Abstract:In order to optimize the performance of the fuel cell system for vehicles and decrease the fluctuation of the DC bus voltage, proposes a hierarchical energy management strategy based on adaptive moving average filter, equivalent cost minimum and nonlinear control of the energy state of super capacitors. Firstly，construct the fuel cell vehicle power system model and the mathematical model of hydrogen consumption and degradation cost; Then, the adaptive moving average filter and the equivalent cost minimum strategy are used to optimize the output power of fuel cell ; And the super capacitor energy state is controlled by a nonlinear control strategy within a reasonable range, to improve the dynamic power output capability of the super capacitor and suppress the fluctuation of the DC bus voltage. The simulation experiment results show that compared with the power following strategy, the hydrogen consumption is reduced by 12.94%, the durability of the fuel cell is improved by 12.6%, the total cost is reduced by 12.63%, and the DC bus voltage fluctuation is significantly decreased, which indicates that the proposed energy can optimize the hydrogen consumption and durability of the fuel cell, and can improve the stability of the bus voltage.

Abstract:In order to solve the problems of slow speed, low efficiency, and CPU computing resources in the traditional SM4 encryption and decryption methods, a reconfigurable high-speed data encryption system is proposed. The system is based on Xilinx Virtex UltraScale VU9p FPGA, using PCIe hot-swappable features, can be quickly applied to office hosts or servers, fast data transmission through PCIe high-speed interface, parallel and schedulable SM4 algorithm logic in FPGA, and a dedicated DMA design The module realizes bypassing the host CPU to transmit plaintext ciphertext, reducing the resource occupation on the host side; the encryption and decryption system implemented by FPGA is reconfigurable, which greatly reduces the hardware cost of algorithm iteration. System analysis, testing and experimental results show that the system achieves high-speed and reliable data transmission and encryption, and the bus rate reaches 8GT/s, which can effectively meet the needs of fast encryption and decryption of large-capacity data; it adopts parallel schedulable pipeline encryption and decryption, which is better than traditional software. In this way, the encryption and decryption rate is increased by approximately 25.78 times.

Abstract:Voice coil motors (VCM) have higher requirements for the accuracy and robustness of the control system. Aiming at the problem of non-linear electromagnetic characteristics caused by changes in motor parameters and mechanical friction in its work, the principle analysis of voice coil motors and the establishment of corresponding mathematical models are carried out, and the unmeasurable and unknown interference in the mathematical model of the system is defined as the total interference , Utilize the nonlinear extended state observer (NLESO) to estimate the system position and the total disturbance, and the estimated total disturbance is fed-forward to compensate the control loop, thus a non-linear extended state observer is proposed. The closed-loop complementary sliding mode position control algorithm. According to the deduced mathematical model, a system simulation model is established in MATLAB/simulink. The simulation experiment shows that the NLESO+CSMC solution can reduce the steady-state error by up to 95.5% compared with the traditional sliding mode control, and the full tracking time can be reduced at the maximum. It is as small as 0.158s, and it still maintains good tracking performance after changing the desired signal, which improves the accuracy and robustness of the control system.

Abstract:This paper is aimed at the problem of real-time fault diagnosis of the aircraft dynamic acquisition integration, timing module fault fluctuations, and the aid of flying testivity verification integrated platforms, based on quantum neural network and trend The self-correction safety warning model designed a real-time flight safety monitoring system that adapts to a multi-machine type, which uses nine different functional modes to compare analysis. Simulation effects are integrated with three evaluation criteria, and the simulation results show this article. Flight Safety Integrated Monitoring Systems regardless of output accuracy (less than 10% of the error), or the stability of the prediction effect is superior to the market civilian FDMS system.

Abstract:Aiming at the characteristics of unmanned formula racing cars requiring precision and real-time performance of the sensing system, a target detection system based on the fusion of lidar and camera information was designed. The sensor perception model is established. The visual information uses the yolo v4 image recognition algorithm. The distance information is obtained by filtering and clustering the point cloud. The tilt of the point cloud is corrected by a rotation processing algorithm based on ground fitting. Coordinates and point cloud coordinates adopt Euclidean distance strategy for target information fusion. The test results show that the radar algorithm can complete three-dimensional target detection. The average accuracy of the yolo v4 (you only look once-4) image recognition algorithm for traffic cone detection is 97.5%, the average pixel error rate in the direction is 1%, and the target detection The system meets actual driving requirements in terms of accuracy and real-time performance.

