The research status of friction stir welding（FSW）temperature field finite element simulation method and experimental measurement method are summarized systematically. The statistical analysis of papers based on finite element simulation method is carried out from three aspects ： heat source model， computational solid mechanics（CSM）， and computational fluid dynamics（CFD）. For experimental measurement method， the characteristics of different temperature measurement methods are summarized by discussing the temperature measurement principle and research routes of thermocouple and infrared thermal camera. This paper analyzes the advantages and disadvantages of the above methods， and puts forward the future research direction.

The research progress of carbon fiber-reinforced thermoplastic composites was introduced systematically， including interface modification （physical modification and chemical modification）， structural design （biomimetic structure， three-dimensional woven structure and mechanical structure） and molding process （resin transfer molding， automated fiber placement， long fiber reinforced thermoplastics，hot press and 3D printing）. Finally， the application and development of carbon fiber reinforced thermoplastic composites was introduced in the transportation industry， and looks forward to the future development.

Based on a modified gasoline engine， this paper studied the effects of EGR rate and ignition timing on spark assisted homogeneous compression ignition （SACI） combustion and emission performance， and analyzed the relationship between the spark ignition/compression ignition（SI/CI） switching timing and combustion parameters. The results show that external exhaust gas recirculation （EGR） can effectively reduce pressure rise rate， and suppress knocking， it also plays a role in improving fuel economy and NO _x emissions. By adjusting the ignition timing to fit CA50 and SI/CI switching timing within the range of 9~10°CA ATDC and 2°CA ATDC respectively， the engine can obtain the best torque and BSFC.

To control the influence of in-wheel motor eccentricity and other unforeseen factors on the ride comfort of electric vehicles， the problem of improving ride comfort of in-wheel motor electric vehicle by H_∞ control of active suspension constraint state is studied on random road. Based on the standard state feedback H_∞ control， the linear matrix inequality representation of constrained state feedback H_∞ control is established. Considering the interaction of road and motor eccentricity， a four degree of freedom plane model of in-wheel motor electric vehicle with active suspension is built， and the constrained state feedback H_∞ control design of in-wheel motor electric vehicle active suspension is realized. The H_∞ control simulation model of active suspension of in-wheel motor electric vehicle with constraint state feedback is developed by using Matlab/Simulink. The ride comfort simulation and comparison of active suspension and passive suspension are completed through the model on random road. The results show that the eccentricity of in-wheel motor will affect the improvement ability of the active suspension， and the constraint state feedback H_∞ control of the active suspension improves ride comfort of in-wheel motor electric vehicle on random road.

Active hydraulic mount with an inertia track， decoupling membrane and oscillating coil actuator is taken as a research object， this paper focuses on the concise and effective expressing for the mid-high-frequency dynamic characteristics and corresponding parameters identification method. Based on the intrinsic properties that the natural frequency of the actuator mover is much higher than the one of the inertia track， and the fluid channel is closed due to the increase of fluid resistance in the mid-high-frequency bands， the degree of freedom （DOF） for the fluid channel is ignored， and a 1-DOF lumped parameter model is established. The concise expressions of the active and passive dynamic characteristics are obtained for the active hydraulic mount with oscillating coil actuator. It is revealed that the passive dynamics have three horizontal sections and the active dynamics have two horizontal sections， and these conclusions have been validated by experiments. Based on a high-frequency elastomer test system， combined with an appropriate clamping， a dynamic test scheme is provided， which avoids the development of a high-frequency test bench， and which can be used to obtain effective active and passive characteristics up to 400 Hz.

To maintain NO _x conversion efficiency and reduce urea consumption of urea-SCR system， this paper designed a novel hierarchical controller. The up-level controller was a MPC controller to optimize the target ammonia coverage ratio online. The low-level controller was a triple step method controller to fast track the reference using steady-like control， reference feedforward control and gain-scheduling feedback control. Compare the controller performance in MIL environment and all key indicators improved with the cascaded controller. Ammonia slip decreased over 85%， and specific NO _x emission declined 27.4%， while the urea consumption also decreased by 5%. The results show that the optimized strategy proposed in this paper has strong application value.

