Abstract:For the demand of wind turbine blade adaptive grinding device, this paper proposes an adaptive constant force grinding device with fuzzy PID control based on online sequential extreme learning machines (OS-ELM), by combining OS-ELM to rectify the control parameter input of fuzzy PID controller more quickly, and then get the suitable KP、KI、KD input initial values by fuzzy rules to achieve the PID control parameters are adjusted online. The simulation model of the grinding head control system is validated and optimized by MATLAB/Simulink simulation software, and finally the system control efficiency, stability and grinding effect are tested by the device prototype experiment. The experiments concluded that the device is capable of constant force bright surface grinding of wind turbine blades, the grinding efficiency is significantly improved, and the roughness of the product after grinding is between 10~12 μm, which meets the requirements of the enterprise after grinding blade roughness.

Abstract:Aiming at the problems of poor noise reduction effect, failure to distinguish the importance of multi-scale features, and insufficient extraction of temporal features in the current fault diagnosis of chemical processes, this paper proposes a chemical fault diagnosis method based on multi-scale fusion model. In this method, the attention mechanism is combined with soft threshold method and multi-scale learning respectively, and a multi-scale deep residual shrinkage network (MDRSN) is constructed. Moreover, the extracted multi-scale spatial features are sent to the bidirectional gated cyclic unit (BIGRU) to further extract temporal features. Compared with the single-channel network, BIGRU can not only complete the learning of past information, but also complete the learning of future moment information, so more temporal correlation information can be obtained. Finally, the modified Tennessee-Eastman process data were used to verify the classification accuracy of 95.08% and the recall rate of 94.76%, which was obviously better than the comparison method, and the effectiveness of the method was proved.

Abstract:Constructing diagnosability model is the prerequisite for diagnosability design, and diagnosability model can analyze the diagnosability of faults in the system. Most of the current research results focus on qualitative research, i.e., the evaluation of diagnosability focuses on the qualitative evaluation of "whether the fault can be diagnosed" without further considering the difficulty of quantifying the fault detection or isolation. To address the above problems, this paper proposes a quantitative diagnosability evaluation model based on earth mover′s distance (EMD), which transforms the quantitative evaluation problem of fault diagnostic difficulty into a distance measurement problem between multiple distributions of test data. First, a qualitative model of multi-signal flow diagram for diagnosability analysis is established to describe the functional composition structure of the system and the fault-test-signal correlation relationship; then, based on the qualitative model, a quantitative diagnosability evaluation model based on EMD is established in combination with the data-driven framework, which can not only qualitatively describe the correlation relationship between fault and test, but also quantitatively measure the difficulty of diagnosing In addition to qualitatively describing the correlation between faults and tests, it can also quantitatively measure the ease of diagnosing faults. Finally, the experimental case of a switching power supply of a certain type of equipment is used to verify the effectiveness of the method in this paper, which provides a feasible idea for diagnosability modeling.

Abstract:The three-phase three-level VIENNA rectifier causes the current zero-crossing distortion due to the uneven voltage stress and voltage imbalance at the midpoint of the switch transistor, which reduces the overall performance of the system. Therefore, the causes of midpoint voltage imbalance and the influence of vectors on midpoint potential in the seven-segment algorithm are analyzed from the perspective of space vector, and based on the traditional SVPWM modulation, an improved space vector modulation strategy is proposed to reduce the output voltage fluctuation: In the same sampling period, a regulating factor is used to reasonably allocate the action time of positive and negative small vectors, thereby canceling the middle vector and maintaining voltage stability. In addition, the algorithm implementation is optimized by analyzing the relationship between the adjustment factor and the system modulation ratio. Finally, simulation shows that the midpoint voltage fluctuation of the control strategy is reduced from ±12 V to ±4 V compared with the traditional control strategy, which proves the correctness and effectiveness of the proposed strategy.

Abstract:Active power filter (APF) is widely used in distribution network because of its fast tracking and harmonic elimination function. Among them, LCL type APF has better ability to suppress high frequency harmonics. With the increasing number of APF connected to the grid, the interaction between the APFs can not be ignored. For the problem of interaction among multiple APFs, firstly, the equivalent model of multi APF grid connection is derived based on Norton theorem. Secondly, the interaction is quantitatively analyzed by using the generalized dynamic relative gain array (GDRGA). Finally, the suppression of interaction is transformed into a multi-objective optimization problem. In view of the defects of traditional sparrow algorithm, such as poor global optimization ability and easy to fall into local optimization, tent chaos and dynamic random Cauchy mutation are introduced to improve, and the improved sparrow algorithm is used to coordinate and optimize the control parameters of APF. The results show that the improved sparrow algorithm can get pareto solution sets with better distribution and better performance. After multi-objective optimization, the interaction between APFs can be effectively suppressed, which verifies the effectiveness of the proposed method.

