To enhance the system's response speed and anti-interference capability and reduce the speed fluctuations of the range extender， a method of using generator torque to compensate for engine torque vibration is peoposed. The output torque of the engine to determine the influence of the harmonic components in the torque on the speed is analyzed. Then， a control strategy combining an adaptive backstepping method and a repetitive controller is designed. The repetitive controller is parameterized based on the fundamental frequency of the periodic torque disturbance of the system， and the stability and robustness of the system are analyzed. Finally， simulation experiments are conducted under different operating conditions， and the results show that this control strategy has good effectiveness and practicability in suppressing the speed fluctuations of the range extender.

In order to improve the power density， the aviation gear is usually made of thin wall structure such as thin web and thin rim， but this will make the gear highly flexible， resulting in decreased dynamic performance. In high-speed thin-walled gear design， how to balance lightweight and dynamic performance to make the system performance optimal is a big problem. Therefore， firstly considering the influence of thin-walled gear， shaft， housing and other components flexibility， gear 6 degrees of freedom， centrifugal force， inertia force， the flexible multi-body dynamics model of high-speed thin-walled gear system is established， and verified by experiments. Secondly， a multi-objective lightweight optimization design method for high-speed thin-walled gears based on flexible gear dynamics is proposed. In the optimization design， based on the dynamic simulation of high-speed thin-walled gear， the evaluation indexes of dynamic performance， strength and fatigue life are established. The parameter-performance correlation analysis method is used to select the most influential parameters for optimization. Finally， the multi-objective lightweight design research is carried out， and the high-speed thin-walled gear system with lower vibration noise and lighter weight compared to traditional designs is obtained.

In response to the causes of unclear and difficult reduction of coupling errors in force measurement platforms arranged with multiple sensors， this article takes the piezoelectric six axis force measurement platform as the research object， analyzes the coupling causes from the entire process of sensor coupling， assembly error and calibration device. Based on measurement principles， establish an inter dimensional coupling model containing 34 influencing factors， use sensitivity algorithms to screen and quantify the degree of influence of each factor， and accuracy tolerance allocate is performed with a coupling error ≤ 5% of the range as the target. The sampling simulation shows that the 99% confidence upper limit of each coupling error after allocation meets the requirements. Experiment shows that the coupling error of platform force output is ≤ 3.64% of the range under tolerance， and the coupling error of torque output is ≤ 4.80% of the range， which verify the effectiveness of the accuracy range and their applicability in coupling error control.

To achieve the extrapolation of the full-cycle load spectrum from limited load data， this paper proposes a load extrapolation method based on local adaptive bandwidth diffusion kernel density estimation， in view of the limitations of traditional adaptive bandwidth optimization algorithms in parameter selection. This method first reduces the two-dimensional rainflow matrix to one-dimensional equivalent amplitude， then optimizes the local bandwidth based on the local integral mean square error， and uses the local optimal bandwidth to construct a probability density distribution model through diffusion kernel density estimation. Finally， the target frequency load is extrapolated by combining the Monte Carlo simulation method. The preprocessed load data of a certain special vehicle's comprehensive transmission device are compared and verified. The results show that compared with the traditional method， the probability density distribution curve and cumulative frequency curve obtained by the method proposed in this paper are closer to the actual equivalent amplitude. The correlation coefficients and determination coefficients are both closer to 1， with the correlation coefficients being 0.983 8 and 0.999 6 respectively， and the determination coefficients being 0.967 9 and 0.999 1 respectively. The root mean square errors are also smaller， being 5.05×10^-5 and 15.9 respectively.

A more accurate bolt connection stiffness model is proposed for the single lap joint structure of composite laminates under secondary bending. Firstly， the bending stiffness is calculated by considering the secondary bending caused by eccentric loads. Secondly， the carbon-aluminum single lap model is established by ABAQUS finite element software and compared with the M-G model. Finally， the influence of friction coefficient， bolt hole clearance， metal plate thickness and ply sequence on the stiffness of bolted connection structure at each stage is analyzed， and the influence of contact state of connection contact surface on bolt connection structure is revealed. The results show that the increase of the friction coefficient will lead to the increase of the critical friction force in the viscous stage， but it does not affect the stiffness of the bolt connection structure. The increase of the bolt hole clearance will lead to the decrease of the contact area between the bolt hole and the screw， thus reducing the stiffness of the bolt connection structure. The metal plate thickness has a more significant effect on the stiffness of the bolt connection structure in the viscous stage and the contact stage. The 0°ply can improve the stiffness and bearing capacity of the laminate， and the 45°ply can improve the compressive strength around the hole.

