The Crop Science Society of China

Q-learning belongs to reinforcement learning and artificial intelligence learning algorithm. Reinforcement learning does not need external guidance; it interacts with the external environment through its own sensors. It maps the state of the external input environment to output action through continuous learning, and makes the corresponding reward value of this action the maxi-mum. In order to make the submersible have the ability to adapt to the environment independently, it can adjust the path automatically through its own learning. This paper proposes to introduce Q-learning mechanism in reinforcement learning to complete the adjustment of fuzzy rule strategy in un-known environment.

Aiming at the lack of robustness of existing direction of arrival (DOA) estimation algorithms in impulsive noise environment, a new robust DOA estimation method in impulsive noise is proposed in this paper. In this method, the hyperbolic tangent function with good amplitude limiting characteristics and the sign function with phase holding performance are selected to solve the effective suppression of impulse noise and the efficient and reliable retention of signal information. Combined with MUSIC algorithm, the robust DOA estimation under impulse noise is achieved. Simulation results show that the proposed method has better and more robust DOA estimation performance than existing methods.

This study focuses on the power systems of inland waterway vessels in Chinese Yangtze River, systematically outlining the low-carbon technology pathways for different power system types. A comparative analysis is conducted on the technical feasibility, emission reduction potential, and economic viability of LNG, methanol, ammonia, pure electric and hybrid power systems, revealing the bottlenecks hindering the large-scale application of each system. Key findings indicate that: (1) LNG and methanol fuels offer significant short-term emission reductions in internal combustion engine power systems, yet face constraints from methane slip and insufficient green methanol production capacity, respectively; (2) ammonia enables zero-carbon operations but requires breakthroughs in combustion stability and synergistic control of NOX; (3) electric vessels show high decarbonization potential, but battery energy density limits their range, while PEMFC lifespan constraints and SOFC thermal management deficiencies impede commercialization; (4) hybrid/range-extended power systems, with superior energy efficiency and lower retrofitting costs, serve as transitional solutions for existing vessels, though challenged by inadequate energy management strategies and multi-equipment communication protocol interoperability. A phased transition pathway is proposed: LNG/methanol engines and hybrid systems dominate during 2025–2030; ammonia-powered systems and solid-state batteries scale during 2030–2035; post-2035 operations achieve zero-carbon shipping via green hydrogen/ammonia.

When using genetic algorithm to generate path coverage test cases, test cases are usually input into the instrumented program to obtain program path representation, and fitness values are calculated according to the path representation.This paper uses deep neural network to predict the fitness value of genetic algorithm. Firstly, a batch of test cases are generated by random method, then the fitness value of test cases is calculated, and the training set of neural network is obtained by combination. Then the trained neural network is used to predict the fitness value of the population in the process of evolution to reduce the running time of the population in the process of evolution and improve the efficiency

In response to increasingly stringent emission reduction requirements, methanol holds great potential as an alternative fuel for marine applications. This study proposes an aftertreatment system targeting pollutants emitted by methanol/diesel low-speed engines, which consists of methanol oxidation catalyst (MOC) and selective catalytic reduction (SCR). By establishing models of the methanol/diesel engine and its aftertreatment system, this research systematically analyzes the impact of the MOC + SCR aftertreatment system on engine performance and emissions under both diesel mode and methanol mode. Then, a comparative analysis of the energy balance of the engine with and without the MOC + SCR system was conducted. The results indicate that at 25%, 50%, 75%, and 100% load, the brake specific fuel consumption (BSFC) in pure diesel mode is significantly lower than that in methanol mode, while the exhaust temperature is higher than in methanol mode. After adding the MOC + SCR aftertreatment system in methanol mode, the BSFC generally increases, while the exhaust temperature generally decreases. Except at 25% load, the conversion efficiency of MOC for HC and CO can reach 100%. Under all four engine loads, the conversion efficiency of SCR for NO X can reach 100%. It can be seen that the effect of combined aftertreatment system on pollutant removal can meet ideal requirements. After installing the aftertreatment system, the engine’s brake efficiency is reduced by about 3.7% under medium and high loads, which is within an acceptable range. This study can provide fundamental theoretical and technical support for achieving ultra-low emissions in marine methanol/diesel low-speed engines.

