Abstract:Aiming at the problems of attenuation and distortion of the discharge pulse signal and susceptibility to noise interference in the traditional generator partial discharge(PD) monitoring method. In order to improve its accuracy and rapidity, this article uses broadband detection technology to effectively capture the PD signal, and uses the modal decomposition method to extract the inherent modal components of the discharge pulse signal. Correlation analysis, according to the kurtosis criterion, obtains the effective components containing the characteristic information of the PD pulse, thereby decomposing and separating the interference signal from the PD signal, and accurately grasping the real-time operating status of the generator. The results show that compared with the soft threshold wavelet denoising, the signal-to-noise ratio of this method is improved by 52.5%, and the mean square error is reduced by 30.7%, which are 6.75 and 0.04, respectively. It performs well in the smoothness of the waveform. At the same time, the calculation time of this method is only 8.62s, which greatly reduces the number of iterations and reduces the iteration calculation time during signal analysis. Effectively improve the denoising effect and collection efficiency of the generator PD signal.

Abstract:In order to solve the problems of low transmission rate, small bandwidth and poor reliability in large data transmission system, this paper proposes a design scheme of high speed data transmission between FPGAs based on SRIO protocol. The design uses SRIO IP embedded in FPGA to check the transmission data for sending, parsing and receiving; The FPGA internal GTP high-speed serial transceiver is used as the physical layer transmission basis. Photoelectric conversion module is used to convert optical and electrical signals to complete data long-distance transmission. The data transmission rate can reach 280MB/s, and the scheme has been successfully applied to the telemetry system memory ground test platform project, which can realize high-speed and reliable data transmission between two FPGA devices.

Abstract:Rolling bearing is one of the important parts of traction motor, and the accuracy of its fault diagnosis is of great significance to ensure the normal operation of traction motor.In order to improve the accuracy and effectiveness of bearing fault diagnosis, the method of maximum correlation kurtosis deconvolution (MCKD) combined with Hilbert-Huang transform (HHT) is used for diagnosisIn view of the selection of MCKD algorithm subject to shift number (M), filter order (L) and shock signal period (T), it is particularly dependent on the choice of experience. The dynamic particle swarm algorithm is selected to optimize it to reduce noise signal interference. Pulse signal triggered by fault.Then use the HHT algorithm to get the signal envelope spectrum, which can better identify different types of faults. Combining VS with MATLAB can realize the application of diagnostic algorithms to the high-level development language environment.The algorithm was verified using the CWRU bearing data set. The verification results show that the method can effectively enhance the fault characteristics. The fault frequency of the bearing inner ring is 162Hz and the bearing inner ring fault frequency is 108Hz, which can accurately identify the fault type of the bearing.

Abstract:Aiming at the problem of tail energy consumption and application object limitation in APP energy bug model in traditional research, a system call based energy bug model is constructed. First, mix-cross-recursive feature elimination method is used to select important system calls that affect the energy consumption of each category of APP as features to improve the feature refinement granularity. Then, multiple regression models are constructed for each category of APP. By comparing the average absolute error and coefficient of determination of different models, linear kernel support vector machine regression is selected as the energy bug model of the classified APP. Finally, based on the test set, the average absolute error of the model constructed by the mix cross recursive feature elimination method and the cross recursive feature elimination method is compared. The result shows that the accuracy of the model constructed by the mix recursive feature elimination method is increased by up to 4.4%. At the same time, the classification model is compared with the classification model based on the test set. The average absolute error of the unclassified model shows that the accuracy of the classification model is increased by up to 6.7%, and the classification model can accurately detect the energy bug of the historical version of the APP.

Abstract:In line power flow regulation, in order to solve the problem that UPFC bus point voltage and DC capacitor voltage fluctuation, by setting up UPFC mathematical model, current decoupling controller of parallel converter are designed, according to the principle of power balance, the current feedforward compensation of voltage stabilization control is derived. and an improved coordinated control strategy based on shunt converter is proposed. The obtained feedforward compensation amount is introduced into the active and reactive power control system of the parallel converter，in order to realize unit feedforward, the methods of moving feedforward compensation points and adding feedforward compensation functions are adopted to improve the compensation effect during power flow regulation.. MATLAB is used to build a simulation model of 500KV AC transmission system with UPFC, the results show that compared with the coordinated control strategy before improvement, the proposed control strategy not only meets the requirements of coordinated control of active and reactive power flows, but also improves the voltage supporting capacity of the transmission system (the fluctuation of access point voltage and DC capacitor voltage has been reduced to 6.7 KV and 1.1 KV respectively), the response time of the system to the output of DC capacitor is also shortened by 0.5s.