Abstract:Aiming at the defects of the existing power grid micro meteorological measurement methods, a new micro meteorological measurement method based on MEMS is proposed by using the relationship between wind speed and wind pressure, and its feasibility is verified by experiments. The experimental results show that when the wind speed is lower than 6m / s, the error is greater than 2m / s. when the wind speed is greater than 6m / s (soft breeze), the accuracy of the design is high, and the error does not exceed 2m / s, which can meet the needs of use under strong wind conditions. The structure of the wind speed sampler is studied. The research shows that the structure of the wind speed sampler has a great impact on the accuracy of the collected wind speed data. The structure of the wind speed sampler is studied through experiments, the data obtained by using a silica gel sampling tube with an aperture and wall thickness of 3×5mm and a length of 30cm is the closest to the wind speed sampling device used for comparison.

Abstract:Aiming at reducing the bistatic radar cross section and increasing the stealth function of the antenna system, a new two-dimensional hypersurface design method with low absorption and high permeability is proposed. The proposed frequency selective rasorber (FSR) structure is composed of the resistance layer on the upper layer and the frequency selection layer on the lower layer. Two layers of substrate are isolated by nylon bolts and nylon columns to ensure a distance of 11 mm between both substrates. The two layers are designed with the zhong-type structure. In this paper, the principle of equivalent circuit absorption and wave penetration of FSR is analyzed, and the absorption effect of FSR with different resistance values is checked by parameter optimization. By analyzing the current distribution at different frequencies, the penetrating and absorbing functions are explained from the physical level. The simulation results show that the -3dB bandwidth of the transmission band ranges from 9.6 GHz to 12 GHz. The insertion loss is only 0.15 dB at the center frequency of 10.7 GHz of the passband, and the absorption bandwidth ranges from 3.84 GHz to 7.84 GHz, with the relative bandwidth reaching to 69%. The FSR has a smaller unit size, i.e. the unit size is 0.026λ*0.026λ.

Abstract:In the process of overhead cable installation and construction, the tension monitoring of steel wire rope used for tower lifting is of great significance to construction safety. Therefore, a tension detection device based on the principle of three-point bending method is designed. The single chip microcomputer system in the device amplifies the signal obtained by the pressure sensor into the computer and converts it into electrical signal output to realize monitoring. However, there must be the interference of nonlinear factors in this method. Through experimental calibration and error compensation, the error of the device can be controlled within 2%; In practice, it is found that the measurement error caused by ambient temperature can not be ignored. Therefore, temperature compensation is carried out through temperature calibration test. The results of practical engineering data show that the measurement error can meet the engineering needs.

Abstract:Global satellite navigation system is a huge and complex system. If the test verification and ground operators’ training are carried out directly in the real environment, there may be a huge risk in addition to the huge system resources. By establishing the high-fidelity model of space segment, ground segment and user segment of the satellite navigation system, developing the system-level function and performance evaluation algorithm, and realizing the system indirect port simulation defined by ICD file, developing a global satellite navigation system level simulation and analysis software to support “BeiDou III” -- BDSim. This paper focuses on BDSim as a system-level test simulation platform for system testing and evaluation and ground operator training. Firstly, the method of constructing satellite navigation high fidelity model system is proposed, and the model system of BDSim is given. Then, according to the characteristics of “BeiDou III” step-by-step implementation and stepwise networking, the system level test and evaluation method of integrating heaven and earth with "virtual and real integration" is proposed. Finally, according to the characteristics of high complexity and high accuracy of “BeiDou III” ground segment, an immersion operators’ training and assessment method based on BDSim high-fidelity model system and system real data drive is proposed. All relevant methods have been effectively applied in engineering practices. The results show that the methods proposed in this paper can provide technical support for the joint debugging test of each subsystem of the satellite navigation system, the functional and performance test of the system level, and the training and assessment of ground operators, which is of reference significance for the construction of “BeiDou III” related projects.