SA516Gr.70 steel has excellent strength and toughness and is widely used in the manufacture of pressure vessels， but the traditional arc welding is prone to lead to defects， such as crack and hydrogen embrittlement in the weld. In this paper， a W-25Re（%） alloy stirring tool was used to carry out friction stir welding of 4 mm thick SA516Gr.70 steel plate， and the thermal cycle curve during welding was measured by using a type K thermocouple to study the effect of thermal cycle on the microstructure and mechanical properties of the welded joint. The results show that defect-free welded joints can be obtained at a rotation rate of 800 rpm and a traverse speed of 90 mm/min. The maximum value of the peak temperature of the thermal cycle appeared in the advancing side （AS） reaching 1 018 ℃ and in the retreating side reaching 998 ℃. The microstructure of the weld nugget zone （WNZ） produces fine martensite and bainite， with an average grain size of 5.4 μm， which is a significant effect of grain refinement in the WNZ compared with the base metal. The microhardness of welded joints is negatively correlated with the distance from the center of the weld， and the microhardness of the WNZ reached 449.7 HV. Tensile test fractures are all in heat affected zone （HAZ）， and different areas of the joint show different fracture morphologies.

To improve the corrosion resistance of GO/PANI/BTA/EP coating on the surface of 6061 aluminum alloy and explore its corrosion inhibition mechanism，graphene oxide （GO）/polyaniline （PANI） nanoparticles were prepared by in-situ copolymerization method. The corrosion inhibitor benzotriazole （BTA） was adsorbed on the surface of GO/PANI nanoparticles through physical adsorption， resulting in GO/PANI/BTA.The infrared spectrum shows that a layer of BTA has been successfully loaded onto the surface of GO/PANI. The adsorption rate of the loaded BTA on the surface of GO/PANI can reach 18.36%， and the release rate after 5 hours can reach over 90%， providing effective protection for aluminum alloys. The polarization curve and immersion experiment results show that the epoxy coating （GO/PANI/BTA/EP） with 1% GO/PANI/BTA added has excellent corrosion resistance. The self corrosion potential of 1% GO/PANI/BTA/EP is -0.267 V， which is 0.247 V higher than that of GO/PANI/EP； The self corrosion current density is 9.795×10^-9A?cm^-2， higher than GO/PANI/EP （self corrosion current density of 7.638×10^-7A?cm^-2） is two orders of magnitude lower. The 1% GO/PANI/BTA/EP composite coating has excellent corrosion resistance， and its corrosion inhibition mechanism comes from the organic combination of graphene oxide shielding， polyaniline passivation， and BTA corrosion inhibition， achieving a synergistic effect.

Based on the experimental results of interfacial temperature and sonotrode displacement， the finite element method was employed to investigate the friction and material softening process in pure copper ultrasonic welding， and the distribution of the temperature fields and plastic strain during the welding process were also analyzed. The results show that the established model can better simulate the dynamic friction coefficient and ultrasonic softening variation processes during welding. The average friction coefficient of the workpiece contact surface increased rapidly in the early stages of welding， and then decreased sharply， and presented smooth fluctuations after welding time of 0.2 s. The ultrasonic softening coefficlent of copper increased sharply in the early stages of welding， then slowly increased， and began to increase significantly when the welding time was 0.3 s. Ultrasonic softening is the main physical mechanism of materials plastic deformation.