Abstract:Aiming at the increasingly complex operation and maintenance problems of power system, this paper combines Beidou precise space-time technology with the theory of digital twin system, and puts forward the method of building the operation and maintenance platform of smart grid. By combining Beidou and Internet of Things sensors, this method can obtain real-time information of power system operating state, equipment working condition and equipment parameters, and take these information as input of digital twin system. By interacting with the actual power system, the digital twin system can form a digital twin operation and maintenance platform that can perform state prediction, fault diagnosis, system self-healing and other functions. The research results show that the method proposed in this paper has reference value for improving the operation and maintenance capability of smart grid.

Abstract:In order to realize the rapid traceability detection of coffee origin, a detection method based on the combination of electronic tongue and generative adversarial networks (GAN) - convolutional denoising autoencoder (CDAE) – extreme learning machine (ELM) is proposed. Aiming at the problems of low accuracy and poor generalization ability of deep learning model caused by the insufficient number of original data samples of electronic tongue detection, generative adversarial networks (GAN) is used to expand the data scale of training samples and improve the robustness of the system; According to the characteristics of complex electronic tongue output signal, large dimension and many noise, convolutional denoising autoencoder (CDAE) is used to extract the features of electronic tongue signal in low dimensional feature space to improve the expression ability of key features; Finally, extreme learning machine (ELM) is used to classify and identify the extracted feature information, and a coffee origin traceability detection and analysis model is constructed. The results show that compared with the traditional machine learning models such as discrete wavelet transform (DWT), support vector machine (SVM) and extreme learning machine (ELM), as well as the deep learning models such as VGG16 network, GAN-CDAE-ELM has better discrimination effect on coffee from different producing areas, and its test set has higher accuracy, precision, the recall and F1 score reached 99.00%, 99.03%, 99.00% and 0.9901. This study provides a new idea for rapid identification and detection of coffee producing areas based on intelligent sensory system.

Abstract:Rolling bearing is one of the important parts of rotating machine. Aiming at the problem of rolling bearing fault diagnosis, this paper proposes an algorithm combining multiscale permutation entropy (MPE) and support vector machine(SVM) optimized by particle swarm optimization (PSO). The fault characteristics of the bearing fault data was obtained by the MPE method, fitting as a feature vector into the PSO-SVM model, using Case Western Reserve University bearing dataset for verification. It is found that this method can effectively identify the fault of the rolling bearing. This method is compared with the fault classification results obtained by combining the multi-scale permutation entropy with the traditional SVM method and the SVM method optimized by grid search. It is found that the method proposed in this paper has certain advantages in the efficiency and accuracy of rolling bearing fault diagnosis.

Abstract:Electrocardiogram (ECG) detection is the most commonly used diagnostic method of heart disease. However, in the process of ECG signal acquisition, it is often disturbed by noise, which greatly affects the accuracy of ECG signal classification and diagnosis. In order to improve the accuracy and anti noise ability of classification diagnosis, this paper improves and designs an ECG classification and diagnosis model which use deep residual shrinkage network (DRSN) to resist noise automatically and integrate spatial information by global average pooling (GAP). The classification performance of the model is verified on MIT-BIH arrhythmia data set, and its anti noise performance is analyzed and compared with the ordinary convolutional neural network (CNN) model. The experimental results show that the classification accuracy of the designed DRSN + GAP diagnostic model based on AAMI standard is up to 99.3%, and its anti noise performance is better than ordinary CNN model for power frequency and Gaussian noise with different intensity.