Abstract:The guide wheel is an important part of the hydraulic torque converter, and its 3D model plays an important role in the study of the performance parameters of the guide wheel. In order to solve the problem of detecting and reconstructing the complex guide wheel blade, the 3D point cloud data of the guide wheel is obtained by using the line laser scanner, the point cloud is processed by noise reduction using K-D Tree combined with bilateral filtering algorithm, the point cloud is refined by introducing the Mean Shifting clustering algorithm based on Gaussian sphere on the basis of k-means clustering algorithm, the refined data is reconstructed by surface reconstruction and The proposed method is used in conjunction with CATIA software to reconstruct the 3D model of the hydraulic torque converter guide wheel. The results show that the proposed method has good filtering and noise reduction effects, and the streamlined point cloud not only preserves the geometrical features of the guide wheel, but also improves the computational efficiency of the reconstruction algorithm. Analysis of the alignment with the point cloud obtained from the contact measurement with an error of ±0.1mm, with the number of point clouds greater than the threshold deviation not exceeding %5 of the overall point cloud.which can obtain a three-dimensional reconstruction model that meets the accuracy requirements and provides a basis for the three-dimensional smooth numerical simulation of the guide wheel and the optimization of performance parameters.

Abstract:Aiming at the problems of weak fault feature information and low recognition rate of diesel engine system, a fault recognition method of key components of diesel engine based on multi feature extraction and  kernel entropy component analysis (KECA) is proposed. Firstly, the collected signal is reconstructed and denoised by ensemble empirical mode decomposition, and then the variance, kurtosis, square root amplitude, peak factor and arrangement entropy are extracted as the characteristic parameters, which are reduced by KECA. Finally, support vector machine is used for fault identification and classification, and the classification results of other dimensionality reduction methods are compared. The results show that the classification results of this paper are obviously better than the other two, and the correct rate of fault identification is 96.67%, which shows that this method can effectively diagnose the fault of key components of diesel engine system, and has a good application prospect.

Abstract:Aiming at the problem that the traditional performance prediction method does not treat the importance of each attribute feature to student’s scores and the low completion rate of student’s online learning differently, a convolutional neural network of relational graph and gated recurrent unit integrated into the attention mechanism（AR-GCN-GRU）score prediction method for students is proposed.The integrated attention mechanism is used to capture the relationship attribute characteristics between students, and at the same time, extract the important attribute characteristics of students and visualize them,And the method integrates the advantages of the convolutional neural network of relational graph (R-GCN) and the Gated recurrent Unit (GRU),and can not only capture the internal correlation between nodes,but also extract the representative information of students' behavioral attributes well.The model was contrasted and ablation experiment on a public data set,The F value and accuracy of the model reached 99.00% and 99.73%, The experimental results show that the method has been significantly improved than other algorithms, and the effectiveness of the attention mechanism is verified.

Abstract:Accurate simulation of wind turbine radar echo signals and analysis of its Doppler characteristics are of great significance to solving the problem of passive interference from wind turbine to radar stations. Aiming at the problem that existing methods can only consider monostatic scattering, this paper derives the bistatic echo formula of the wind turbine blade in the presence of ground based on the multipath effect. The equivalent calculation model of the scattering paths between the wind turbine blade and the ground is deduced based on the principle of mirror image and the simulation method for the bistatic echo of wind turbine blade in a half-space is proposed. In the case of the Vestas-V82 wind turbine, the wind turbine blade bistatic echo in the presence of ground is simulated and compared with the monostatic echo. The results show that when the position of the transmitting radar and the receiving radar are different, it will cause the echo to increase a set of Doppler flicker and reduce the maximum Doppler frequency of the echo. It provides a reference for the identification of wind turbine radar echo signals.

Abstract:Non maximum suppression algorithm(NMS) is the main algorithm to select the accurate positioning box in object detection. The algorithm only takes the classification score as the standard, which may remove the prediction frame with low score but accurate positioning, and is more unfriendly to the situation with occlusion. A-NMS method is proposed, which integrates the attention mechanism into the non maximum suppression algorithm, and adjusts the final score of the box by combining the position information with the score information of the box. In addition, an improved distance based intersection union ratio loss function is proposed, the loss term is redefined, and it is introduced into non maximum suppression to calculate the intersection union ratio between frames instead of IOU. Finally, the two improved algorithms are integrated into three classical target detection. The above two algorithms are verified on Pascal VOC 2012 and MS-COCO 2017 data sets. The results show that the detection accuracy has been improved by 1% ~ 2%.