Laser ablation is an effective interfacial treatment to improve the joint bonding performance， whereas the effect mechanism of the laser treatment process on the shear strength of bonded joints is not clear. This study developed a test fixture for the shear strength of bonded joints which addressed the out-of-plane bending during the tensile shear of thin plates. The effect of laser surface treatment process on the shear strength of aluminum bonded joints was elucidated through the interfacial microstructure， wettability， and fracture mechanism. The results show that the shear strength of bonded joints is affected by the combination of interfacial roughness and surface wettability. The highest shear strength （24.51 MPa） was obtained using a laser energy density of 82.6 J/cm²， a 45°+135° morphology， and overlap rate of -25%. Compared to the rolled surface （11.74 MPa）， the shear strength of the laser-machined joints increased by 108.7%.

Al-5Ti-1B （wt.%） master alloy refiner and Al-10Ce （wt.%） master alloy modifier were used to composite refine and modify ADC12 Al alloy. The phase structures， microstructures， distribution characteristics of the second phases， and interfacial bonding characteristics of the Al alloy were analyzed by X-ray diffraction （XRD）， optical microscopy （OM）， scanning electron microscopy （SEM）， and transmission electron microscopy （TEM）. The tensile properties and Brinell hardness of the Al alloy were also tested， and the correlation between microstructures and mechanical properties was established. The results showed that the refinement effect of α-Al matrix of ADC12 Al alloy after composite refinement and modification was remarkable， and the eutectic Si transformed from rough long needles to fine round short rods. After adding 0.75%AlTiB-0.75%Ce， the secondary dendrite arm spacing was reduced to11.4 μm， CeCu_0.5Si_1.5 phase was dispersed in the matrix and CeCu_0.5Si_1.5 rare earth phase was well boned with Mn_4.6Fe_0.4Si₃ Si-rich phase at the interface. The tensile strength， elongation， and Brinell hardness of the Al alloy are 272.7 MPa， 3.1%， and 87.7 HB， respectively， which were increased by 79.2%， 287.5%， and 11.2% compared with as-received Al alloy. The comprehensive mechanical properties are applicable for die-casting requirements that need high-strength， heat treatment-free cast aluminum alloys， such as for manufacturing automotive engine parts， transmission casings， and similar components.

Convolutional neural networks （CNN） are the key modules for extracting traffic features in deep learning algorithms for traffic state estimation. However， their computational instability on sparse data and multi-modal traffic states limits the accuracy of deep learning for state estimation. To improve the accuracy of CNN-based traffic feature analysis， this paper proposed an encode-decode traffic adaptive convolutional network. Firstly， the proposed network constructed a traffic feature encoding CNN， which uses down-sampling operations to aggregate neighboring traffic information and extracts effective traffic features from sparse data. Secondly， a traffic state adaptive reconstruction CNN is constructed to utilize the extracted features to accurately reconstruct different patterns of traffic states. To represent diverse spatio-temporal structure of traffic states， this CNN introduced prior knowledge to guide the deformation of convolutional kernels. Finally， the performance of algorithm in traffic state estimation under sparse data and different scenarios was validated using taxi GPS data in Changchun City. Experimental results show that compared with advanced algorithms such as LSTM and GAN， the improved CNN alogorithm in this paper improves the estimation accuracy by 6.05 km/h RMSE. Besides， the proposed algorithm has an accuracy difference of only 3.34% RMSE in different traffic scenarios， can provide robust traffic festure analysis support for deep learning-based traffic state estimation.

To study the characteristics of passenger-cargo mixed traffic flow in vehicle-connected environment and the lane change strategy suitable for the mixed traffic flow environment， the relative entropy is introduced to quantify the order of passenger-cargo mixed traffic flow， and on this basis， conditional agglomeration lane change strategy is proposed. The effect of conditional lane change strategy on the capacity and order of mixed traffic flow was studied by cellular automata simulation. The results show that compared with non-agglomeration lane change strategy， CDA（Conditional aggregation） lane change strategy can improve the maximum road capacity by 5.5%~10.5% in mixed traffic flow with different CAV（Connected and autonomous vehicle） penetration rate and passenger-cargo ratio. It can significantly reduce the relative entropy of traffic flow， increase the number of vehicles in CACC（Cooperative adaptive cruise control） queue and improve the orderliness of traffic flow. When the CACC minimum queue size of the conditional lane change strategy is 4~5 vehicles， the road capacity can be improved the most.

