In order to study the prediction and health management of PEMFCs(proton exchange membrane fuel cells) for vehicles, a method combining GWO(grey wolf optimizer) and RBF(radial basis function) neural network with relative power loss rate as a health indicator was proposed to predict the remaining useful life of vehicular PEMFCs. Firstly, by analyzing the polarization curve of the fuel cell at the initial moment, a calculation method based on the relative power loss rate as a health indicator was constructed, and its feasibility was verified using the grey correlation analysis method. Then, the RBF neural network optimized by GWO algorithm was applied to predict the remaining useful life of vehicular PEMFCs. Finally, the proposed method was validated using two datasets. The results show that compared with other methods, the GWO-RBF method proposed in this paper has the smallest average absolute percentage error and root mean square error, the largest coefficient of determination, and a relative error of less than 1%. It is concluded that the proposed method can be used to predict the remaining useful life of vehicular PEMFCs with fewer datasets and better accuracy.

The efficient utilization of electromagnetic spectrum resources has become a significant concern in the domain of wireless communications, with EMSM(electromagnetic spectrum map) playing a crucial role in visually representing spectrum usage within a specific task area and providing valuable support for the optimization of wireless networks. To address the challenges associated with generating fine-grained EMSMs under conditions of complex scenes and limited spatial point monitoring data, an improved DRN(deep residual network) model, ES-AFB(enhanced with a spatial attention feature block), was proposed. This model drew inspiration from image super-resolution techniques and leveraged the strong spatial characteristics of EMSMs to design a deep residual network capable of extracting the correlation and spectral features of EMSMs. The enhanced spatial attention feature block was utilized to mine the intrinsic implicit spatial features of coarse-grained EMSMs. Subsequently, the data size was reconfigured through the network’s multilayer up-sampling module, enabling the achievement of a more effective fine-grained image restoration. This approach allows for the generation of high-quality fine-grained EMSMs using limited coarse-grained monitoring data. The effectiveness of the algorithm is validated through simulation experiments, with the root-mean-square error of the EMSMs generated from actual data being found to be no more than 3%.

A simplified lower-limb exoskeleton model was established for the prototype, and the D-H parameter method was used to perform dynamic analysis. Joint angles were measured experimentally and used as inputs for the controller. To address the robot's trajectory tracking problem, traditional PID control was employed, showing good tracking performance but slow response and parameter tuning speed. Although PSO(particle swarm optimization) accelerated the parameter tuning, issues with low convergence accuracy and local optimum traps persisted. Therefore, a PID control based on a chaotic-mapping improved PSO algorithm was designed. The results show that the randomness was enhanced, the parameter tuning speed was increased, and the tracking error was reduced. Simscape was used for visual simulation of joint angles, and the control performance was further validated through various experiments.

Considering the oscillation phenomenon of the original DWA(dynamic window approach) in path planning, an improved DWA path planning algorithm was designed, which is integrated with the artificial potential field method. Firstly, the safety constraint of the DWA algorithm is improved, and the linear obstacle distance evaluation function in the original DWA was improved to the nonlinear obstacle potential field function in the artificial potential field method. Secondly, the improved DWA was combined with the smooth A^* path of the gradient descent method to solve the problem of poor global planning of the traditional algorithm. Finally, the feasibility of the algorithm was verified by simulation experiments and physical experiments. In the simulation experiments, compared with the original algorithm, the improved algorithm in this paper reduces the path of the designed obstacle scene by 9.84%, reduces the running time by 31.71%, and improves the smoothness by 6.49%. Meanwhile, compared with the results of related literatures, the results of this paper have been improved to different degrees in different scenarios. In the physical experiments of automated guided vehicle, the path length is reduced by 10.76% and the elapsed time is reduced by 13.09%. Therefore, the improved DWA generates better path smoothness, shorter path length and shorter elapsed time.

In order to do a good job of emergency management in the airport sector, strengthen the construction of the emergency response system, and improve the emergency response capability, a decision support method based on hybrid reasoning was proposed for the disposal of airport emergencies. Firstly, the ontology model of airport emergencies was constructed by abstracting the emergencies and the disposal process in the airport based on the actual scenarios and official documents. Secondly, the hybrid reasoning combining rule-based reasoning and case-based reasoning was introduced for case retrieval, and case representation was performed for the constructed ontology model to construct a case database. Lastly, the retrieval results are corrected using a feature weighting algorithm for attribute trade-offs, and the attribute parameters were adjusted using a neural network-based weight parameter optimization strategy. The advantages of the Bert+LSTM combination in this task scenario were verified by comparing it with commonly used deep learning models, and the final example proves that when an emergency occurs, the model can focus on the emergency itself, refer to historical cases and disposal standards, and obtain a structured data that comprehensively describes the information and disposal measures of the emergency, which provides support for emergency disposal decision making.