Abstract:Aiming at the problem of three-dimensional measurement of the inner surface of deep-hole parts, this paper proposes a new three-dimensional reconstruction algorithm for the inner surface of deep-hole parts. This method uses a telecentric lens and a multi-line structured light generator to obtain structured light images of the inner surface of deep-hole parts, combined with the morphological characteristics of deep-hole parts, establishes a three-dimensional projection model of a telecentric lens based on a cylindrical coordinate system, calculate the radius of the measurement point with the offset in image, and the 3D reconstruction point cloud of the measurement image is obtained on this model. The experimental results show that the root-mean-square error of this method is within 0.03mm in the measurement of a deep hole with a diameter of 155mm, and the measurement accuracy meets the requirements of the system, which provides a data basis for accurately analyzing the state of the inner surface of deep-hole parts.

Abstract:Analyzed the changes to the Ultra-Wide Band (UWB) physical layer of the latest IEEE 802.15.4z protocol, and designed three types of Low Frequency Pulse Ultra-Wide Band (LRP UWB) based on MATLAB Working mode and high-frequency pulse ultra-wideband (High Frequency Pulse Ultra-Wide Band, HRP UWB) communication ranging integrated system, and under the conditions of 802.15.4 CM1 channel simulation and comparison of the bit error rate of each system (Bit- Error-Rate, BER) and ranging accuracy to analyze the UWB system performance characteristics under different operating modes. The bit error rate simulation result is that LRP UWB is lower than HRP UWB, and the LRP long-range mode has the lowest bit error, and the signal-to-noise ratio is 10-4 above 15, indicating that the long-range mode communication link is the most stable. The range simulation results show that the three modes of the LRP UWB system are comparable to the HRP UWB system in terms of range performance within 20m, with an accuracy of 0.1m; while the LRP UWB long-range mode has a better range accuracy than 0.1m over 100m, and the other modes have an accuracy of 0.5m or more, indicating that the long-range mode is more suitable for long-range range. Finally, the range measurement capability of the pulse accumulation ranging system was designed and simulated, and its range accuracy was better than 0.4m over a distance of 100m, which is more than twice the accuracy of a single pulse in long-range mode, indicating that the use of pulse accumulation ranging technology can further improve the system's long-range range measurement capability.

Abstract:The common decoding algorithm for TPC codes (Turbo Product Code) is the Pyndiah-Chase-II algorithm, but the Pyndiah-Chase-II algorithm involves a large number of sorting operations, complex branching structures and storage scheduling in the process of searching for the least reliable input bit positions and shortest Euclidean distance code words making it very unfavorable for integrated circuit hardware implementation. In order to solve these problems, proposing a TPC decoding algorithm based on probabilistic computation, the algorithm includes information input layer, random bit stream generation layer, BCH hard judgment layer, BCH&CRC check layer, and output layer, and the sub-code of TPC code adopts BCH code, program design of decoding algorithm and simulation of decoding performance and decoding delay by MATLAB software. The simulation results show that the decoding algorithm can achieve the same decoding performance as the traditional Pyndiah-Chase-II algorithm, and it only needs two iterations on average to achieve correct decoding, which can effectively reduce the decoding delay. Finally, the FPGA-based hardware design is completed. The BCH hard judgment layer is implemented by the lookup table method, and the logic structure of other layers is simple and all are gate-level operations, so it can significantly reduce the hardware overhead and power consumption, and is easy to implement with integrated circuits.

Abstract:The radar cross section (RCS) is an important physical parameter to characterize the scattering characteristics of targets. This work aims at the problem that the typical fluctuation models have low precision in describing the RCS distribution characteristics of stratospheric aerostats, and the Gaussian mixture model is applied to fit the dynamic RCS distribution for aerostats. Firstly, the line-of-sight (LOS) angle for radar in aerostat’s body coordinates is formulated. Secondly, a superposition of several Gaussian models is used to characterize the probability density distribution of aerostat’s RCS, and the model parameters are approximated via the expectation maximization algorithm. Finally, the fitting effect of Gaussian mixture model on RCS distribution is studied and tested with some aerostat's RCS measurements and compared to typical fluctuation models. The results show that the Gaussian mixture model can improve the fitting effect of RCS probability distribution by up to 96.87% compared with other typical models under the least square criterion, thus demonstrating the effectiveness of Gaussian mixture model.