Abstract:This paper uses ultra-wide-band (UWB) technology to obtain measurement information, based on the multi-dimensional scale (MDS) algorithm, a super multi-dimensional scale positioning algorithm (TA-SMDS) is proposed, which combines time of arrival (TOA) and angle of arrival (AOA) information, with the goal of solving the localization problem in the environment of the global satellite navigation system signal complete rejection. Based on the TA-SMDS algorithm, the kernel matrix in complex domain is further constructed, a super multi-dimensional scaling location algorithm in complex domain (TA-CDSMDS) is proposed. Comparing the positioning results of the three algorithms and the errors under the angular of 10°, 15°, and 20° as well as different ranging errors, it can be obtained that the TA-CDSMDS algorithm has smaller positioning errors and is closer to the CRLB than the TA-SMDS and MDS. Analyzing the calculation time of the algorithm under different nodes, the TA-CDSMDS algorithm reduces the time by 28% to 48% based on the optimized TA-SMDS, and has better localization performance.

Abstract:With the progress and large-scale application of the drilling and completion technique in horizontal well, the thru-drilling memory logging technology develops rapidly because of its efficiency and security. It has became the main logging technology in horizontal well. New problems and challenges appear in the horizontal well′s field application of the thrudrilling acoustic logging tool because of the acoustic logging tool′s special structure and the special process with which the logging tools go through the hole of the drilling tool into the bottom of the well. In this paper, the phenomenons of large slowness measurement error in the thru-drilling memory acoustic logging tool’s field applications are summarized. In view of these problems, laboratory tests and forward simulation are carried out.Then we know that the reason of measurement error is that the receiving array is not parallel to the wall of the well in the horizontal well and the highly-deviated well.The corresponding optimization schemes of the tool design and the dynamic balancing logging process are proposed,which are based on the principle of leverage. The effect of the optimization schemes is verified through field tests. This paper can provide reference for the research and development of the thru-drilling acoustic logging tool and the analysis of the tool’s measurement influence factors.This paper has important significance for the popularization, application and technological development of the thru-drilling system.

Abstract:A robust resource scheduling method of the multi-jammer joint jamming beam and transmitting power is proposed for the problem that some parameters of multi-station radar systems can only be obtained through parameter estimation, and the performance of cooperative jamming is degraded due to the accuracy of parameter estimation. First, the sum of the PCRLB of multiple targets in jamming environment is used to evaluate jamming performance. Then, based on the jamming signal model, radar echo signal model in the suppressive jamming environment, and the geometry of jamming support formation position, the generalization error caused by the uncertainty of radar system parameter estimation is combined with the jamming beam and transmit power to establish a robust resource scheduling method with jamming resource constraints. Finally, due to the coupling of resource parameters, a two-step decomposition solution based on genetic algorithm is used to solve it. The simulation results show that the proposed robust resource scheduling strategy can effectively reduce the performance of multi-target tracking in radar systems and enhance the cooperative jamming performance of multi-jammers to the radar systems.

Abstract:In order to solve the problem that power plant boiler operators rely on experience to adjust boiler operating parameters to reduce the NOx concentration at the SCR inlet and improve the denitration performance, a method for predicting the NOx concentration at the SCR inlet is proposed. This method establishes a CNN(1D)-LSTM model based on convolutional neural network and long short-term memory neural network. The NOx concentration at the SCR inlet can be predicted after 5 min. Power plant operators can use the prediction results of the model as an important reference for the NOx concentration at the SCR inlet, and more effectively adjust boiler parameters for denitrification optimization. The results show that the LSTM model for predicting the NOx concentration at the SCR inlet after 3 min is better than CNN(1D)-LSTM; the CNN(1D)-LSTM model for predicting the SCR inlet concentration after 5 min has a great prediction accuracy compared with the LSTM model. The Emape on the test set is 7.05%. The desired effect was achieved.

