Aiming at the thermal runaway problem of power battery during the use of electric vehicles， a power battery emergency thermal management system is proposed in this paper， which sprays high-pressure liquid refrigerant onto the surface of overheated battery to cool the battery in seconds. Experiments were conducted to analyze the effects of spray time， spray duty cycle and spray frequency on the cooling capacity of the system. The results show that when the spray time is 5 s， the average temperature drop of unit mass refrigerant is much lower than that of spray time is 6 s， which can be －41.94 ℃/kg， and the utilization efficiency of refrigerant is improved； when the duty cycle of pulse injection is 90%， the cooling capacity of the system can be enhanced by 10% than that of continuous spray， and the utilization efficiency of refrigerant also improves slightly； when the frequency of pulse injection is 2 Hz， the cooling capacity of the system exceeds that of continuous spray， and the cooling capacity of the system can be enhanced by 19.4%.

Aiming at the multi-objective optimization（MOO） problem of multi-point control strategy for auxiliary power unit（APU） of the range-extended electric vehicle， an energy management control strategy for APU based on MOO results is proposed. Firstly， the vehicle simulation model was established on AVL-Cruise and Matlab-Simulink software， and a MOO model was built with the system energy consumption， emissions and battery capacity attenuation rate as the objective functions based on NSGA-Ⅱ algorithm， the minimum continuous working time of the engine was taken as the optimization variable. In off-line optimization， Pareto optimal solution was obtained under the comprehensive objective. An real-time adaptive fuzzy controller was designed and the minimum continuous working time of the engine was adjusted online. Simulation results show that the proposed strategy can effectively balance the relationship among energy consumption， emissions and battery capacity decay rate， while effectively reducing energy consumption and emissions while maintaining a small battery capacity loss rate.

In the emergency braking process， the total wheel demand braking torque is calculated and then the regenerative-frictional braking torques are distributed by the distributed driven electric vehicle. However， it not only increases the control complexity， but also is failure to fully utilize the motor energy recovery potential. To improve the anti-lock braking control and energy recovery effects， an optimal sliding mode （OSM）-ABS control system was designed. The regenerative braking torque to achieve a maximum feedback power was regarded as one element of the disturbance vector， and the frictional braking torque was regarded as the only element in the control vector. The control characteristic of the OSM control algorithm is given full utilized， namely， the influences of the disturbance vector in the control system can be compensated by the reaching law solution. On the premise of ensuring the recovery effect of braking energy， the secondary distribution process of braking torque is omitted. The simulation results show that， compared with the general sliding mode ABS control strategy with a regenerative-frictional braking torque distribution process， the proposed OSM-ABS control strategy has satisfactory effects on anti-lock control.

Based on the current energy utilization status， taking into account the organic combination of solar energy and gas， the gas engine heat pump system with solar energy was proposed and thermodynamic model was established. Exergy for the components of the system was researched， and the conclusion was drawn. With the water temperature increasing， the exergy of the condenser and the evaporator increasing， the exergy of the compressor reduced， but the exergy of the engine and heat exchanger have little change hardly. With the change of engine speeding， the exergy loss of compressor becomes larger， and the exergy loss of condenser and evaporator also showed a similarity law， but other parts of exergy loss almost unchanged. There is little difference in the variation law of useful energy loss of gas engine heat pump with or without composite solar energy. However, after the solar energy is compound, the configuration power of the gas engine can be reduced, the gas quantity can be reduced, and the useful energy loss value of each component except the solar collector plate is reduced, which is of great significance for reducing the total energy consumption of the system.

The non-stationary characteristics of rotating machinery equipment increase the difficulty of predicting the operating state. Therefore, based on the neural network technology, a vibration frequency time series prediction method is constructed. Combining the gradient descent method and Newton method to optimize the back-propagation neural network， aiming at the seasonality and trend of the time series of actual mechanical vibration frequency， the first-order backward difference is processed by the difference method， the autoregressive sequence is deduced， and the time series prediction model of rotating machinery vibration frequency is obtained. In the experimental link， the vibration frequency time series within one hour is predicted for the rotor of a steam turbine generator unit in a power plant， and the experiment is completed on the basis of setting the parameters such as the number of network layers. From the absolute error and relative error values， the proposed method has the ability to reflect the vibration frequency trend， and the prediction accuracy is ideal.