Abstract:In order to quickly obtain the complete feature information of the human pulse signal, and quickly and accurately identify the correlation between the pulse feature information and the human disease.The study uses the time-domain characteristics of multi-period pulses and the Hilbert-Huang-Transform(HHT) based on Ensemble Empirical Mode Decomposition（EEMD） to obtain the instantaneous frequency and amplitude as the frequency-domain characteristics. The time domain and frequency domain features as input fused convolutional neural network to identify and classify the pulse characteristics of the human body.The pulse signals of three clinical symptoms were obtained from the MIT-BIH standard database for experimental analysis. Finally, through experiments, the accuracy of pulse feature recognition and classification is 91.88%. Using EEMD-based HHT as a supplement to time-domain features, time-frequency feature mixing can make the PPG pulse signal complete characterization, and perform classification experiments on the convolutional neural network to obtain better classification results. Methods willing clinical diagnosis of intelligent development, improve the accuracy and efficiency of clinical diagnosis to provide a good role in promoting.

Abstract:For the purpose of accurately and efficiently extracting grid related value information from mixed big data, an automatic power grid data summarization algorithm based on long and short term memory network and artificial colony optimization algorithm is studied. Design the contextual information of the bidirectional LSTM learning target words, increase the attention mechanism, and extract the electric power category words and terms. The conditional random field model performs training tasks on embedded sequences to predict whether sentences can be classified into the category of electricity. With the support of the improved artificial clustering optimization algorithm, the problem of extracting power abstracted from big data is optimized, and the most valuable power related data is determined from the mixed big data. The proposed algorithm is validated based on actual grid data, and the results show that the proposed algorithm achieves good results.

Abstract:Aiming at the phenomenon that the wireless signal penetrates the wall under the influence of non-line-of-sight factors in indoor positioning, the positioning is offset. Because the radio signal propagation has the characteristics of strong volatility and easy interference, the use of strong anti-interference, high penetrating power and Ultra-wideband technology with excellent ranging value collects measurement data. In this paper, the signal through-wall derivation formula is combined with the wireless signal logarithmic attenuation model to process the collected distance value data. The attenuation model reduces the energy loss of the signal when passing through the wall, improves the accuracy of the ranging value, and reduces the measurement by the least square method. Accumulated error of distance to improve positioning accuracy. Through the test of actual experimental scenes, it is concluded that the positioning accuracy of this method can reach 0.28m in a through-wall environment, which meets the accuracy requirements and can provide services for general location requirements.

Abstract:Aiming at the inconvenience of current millimeter-wave radar data acquisition, a scheme of millimeter-wave radar data acquisition system based on Zynq-7000 is proposed. The system is equipped with the AXI4 bus to send radar parameter commands to the radar equipment at the processing system (PS) end, so that the radar equipment can be powered on normally. After the radar data is received through the custom UART module, the data is synchronized and processed to retain the valid data. At the same time, the valid data is converted into AXI-Stream data stream into the S2MM port of the DMA, and finally the data is flushed to the DDR (Double Data Rate). This scheme separates the computer (PC) and directly buffers the effective radar data into ZYNQ's DDR, effectively solving the problem of high cost and single function of the existing acquisition system. It provides a low-cost data collection system for gesture recognition, parking assistance and UAV obstacle avoidance.