Aiming at the key problem of the influence of thermal error on the static accuracy of machine tool spindle， a thermal error prediction model based on modified chicken swarm optimization algorithm （MCSO） and support vector machine （SVM） is proposed. Spectral clustering based on unsupervised learning and Spearman correlation analysis are used to identify the key sensitive temperature measuring points of the spindle， so as to reduce the number dependence of temperature data distribution and weaken the multicollinearity among temperature variables. Levy flight strategy is adopted in the local search of hens， and a nonlinear dynamic adaptive inertia weight for updating chick strategy is constructed to realize the global optimization of kernel function， and then MCSO is used to optimize the kernel function， penalty factor and deviation of SVM. MCSO-SVM， BP-GA， GA-SVM and CSO-SVM are used to establish thermal error models respectively. Meanwhile， the prediction ability of models under different working conditions is compared and analyzed. The experimental results of thermal error show that spectral clustering and Spearman correlation analysis can effectively reduce the coupling effect caused by collinearity of temperature variables； MCSO-SVM can achieve high-precision prediction of spindle thermal errors under typical speed， and the model has good generalization ability and robustness.

A deep reinforcement learning based eco-driving strategy for connected electric vehicle （EV） was proposed to improve its energy efficiency at signalized intersection. Firstly， the dynamics of the EV is modelled， and the simulation environment of signalized intersection crossing scenario is established. Secondly， the reward function including multiple objectives is designed considering energy consumption reduction and travel efficiency improvement. The Deep Determinate Policy Gradient （DDPG） is developed to control the vehicle acceleration in continuous action space. Finally， a Monte Carlo simulation is conducted to verify the effectiveness and robustness of proposed method in different driving conditions. The simulation results show that the proposed strategy can improve the vehicle energy efficiency while ensuring travel efficiency in both single and multiple intersection scenarios， compared to a conventional accelerate-constant-brake strategy. In addition， a field test is conducted based on a developed connected automated vehicle digital twin platform. The experiment results show that the proposed reinforcement learning based eco-driving strategy has the potential to improve the vehicle energy efficiency and travel efficiency， simultaneously.

In order to study and evaluate the impact of "oil-to-gas" in traditional gasoline taxis on emissions， two typical China Ⅳ and China Ⅴ "oil-to-gas" dual-fuel taxis were selected as test vehicles in Tangshan City， the actual road gaseous pollutants （CO， CO₂， NO _x ） and particulate matter （PN） emission were tested by PEMS （SEMTECH-DS and ELPI）， and the emission characteristics and differences of "oil-to-gas" dual-fuel taxis under different road types and speed conditions. The results show that compared with gasoline fuel， the CO， CO₂ and PN emission factors of two test vehicles fueled by CNG were reduced by 31.0%~88.6%， 19.8%~23.8% and 79.3%~85.7% respectively， while NO _x emission factors were increased by 4.4 and 1.7 times respectively. Meanwhile， the CO₂， PN， and China Ⅳ test vehicle CO emission factors in different road types/speed ranges were significantly reduced， while China Ⅴ test vehicles CO emission factors were reduced except for urban roads/low-speed ranges （vehicle speeds <50 km/h）， it was relatively higher in the suburbs and in the middle and high speed range where the vehicle speed higher than 50 km/h. The "oil-to-gas" dual fuel effectively reduced the PN emission in each particle size section， and the peak PN concentration at the particle size 9 nm was reduced by 79.7% to 82.5%. Under the two fuel conditions， the PN emissions of test vehicles with a particle size of less than 23 nm accounted for 71.6%~96.9% of the total PN emissions， and the China V was significantly higher than the China Ⅳ， and the natural gas fuel was higher than the gasoline fuel.