Abstract:Limestone-gypsum wet flue gas desulfurization (WFGD) is the main method of flue gas desulfurization in thermal power plants and plays an important role in atmospheric environmental protection, but it can also suffer from corrosion and fouling problems that affect the operational efficiency. In order to optimize the operation of the wet WFGD system, a data-driven approach is used to model the SO2 flue gas emissions dynamically. Firstly, the SO2 emission data are decomposed using the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) to obtain several intrinsic mode functions (IMFs). The intrinsic mode functions containing noise are denoised using wavelet threshold denoising to obtain the pure components. Then a deep learning model combining convolutional neural network (CNN) and bi-directional gated recurrent unit (BiGRU) is designed for SO2 emission prediction. After comparing the two schemes of predicting the components separately and then reconstructing them and reconstructing the components and then predicting them, it is found that the root mean square error and the mean absolute error of the former are reduced by 0.135 7 and 0.284 3, respectively, compared with the latter. Experiments are conducted based on the first scheme in comparison with other benchmark models, the root mean square error and the mean absolute error of the proposed model are 0.699 6 and 0.355 3, which are the lowest. The results indicate that the proposed model has significant advantages in predicting SO2 emission concentration.

Abstract:VHF data exchange system is one of the main communication methods in the e-navigation strategy led by the international maritime organization. Among them, VDE-TER business divides many kinds of business logical channels according to physical channel characteristics. In order to improve the data transmission efficiency under the premise of ensuring the transmission quality of the VDE-TER channel, this paper proposes an improvement of Max-Log-Map based on a priori correction factor based on an in-depth analysis of the Turbo encoding and decoding algorithms under different service logic channels of the VDE-TER algorithm. According to the Turbo coding characteristics of VDE-TER different service logical channels, the algorithm realizes the determination of the optimal correction factor value of the specific logical channel of VDE-TER through simulation and the performance test of large samples of actual physical channels. The results show that the performance of the decoding algorithm with correction factor is improved by 0.6~0.8 dB gain compared with the traditional decoding algorithm, which lays a theoretical and technical foundation for the development of high-performance VDES products.

Abstract:In view of the problems of color distortion and unsatisfactory crack detection effect in aerial image of insulators in fog, an improved Hough transform method is adopted to locate and detect the cracks of insulators in fog. Aiming at the disadvantage of color distortion, the formula of perspective is improved by the absolute value of the difference between atmospheric light intensity and dark channel, so as to improve the color distortion of bright areas such as blue sky and white clouds after defogging. For the obstacle of insulator positioning and crack detection, the feature that the center of the ellipse is the point with the minimum maximum length to the edge of all points inside and outside the ellipse is proposed, and the center of the circle is calculated quickly to reduce the Hough parameter dimension and calculation amount, so as to realize insulator positioning in the image. Finally, the combination of peak detection and zeroing in Hough space is used to solve the problem of time-consuming and low accuracy of traditional Hough transform linear detection algorithm. By setting the threshold value, the generation of false cracks and over connections can be reduced, and the cracks can be detected quickly and accurately. The simulation results show that the improved Hough transform proposed in this paper improves the detection rate of insulator cracks in fog by 1.4 times and the accuracy rate by 5.5% compared with the traditional Hough transform.

Abstract:Aiming at the problems of small target size and complex background in remote sensing vehicle detection tasks, a lightweight YOLOv5 algorithm based on multiple pyramids and multiscale attention is proposed. In the backbone network, the number of downsampling is reduced, the small target detection ability is improved, and light weight is achieved; in the neck, the information of different feature layers is fully utilized through the redesigned multi-pyramid network to enhance the feature fusion ability. And introduce an improved multi-scale attention module to obtain a larger receptive field and area of interest for the shallow feature map; finally, the K-means++ clustering algorithm is used to cluster and analyze the target size, and an anchor frame scale suitable for the target is designed. and aspect ratio. In the self-built remote sensing vehicle dataset, the target detection accuracy is not only improved, but also the parameter quantity is greatly reduced. Compared with YOLOv5s, AP0.5% is increased by 2.3%, AP0.5:0.75% is increased by 4.3%; the number of parameters is reduced by 65%, and the model size is reduced by 60%. It effectively improves the detection accuracy of small targets while reducing weight.