This paper focuses on a recognition method of braking intention based on the test data of real vehicle and the support vector machine（SVM）. Brake pedal displacement， brake pedal force and braking deceleration were selected as the identification parameters of brake intention， and the braking conditions were divided into emergency braking， continuous braking and conventional braking. The vehicle test system of braking intention recognition was built， and several groups of brake tests were carried out to obtain the test data of identification parameters. The SVM model of brake intention recognition was constructed， and the RBF kernel function was selected as the SVM kernel function. Based on k-fold cross validation method and grid search method， the penalty factor C and kernel function parameter σ were optimized. Based on the actual test data， three braking conditions， gentle braking， conventional braking and emergency braking， were selected to verify the SVM model. The results show that the SVM model has high recognition accuracy of braking intention， which provides a theoretical basis for further application.

To solve the problems existing in geometric modeling and simulation calculation of shell aluminum foam model， based on the theory of Voronoi model， the modeling method and its implementation process of 3D solid aluminum foam model with random cell wall thickness are put forward. The model of solid aluminum foam is established by using ABAQUS foam aluminum modeling plug-in program written in Python language， and the quasi-static compression finite element simulation analysis of aluminum foam is carried out. The mechanical behavior， deformation mode and the calculated stress-strain curve of solid model and shell model in simulation are compared and analyzed. The results show that the simulation results and the deformation mode of the solid model are closer to the experiment， the stress-strain curve of the solid model is smoother and it has smaller fluctuations， the solid model is more reasonable than the model of the shell model in the compression process.

The VISSIM simulation software is used to analyze the influence range and driving risk variation rule of moving work zone on right lane of a three-lane urban expressway. The results show that the leftmost lane is only indirectly affected by the moving work zone when the traffic volume reaches 2000 pcu/h. The influence of moving work zone on the middle lane is controlled by the traffic volume and the speed of moving work zone. A range of 200 m behind moving work zone on the right lane is directly affected by moving work zone. When the traffic volume reaches 1600 pcu/h or lower， with the decrease of the traffic volume and the increase of the speed of moving work zone， the average driving risk of the road section decreases gradually. The research results can provide a scientific basis for the operation and management of urban road working vehicles.

In this paper， the volatility index of travel time based on bus headway is firstly proposed， and the method of dividing bus operating environment is established by using fast search and density peak clustering algorithm， which can be divided into low volatility， medium volatility and high volatility. The influencing factors of bus travel time between stations were analyzed， and the selection method of model input variable set based on embedding method was designed， and the deep neural network prediction method （CFDP-DNN） considering bus operating environment was established. In order to verify the effectiveness of this method， it was compared with SVM， ANFIS， BP neural network and other methods. The experimental results show that the correlation of CFDP-DNN prediction results is 0.9354， and the MAPE error range is 11%~22%， indicating that the division of traffic operating environment based on bus headway can effectively improve the prediction accuracy of travel time. The prediction method proposed in this paper can realize real-time and accurate bus travel time prediction between stations and provide theoretical support for bus dynamic scheduling.

To reduce traffic accidents in an environment where intelligent connected vehicle and non-connected vehicle are mixed， a cognitive distraction recognition model based on bi-directional long short-term memory（Bi-LSTM） with attention mechanism at unsignalized intersections was proposed in a mixed traffic environment. The driving simulator data of 60 drivers in the mixed traffic environment was collected and support vector machine recursive feature elimination algorithm（SVM-RFE） was adopted to extract the optimal feature subset as the input of the model. The results show that the recognition accuracy of the model is as high as 96.58% and the F₁-scores is 96.24%. Compared with SVM and decision tree distraction recognition models， this model has the best performance in terms of accuracy， recall， the F₁-scores and the ROC curve. The model can be applied to the autonomous driving distraction alarm assistance system， which is of great significance to improving road safety.