To improve the safety of intelligent vehicles， intelligent driving systems must be able to accurately predict the evolution of traffic scenarios and accurate evaluate the risk of potential collisions， especially considering the uncertainties of future vehicle status and movement. To address this problem， this paper proposes a collision risk assessment method considering dual uncertainties， including the uncertainty of vehicle multi-state perception and the uncertainty of vehicle motion model considering driver control input and road geometry factors. Firstly， the extended Kalman filter is used to complete the vehicle state estimation， integrating road geometry and driver behavior to optimize model matching. By determining the model switching threshold and allocating the yaw rate weight，the switching between diffenrent kinematic models to make the model more suitable for the current driving scenario and more accurate uncertain vehicle trajectory prediction. Secondly， the heuristic Monte Carlo simulation is used to test the potential collisions of the sampled trajectory points， and the collision risk is quantified as the probability of future collisions. The deterministic detection is used as heuristic information to significantly improve the estimation efficiency. Finally， according to the China-new car assessment program， four typical collision scenarios are built in Prescan， including rear-end collision driving in a straight line， lateral collision at an intersection， forward collision during lane change overtaking， and rear-end collision driving in a curve， for validation and assessment of the proposed method in this paper. The results show that in a variety of collision scenarios， the proposed method in thi paper can accurately perceive the risk of collision 2 seconds before the occurrence of a collision and provide a 100% collision probability warning， effectively avoiding or mitigating collisions.

To explore the reasonable length of the auxiliary lane at the exit of the two-lane highway， the operating status of vehicles was analyzed， the maximum distance formula for vehicle lane change in the auxiliary lane segment was derived； by introducing reliability theory， a functional function of the length of the auxiliary lane was established， and the relationship between the length of the auxiliary lane and the reliability of the auxiliary lane was analyzed， the recommended values of the length of the auxiliary lane under different design speeds were calculated； finally， the rationality of the recommended values was analyzed through VISSIM simulation in this paper. The results show that the length of the auxiliary lane is positively correlated with the reliability probability of the auxiliary lane. In different simulation scenarios， the recommended value of the simulation section has good traffic flow compared to the standard value， and will not cause material waste. The research results can provide reference for further improvement of the regulations.

To analyze the influence of steel truss web deformation on the deflection of the steel truss composite box beam. Firstly， establishing displacement functions based on the rotation angle of the top flange， bottom flange， steel truss web，and composite box beam section， the expression of deflection displacement of composite box beam was obtained by energy variational method. Secondly， the beam segment analysis element stiffness matrix and nodal load array were derived based on the finite beam segment method， and the deflection displacement of the composite box beam was solved. Thirally， the deflection of the composite box beam under different load conditions for the compasite box beam were analyzed and compared with the elementary beam theory. Lastly， the influence of structural parameters such as height-span ratio， diameter of steel truss web， wall thickness of steel truss web and tilt angle of steel truss web on additional deflection were analyzed. The results show that， the influence of the deformation of the steel truss web deformation on the deflection of the composite box beam cannot be ignored， the maximum deflection error of the composite box beam was 27.75% by using the elementary beam theory. Among the structural parameters influence the deformation of the steel truss web， the wall thickness of steel truss web has the greatest influence on the additional deflection， followed by the diameter of steel truss web， the height-span ratio and the tilt angle of steel truss web. In addition， the additional deflection ratio was positively correlated with the height-span ratio and tilt angle of steel truss web， negatively correlated with the diameter of steel truss web and wall thickness of steel truss web.

Photovoltaic aprays are affected by external environmental factors， and local shading effects limit the system's power generation efficiency. To address this issue， this paper proposes an optimization control method that utilizes Logistics chaotic sequence initialization and adaptive weighting to construct an adaptive search aquila optimizer. This algorithm can dynamically track the maximum power point of the photovoltaic array， thus achieving optimal control of traffic self consistent energy system. Experimental results under different lighting conditions on a simulation platform demonstrate that the algorithm proposed in this paper has better tracking accuracy and speed compared to perturbation observation methods， particle swarm algorithms， and traditional aquila optimizer， and is less likely to get stuck in local optimal solutions. The proposed method has a certain reference value for improving the power generation efficiency of photovoltaic power generation systems and reducing operational costs in transportation.