Abstract:Aiming at the problems of low detection accuracy and slow sorting speed of coal gangue sorting tasks in industrial scenarios, a coal and gangue target detection algorithm based on improved YOLOv5s is proposed. A lightweight attention mechanism CBAM is added to convolutional layer of the backbone network to improve the ability of target feature expression in complex cinder environment. Secondly, the BIFPN structure is added to the feature fusion layer. The bidirectional cross-scale connection and weighted fusion are carried out in the BIFPN structure to strengthen the feature information of shallow layer of coal gangue and the location information of high-rise coal gangue, and solve the problem that the color and texture of coal gangue are similar and difficult to classify; Finally, on the basis of the original algorithm DIoU, the aspect ratio of the bounding box is added to improve the accuracy of the inspection box detection. The proposed method is tested by using 10 000 coal gangue images collected in an industrial production environment as a dataset. Experimental results show that in comparison with YOLOv5s model before the improvement, on the premise that the detection speed remains basically unchanging, average precision mAP_0.5 of the improved algorithm reaches 93.3%, and average detection precision is increased by 5.1%, which realizes the requirements for target detection of coal gangue.

Abstract:Insulator is one of the important components on the transmission line. It is an important means to ensure the safe transmission of power to accurately detect the insulator and its defects using UAV patrol inspection. In order to solve the problem that the main target detection network directly scales the original image when processing high-resolution images, which leads to the loss of target details or re detects the original image by cutting it into blocks, which leads to the loss of the overall information of the target, a dual branch structure backbone network (RC Net) is designed based on the residual network (ResNet 50), which can reduce the loss of target context information and local information. At the same time, the deformable convolution is introduced to replace part of the conventional convolution to change the sampling points, so that the sampling points can more closely fit the geometric shape of the target itself, improve the feature expression ability of the network, and redesign the parameters of the anchor frame according to the size and shape of the insulator itself, so that the anchor frame is more suitable for the scale of the target itself, and the frame regression is more accurate. The experimental results on the expanded Chinese transmission line insulator dataset (CPLID) show that the average accuracy of the algorithm proposed in this paper reaches 88.3%, which is better than the current mainstream detection algorithm in the high-resolution image background.

Abstract:Frit is a dot matrix pattern consisting of densely arranged small black dots, which can be found around the edges of the auto windshield. In the process of frit printing and sintering, there may be some defects such as adhesion and deformity, which tend to be false detection or missed detection in manual detection. And that manual data is difficult to be collected for deep analysis. To solve this problem, this paper tries to use machine vision technology to extract the outline of the frit as the camera moving track. Then the PLC is used to control the movement of the two cameras to take the frit pictures in sections from the four edges of the windshield to catch the clear images of the little black dots. Finally, the YOLOv5s algorithm is used online to identify and locate the frit defects. Compared with other different algorithms, the YOLOv5s algorithm is proven to be superior and robust in frit defect detection.

Abstract:In order to better judge the corrosion degree of metal corrosion images, in view of the problems of low brightness, low contrast and blurred details in corrosion images, an improved corrosion image enhancement method based on homomorphic filtering and multiscale fusion is proposed. First, the original corroded image is divided into the base image and the detail image by guided filtering and then weighted and fused to obtain the detail contrastenhanced image. Secondly, the original eroded image is converted into HSV color space, and only the luminance component is subjected to the improved single-parameter block homomorphic filtering to obtain a luminance-enhanced image, which can reduce the homomorphic filtering parameters and improve the phenomenon of excessive brightness enhancement of homomorphic filtering. Finally, using three weights of Laplacian contrast, saliency and saturation to perform multi-scale fusion on the two images with dominant features after processing to obtain the final enhanced image. The experimental results show that the information entropy, mean, average gradient and standard deviation of the algorithm in this paper are improved by 7.4%, 9.8%, 43.34% and 29.8% respectively compared with the original image. Among them, the average value of information entropy, average gradient and standard deviation are better than the other three algorithms. The algorithm in this paper can effectively improve the overall brightness of the corroded image, improve the contrast of dark details, and improve the image quality.

Abstract:A convolutional neural network (CNN) for fractional interpolation of inter prediction is proposed because of the poor interpolation effect of traditional interpolation filters and the deep learning methods, which only generate half pixel samples, or need to train the corresponding model for each pixel position and quantization parameter (QP), or introduce additional information as input. Based on the dense residual network, the model combines multi-scale distortion feature extraction structure and sub-pixel convolution to increase the accuracy of feature extraction and generate fractional pixels. The characteristics of fractional interpolation task are analyzed and the data set with true distortion is constructed. The model directly generates fractional pixel samples and can adapt to arbitrary quantization parameters (QP). Experimental results verify the efficiency of the method. Compared with H.265/HEVC, this method achieves 2% in bit saving on average under low-delay P configuration. Compared with similar methods, the overall performance has also been improved.