A theoretical model is proposed for the wheel-load diffusion to investigate the transmission mechanism of vertical loading on the steel-concrete composite bridge deck system under local wheel loads. An existed full-scale test model was referred and the corresponding finite element numerical simulation were also conducted to verify the presented model and analysis the key parameters. The results show that， rather than the traditional assumption of 45 degree diffusion effect， the pressure on the steel top plate of the composite deck under vehicle loads has a significant local effect along the transverse direction of U-rib， and the sharing law of the wheel load could be obtained from the 3-span beam simply supported by the webs of U-rib. The paramelric analysis shows that， the proposed theoretical model is conservative in the common value range of engineering parameter of orthotropic steel-concrete composite bridge deck， and it could be referred to the design of orthotropic composite bridge deck.

The pseudo-static test was conducted to investigate the seismic behavior of earthquake-damaged precast concrete（PC） frame structures （the earthquake-damaged frame structures are within moderate damage） strengthened with buckling-restrained braces（BRBs） and to analyze the hysteretic energy dissipation of BRBs. A half-scale PC frame structure was fabricated. After it was pre-damaged by quasi-static loading， it was strengthened with BRBs， and then tested under pseudo-static loading again. The structural experimental phenomena and failure characteristics were observed and recorded. The structural seismic behavior， including bearing capacity， ductility， and energy dissipation， was evaluated. The test results show that after repairing with BRBs， the bearing capacity and energy dissipation can be significantly improved （the ultimate bearing capacity and energy dissipation increased by 51.3% and 68%， respectively）， other seismic behavior like the ductility coefficient can be recovered well （ductility coefficient restored to 88% of the initial ductility of the structure）； the earthquake-damaged PC frame structure strengthened with BRBs under pseudo-static loading exhibited a failure mode of beam-hinge with BRBs. The earthquake-damaged PC frame structures still had retrofit value and that the above retrofitting method was effective.

A scale model of a large span structure with a circular hyperbolic-paraboloid roof was designed and made， and an impinging jet device was used to simulate the downburst. The test was designed to obtain the roof wind pressure coefficient and explore the influence of radial distance and wind direction angle on the wind pressure of circular hyperbolic-paraboloid roof. Then， by changing the width of the building， the oval hyperbolic-paraboloid roof was designed and made under the same conditions. The similarities and differences of wind pressure on the two roofs were compared and analyzed. The conclusions are as follows： ①The maximum wind pressure values of the high points diagonal of the circular hyperbolic-paraboloid roof first increase and then decrease with the increase of the radial distance. When the wind pressure is 1.25D_jet， the wind pressure reaches the maximum， and the wind pressure coefficient is -0.81. ②Wind direction angle has a great influence on the wind pressure of low point areas of circular hyperbolic-paraboloid roof. When airflow frontal hits the low point area， the wind pressure on the windward area increases rapidly， and the change gradient of the wind pressure is very large. ③Due to the difference of lengths and widths， the wind pressure of the area which is 1/3L from roof center on high point line and the low point areas of the oval hyperbolic-paraboloid roof is slightly greater than that of the circular hyperbolic-paraboloid roof. These areas need to be strengthened during structural design.

Emotions can be recognized through the hidden patterns in the EEG signals. The large number of EEG features extracted based on numerous EEG channels makes the task of emotion recognition very complex. To solve above problems， An EEG emotion recognition algorithm (SEE) based on three-stage feature selection strategy is proposed. EEG features were systematically extracted from time domain， frequency domain and spatial domain in this study. Based on the extracted EEG feature set， the features with no significant difference between classes were removed by t-test firstly， and then the recursive feature elimination strategy was used to select target related features. Finally， the final feature set was determined by the sequential backward feature selection strategy for the emotion recognition. The experimental results show that the model constructed in this study has better emotion recognition ability than other methods. Compared with the existing feature selection algorithms， SEE can filter out better feature subset with a low time complexity. In addition， emotion-associated EEG channels and frequency bands were detected. The experimental results show physiological significance of emotion， which may facilitate the development of emotion-targeted EEG devices.