Based on the problem of fatigue fracture of the T joint of the typical parts of the aircraft metal integral air inlet due to the resistance of air circulation during the service process， the TC4 titanium alloy was heat treated after the T joint welding with the laser double beam welding technology， and different heat treatments were studied. The effects of different heat treatment methods on microstructure， fatigue properties and fracture mechanism of welded joints were investigated. The results show that the fatigue life of the specimens with double annealing treatment is 1.89 times that of the specimens without heat treatment， and the fatigue life of the specimens with one annealing treatment is 1.43 times that of the specimens without heat treatment. The fatigue fracture analysis shows that， compared with the specimens without heat treatment， the fatigue stripe spacing in the stable fatigue crack growth zone of the specimens after heat treatment decreases， indicating that the fatigue crack growth rate decreases， so the fatigue life increases. After heat treatment， acicular α' martensite precipitates in the microstructure of the net basket， which increases the resistance to crack initiation and propagation and promotes the fatigue life of the specimen.

In order to alleviate the contradiction between motor vehicles and pedestrians， the connotation of bad pedestrian crossing behavior was proposed， and the correlation between pedestrian crossing behavior and pedestrians' psychological expectation was analyzed based on the cost-benefit theory. Spearman correlation coefficient and VIF were used to pre-screen the key factors， and then binary logistic regression model was used to construct the decision-making model of street crossing behavior in three scenarios to screen out the key factors that can be changed， and the corresponding technical correction measures were proposed. In order to verify its effect， the number of bad pedestrian crossing behavior before and after partial correction was compared， indicating that colored zebra crossing and other measures can reduce the probability of bad pedestrian crossing behavior， and the correction measures can improve the pedestrian crossing behavior.

To improve the efficiency of urban traffic signal control system， this paper proposes a sub-area division method based on vehicle trajectory data and density peak clustering. Firstly， the correlation index between adjacent intersections is calculated by combining the influence of distance between intersections， vehicle delays and platoon dispersion based on vehicle trajectory data. Secondly， the distance matrix is obtained according to the correlation indexes， which is used as the input of the density peak clustering algorithm. For the hyperparameter determination in density peak clustering， the concept of potential entropy in the data field theory is introduced to optimize. Simultaneously， the elbow rule is used to determine the number of clusters. Finally， the division of sub-areas is completed by using the improved clustering algorithm. The experiment on real-world vehicle trajectory data in Zhongguancun West District of Beijing shows that the proposed method could divide the road network into sub-area effectively and reasonably based on vehicle trajectory data only.

In cold mega-competition events， in order to alleviate pedestrians suffering from severe weather such as low temperature， wind and snow while waiting for a long time in the bus station， and improve the quality of travel service in the area， a decision-making guidance and passenger flow control method was proposed based on mobile information intervention from the perspective of information service. Based on the results of questionnaire survey， the characteristics of waiting behavior in cold environment were analyzed， and a decision-making behavior model for pedestrians waiting for bus was established based on the Nested Logit model. With the model， the coupling influence of multiple factors on pedestrians' decision-making behavior was discussed and the sensitivity analysis of key variable was carried out. The management suggestions were put forward from the content and frequency of information release. The results show that， compared with the traditional stop sign information， mobile phone information intervention can help to improve the probability of temporary heating and improve the service experience of pedestrians. With the interference of mobile information， pedestrians' decision-making behavior is mainly affected by the layout of temporary heating places and the time from the start of competition. At this time， it is suggested that managers should focus on the information about temporary heating and competition when formulating the content of mobile information. When the frequency of information release is twice per hour， the probability of pedestrians receiving information guidance reaches the highest， which is about 36.6%， indicating that the control effect of information intervention reaches the best. The research can provide theoretical basis and data support for the intervention of passenger flow and the improvement of information service level in the competition area.