Abstract:It is an important means to search the suspected shoes in the surveillance video at the scene of crime. Aiming at the problems of low automation and manual screening, this paper proposes a shoe recognition algorithm based on improved deep residual network and data augmentation. In order to enhance the ability of network feature extraction, the deep residual network is studied. The bottleneck structure is improved without adding any parameters to improve the accuracy of the algorithm; Aiming at the problem of down sampling operation in bottleneck structure, the down sampling module is improved to alleviate the problem of information loss in network down sampling; Mixup and optical transform data augmentation algorithm are introduced to establish the linear relationship between data, enrich the diversity of data, and enhance the robustness of network model; Finally, the combined training method of center loss function and softmax loss function is adopted to make the training data achieve better clustering effect. In order to verify the effectiveness and practicability of the proposed algorithm, the proposed algorithm is tested on multi background shoe data sets. The test results show that the accuracy of map and rank-1 of the proposed algorithm is 66.83% and 86.77% respectively, which can effectively improve the accuracy of network recognition.

Abstract:Aiming at the disadvantages of large amount of calculation and low time efficiency of traditional maximum inter class difference method (OSTU) in image segmentation, a sparrow optimized OSTU segmentation method (SRWSSA) based on singer chaotic map and random walk strategy is proposed. Firstly, singer chaotic map is used to improve the initialization sparrow population, increase the diversity of initialization population and improve the global search ability; Secondly, the random walk strategy is used to perturb and mutate the updated optimal sparrow, so as to further increase the population diversity and enhance the local search ability; Finally, the standard image is segmented by two-dimensional OSTU using the proposed optimization algorithm to obtain the optimal threshold segmentation image. The SRWSSA algorithm proposed in this paper has significantly improved the optimization ability and iteration time. Compared with PSO-OSTU and SSA-OSTU, the number of iterations is reduced by 83.3% and 76% respectively. The image peak signal-to-noise ratio is increased by 8.2% and 11.3% respectively, and the running time is also improved. Practice shows that this method is feasible.

Abstract:In order to improve the calibration accuracy, a method of automatically and accurately extracting the center of the circular marker point is proposed by the paper, and the research of camera calibration is carried out. Firstly, the calibration image is preprocessed, and then the center of four circular markers on the calibration board are automatically extracted by the program to perform image correction. Secondly, an automatic iterative projection method is applied to iteratively process the corrected image, and finally the center coordinates of the mark point on the original calibration image are obtained through anti-perspective projection. Two extraction methods including manual and automatic are designed. The center coordinates of the last extracted mark points are substituted into Zhang Zhengyou's calibration algorithm, and the tangential distortion is considered to obtain the calibration parameters of the camera. The experimental results show that as the number of iterations increases, the reprojection error becomes larger and larger. When the number of iterations is one, the accuracy of calibration is better than that of multiple iterations, and it is slightly higher than the accuracy of manual extraction. Compared with the 0.2pixel calibration standard of Microvision commercial software, the accuracy has been improved by about 58%.

Abstract:In view of the problems that indoor positioning systems based on Bluetooth low energy technology (BLE) use machine learning algorithms such as multi-layer perceptron (MLP) as positioning algorithm, which leads to the problem of poor positioning accuracy. An indoor positioning method based on long short-term memory network (LSTM) is proposed in this article, which uses the time domain information in the positioning process to improve the positioning accuracy. First of all, a fingerprint database is built by collecting received signal strength indication (RSSI), and then rely on the mapping relationship between RSSI and two-dimensional coordinates for network model training to obtain weight coefficients. Finally, use the trained neural network model to build an indoor positioning system. The test results show that the average positioning error of this system is 1.41m, improved by 49% and 16% respectively when it is compared with MLP and RNN algorithms, and the positioning accuracy is significantly improved, which can meet the needs of indoor positioning.

Abstract:To study the surface defects classification method of strip steel, a small sample classification method of strip steel surface defects based on improved relational network is proposed. In this method, firstly, the net-work-in-network model was used as reference to enhance the characteristics of the network's feature recognition ability and non-linear expression ability of the local receptive fields; secondly, the model was combined with the relational network model; thirdly, a new self-normalized non-monotonic function was used as the activation function and the modified average absolute error was used as the loss function to allow more information to flow into the neural network. In this way, the model is enabled to learn more refined feature expression capabilities, so as to have better accuracy and generalization ability. The new model is tested on the NEU-DET data set, and the test results show that the defect classification accuracy rate obtained in the 5-way 1-shot task is 79.95%, which is 7.22% higher than the original model; the defect classification accuracy rate obtained in the 5-way 5-shot task is 92.04%, which is 2.15% higher than the original model.