Aiming at the problem that the traditional particle swarm optimization（PSO） algorithm has premature convergence and unable to find the global optimal solution in solving the optimization problem, a particle swarm optimization algorithm based on kinship selection is proposed, which improves the global search ability of the algorithm. In addition, the communication mechanism of multiple populations and the elimination mechanism between each subpopulation are introduced, which effectively avoids individuals falling into the local optimum in the process of optimization. In the experiment part, the single objective optimization function set is compared with the traditional particle swarm optimization algorithm and the results of some competitive algorithms. obvious advantages; then, the new algorithm is applied to the optimization problem of truss dome, and compared with the traditional particle swarm optimization algorithm, a feasible solution to this practical problem is obtained.

In order to solve the problem that supervised learning needs a lot of time to complete data set annotation in the field of image classification， a self-supervised image classification framework， SSIC framework， is proposed. SSIC framework is a self supervised learning method based on contrastive learning， which has better performance than the existing unsupervised methods. A new framework is designed and a more effective pretext task is selected to improve the robustness of the model. In addition， a targeted loss function is proposed to improve the performance of image classification. experiments was conducted on UC Merced， NWPU and AID data sets. Experimental results show that SSIC framework has obvious advantages over the latest technology， and it also performs well in low resolution image classification.

In this paper， a novel convolutional neural network（CNN） for multi-focus image fusion is proposed. Compared with existing image fusion methods based on CNN which decompose the source image into several patches and adopt a classifier to estimate whether the patch is focused or defocused， the method in this paper directly converts the whole image into a decision diagram. The pixel-level regression strategy can make use of the complementary information and address the difficulty of estimating blur level around the focused/defocused region. Furthermore， the ringed residual network（RResNet） block is utilized to extract more semantic information from the focused region in image fusion field. In the meanwhile， the structural similarity index（SSIM） loss is utilized to estimate the structural similarity between the generated fusion image and the ground-truth reference to improve the quality of the fused images， and the edge preservation loss function is applied to preserve more gradient information from source image. Experimental results demonstrate that the proposed method is superior to other fusion algorithms in subjective visual effect and objective assessment.

The mean distance ordered partial codebook search（MPS） algorithm is a fast codebook search algorithm for image vector quantization. However， in order to find the initial matching codeword， the MPS algorithm needs to calculate the mean squared distance between all vectors and all codewords. For a codebook with n vectors and length k， the calculation amount of this part is O(nk+k\mathrm{l}\mathrm{o}\mathrm{g}k)， which limits the acceleration effect of MPS. To solve this problem， an image codeword fast search algorithm based on neighborhood similarity was proposed. The algorithm first sorts the original codebook from small to large according to the codeword components and values to obtain the sorted codebook. Then， the candidate initial matching codeword is determined in the neighbor vector of the current image vector， and the final initial matching codeword is determined by distance comparison. Finally， a codeword search based on the sorted codebook is performed with the initial matching codeword as the starting search point. The algorithm reduces the calculation amount of initial matching codeword selection in the MPS algorithm to O(n+k\mathrm{l}\mathrm{o}\mathrm{g}k)， and has the same results as the full search algorithm and the MPS algorithm. The results of comparative experiments on the test images show that the FSNS algorithm has the highest speedup ratio， and the average time speedup ratio ranges from 4.38~11.24， while the MPS algorithm and ITIE algorithm are 3.19~6.01 and 1.49~2.99.

The dispute focus is the focus of dispute between the plaintiff and the defendant， which is the main line and hub of leading the trial and settlement of disputes. Accurate and rapid induction of the focus of disputes is conducive to improve the quality and efficiency of the trial， and achieve the effect of supporting the construction of 'intelligent justice'. An innovative end-to-end model was proposed to solve this problem. Based on deep neural network， this model deeply understood the semantic information of the text between both parties. By combining word level and sentence level information， this study carried out sentence level contradiction detection， classification， and complete paragraph level contradiction classification. Through certain rules， this method combined the results of the two parts， finally identified all the dispute focuses in the pleading text. Experiments on real datasets show that the proposed model can identify the focus of dispute between the plaintiff and the defendant accurately and quickly. The recognition accuracy is improved effectively compared with the existing methods.An effective new path is proposed for the intelligent identification of the dispute focus of the defense text.