To reveal the influence of urban public transportation on the dispersion of COVID-19 within the city， an F-SEIR model considering migrants is established to analyze the impact of inter-regional population movement on the spread of the epidemic. The epidemic transmission characteristics within enclosed public buses were considered， on which， an COVID-19 influenced model of urban public transportation communication was established. Using IC card data and GPS data of public transportation in Wuhan， the empirical case study was simulated with each of the 160 towns in Wuhan as the analysis units. The results demonstrated that under the influence of public transportation， the spread of urban epidemic is accelerated and the phenomenon of "flying point transmission" appears in many towns， and the number of infected people attains 70 446. Timely countermeasures to suspend public transportation have successfully reduced the number of infected persons by 18.64% to 28.34%， and the earlier the measures are taken， the larger total number of infected persons are reduced. The model accurately fits the evolutionary trends of COVID-19. By comparing the simulation results from different dates after the shutdown of publication transportation service on Jan. 23， 2020， the goodness-of-fit between the fitted data and the official epidemic data attains as high as 0.9675. The model is helpful to understand the impact of public transportation on the spread of the epidemic within cities， which may further provide theoretical and decision-making guidance for risk evaluation and policy stipulating during the different stages of other major public health events.

Mixed traffic flow will consist of cooperative adaptive cruise control （CACC） vehicles and manual driven （MD） vehicles under connected and automated environment. Then this paper focuses on CO₂ emission of such mixed flow. Firstly， the mixed traffic flow in this paper was defined according to characteristics of connected and automated environment. Car-following behaviors of each type of vehicles in such mixed flow were described by car-following models calibrated via experimental data. Then numerical simulations were performed considering periodic boundary condition. Based on trajectory data in simulations， CO₂ emission model was used to calculate impacts of CO₂ emission for the mixed traffic. Finally， effect mechanism of CO₂ emission was explored from the perspective of mixed flow stability. An improving method for reducing CO₂ emission was also proposed. Results show that CO₂ emission of such mixed flow would increase and then decrease， with the increase of CACC penetration rate p， which has qualitative relation with traffic flow stability. By using the proposed improving method based on variable headway strategy， CO₂ emission of such mixed flow would decrease with the increase of p. Compared with MD vehicular flow （p=0）， CACC vehicular flow （p=1） could reduce CO₂ emission by 19.35%. Results obtained in this paper can provide reference for management strategy of such mixed traffic flow， from the perspective of reducing CO₂ emission.

Considering the influencing factors of the damping coefficient， the influence of the selection of the target damping ratio and the sensitive frequency range on the pore water pressure， the amount of buoyancy and the internal force of the underground structure under seismic loads is analyzed. The results show that the larger the target damping ratio， the slower the increase of the excess pore water pressure of the soil， and the smaller the floating and the internal force increment of subway structure. There is a threshold for the influence of the upper sensitive frequency.When the effect range exceeds the threshold， the excess pore water pressure， the floating and the internal force of the structure are basically unaffected. The analysis shows that the influence factors of Rayleigh damping should be fully considered and the damping coefficient should be determined reasonably in the analysis of liquefaction dynamic of subway structure.

Through the axial compression test of the thin-walled concrete-filled square steel tube short column confined by the stirrup， the effects of the concrete strength， the volumetric stirrup ratio of the stirrup and the connection mode of the stirrup and the steel pipe wall on the ultimate bearing capacity， stiffness， ductility and transverse deformation coefficient of the specimen are studied. The research results show that the built-in reinforcement can help to improve the ultimate bearing capacity and ductility of the specimen. This improvement effect decreases with the increase of concrete strength， and increases with the increase of the volumetric stirrup ratio of the stirrup. After the stirrup is welded to the steel pipe wall， the mechanical properties of the test piece have not changed significantly. The ultimate strength theory is used to mechanically analyze the specimens in the ultimate state， and the formula for calculating the ultimate bearing capacity of the concrete-filled square steel tube short columns constrained by stirrup is deduced， and combined with the test data of similar specimens in other documents to verify the formula， which proves the correctness of the formula.

