolutional networks (GCN) was proposed for grid fault diagnosis. Voltage and current signals were decomposed in time and frequency domains by FFT to extract fundamental waveform amplitude and phase. CNN extracted the temporal features of the decomposed data, and layer normalization was introduced to enhance the model stability. The spatial topology of the grid was processed with GCN to extract and integrate spatial features. The model’s effectiveness was verified through modeling and simulation of the IEEE 39-bus power grid system. Experimental results show that the proposed model possesses strong generalization capabilities, and the fault diagnosis accuracy under various tasks, sampling intervals, and noise conditions outperforms existing models.

A tomato leaf disease detection model based on the improved CenterNet algorithm was proposed in order to address the false detection and missed detection phenomena in traditional tomato leaf disease detection. A feature fusion module that integrated the attention mechanism was constructed in order to enhance the model's cross-scale feature fusion capability. The multi-branch convolutional module RFB was added to the backbone network in order to expand the receptive field and enhance the ability to extract target features. The pyramid convolution PyConv was introduced into the backbone network to enhance the extraction of multi-scale features by calculating receptive fields of different scales and reduce information loss. Pruning optimization strategies were designed in order to reduce the impact of introducing modules on the number of model parameters and computational load. The test results showed that the accuracy rate, recall rate, mAP₅₀ and mAP_50:95 of the improved model reached 96.3%, 80.2%, 91.4% and 78.7% respectively. The proposed model can effectively improve the accuracy of tomato leaf disease detection, and the model has good generalization.

Existing anomaly detection methods of time series data focus on extracting the temporal variation features, while the spatial dependency features between multiple variables are ignored. To address this problem, a detection method based on a spatiotemporal graph attention network was proposed. The original multivariate time series data were transformed into a time-series graph with spatiotemporal dependencies, and a spatiotemporal graph attention network was designed to separately extract the temporal variation features and spatial dependency features. The periodic patterns of fused spatiotemporal features were learned by a multilayer perceptron, and an anomaly detection was performed based on the anomaly scores between prediction values and observation values. Experimental results on public datasets showed that the proposed method significantly outperformed state-of-the-art baseline methods in terms of anomaly detection accuracy and robustness.

A bi-level optimization strategy for active distribution networks with smart soft open point (SOP) considering carbon-guided electric vehicle (EV) clustering was proposed, in order to improve the consumption of wind-solar new energy and fully exploit the potential of EV clusters in optimal operation of active distribution networks for carbon emission reduction. Firstly, under the premise of EV cluster integration, the active/reactive power outputs of distributed generators and SOPs were coordinated and optimized by the upper layer to minimize system operation costs. Secondly, the lower-layer generalized energy storage optimization model for EV clusters based on Minkowski sums was constructed to minimize charging-discharging costs. A dynamic tariff mechanism based on dynamic carbon emission factor was proposed to guide vehicle-grid energy interaction and achieve win-win benefits for both parties. Finally, simulation verification on the improved IEEE 33-node system showed that the strategy could effectively promote friendly interaction between active distribution network and EV cluster, reducing the risk of system voltage overruns.

A hierarchical architecture coordinated control strategy considering driving styles for active four-wheel steering (AFWS) and direct yaw moment control (DYC) systems was proposed to improve the handling stability of distributed drive electric vehicles and accommodate the driving styles of different drivers. This strategy employed a three-layer control architecture, including the upper controller, the middle controller, and the lower controller. A reference model for handling stability considering driving styles was established in the upper controller. The stability factors of vehicles with different driving styles were determined through driver-in-the-loop experiments, and the vehicle states were categorized into stable, transitional, and unstable regions based on the phase plane theory. A hybrid game control model for AFWS and DYC based on Stackelberg leader-follower game and Pareto cooperative game was established in the middle controller to improve the vehicle’s handling stability under complex driving conditions. The lower controller was used to optimize the wheel drive torque distribution with the goal of minimizing the tire load rate. The driver-in-the-loop test platform was built based on the Simulink simulation software and the Logitech G29 driving simulator, and open-loop and in-loop tests with drivers were conducted. The results indicated that the proposed control strategy can adapt to the driving styles of different drivers and meet their personalized needs, thereby improving the vehicle’s handling stability.