Electricity prices exhibit both stationary and nonlinear characteristics. Wavelet decomposition can reduce noise, stabilize variance, and separate periodic/random components, thereby improving forecasting accuracy. Modeling different sub-series using ARMA (Auto Regressive Moving Average) and Kernel Extreme Learning Machine (KELM) respectively yields better performance than a single model. Therefore, a hybrid wavelet-ARMA-KELM forecasting framework is proposed.

For the power generation-side high-proportion renewable energy system based on electro-hydrogen collaborative technology, the contradictory relationship is hard to balance between economic benefits and environmental benefits in the optimal design. Detailed analysis and evaluation of its comprehensive performance is a key scientific issue that needs to be urgently addressed. The sensitivity of the constructed integrated energy system to government policies (such as energy price policies and carbon tax policies) are an important factor. Therefore, this paper builds a photohydrogen-gas integrated energy system model coupled with electro-hydrogen collaborative technology firstly. Secondly, a multi-objective hierarchical optimization method is proposed considering different priority strategies. Thirdly, the feasibility of promoting emission reduction by formulating natural gas and carbon tax pricing schemes is analyzed. Finally, the simulation example proves that the proposed system, based on the electro-hydrogen collaborative strategy and making rational use of the energy storage advantages of green hydrogen, has strong renewable energy consumption capacity, providing a theoretical basis for green hydrogen technology to help the consumption of renewable energy on the power generation side.

Glass fiber-reinforced polymer (GFRP) composites exhibit significant advantages over conventional structural webbing materials, including lightweight and corrosion resistance. This study investigates the localized compression performance of the proposed GFRP grid web–concrete composite beam through experimental and numerical analyses. Three specimen groups with variable shear-span ratios (λ = 1.43, 1.77) and local stiffener specimens were designed to assess their localized compressive behavior. Experimental results reveal that a 19.2% reduction in shear-span ratio enhances ultimate load capacity by 22.93% and improves stiffness by 66.85%, with additional performance gains of 77.53% in strength and 94.29% in stiffness achieved through local stiffener implementation. In addition, finite element (FE) analysis demonstrated a strong correlation with experimental results, showing less than 5% deviation in ultimate load predictions while accurately predicting stress distributions and failure modes. FE parametric analysis showed that increasing the grid thickness and decreasing the grid spacing within a reasonable range can considerably enhance the localized compression performance. The proposed analytical model, based on Winkler elastic foundation theory, predicts ultimate compression capacities within 10% of both the experimental and numerical results. However, the GFRP grid strength adjustment factor βg should be further refined through additional experiments and numerical analyses to improve reliability.

Bursting disc is a safety relief device which is widely used in industrial production, especially in the arms industry, energy and other fields because of its' simple structure and the high safe capability which is better than safety valves, so the safe and stable operation of bursting discs has attracted widespread attention. However, the research both at home and abroad about bursting discs are concentrated on design and abnormal failure of bursting discs, and the research on the pressure relief test devices of bursting discs has less been published publicly. In order to solve problem of the pressure relief test of bursting discs, in this paper, the RBF neural network is used to compensate the temperature of the pressure sensor, it is of significance in strengthening the safety test of bursting disc.

In response to the development requirements of high-performance capillary pumps, a three-dimensional numerical model for analyzing the performance of capillary pumps was established. The model was used to calculate the flow and heat transfer of capillary pumps, utilizing momentum source terms to handle capillary forces, and comparing the differences in performance of capillary pumps with different structures. The research results indicate that the ratio of outer diameter to inner diameter of the capillary core has a significant impact on the performance of capillary pumps. There is often a trade-off between reducing the maximum temperature of the capillary pump and the maximum temperature difference of the heated surface, as well as reducing the inlet and outlet flow resistance. It is necessary to comprehensively consider the impact of structural dimensions on performance and the relationship between actual requirements in order to select appropriate structural dimensions.