To reveal the spatiotemporal characteristics of the dockless bike sharing connected to the metro and analyze the impact of built environment factors on connection demand， the built environment around the entrances and exits of Shenzhen metro stations was considered in this paper， a method for identifying connection travel was constructed by employing Thiessen polygons. This method explores the characteristics of four types of connection traffic on weekdays from both temporal and spatial perspectives， including access transfer （AT） of morning rush hour， egress transfer （ET） of morning rush hour， AT evening rush hours and ET of evening rush hours. The geographically and temporally weighted regression （GTWR） model was utilized to analyze the impact of four types of built environment factors on connection demand. The results indicate that the distribution of connection traffic exhibits spatial heterogeneity， with higher demand during the morning rush hour than during the evening rush hour. Business residential and scientific educational cultural service POIs have a positive impact on connection traffic in the urban core area， while it is negatively correlated in the suburban areas. However， the impact of transportation facility service POIs on connection traffic varies in the opposite direction. Corporate enterprise POIs have a positive impact on connection traffic in the western urban areas， while it is negatively correlated in the eastern areas.

The space of multi story buildings is usually limited， and how to reasonably layout public facilities within the limited space to meet the needs of different user groups is a challenge. Therefore， this paper proposes an optimization design method for the utilization of public facilities in multi story buildings based on firefly algorithm. Establishing a public facility utilization optimization model with the optimization objectives of maximizing efficiency， fairness， and coverage， and using firefly algorithm to obtain the optimal solution of the model， in order to achieve the optimization design of public facility utilization in multi story buildings. The experimental results show that the proposed method has high efficiency， fairness， and coverage， indicating that the planning results of this method can ensure the rational utilization of public resources， benefit the target population， and cover the widest possible geographical area.

This paper prepared a controlled low-strength material for backfilling in the narrow area. The variation rules of its working performance， mechanical properties and durability with the water-solid ratio， the gray-water ratio and the content of fly ash were studied. The results showed that the fluidity of CLSM was positively correlated with the ratio of gray-water and water-solid. The compressive strength was between 0.19 and 0.87 MPa at 4.1%~12.9% of the cement content when the design fluidity was 25 cm， highly positively correlated with the cement content. The CBR value and the modulus of resilience have a good linear relationship with the compressive strength. The compression deformation of mix proportion was much lower than that of silty sand， which can improve the construction efficiency and tamping effect at the narrow backfill surface of the airport foundation pit and pipe gallery.

In view of the poor energy absorption and buffering capacity of existing concrete rigid guardrails， a new type of composite guardrail with U-shaped steel and concrete encased by foam aluminum is proposed. Using ABAQUS to establish a guardrail vehicle collision coupling system， simulate the collision process of a 49 t heavy-duty truck with a guardrail， and analyze the guiding function， buffering function， and energy absorption function of the guardrail. The results of finite element analysis show that the ultimate strength of the new composite guardrail with wrapped U-shaped steel and concrete assembled with foam aluminum is significantly improved compared with the existing ordinary concrete guardrail. At the same time， during the collision of the truck with the guardrail， the loaded truck did not cross the guardrail， and the new combination guardrail has good guidance performance； the lateral acceleration and longitudinal acceleration at the driver's position are both less than 200 m/s²； the protection energy of the new composite guardrail is 431 kJ， and the impact efficiency is 91.25%. The new composite guardrail structure of wrapped U-shaped steel concrete with foam aluminum has a good energy absorption and cushioning effect， which can reduce the degree of injury to passengers， has a good anti-collision performance， can effectively improve the safety of riding， and can provide a reference for the energy absorption design of concrete guardrails.