An improved slime mold algorithm with artificial bee colony (ISMABC) was proposed aiming at the problem of cooperative path planning for multiple unmanned aerial vehicles (UAVs). A path planning cost model was established, and the path planning problem in a three-dimensional environment was transformed into an optimization problem by introducing a fitness function and constraint conditions, which was solved by the proposed algorithm to obtain the optimal path. The slime mold algorithm was improved by employing the good point set strategy and a nonlinear convergence factor. Then the population diversity was increased, and the convergence speed of the algorithm was accelerated. An elite opposition-based learning strategy was applied to the global best individual in order to enhance population quality. A global best position guidance was introduced based on the exploration capability of the artificial bee colony in order to improve the exploitation capability of the slime mold algorithm. Comparative analysis of optimization on 14 test functions and some functions from the CEC2017 test suite showed that the optimization ability and convergence speed of the ISMABC algorithm were significantly enhanced. The algorithm was applied to solve the problem of cooperative path planning for multiple UAVs in order to verify the feasibility of the ISMABC algorithm. Comparative analysis shows that the ISMABC algorithm can be used to plan paths with minimal cost that satisfies the constraints for each UAV.

To address the issue of poor data quality in wind turbine operational data collected by the supervisory control and data acquisition system, a method combining an improved imputation diffusion model and long short-term memory (IDM-LSTM) was proposed. A dual-mask collaborative strategy was employed in the training process of the imputation diffusion model, which helped the model focus on key abnormal distribution regions and enhanced its robustness against abnormal disturbances. A hierarchical residual inverted Transformer (HRIformer) was used as the denoising model, combining the iTransformer with residual connections to improve the model’s ability to capture complex features. During the inference phase of the imputation diffusion model, the periodic visibility reconstruction mask (PVRM) strategy was applied, controlling the mask range by setting an appropriate mask cycle, ensuring the consistency of sequence reconstruction and temporal integrity. The imputation diffusion model is responsible for anomaly detection, while LSTM handles the correction, resulting in an integrated data cleaning framework for unlabeled wind power data. Experimental results from a real wind farm show that IDM-LSTM cleaning improved the Pearson correlation coefficients for wind speed-power and rotational speed-power by 3.78% and 3.43%, respectively, compared with the original data, significantly enhancing wind power data quality.

Significant progress in multimodal large language models (MLLMs) has driven advances in visual question answering, visual understanding, and reasoning tasks, and their potential for deployment on resource-constrained edge devices is increasingly recognized. However, large model sizes and the substantial costs of deployment and inference remain major barriers to practical adoption. Optimizing MLLMs for edge devices has become a critical research direction in this field. A comprehensive survey of recent advances in optimizing MLLMs for edge deployment was presented, along with the associated challenges and development trends. The research evolution of MLLMs on edge devices was reviewed, with particular emphasis on model architecture optimization and inference scheduling strategies. In model architecture optimization, techniques including visual information compression, sparse attention, and mixture-of-experts models were specifically analyzed. System-level optimizations involving computation scheduling, hardware adaptation, compilation optimization, and cloud-edge collaboration were investigated to enhance inference efficiency and energy efficiency. Furthermore, the key challenges of these models in practical applications were discussed, and a variety of task scenarios ranging from assistive to collaborative and autonomous types were covered, categorized by the perspective of autonomy levels. Finally, current limitations were summarized and future research directions regarding standardized deployment, efficient computing and storage, and multi-modal fusion optimization were outlined.

To improve the cooling performance of power modules in new energy vehicles, a fluid–thermal–solid coupling numerical method was used to analyze the thermal management of an insulated gate bipolar transistor (IGBT) power module. A three-stage design optimization method, including contribution quantification, surrogate modeling, and overall optimization, was proposed. A numerical model of the IGBT power module was established in ANSYS Fluent, and the resulting relative error between the simulation and the experimental data was 3.7%. The effects of substrate ceramic material, coolant flow rate, and Pin-Fin geometry on thermal performance were analyzed, showing that convective thermal resistance and ceramic layer resistance are the main factors affecting chip thermal resistance. Based on surrogate modeling and multi-objective optimization, the design of a 750 V/820 A H-Boost IGBT power module was optimized. The optimized design reduced chip thermal resistance by 21.1%, pressure drop by 39.3%, and module mass by 6.1%.

A day-ahead market economic dispatch model that incorporated the flexible ramping products (FRPs) provided by energy storage systems was proposed to fully harness the flexible regulatory capabilities of various resources to ensure real-time flexibility, thereby addressing the problem that conventional thermal power units cannot meet the power system’s flexibility demands against the backdrop of the construction of a new type of power system. The demand composition and the opportunity cost of FRPs were introduced. A decision tree model for FRPs participation in flexible ramping deployment in the day-ahead and real-time markets was obtained by conducting the probabilistic analysis of the decision-making schemes for FRPs in both the day-ahead and real-time markets, and the cost and revenue analyses were performed under the scenarios where FRPs are either abundant or scarce. Based on this, a day-ahead market economic dispatch model that considered energy storage providing FRPs was established. After the economic dispatch in the day-ahead market, the cost and FRPs revenue settlement was achieved based on the probability of accepting FRPs in the day-ahead market and the expected deployment probability in the real-time market. A case study analysis using an improved IEEE 30-bus system was conducted to validate the superiority of the model with energy storage participation, and the impact of the acceptance probability in the day-ahead market and the expected deployment probability in the real-time market on the system was discussed.