Aiming at the problem that ViBe (visual background extractor) algorithm is prone to ghosting during moving target detection, an improved algorithm, ViBe-BR (ViBe with background restoration) was proposed by adding a background restoration stage to the original algorithm. First, the foreground region within the background image was pre-extracted by combining three-frame differencing. Then, the interior of the region was filled using the background pixels around the foreground region to obtain the restored image; Finally, the restored image was corrected and ViBe detection was performed based on the reduced background to achieve the effect of suppressing ghosting. The experimental results show that the ViBe-BR algorithm achieves good detection results in four different scenes, and compared with the ViBe algorithm, the average precision, recall, and F₁ value of foreground detection have been improved by 0.222, 0.03, and 0.123 in that order, which effectively eliminates the influence of ghosting, and it can be applied to practical geo-localization tasks in order to obtain the geographic location information of the moving targets.

To solve the problem of high memory and computational resource demands in obstacle detection models within autonomous driving perception domain controllers, a lightweight obstacle detection method based on improved YOLOv8 was proposed. This method reconstructs the YOLOv8 backbone network using FasterNet, which utilizes less memory access and computational resources. To mitigate the accuracy decline and the insufficient detection capabilities for small objects caused by model lightweighting, three main improvements were made to YOLOv8: SPD-Conv (space-to-depth convolution) was used to replace traditional stride convolution in the neck network to enhance small object feature extraction. IPIoU(inner powerful IoU), combining the concepts of IIoU(inner IoU) and PIoU(powerful IoU), is introduced as the bounding box regression loss to accelerate loss convergence and improve small object detection performance. SimAM (simple attention module) was incorporated to further enhance model detection accuracy. Experimental results demonstrate that, compared to the original model, the improved model achieves a reduction of 29.1% in parameters, 20.5% in computational load, and 28.8% in model size, while increasing mAP@0.5 by 1.2%. Once deployed in autonomous driving vehicle controllers, the model effectively detects obstacles on the road ahead.

Aspect-level sentiment analysis detects the sentiment polarity of given aspect terms from a fine-grained perspective, providing decision support for e-commerce, consumers, and other groups by mining textual aspect sentiment. Different syntactic dependencies were treated equally in existing methods resulting in the influence of relation types in convolutional networks and the global information from semantic perspective being overlooked. To address these issues, considering the flexibility and complexity of graph structures, the excellent performance of auxiliary tasks in capturing aspect sentiment based on global semantic information and completing fine-grained aspect information, the model WRCN-CL (weight relational convolutional networks and complementary task) which incorporates two tasks: WRCN(weighted relational convolutional networks) and CL (complementary learning) was proposed. Specifically, Bi-LSTM (bidirectional long short-term memory network) was used to extract textual features, which were entered into WRCN and CL tasks separately. Aspect-related semantic information was collected from a global perspective to enhance knowledge, while the aspect representations from CL combined with GCN (graph convolutional networks) to deeply explore syntactic information based on positional and type-aware relational information in WRCN. The fused global and local features were then input into a pooling layer to obtain comprehensive information representation for improved classification performance. Experimental results demonstrate significant improvements with the accuracy of 83.49%、78.19%、75.89% on three public datasets compared to baseline models, proving the effectiveness of the proposed model in aspect-level sentiment analysis classification task.

In order to understand the properties of char which is the main byproduct of coal pyrolysis in the reactor with internals, especially the char prepared at high temperatures, the char samples were prepared by pyrolysis of Yulin bituminous coal under vacuum condition between 1 000 ℃ and 1 500 ℃ in the fixed-bed reactor with internals. The variations of pore structure, composition, calorific value, and carbon structure of chars with pyrolysis temperature were studied. The isothermal combustion characteristic and kinetics were investigated by MFBRA(micro fluidized bed reaction analyzer). The results show that the increase in pyrolysis temperature leads to a deeper degree of pyrolysis. As the pyrolysis temperature increases, the C content and C/H molar ratio of chars gradually increase, H and O contents of chars gradually decreases, and the calorific value increases slightly. The specific surface area and pore volume of chars firstly decrease with the increase of pyrolysis temperature, and then remain essentially constant after the pyrolysis temperature exceeds 1 300 ℃, which is consistent with the trend of graphitization degree. The three-dimensional diffusion model G(x)=[1-(1-x)^1/2]² can well describe the combustion process of chars. The activation energy and pre-exponential factor increase with the increase of pyrolysis temperature, and vary in the range of 30.49~90.17 kJ/mol and 0.69~352.25 s^-1, respectively. The significant decrease in combustion reactivity of char prepared above 1400 ℃ is mainly due to the sharp increase of C/H.

Pressure drop plays an important effect on the performance of fiber filter-stick, and is determined by the materials and geometry structure. In order to develop a method to design and guide the production of the fiber filter, the multiple regression method uses the density of fiber tow, length, and circumference of the sticker as the argument to train the pressure drop model based on the production data. The fiber filter is modeled as multizone represented as the fiber tow and the forming paper. The flow dynamics in these zones are simulated based on the porous media model. The osmotic coefficients represent the pressure drop in the zone packed with the fiber tow. The simulation results show that the pressure drop is positively correlated with the length of the fiber rob, and the type of filter tow has a greater influence on the pressure drop than the circumference. For the design of the fiber filter, the regression model is first used to obtain the consumption of the fiber tow based on the design pressure drop. Then the simulation based on the porous media model is carried out to validate the prediction. If the error between the two methods is within 10%, this predicted fiber stick can be produced. By analyzing the production data and the prediction from models, it is concluded that the method proposed in this work is sufficient to direct the design and production.