The MFF-STDC network model based on multi-level feature fusion is built in this paper to solve the problem that lightweight networks have weak segmentation effect for small target category objects in complex environment. Firstly，by superimposing the feature extraction module based on group convolution many times， the feature extraction capability of the network is improved. Secondly， the combination ability of multi-scale feature information is improved by hierarchical attention module and CA mechanism. Lastly，A-Cityscapes dataset A-IDD dataset and Field dataset were built based on adaptive replication algorithm， the number of small target categories in the dataset was increased， and training and testing were completed. The MFF-STDC network improves the mIoU by 4.01%， 3.65%， and 2.94% respectively comparing with the STDC， and segmentation of the small target categories in the complex environment is much better than that of other networks. A real-world testing experimental platform is built， and the test results show that the MFF-STDC network effectively improves the semantic segmentation accuracy and classification ability of small target categories， and meets the real-time requirements.

Aiming at the problems of incomplete use of patient data and rough data combination in deep learning methods for survival prediction of patients with stomach cancer， a multimodal data fusion survival prediction model for patients with stomach cancer was proposed. Firstly， multimodal data including clinical data， gene expression data and medical images of the same patient were preprocessed. Secondly， the multimodal data was input into the graph attention network （GAT） to make the multimodal data merge with each other under the attention mechanism. Thirdly， the medical images processed by convolutional neural network were introduced to work with the output of graph attention network to predict the results. Finally， ten-fold cross-validation was used to prove the stability of the model performance， and the results were compared with other methods using the same dataset. The results showed that the model proposed in this paper achieves a leading accuracy.

In order to solve the problems of high time complexity and insufficient global exploration ability of the battle royale optimization algorithm， this paper proposes an improved battle royale optimization algorithm based on chaos mapping， center selection and elite adaptive strategy. Compared with the test results of particle swarm optimization algorithm， whale optimization algorithm， and the battle royale optimization algorithm， the improved battle royale optimization algorithm significantly reduces time complexity and notably enhances convergence precision， speed， and stability. In the application of solving the inverse kinematics problem of robots， the accuracy and stability of the improved battle royale optimization algorithm are better than those of the traditional battle royale optimization algorithm， proving its practicality and potential for development in solving robot inverse kinematics problems.

To address the shortcomings of existing single carrier 3D model information hiding algorithms in terms of capacity， robustness， and invisibility， a information hiding algorithm based on 3D model multi carrier surface normal vectors is proposed. Firstly， divide the 3D model carrier set into m classes， correspondingly， the binary sequence of secret data is also divided into m segments， then， obtain the feature points of the 3D model， find the distance from the centroid of the model to each triangular patch and sort the triangular patches of the 3D model， determine the cosine value interval of the angle between the normal vector of the triangular surface of the 3D model and the direction vector of the connection between the vertex and the centroid， secondly， the original hidden data is transformed into the corresponding binary encoding sequence B through Hilbert curve scrambling processing， finally， the embedding of hidden data is achieved through the matching degree between the cosine value interval and the binary encoding sequence of the hidden data， through simulation experiment analysis and comparison， it can be concluded that this algorithm has good robustness and invisibility， and can effectively resist universal attacks during the transmission process of dense carriers.

In order to solve the problems of color distortion， semantic ambiguity， unclear texture and shape characteristics in the process of near-infrared image colorization， we propose a method of infrared image colorization （RAUGAN） . Firstly， the CycleGAN network generator is improved， and a Res-ASPP-UNet network is designed and fused， which connects atrous spatial pyramid pooling（ASPP） to Skip connection structure of the original UNet， the output characteristic graphs of different scales in the decoding branch can be combined with the corresponding output characteristic graphs in the encoder. Secondly， a deep bottleneck layer composed of residual block， convolutional block attention module （CBAM） is designed to replace the bottleneck layer in UNET network to enhance the local area feature， improve its recognition ability and prevent gradient explosion. Finally， the perceptual loss function is introduced in the discriminant network to solve the problem of color restoration distortion.Experimental results show that the proposed method is superior to the original CycleGAN network.