Abstract:Aiming at the problem that the full flow detection mode is easy to cause the performance bottleneck of the security detection equipment, an improved dingo optimization algorithm is given to optimize the radial basis function RBF neural network for normal traffic filtering. First, the wild dog optimization algorithm was improved using Singer chaotic mapping and search balance strategy; second, the output weights of the RBF neural network were optimized with the improved wild dog optimization algorithm, and the network was trained using the CSE-CIC-IDS2018 dataset to construct a normal traffic filtering model. Finally, before the network traffic entered the security detection device, filter out as many normal traffic as possible to reduce the workload of the security detection device. The experimental results show that compared with the existing models, the normal traffic filtering model of IDOA-RBF neural network has a great improvement in modeling time, while maintaining a high recognition accuracy, and can filter out 72.9% of the normal traffic in the traffic to be detected.

Abstract:To address the problem that ground infrastructure cannot provide emergency communication effectively after natural disasters, a sum rate optimization scheme for unmanned aerial vehicles swarm-assisted emergency communication system based NOMA technology is proposed. Firstly, the scheme constructs a unmanned aerial vehicles swarm-assisted emergency communication model, the model objective of maximizing the total sum communication rate of ground users under the constraints of maximum UAVs transmitting power and quality of service for ground user; Secondly, the power allocation under the NOMA mechanism is implemented by improving the simulated annealing algorithm; Finally, the ground users are clustered by improved K-means algorithm, and then the path loss and line-of-sight link probability between unmanned aerial vehicles and users are optimized for completing the three-dimensional deployment and maximizing the system sum rate. The numerical simulation results verify the effectiveness of the proposed scheme.

Abstract:：Aiming at the problem of low target contrast and multi-scale underwater images in underwater target detection, an underwater target detection algorithm based on the fusion of high-frequency enhanced network and Feature Pyramid Networks(FPN) is proposed. The algorithm improves the extraction of underwater target edge, contour information and target underlying information. Firstly, octave convolution is introduced to decompose the output features of the convolution layer by frequency, and the feature maps extracted by the backbone network are separated from high-frequency and low-frequency information. Since the contour information and noise information of underwater targets are contained in high-frequency features, Squeeze-and-Excitation Network with adaptive enhancement characteristics is introduced into the high-frequency information channel, and a high-frequency enhanced convolution is formed. It can achieve the purpose of enhancing useful contour feature information and suppressing noise. Secondly, the enhanced high-frequency feature components are integrated into the shallow network of FPN. It improves the feature representation ability of the original FPN for underwater multi-scale targets and alleviates the problem of missed detection of multi-scale targets. Finally, the proposed method is fused with the baseline algorithm Faster R-CNN, and the experiment is carried out on the dataset provided by the National Underwater Robot Competition. The results show that the recognition accuracy of the improved algorithm reaches 78.83%, which is 2.61% higher than the baseline. Compared with other types of target detection algorithms, it still has advantages of accuracy and real-time detection. The effectiveness of improving foreground and background contrast from the perspective of feature map frequency domain is demonstrated.

Abstract:In electroencephalography (EEG) data acquired in synchronization with functional magnetic resonance (fMRI), gradient residual spiking artifacts persisted after preprocessing using average template subtraction (AAS). There is a need for more accurate removal of residual spikes, so as to decrease the interference from frequency-based activity inferences, and less spurious correlations between time series.Aiming at the characteristics of spike artifacts in EEG data, this paper first uses the Schr-dinger method to decompose and identify the EEG data containing spikes, automatically subtracts most of the spike components with a large amplitude difference from the EEG, and then uses the amplitude threshold method to compensate the error by inverse compensation. Residual spikes with the same amplitude as the EEG are located to realize the location and removal of spike artifacts. For simulated signals, the signal amplitude error (Er) obtained by this method is 24.95% higher than that of the Schr-dinger method on average, and the signal-to-noise ratio (SNR) is 27.13% higher than that of the Schr-dinger method. For real signals, the Pearson correlation coefficient obtained by this method is significantly less than For the other four methods, the denoising effect is 11.42% higher than that of the Schrodinger method. Compared with other methods, the use of Schrodinger combined with threshold algorithm, significantly improved the peak recognition accuracy and the denoising effect, whether the peak is located in the trough of the waveform, or the high-frequency fluctuation amplitude is comparable to the peak. This denoising method provides strong support for the fusion study of EEG-fMRI.