A new 3D underwater AUV path planning method (IADQN) was proposed due to the low quality of the generated path and poor dynamic obstacle avoidance ability of AUV path planning methods in complex marine environments. In order to resolve the problem of insufficient obstacle recognition and avoidance ability of AUVs in unknown underwater environments, an adaptive potential field method was proposed to improve the efficiency of action selection of AUVs. In order to address the problem of low sample selection efficiency in the traditional deep Q network (DQN) experience replay strategy, a priority experience replay strategy was adopted to select samples with higher contributions to training from the experience pool to improve the efficiency of training. AUV dynamically adjusts the reward function according to the current state to accelerate the convergence speed of IADQN during training. Simulation results show that, compared with the DQN scheme, IADQN plans a time-saving and collision-free path efficiently in a real ocean environment; the AUV running time is reduced by 6.41 s, and the maximum angle with the ocean current is reduced by 10.39°.

Existing scheduling models for Internet data center (IDC) commonly have two key shortcomings: insufficient consideration of the voltage fluctuations in distribution networks and the impact of IDC storage data constraints on the spatiotemporal transfer of the workloads. An optimized operation method for flexible distribution networks was proposed considering the flexible IDC scheduling. Uncertainty scenarios for renewable generation and load demand were generated using Monte Carlo simulation and the K-means clustering algorithm. Differentiated scheduling strategy was established to distinguish between sensitive and tolerant loads. An IDC power consumption model along with a data storage constraint model were constructed. A robust optimization model was established with the objective of minimizing both distribution network operating costs and IDC electricity purchasing costs. The dynamic regulation capability of soft open point (SOP) and voltage fluctuation constraints were also incorporated into the model. Simulation results demonstrated that the proposed method significantly enhanced voltage quality, markedly increased the renewable energy accommodation rate, and reduced the electricity procurement cost for IDC by collaboratively optimizing the spatiotemporal load shifting of IDC workload along with the power regulation and reactive power support capabilities of SOP. It also validated the impact of storage capacity constraints on the efficiency of spatiotemporal workload shifting.

An improved sliding mode active disturbance rejection control strategy based on the quasi-continuous algorithm was proposed by utilizing the virtual synchronous generator (VSG) technology to address the issue that the traditional linear control strategy of photovoltaic hybrid power conversion system (PV-HPCS) was difficult to achieve ideal control effect in terms of power and voltage fluctuation suppression, and to improve the performance of PV-HPCS, to stabilize the power fluctuation of the grid and maintain the stability of bus voltage. The VSG was integrated with the dual closed-loop control strategy of voltage and current. The PI control was adopted in the voltage outer loop to provide reference values for the current inner loop. A high-order super-twisting sliding mode observer (HO-STSMO) which realized higher response speed and tracking accuracy was adopted in the current inner loop. The quasi-continuous integral terminal sliding mode controller (QC-ITSMC) was introduced in the form of dual sliding mode switching control law, and an improved exponential reaching law was designed to smooth the system control signal and improve the robustness of the system. The simulation model and experimental platform were built, and the experimental results showed that the improved control strategy effectively suppressed the power fluctuation of the grid and improved the stability of the system.

A traffic scene perception algorithm (SDFormer++) based on the principle of cross-task bidirectional feature interaction for autonomous driving in urban street scenarios was proposed by leveraging the explicit and implicit correlations between the semantic segmentation tasks and the depth estimation tasks to improve the overall performance of traffic scene perception algorithms. An interaction-gated linear unit was added into the cross-task feature extraction stage to form high-quality task-specific feature representations. A multi-task feature interaction module that used the bidirectional attention mechanism was constructed to enhance the initial task-specific features by utilizing the feature information of shared cross-domain tasks. A multi-scale feature fusion module was designed to integrate information at different levels to obtain fine high-resolution features. Experimental results on the Cityscapes dataset showed that the algorithm achieved a mean intersection over union (mIoU) of 82.4% for pixel segmentation, a root mean square error (RMSE) of 4.453 for depth estimation, an absolute relative error (ARE) of 0.130 for depth estimation, and an average distance estimation error of 6.0% for five typical traffic participants, all of which outperformed the existing mainstream multi-task algorithms such as InvPT++ and SDFormer.