A text-guided face image inpainting method is proposed to address the problems of structural distortion， texture blurring， and uncontrollability in current face inpainting methods. The method reconstructs the missing regions in an image by fusing image features and corresponding text features. In the network training， a visual-textual modal fusion module is designed for associating image and textual features so that the reconstruction of missing regions of the face is not only based on the visual semantics visible in the image， but also guided by textual semantics with rich text. An attention-aware layer is added between the encoded and decoded features to improve the consistency of the appearance of the visible and generated regions. Experimental results on the CelebA-HQ face dataset show that the method in this paper is able to obtain restoration results that are more natural and consistent with the textual semantics in terms of texture and structure， and its visual effect and evaluation metrics are better than those of the comparison algorithms.

The scene in a video sequence changes in real time， sometimes the foreground target changes together with the background， and sometimes the foreground target changes while the background remains unchanged. It is very difficult to achieve detection and tracking of foreground targets. To this end， a high-speed moving object detection algorithm based on Gaussian kernel density estimation is proposed. Using Gaussian kernel density estimation to establish a background model and obtain the probability density distribution of each pixel point； Extract keyframes containing high-speed moving targets from the video sequence and calculate the weight of each keyframe's grayscale value； Using a full sample timing and real-time selective update strategy to update the background model， and using the updated model to achieve accurate detection of high-speed moving targets. The high-speed motion object detection was carried out on a certain video segment in the highwayI-raw standard test sequence， and the results showed that the proposed method has high detection accuracy.

In the software defined Internet of Things （IoT）， multi-source heterogeneous data comes from different devices and has different formats， structures， and qualities， which increases the complexity of feature detection. Therefore， a mixed attribute feature detection method for multi-source heterogeneous data in software defined IoT is proposed. Using the joint Kalman filtering algorithm to fuse multi-source heterogeneous data in the software defined Internet of Things， completing the initial integration of heterogeneous data. Combined with evidence classification algorithms， network data with the same mixed attributes are divided into the same dataset to achieve classification of multi-source heterogeneous data. Based on the inverse similarity characteristics of multi-source data， an edge operator calculation method is introduced to split the classified data attribute features， and combined with support vector machines， accurate detection of multi-source heterogeneous data attribute features is achieved. The experiment shows that the covariance calculation results of the proposed method are always below 0.15， and the distinction between different attribute features is more obvious， with a detection probability of over 0.8. This method can achieve precise partitioning of mixed attributes of multi-source heterogeneous data in software defined IoT.

The evaluation reform of university faculty is an important component of educational evaluation reform and a crucial guarantee for stimulating the intrinsic motivation of university faculty. Self-evaluation by faculty is a key aspect of the comprehensive evaluation system for university faculty. Currently， self-evaluation largely relies on manual statistics and analysis， which presents issues such as strong subjectivity in the evaluation process， low work efficiency， and poor accuracy. Therefore， this paper proposes a method for extracting keywords from university faculty self-evaluation based on an improved PositionRank algorithm. First， faculty self-evaluation data is collected and cleaned to remove redundant and anomalous data， improving data quality. Second， a graph-based keyword extraction algorithm is used to generate high-quality label data. Finally， the improved PositionRank algorithm is employed to adaptively learn the attention weights between phrases， achieving accurate keyword extraction. Experimental results show that this method efficiently identifies key content in faculty evaluations， significantly improving the accuracy of keyword extraction while demonstrating strong evaluation consistency. It helps reveal the core strengths and areas for improvement of faculty and provides strong technical support for improving the university faculty evaluation system.

To address the issues of unreasonable project activity scheduling and high expenditure， this paper proposes an improved differential evolution algorithm based on directional crossover mechanism （DirDE）. The mechanism enhances the algorithm's convergence speed by guiding the global exploration and local exploitation directions of the population. Meanwhile， this mechanism helps the algorithm jump out of the local search and avoid falling into the local optimum based on the gene guidance of the parent individuals. In the experimental section， benchmark function experiments are designed to verify the optimization ability of the DirDE. The experimental analysis results demonstrate that DirDE exhibits better convergence， accuracy， and the ability to avoid falling into local optima. Finally， the method proposed in this paper was tested through simulation experiments on a real-world linear programming model for multi-project party affairs activity scheduling optimization. The algorithm demonstrated its competitiveness and can serve as an effective tool for solving real-world party affairs activity scheduling problems.