With the continuous expansion of new energy installed capacity, high-power hydrogen production power supplies for water electrolysis require multi-unit parallel connection to satisfy high power demand. It leads to the new issue of power balance control and reactive power compensation in multi-unit parallel systems. This paper first analyzes the causes of system power module imbalance and grid voltage drop based on the proposed topology. It establishes the circuit equivalent model for power imbalance in parallel hydrogen production power supplies and grid-connected rectifiers, and then analyzes the characteristics of parallel balance and grid connection under different control modes of the hydrogen production power supply. Secondly, a solution is proposed using communication methods to achieve adaptive system impedance adjustment and adaptive reactive power compensation for the hydrogen production power supply. Finally, a simulation model of the parallel hydrogen production power supply is built to verify the feasibility and effectiveness of the proposed method.

In the upcoming area of 6G, integrating sensing and communication in complex environments, such as satellites-denied sea surface, evolving into a key point for unmanned surface vehicle(USV) environment awareness. However, accurately perceived heading velocity and position of high speed USV in maritime scenario remains a challenge due to the limited availability of information and hardware cost. To address this issue, the newly released Orthogonal Time Frequency and Space (OTFS) modulation, which extracts Doppler shifts in the spatial domain and time delays, has been proposed to lift relative velocity and position accuracy. In this paper, we introduce a solution that leverages the OTFS waveform to continuously receive communication signals and identify delay-Doppler domain indices simultaneously. Specifically, we design a factor graph to realize belief propagation(BP), which has the potential to retrieve enough information to complete perception. Simulation results show that the proposed method yields significant improvements in velocity and position accuracy compared to the classic methods.

The sub-array is an important part of phased array radar, and its performance directly determines the performance of the phased array. For the problems of low integration, large size, high power consumption, complicated installation and poor maintenance of the brick-type module used in the traditional phased array, a novel digital sub-array is designed. It uses a sheet cold plate and a shunt structure to design the sub-array as a front “brick” and a back “Tile” structure. In view of the cooling problem of an extensible digital sub-array of a three-dimensional laminated structure, this paper uses a liquid cooling method to take away the internal heat of the sub-array at the same time, the cold plate was optimized and designed, the main heat source was theoretically calculated and simulated. Finally, the optimized digital sub-matrix was simulated and tested to achieve a better cooling effect. Through simulation and testing, the maximum temperature of the chip is 60.7°C when the digital sub-matrix is working, which better meets the requirements.

Artificial intelligence plays a critical role in high-precision inspection of new energy vehicle components, addressing micro-defect identification and 3D geometric feature quantification to ensure driving safety. To achieve accurate defect detection while balancing computational efficiency, this study integrates Digital Twin and 3D vision technologies: high-fidelity defect samples are generated via virtual-physical interaction platforms, with point cloud data compressed using voxelization algorithms. A dedicated component inspection system was developed. Experimental validation against Canny edge detection and morphological gradient methods demonstrated performance superiority. System efficacy was evaluated using actual component data, showing the improved algorithm reduced average relative error to 3.5%, with crack and holeposition errors at 3.15% and 2.78% respectively—70.8% lower than conventional methods. The system supports 12 component types and maintained 120-hour fault-free operation under high temperatures. Through virtual-physical data synergy and algorithm optimization, this research delivers high-precision defect detection, offering a referential technical solution for new energy vehicle quality control.