Abstract:The acoustic emission detection technology has used widely, and the good sensitivity of the acoustic emission instrument system is an important guarantee for the acoustic emission detection. In this paper, a set of sound source signal detection platform is built, and the response characteristics of acoustic emission signals of two different sound sources in a circular steel plate are studied through experiments and numerical analysis. The experimental results show that with the increase of the distance from the sound source, the variation trend of the response amplitude of the signal in different directions is basically the same, and the difference of the response amplitude between different directions is mainly distributed in 2.1~8.8 dB. When the distance from the sound source is 0~350 mm, the overall decrease of the signal response amplitude is faster, and the signal attenuation rate of the pulse excitation is about 3% lower than that of the lead-break excitation. The sound source generated by the pulse signal excitation sensor has better stability and reproducibility. This study provides a reference for the selection of the sound source in the sensitivity test of the acoustic emission instrument.

Abstract:The data-driven approach to rolling bearing remaining useful life (RUL) prediction shows great potential, but there is still room for improvement. Therefore, a prediction method of rolling bearing RUL based on Autoformer model is proposed. Combined with the expert knowledge in this field, the original signal of rolling bearing is artificially extracted and optimized, and the complex mapping relationship between input features and RUL is mined by using the powerful multi-dimensional feature extraction capability of Transformer models. According to the periodic characteristics of the vibration signal of rolling bearings, the Autoformer model is used to decompose the time series to deal with the trend term and the periodic term separately. Experimental results show that the average scores of the proposed prediction method on the PHM2012 dataset is improved by 50.03%, 21.31% and 19.93% respectively, compared with other methods in the literature. Proves the superiority of this method.

Abstract:At present, the existing line selection method using single fault feature is affected by the transition resistance, fault closing angle and fault location, which generally has the problems of poor noise resistance and low accuracy. In view of this, a new small current grounding line selection method of extreme learning machines (ELM) is proposed, which integrates transient high frequency energy and waveform correlation fault characteristics. The transient high frequency energy in zero sequence current is extracted by using variable mode decomposition and hilbert transform. Through zero-sequence current correlation analysis, the comprehensive correlation coefficient of transient waveform is extracted, and the fault feature vector is formed by the two. The threshold free ELM model is input to realize fault line selection. The effectiveness of the proposed method is verified by extensive simulations, which show that the method is noise-resistant and largely unaffected by factors such as transition resistance, fault closure angle and fault location. The method is verified by the example of substation ground fault data in the paper, and the results show that the method has an accuracy of 100% in line selection.

Abstract:To meet the demand for rapid detection of battery health status in the process of retired power battery recycling, this paper takes soft pack lithium iron phosphate batteries as the research object and proposes a rapid detection method of lithium battery health status based on infrared thermal imaging. By changing the battery charging and discharging current multipliers, the temperature changes of batteries with different aging degrees during the discharge process are studied, and the infrared thermographic video during the discharge process is collected to establish the correspondence between the battery health state and the infrared thermographic features, which is used as the health factor for battery health state detection; an improved video recognition algorithm based on SlowFast-LSTM deep learning network model is constructed for battery health state detection. The improved video recognition algorithm achieves an average recognition rate of 80.78% for the six categories of battery health state 0~40%, 40%~50%, 50%~60%, 60%~70%, 70%~80% and 80%~100%, and a single battery detection time of 3 minutes, which enables fast detection of battery health state.

Abstract:In order to achieve high-performance radiation in the specified direction of a broadband antenna, a wideband microstrip equiangular spiral antenna was designed. To achieve impedance matching, an exponential gradient microstrip balun structure is designed for the antenna feed section. After physical testing parameters, the results show that the antenna with exponential gradient Barron structure achieves good impedance matching and has good broadband characteristics. Its operating bandwidth is 1.74 GHz to 4.82 GHz, with a minimum return loss of -30 dB. At the same time, a flat bottom metal reflection cavity is designed to reflect back electromagnetic waves. While maintaining broadband operating characteristics, the antenna has good unidirectional radiation characteristics. The gain of the antenna in the entire operating frequency band is greater than 6 dB. Compared with traditional spiral antennas, it has the characteristics of broadband, directional, and high gain, and has certain application prospects.