Text input is the most common interaction behavior in viture reality（VR） environment， and the mainstream text input is currently realized by laser pointing. However， the existing methods have many drawbacks， such as low efficiency， large jitter， and easy false trigger， which cannot be used to frequently input text in VR environment. Therefore， a novel text input method for VR environment is proposed. First，First, partition the keyboard, use the handle to select the area where the characters are located， and use the word disambiguation algorithm to realize text input in units of words； secondly， perform cluster analysis on the user's click coordinates to do one-key multi-word processing； Finally， three keyboard layouts that conform to user habits are designed， and the optimal layout is determined The experimental results show that the typing speed of the optimal layout is 13.44 WPM（Words Per Minute） with an accuracy of 92.26%， which is a significant improvement compared with other input methods.

In unbalanced text classification， the classification results tend to the majority and ignore the minority， which leads to poor classification effect. The unbalanced text classification method based on deep learning is studied. DA method is used to select unbalanced text features. DA method sets the scoring standard to the minimum value of the difference of document probability correlation， so that the selected text features are evenly distributed in most classes and a few classes to improve the balance of text features. The subset obtained by feature selection is used as the input of the depth belief network composed of multiple constrained Boltzmann machines. The constrained Boltzmann machine obtains the optimal probability distribution of training samples through pre training. The weight of the constrained Boltzmann machine is determined by contrast bifurcation algorithm. After the parameters of the constrained Boltzmann machine are set， the greedy algorithm is used to train the constrained Boltzmann machine iteratively until the whole process is completed text classification. Experimental results show that this method can effectively classify unbalanced text， and the classification accuracy is more than 99.5%.

In order to improve the effectiveness， a time-varying grid model is proposed to divide the spatial and temporal dimensions of the region according to the satellite earth observation width and the maximum vessel speed. The vessel track prediction problem is transformed into a time-varying grid transition probability prediction problem， which effectively reduces the complexity of multi-step prediction. The multi-step time-varying grid transfer prediction with high accuracy is realized by the improved sequence to sequence（Seq2Seq） model， which has learned a large number of historical tracks in the research area. A satellite observation task planning algorithm for time-varying grids is designed and simulation experiments are carried out based on actual AIS data and satellite information. The results of experiments show that the Seq2Seq model based on deep learning has high accuracy in multi-step prediction of time-varying grids， which effectively improves the effectiveness of vessel search by earth observation satellites.

In order to solve the problems of poor accuracy of multi-mode radar signal sorting，a new multi-mode radar signal sorting algorithm based on potential distance graph combined with PCA and improved cloud model through the research of machine learning algorithm is proposed. Firstly， after eliminating the interference points，the potential distance graph is used to cluster. Signal samples can be assigned to each cluster center point after only one traversal without iteration. Then principal component analysis （PCA） is used to reduce dimension and extract key factors of each feature to form new features. Finally， the improved cloud model is used to analyze the new features. The average membership degree between classes is obtained. The evaluation criteria are established to complete the final classification of radar signals. Compared with the existing algorithms， simulation results show that this method has higher accuracy and strong anti-interference ability. To a certain extent， it can solve the problem of wrong sorting of multi-mode radar signals.

In order to simplify the bandwidth allocation process and improve the bandwidth utilization， a bandwidth compensation algorithm is proposed based on effective bandwidth and effective capacity. Aiming at the scene of mixed service arrivals in the communication networks， the bandwidth compensation service mechanism is designed， and the network queuing system with multi-service arrivals and random compensation service is established for analysis. In the model， the arrival process of mixed services is described by the aggregation of different random processes， and the bandwidth compensation service process is composed of basic service and random compensation service. The research evaluates the bandwidth requirements of mixed services under the constraint of QoS（Quality of Service）， and further deduces the compensation bandwidth and compensation probability required to guarantee QoS. It is verified by Matlab simulation that the proposed bandwidth compensation algorithm could guarantee the delay QoS requirements， and the influence of delay QoS parameters and arrival process parameters on compensation algorithm is analyzed.