The main purpose of this study is to investigate the influence of aggregate contact characteristics on shear fatigue life of asphalt mixtures using discrete element method （DEM）. On the basis of the internal structure information of asphalt mixtures that was acquired by CT scanning and imaging technology， the aggregate contact characteristics were obtained using a program written in MATLAB. Three-dimensional （3D） discrete element models for the shear fatigue test of asphalt mixtures were established using a DEM code called Particle Flow Code in Three Dimensions （PFC3D） to simulate the shear fatigue life of asphalt mixtures considering fatigue damage. Simulation results were verified by performing laboratory shear fatigue tests. Research results show that the shear fatigue test of asphalt mixtures can be simulated by 3D DEM considering fatigue damage， and a comparison of simulation and laboratory results shows relatively good agreement. Aggregate contact point and contact length significantly influence the shear fatigue life of asphalt mixtures， and the shear fatigue life increases with the increase in aggregate contact point and contact length. The contact orientation of aggregates on the horizontal cross section marginally influences the shear fatigue life， and the correlation between them is not obvious. Furthermore， aggregate size and void content significantly influence the aggregate contact characteristics in asphalt mixtures. Under the same test conditions， the number of aggregate contact points and total aggregate contact length increase with the increase in aggregate size， while decrease with the increase in void content.

To determine the deformations and internal forces of Continuously Reinforced Concrete Pavement（CRCP）with cracks by the analytical solution based on the state-space method and Euler-Bernoulli beam theory， a three-spring straight beam model was proposed. The vertical soil resistance and the rotation-resisting and shear-resisting capacities of the cracks were taken into account. In addition， the joint model which reveals the stiffness degradation was introduced to simulate the cracks. The analytical solutions for the internal forces and deformations of the pavement structure with cracks under vertical load were obtained by MATLAB programming. The analytical solutions were verified by the existing literature and finite element method. The influences of crack spacing， the magnitude of load and the surface thickness on the deformation response of the pavement structure were discussed. This also validates the analytical approach proposed in this paper.

To investigate the effect of different emulsifier contents on the stability performance of biomass emulsified asphalt， three types of emulsified asphalt with 1%， 3% and 5% anionic emulsifiers were prepared and analyzed by molecular dynamics simulation and macroscopic experiments. Firstly， the Molecular Simulation Software （Material Studio， MS） was used to construct a model of biomass emulsified asphalt with different emulsifier contents and the microscopic mechanism of the emulsifier was analyzed to improve the stability of the emulsified asphalt by the radial distribution function， interaction energy， interfacial layer thickness and solubility parameters of the emulsified asphalt system with different emulsifier contents. The results were validated by macro and micro tests including storage stability， particle size determination and infrared spectroscopy. The results show that at low emulsifier content， the emulsifier can reduce the interfacial tension between the oil-water interface and expand the transition region between the two phases （interfacial layer thickness）， which will prevent interparticle agglomeration and reduce the emulsion particle size， thus reducing the settling rate and ensuring the stability of the emulsion. When the emulsifier content is further increased beyond the critical micelle concentration， the emulsifiers will agglomerate with each other and show larger peaks in the radial distribution function， and the phenomenon of emulsifier agglomeration will appear in the five-day storage stability test， resulting in a corresponding decrease in the proximity of the infrared absorption peak area ratio in the same wavelength band of the upper and lower layers of the biomass emulsified asphalt， and the emulsion stability decreases instead.

In view of the deficiency of the existing research on the effect of gradation segregation on the water damage resistance of asphalt mixture， the effects of various water environment effects and different segregation degrees on the water loss resistance of asphalt mixture were studied based on the moisture sensitivity tester（MIST）. The results show that： the weakening order of water environment on splitting strength and marshall stability is "static immersion + MIST （long-term + short-term effect）" > "MIST （short-term effect）" > "static immersion （long-term effect）"； after the occurrence of gradation segregation， the order of splitting strength and Marshall stability is HC （heavy segregation of coarse aggregate） < LC （mild segregation of coarse aggregate） < D （no segregation） < LF （mild segregation of fine aggregate） < HF （heavy segregation of fine aggregate）； the effect of hydrodynamic environment aggravates the effect of gradation segregation on splitting strength and Marshall stability， and the effect of gradation segregation on water permeability coefficient is much greater than that of water environment.