The main goal of image super-resolution is to create high-resolution (HR) images from low-resolution (LR) ones while maintaining clarity and accurate texture details. However, existing methods still suffer from issues such as lacking fine texture details and excessive smoothing. To address these issues, we propose an adaptive residual scaling generative adversarial network (ARSGAN) for image super-resolution. Specifically, we designed Adaptive Residual Scaling (ARS) mechanism that allows dynamic adjustment of the scaling factor, enabling the model to automatically optimize the fusion of residual information according to the features of the input image. In addition, we coupled Strip Attention to accurately capture image directional features, enhancing the model’s ability to capture long-range dependencies. We have also improved the discriminator by incorporating Linear Deformable Convolution (LDConv), which enhances the model’s ability to handle complex deformations and details. The experimental results demonstrate that ARSGAN achieves superior performance and visually pleasing results.

To address the difficulties in fusing multimodal fault data and the lack of decision accuracy and interpretability in complex equipment maintenance decision-making, this paper proposes a large-model-based Multimodal Fault Feature Fusion-Attention-Enhanced Bayesian Reasoning (MFFA-BR) algorithm and constructs an intelligent decision-making model for complex equipment maintenance. First, a large-model-based multimodal feature encoder is designed and combined with a cross-modal attention mechanism to effectively fuse fault data. Second, attention weights are introduced to optimize the Bayesian inference process, improving decision accuracy and interpretability. Finally, an aircraft engine fault dataset (10,000 samples, covering 10 types of faults) is used as the experimental object, and the algorithm is compared with support vector machines, backpropagation neural networks, and BERT + random forest algorithms. Experimental results show that the MFFA-BR algorithm achieves a fault diagnosis accuracy of 98.2%, a maximum improvement of 12.6% over the compared algorithms. The maintenance plan recommendation accuracy is 96.8%, a maximum improvement of 9.2%. The single-sample decision-making time is 0.32 seconds, meeting real-time requirements. Ablation experiments validated the key role of multimodal feature fusion and the attention enhancement module. This model effectively enhances the intelligence of complex equipment maintenance decisions, providing reliable support for real-world maintenance scenarios.

International audience

With the rapid development of computer technology and artificial intelligence, the lightweight structure of ship phased array radar has become the key to improving its maneuverability and endurance. In order to improve the lightweight effect of the ship phased array radar structure, a finite element model ship phased array radar topological structure optimization method based on the variable density method is studied and proposed for the efficient modeling and solution of structural lightweight. In addition, the research also proposes an ant colony algorithm with improved adaptive factors for the size optimization of phased array radar structures. In the simulation test experiment, the reaction time and memory usage of the proposed algorithm are 33.2ms and 0.30GB respectively. After lightweight size optimization, the ship phased array radar structure has a mass of 340.7kg, a maximum deformation of 3.2mm, and a maximum equivalent stress of 107.9MPa, which is significantly better than the comparison algorithm. The experimental results show that the proposed method has fast response speed, low memory usage, and can make excellent lightweight improvements to the ship phased array radar structure, which provides certain experience and ideas for the development of computer intelligence in lightweight structure.

Based on the analysis of the operating environment characteristics and functional requirements of large enclosed equipment due to its large volume, deep height, heavy daily maintenance work, and reliance on personnel climbing and cleaning, a guide rail cleaning wall climbing robot driven by four-wheel drive magnetic wheels is proposed by integrating the concept of active and passive configuration. It has the functions of vertical climbing, fall prevention, oil cleaning, new oil application, and gap crossing. This article conducts mechanical performance analysis and simulation analysis of robots in their operational conditions, and establishes mathematical models and performs computational analysis for the safe adsorption of robots. In light of the control requirements for the robot, the control system for the robot was designed and debugged. The prototype platform test shows that the equipment can effectively reduce the work intensity of personnel, greatly improve the safety of cleaning, and has strong adaptability, feasibility, and scalability.

This article introduces a design scheme for a new type of tile type transmission component, whichaddresses the shortcomings of traditional components and achieves a highly integrated, low loss, and efficient planar layout through 3D integration technology. It meets the requirements of modern radar and communication systems for lightweight and multi frequency bands.