Abstract:The digital microfluidic system (DMF) based on dielectric wetting (EWOD) effect can realize the movement, splitting, synthesis and other operations of multiple droplets at micro rise level. To reduce the cross contamination between heterogeneous droplets of EWOD devices, the completion time of chemical synthesis and biochemical testing in EWOD devices, and ensure the reliability of devices, this paper makes a spatial modeling of DMF system environment through MATLAB platform. Based on the basic A * algorithm, the inflection point is taken as the influence factor of the estimated function value and integrated into the droplet collision strategy, an improved A * algorithm for the fluid characteristics of EWOD devices is proposed. Through the simulation of chemical synthesis experiment, the algorithm shows good droplet addressing optimization characteristics. Compared with the basic A * algorithm, under the same path length (64 electrodes), this algorithm can effectively reduce the redundant inflection points (reduce 10) and the time for droplets to reach the target electrode (reduce about 23S). In addition, the improved algorithm also shows good adaptability (only needs 24 time stamps) in the experiments of multi-droplets synthesis under complex conditions (there are unknown and known faults). The experimental results show that the proposed droplet addressing algorithm has good applicability and reliability for multi droplet path planning and scheduling optimization in EWOD devices, and can effectively reduce the completion time of droplet applications in EWOD devices.

Abstract:Aiming at the problem of poor adaptability and unsatisfactory enhancement effects of traditional speech enhancement methods in low signal-to-noise ratio environments, this paper proposes a speech enhancement method based on Wasserstein Divergence Deep Generative Adversarial Networks. The DGAN is based on five generators and one discriminator. Five generators are used to enhance the noisy speech signal five times, which effectively improves the enhancement effect of the DGAN in low signal-to-noise ratio environments. At the same time, Wasserstein divergence is used to optimize the network training which can solve the problems in the traditional GAN training process and improve the stability of the DGAN training process. Comparing with traditional speech enhancement methods, the noise adaptability and enhancement effect of this method are significantly improved in low signal-to-noise ratio environments. The experimental results show that, compared with the original noisy speech, SegSNR of the enhanced speech is improved by an average of 6.1 dB. PESQ is increased by an average of 28.9% and STOI is increased by an average of 10.6%.

Abstract:Newly developed or produced pressure transmitters need to conduct a comprehensive test of its technical performance indicators to ensure accurate output of the value. After a period of storage, accumulation, use, or repair of the pressure sensor, its main technical performance must be verified twice to ensure that its performance indicators meet the required standards. The impact of actual production and working environment causes its output to produce nonlinear output. Therefore, an inverse model based on the third-order nonlinear polynomial is proposed for the nonlinear calibration of the pressure transmitter and the WA network model using the improved particle swarm optimization algorithm is used for temperature compensation calibration. It performs temperature compensation calibration. The maximum absolute error of the output after nonlinear calibration is compared with the commonly used end-base translation method. The maximum absolute error is reduced from 6.0265 to 0.3086. The output accuracy after temperature compensation calibration by the improved particle swarm optimization algorithm is reduced from 0.386% to 0.091% , To achieve the high-precision output of the pressure transmitter.

Abstract:Aiming at the problem that the target information is difficult to effectively highlight due to the poor reconstruction effect of ghost imaging, combined with the advantages of autoencoder neural network in noise reduction, a ghost imaging optimization method is proposed. In this method, a handwritten digital data set is used as a sample, and a noise reduction network model is designed based on the second-order correlation of the detection data to obtain the initial ghost image. The network model uses the Leaky relu linear activation function to solve the problem of network oversaturation and cell death, and the effectiveness of the proposed network model is verified through 10,000 testing sample sets. By comparing the quality of ghost images before and after optimization at different sampling rates, the analysis results show that the peak signal-to-noise ratio of ghost images after optimization is increased by 87.02%/93.99%, 81.97%/85.90%, 27.22%/18.16%, respectively; at the same time, the contrast is increased by 479.03%/363.79%, 380.42%/272.91%, 38.76%/31.05%, respectively, compared with CGI, DGI, and CS respectively.

Abstract:In the actual operation of power system, the high temperature of cable conductor often causes power system failure, but because the cable core temperature is not easy to monitor, therefore, a rolling prediction method based on LSTM (Long Short-Term Memory) is proposed to predict the cable core temperature. According to the collected cable core temperature data set, this algorithm is used to train the model, dynamically adjust the parameters of the network model, and study the law of the data change, so as to realize the prediction of cable core temperature. The results show that the RMSE of the algorithm model is 0.1979 °C, and compared with the BP and LSTM algorithm model, the algorithm model can effectively predict the short-term cable core temperature change trend, the results show that the algorithm has practical significance in the safe operation of power system.