In view of the large gap between the theoretical construction temperature of mixture determined by asphalt viscosity-temperature test curve and the actual construction temperature of mixture， theoretical analysis and experimental research were carried out on construction temperature determination method based on the mixing power test of asphalt mixture. Through a modified mixer， measuring the mixing power of the mixture at different temperatures and stirring speeds， the relationship between the three could be obtained and the viscosity model of the asphalt mixture based on the mixing power was established. The model shows that： the generalized plastic viscosity of the asphalt mixture depends on the temperature of the mixture， physical properties of the aggregate， gradation， type and amount of asphalt and other material factors， and has nothing to do with the mixer rotating speeds， which represents the rheological properties of the asphalt mixture. By measuring the generalized plastic viscosity of the asphalt mixture at different temperatures， a curve of the generalized plastic viscosity with temperature is obtained. The temperature at which the gradient of the curve change slows down significantly can be used as the construction control temperature of the asphalt mixture.

The non-compliant interaction between workers， construction machinery and construction appliances in the construction site is an important cause of safety accidents. Therefore， a visual relationship detection method based on construction scene is proposed. Firstly， convolution neural network is used to build entity detection and relationship detection branches to extract entity features and relationship features in construction scene. Secondly， visual module， semantic module and space module are constructed to learn the extracted features， so that the network can fully perceive and understand visual information， semantic information and spatial information. Finally， a graphical contrastive loss function is designed to improve the visual relationship detection performance. The experimental results on the self-made construction relationship detection data set show that the proposed method achieves the R@20， R@50， R@100 recall rate of 75.89%， 77.64% and 78.93%. The proposed method has good visual relationship detection performance， and can accurately detect the objects and their interactions in the construction scene.

Taking the accurate detection of asymmetric defects in small sample images as the research core， a small sample image asymmetric defect detection method based on meta learning is proposed. The adaptive filtering method of small sample image based on modulation is used to remove the noise of small sample image and optimize the quality of small sample image； Through the defect feature extraction method of filtered small sample image based on boundary detection， the defect feature of filtered small sample image is extracted； The extracted features are used as the detection samples of the small sample image asymmetric defect feature detection method based on improved meta learning to realize the small sample image asymmetric defect detection. The experimental results show that the proposed method has good filtering effect on small sample images. When detecting a variety of asymmetric defects， when the number of defect features of small sample images increases， the minimum value of intersection and union ratio of this method is 0.9， and the value of intersection and union ratio is ideal， which can accurately detect asymmetric defects of small sample images.

Aiming at the problem of random Gaussian noise in the process of remote sensing image acquisition caused by different lighting conditions， this paper studies the edge detection algorithm of noisy remote sensing image under different lighting conditions， so as to improve the accuracy of remote sensing image registration and recognition. Aiming at the random Gaussian noise in the process of remote sensing image acquisition under different illumination conditions， a windowed median filtering algorithm is constructed by combining the neighborhood mean filtering algorithm with the median filtering algorithm. The algorithm is used to filter the noisy remote sensing image to eliminate all kinds of noise conditions in the remote sensing image. For the denoised remote sensing image， the Canny operator is used to calculate the image gradient and constrain the maximum value. For the constrained gradient histogram of remote sensing image， the maximum interclass variance method is used to adaptively detect and connect the high and low thresholds of edges， and the edge detection of remote sensing image is completed by tracking edge pixels. The test results show that the algorithm can return the actual remote sensing image to the greatest extent， accurately detect the image edge information， and improve the accuracy of remote sensing image registration and recognition.

Aiming at the low accuracy of building segmentation and boundary blur in high resolution remote sensing images， an improved full convolution network is proposed based on SegNet network. Firstly， GELU， which performs well in deep learning task， is selected as the activation function to avoid neuron inactivation； Secondly， the improved residual bottleneck structure is used to extract more building features in the encoding network； Then， the skip connection is used to fuse the low-level and high-level semantic features of the image to assist image reconstruction； Finally， an improved boundary refinement module is connected at the end of the decoding network to further correct the boundary details of the building and improve the boundary integrity of the building. The experiment is carried out on the Massachusetts Buildings Dataset. The accuracy， recall and F1 score reach 93.5%， 79.3% and 81.9% respectively. The F1 score of the comprehensive evaluation index is about 5% higher than that of the basic network.