Aiming at the problems of redundancy and low registration accuracy of the point cloud data， a bidirectional nearest neighbor distance ratio registration method based on supervoxel is proposed in this paper. Firstly， the target feature points with stable structure are extracted by the supervoxel， and a non-iterative threshold denoising method based on point cloud thickness stratification is proposed； then， using FPFH for feature description， and a bidirectional nearest neighbor distance ratio method is proposed to register the point cloud； finally， an accurate point cloud registration method based on two-level threshold is proposed. The standard database model is used for simulation analysis to verify the effectiveness of the algorithm. The results show that the proposed method in this paper can effectively eliminate the drift noise voxels， and the algorithm has high registration accuracy and strong robustness. Compared with other methods， when the registration time is close， the registration accuracy of this algorithm is improved by 74.2%； When the noise ratio is 6% and 10%， the registration accuracy is improved by more than 67%.

In this paper， a novel nonlinear control strategy based on the nonlinear disturbance observer is developed to compensate the unknown ground effects during the landing procedure. Owing to the intricacy of the ground effects， it is very hard to obtain the precise dynamic model for the UAV's landing procedure. To solve this issue， a nonlinear observer is designed to estimate the unknown disturbance introduced by the ground effects. Then the fast terminal sliding mode control method is combined with the disturbance observer to formulate a new nonlinear robust landing control strategy which is able to suppress the unknown ground effects and drive the quadrotor to its desired landing point accurately. The Lyapunov based stability analysis is employed to prove the stability of the closed loop system， and the finite-time convergence of the UAV's altitude control error together with the disturbance estimation errors are guaranteed. Real-time flight experimental results are presented to show the good landing control performance of the proposed control strategy.

A new method of bipolar field-effect control on DC electroosmosis（DCEO） is proposed from the perspective of electrodynamics. A mathematical model is established in thin-layer approximation and low-voltage limit to verify the feasibility of the bipolar DC-FFET structure for fluid electrodynamic manipulation at the micrometer scale. An integrated device design with two sets of side-by-side reverse polarity gate electrode pairs was used to construct a micro-device model for fully electrically driven analyte processing. By using a smaller gate voltage， an almost perfect liquid mixture can be obtained， resulting in less adverse effects at a small Dukhin number. This technology has great potential in the development of fully automatic liquid phase actuators in modern microfluidic systems.

Aiming at the problem of the current common crawler harvester harvesting first-season regenerative rice with high crushing rate and no mature regenerated rice first-season harvester products， combined with the agronomic requirements of the first-season harvesting of regenerated rice， a triangular crawler-type regenerative rice harvesting was designed. The whole structure and working principle of the chassis of the triangular crawler reclaimed rice harvester are described， the hydraulic walking system is designed， and a theoretical analysis on the performance of chassis is carried out. Field tests were carried out on the chassis of the triangular crawler regenerated rice harvester. The test results showed that the speed range of the triangular crawler regenerated rice harvester chassis in the field is 0~4.5 km/h， and the speed range of paddy field operation is 0~2.8 km/h. The minimum turning radius of paddy field driving is 1780 mm， the maximum climbing angle is 32°， and the maximum ridge height is 215 mm， which can meet the requirements of field walking in the harvesting season of regenerated rice. The actual rolling rate in the field is 31.7%， which is 21.5% lower than that of the common crawler harvester.

A sensor array optimization strategy based on CatBoost algorithm was proposed. Using the self-developed electronic nose testing system based on bionic olfaction，the residual trichlorfon on dandelion was detected，the response characteristic information of the dandelion sample was extracted， and the multi-characteristic data fusion on the sensor array was performed. The CatBoost algorithm was used to perform feature selection on the data matrix. The number of optimized sensors was reduced from 12 to 3， the accuracy rate was increased from 91.69% to 98.03%， and the number of features was reduced by 88%， which was better than correlation coefficient， recursive elimination and other commonly used algorithms. The problem of multiple sensors and data redundancy was solved， and the detection accuracy was greatly improved. The results show that the use of CatBoost algorithm in the detection of trichlorfon residues on dandelion can improve the identification ability of the electronic nose.