A multi-goal multi-agent path planning algorithm was proposed to realize the efficient assignment of tasks to each agent and plan the shortest possible paths for the agents without collision with other agents. The definition of conflict between agents in continuous time and the way of conflict resolution were defined, and the concepts of safety interval and labeling were introduced based on A* algorithm aiming at the problem of low success rate due to the use of discrete time in traditional path planning algorithms. Then the A* algorithm can plan optimal paths that satisfy continuous time constraints. A conflict hierarchical strategy was proposed to reduce the number of nodes extended in the algorithm solving process aiming at the large amount of computation caused by collision detection and conflict avoidance in the multi-agent path planning problem. The experimental results show that the proposed algorithm can solve a better solution and has better applicability, with lower total path cost and higher success rate in the scenario of densely distributed agents.

A new path planning algorithm was proposed to address the limitations of the conventional rapidly-exploring random trees (RRT) path planning algorithm, including excessive redundant sampling points, increased randomness, and lack of smooth paths. A model named the optimal path area prediction network (OPAPN) was developed to predict potential optimal path areas within the map using deep learning techniques. A global feature extraction module, a hybrid attention mechanism, and switchable atrous convolution techniques were incorporated in the model. The network’s understanding of the map’s overall layout and start/goal information was enhanced by the components to reduce unnecessary computational burdens. The number of sampling points was reduced significantly through heuristic sampling in the optimal path regions predicted by OPAPN, and the algorithm’s convergence speed was accelerated via a dual-tree expansion strategy. Both simulation experiments and real-world tests showed that the proposed algorithm performed well in convergence time, node count, and path length, confirming its practical application value.

To address the energy consumption optimization problem in cooperative control of mixed vehicle platoon while ensuring passenger comfort, a collaborative control method combining real-time optimization with distributed model predictive control and an intelligent driver model was proposed. For connected autonomous vehicles in the platoon, passenger comfort constraints were established. Utilizing a precise fuel consumption model, a real-time optimized distributed model predictive control method was designed to reduce real-time energy consumption while ensuring the consistency and stability of the platoon. For human-driven vehicles in the platoon, an intelligent driver-following model that ensures passenger comfort and low energy consumption was adopted. The following stability condition was then derived. Simulation experiments were conducted in the scenarios of constant speed and variable speed leader vehicles to verify the tracking performance of the proposed control method under the constraints of passenger comfort. The average engine power from the initial state to the steady state was used as the energy consumption optimization index, and multiple sets of comparative simulation experiments were conducted. Simulation results show that, compared with the comparative algorithm, the proposed control method can effectively reduce the energy consumption of the mixed vehicle platoon.

A snow scene construction algorithm based on perception estimation was proposed to address the problems of difficulty in collecting data from existing snow scene sample libraries and limited sample sizes during the training of autonomous driving perception performance. The snow scene was divided into two models: snow-covered model and snow-line model, and a snow-covered plane construction algorithm based on spatial perception was proposed to analyze the subtle gradient changes in the image and estimate the preliminary snow area. The preliminary snow areas were refined using connected domain analysis, and the areas were fused with the original image to obtain snow-covered scene images. A snow scene construction algorithm based on a random snow-line model was proposed to generate different motion directions for snowflakes. Snow-covered model and snow-line model were integrated, and various basic snow-line forms were utilized to construct snow scenes. Experimental results from the comparison of multi-frame traffic video data show that, in the snowy scenes constructed using the snowy scene fusion method, as the amount of snow increases, both the interference information and the details of the image are significantly enhanced, subjectively approaching the actual snow scene. Moreover, objective evaluation metrics for snow image quality decrease with the increase of snow cover and snow-line density.

Household energy consumption is growing continuously, with uncertainties arising from photovoltaic (PV) generation and household loads. To address these issues, an optimal scheduling model for household energy management was developed. The energy storage capabilities of electric vehicle (EV) were leveraged, while battery degradation costs caused by frequent charging/discharging were accounted for. The stochastic nature of PV generation and the flexibility in power consumption of smart appliances were also synergized. A real-time energy scheduling algorithm under time-varying tariffs was proposed based on the improved Lyapunov optimization theory. EV charging/discharging was controlled, the operation of different household appliances was scheduled, and the bidirectional power transactions between households and the grid were optimized. The time-varying nature of PV generation and grid tariffs was effectively coped with. Energy utility was maximized while the waiting time delay for each household power demand was ensured not to exceed its tolerable period. Theoretical analysis showed that the proposed algorithm could drive the optimization objective to converge to the optimal value without relying on prior statistical information of the system. The effectiveness and economic efficiency of the proposed optimization strategy were verified through comparisons with existing algorithms and performance analyses under various conditions.