Due to the current single sensor being affected by its imaging performance， it is often difficult to fully reflect all the effective information in the ground object scene， resulting in the problem that the scene information is difficult to accurately identify， so a new infrared and low light image fusion algorithm based on VGG and low-pass filtering is proposed. Firstly， obtain the ground object scene information through infrared and low-level light detectors， use three-dimensional distribution， histogram comparison， and inversion to process the image， analyze the target characteristics of infrared and low-level light images， and study the spectroscopic mechanism of dual-frequency images； On this basis， the low-pass filtering method is used to decompose the infrared and low-light images to obtain their contour information and salient information. The contour part adopts the average weighting strategy for fusion， and the significant part adopts the VGG strategy for multi-layer fusion， and then the reconstructed image is fused； Finally， it is compared with the results of other algorithms， and the performance evaluation method is used to evaluate each fusion algorithm. The experimental results show that the algorithm can enhance the gray level of the scene information in the image， improve the brightness of the scene， and solve the problem of anti background interference of the scene information in the single frequency image.

The State of Energy （SoE） of lithium-ion battery is an important evaluation index for the energy optimization and management of electric vehicles. In order to improve the accuracy and reliability of battery SoE estimation， a SoE estimation method based on the combination of improved Back Propagation Neural Network （BPNN） and Model Predictive Filtering （MPF） was proposed in this paper. Considering the accumulated errors of current or voltage sensor noise， a first-order RC equivalent circuit model was established in this paper. Based on this model， the MPF algorithm was used to estimate the battery SoE. In order to make the estimated results more accurate， the improved BPNN was used to compensate the errors of the estimated results of MPF algorithm. Finally， the accuracy of the proposed method was verified under New European Driving Conditions （NEDC）. The results show that the SoE estimate based on the improved BPNN-MPF algorithm well converges to the real SoE value， compared with traditional MPF algorithm and BPNN-MPF algorithm. The Maximum Absolute Error （MAE） and Root-Mean-Square Error （RMSE） of the estimated values are all within 1%.

The process of driving a car can be affected by many factors. Different drivers have different operating habits and different driving ways under different road conditions. Therefore， in this paper， driving characteristics in snow and ice covered environment were analyzed. Based on Carmaker， the hardware-in-the-loop driving simulation platform， data such as driver behavior parameters and vehicle parameters were collected under this condition （including the bifurcated road surface）， characteristic parameters were analyzed and selected， and data sample database was established. The intention recognition model was established based on neural network， and the model was verified in single condition and compound condition respectively. The performance of the model was compared and analyzed. It can be obtained that the model can accurately identify the driver's intention on the bifurcated road.

In view of the complex problem of clearing and changing seeds in soybean plot breeding， a seed metering device was designed for soybean plot breeding with alternate post changing， the seed feeding and clearing processes of seed metering devices with different structural sizes were simulated by discrete element software. The rotation speed of seed metering plate， the lower limit angle and the width of seed box were taken as experimental factors， and the single grain rate， multiple grain rate and empty grain rate were taken as experimental indexes， and the experiment was carried out by using the experimental method of 3-factor 5-level quadratic regression orthogonal rotation combination design. The results showed that： the effect on single seed rate， rotation speed of metering tray， lower limit angle and width of seed box was extremely significant （P<0.01）；The effect on multiple seed rate， rotation speed of seed plate and lower limit angle was extremely significant （P<0.01）， and the width of seed box was significant （P<0.05）； The results showed that there were significant effects on empty particle rate， rotation speed of metering tray， lower limit angle and width of seed box （P<0.01）. The width of seed box was 9~15 mm and the lower limit angle was 92°~140°， when the rotation speed of seed tray was 45 r/min. The qualified index is more than 90%， the replay index is less than 5%， and the miss index is less than 5%. The results show that the simulation results are basically consistent with the theoretical results. The seed cleaning and exchange of the metering device is simple and efficient， and there is no residual seeds， which effectively solves the complex problem of seed cleaning and exchange.