This article uses the ensemble empirical mode decomposition （EEMD） method to decompose and denoise the magnetic resonance full-wave signal， after obtaining a series of intrinsic mode function （IMF） components， the energy density and average period of each IMF component are calculated based on the principle of adaptive denoising，the IMF components dominated by noise are removed， and the selected IMF components are reconstructed，solved the problem of severe environmental noise interference in magnetic resonance full-wave signals. The simulation experimental data results show that when the signal-to-noise ratio of the magnetic resonance signal is as low as -10 dB， after EEMD processing， the magnetic resonance parameters can still be effectively extracted. The relative error of initial amplitude E₀ extraction is 1.57%， the relative error of relaxation time T_{\mathrm{2}}^{\mathrm{*}} extraction is 2.96%， and the signal-to-noise ratio is improved to 10.31 dB. The noise suppression results of the measured data further validate the effectiveness and practicality of the algorithm studied in this paper， provides technical support for the application of magnetic resonance groundwater detection technology in complex roise environments.

Continuing the research on multi-task balancing strategy for robot manipulators， the use of parallel robot manipulators in automated assembly lines allows for precise grasping and handling of dynamic objects on conveyors. However， the challenge lies in ensuring leakage-free handling of materials， which cannot be achieved by a single robot manipulator. This highlights the importance of multi-robot manipulator collaboration. In this study， a novel approach based on the deep Q-learning （DQN） algorithm is proposed to address the task balancing problem for robot manipulators. The objective is to minimize the cost function associated with the manipulation tasks. To improve the performance of the DQN algorithm， the study incorporates the actor-critic algorithm and introduces the concept of balancing advantage function. This allows for the optimization of the execution strategy， ensuring a balanced allocation of tasks among the robot manipulators. Experimental results validate the effectiveness of the proposed algorithm. It demonstrates a 17.1% improvement in convergence performance compared to the traditional DQN algorithm. Additionally， under predefined experimental conditions， the proposed algorithm achieves approximately a 31.6% reduction in overall cost when compared with the baseline algorithm. This significant reduction in cost enhances the overall execution efficiency of robot manipulators in multi-task scenarios.

A new control strategy was proposed for the air supply system of proton exchange membrane fuel cell， which solves the multiple input and multiple output， strong nonlinear coupling and internal and external disturbance control problem of the air supply system， then realizes the precision accurate tracking control of the air flow rate and cathode pressure. The high-order fully actuated system is derived from fourth-order underactuated system of the traditional state space model， and the extended state observer is designed to estimate the unknown disturbance based on high-order fully actuated system. According to the characteristics of high-order fully actuated system， the control law is designed to realize the cooperative control of the air flow and cathode pressure. Finally， through the simulation result comparison with other control strategies， and verified on the experimental bench， the results show that the air flow rate and cathode pressure can be precise controlled.

To address the issue of high computational complexity and excessive time consumption of intelligent vehicle trajectory planning in three-dimensional space （X-Y-T）， this paper proposes a dual-layer， three-stage trajectory planning method based on adaptive sampling. By using the Frenet coordinate system， the three-dimensional trajectory planning problem is decomposed into two two-dimensional optimization problems： path planning and speed planning. Firstly， in terms of path planning， an adaptive sampling method based on artificial potential fields is introduced to reduce the number of path planning space sampling points. Dynamic programming is used to calculate the path curve that connects each sampling point with the lowest comprehensive cost， and a quadratic programming problem is constructed to further optimize this path to obtain the final planned path. Secondly，in terms of speed planning， the adaptive sampling area for speed planning space is determined according to kinematic constraints. The dynamic programming method is used to calculate the speed curve that connects each sampling point with the lowest comprehensive cost， and a quadratic programming problem is constructed based on this curve to obtain the final speed planning result. Finally， a trajectory tracking error model is constructed， and lateral and longitudinal controllers are designed to verify the trackability of the planned trajectory. Simulation results show that the proposed method in this paper can reduce the planning time to about 0.1 seconds per planning cycle， providing a planned trajectory with a frequency of 10 Hz for intelligent vehicles， achieving coordination between planning and control.