Abstract:Aiming at the difficulty in extracting the feature information of the leakage signal in the process of long-distance oil and gas pipeline leakage detection, a new pipeline negative pressure wave signal feature extraction method is proposed. A complete set of empirical mode decomposition algorithm with adaptive noise is used to denoise the collected negative pressure wave signal, and the Hausdorff distance between the probability density of the component after CEEMDAN decomposition and the original signal is evaluated. Select the effective mode and reconstruct. The cloud model feature entropy and kurtosis of the reconstructed signal are calculated as feature parameters, and the support vector machine is used for classification and recognition. Through laboratory data verification, the method of combining CEEMDAN, Hausdorff distance and cloud model feature entropy can effectively improve the accuracy of oil and gas pipeline leak detection, and realize the identification of small leak signals with a flow rate of less than 4^3m/h. Certain field application value.

Abstract:With the development of sensor technology and microelectronics, it has a wide application value to recognize human motion patterns by wearable sensors. It is of great significance to improve the accuracy of recognition. This paper considered the characteristics of human lower limb motion, and proposed a human motion pattern recognition algorithm based on CNN and Mogrifier LSTM. First, CNN is used to extract local related features of the original data, then Mogrifier LSTM is used to replace the full connection layer to mine the front and back dependencies of local related features. Recognize the six common motion patterns of walking, running, upstairs, downstairs, uphill and downhill. The experimental results show that compared with the traditional LSTM algorithm, the accuracy of Mogrifier LSTM is improved by 1.03%. After combining CNN and Mogrifier LSTM, the accuracy is further improved by 1.17% to 98.18%, which proves the superiority of the algorithm proposed in this paper.

Abstract:In the correction of various errors of TI-ADCs system, most of the current compensation algorithms are targeted for the errors in high-speed (GSPS) and medium-or-low-precision (≤16bit) TI-ADCs systems, and there are relatively few researches on nonlinearity errors. A correction algorithm for the dynamic nonlinearity mismatch errors in high-precision (24bit) TI-ADCs system was proposed. Firstly, the output data containing nonlinearity errors were extracted. Secondly, according to the truncated Volterra series error model, the dynamic nonlinearity errors coefficients were estimated with matrix transformation and LS method. Then, the nonlinearity errors were compensated by a multi-stage correction method based on error reconstruction, and the frequency spectra before and after calibration were obtained. The TI-ADCs system was simulated with single frequency, multi-frequency with equal and unequal frequency interval sinusoidal input, and the SFDR of the system after four-stage calibration was elevated from 22.91dBc, 11.12dBc and 11.14dBc to 104.45dBc, 96.74dBc and 99.25dBc, respectively. The theoretical effectiveness of the proposed method in the whole Nyquist band was verified by simulation results, which indicates a calibration effect for high-accuracy system.

Abstract:At present, in some fields, such as metal wound detection and biomedical, the signal acquisition technology mainly focuses on occasional signal.If the traditional fixed-frequency signal acquisition method is used, it will cause a serious waste of storage resources.To solve this problem, a sampling method based on DDS clock phase synthesis is proposed.The method combines DDS technology and waveform phase synthesis technology and adaptively switches the sampling frequency according to the signal characteristics.While ensuring the integrity of signal information, the data storage of useless signals is reduced. Through simulation and experiment results show that the method on FPGA platform is well on achieving the sampling rate of 12 MHZ and 96 MHZ real-time switching control, a complete reduction of the highest frequency of 10 MHZ waveform detection details, at the same time under the parallel processing of the FPGA using split phase synthesis by sampling rate of at least four times, optimize the resource utilization.

Abstract:In order to realize portable electrocardiogram (ECG) monitoring, a real-time ECG monitoring device based on the BMD101 chip was designed and implemented. The device comparison and selection, circuit filtering and printed circuit board analog-digital separation design were implmented to minimize power consumption and suppress noise interference. Some filter based denoising algorithms are used to eliminate common noises such as possible residual baseline drift, EMG noise, and power frequency interference. Experiments show that the designed system can acquire high quality ECG signals and the algorithm has a good denoising effect. The equipment has the characteristics of low cost, small size, easy operation, and good stability, which can provide a portable equipment for personal ECG monitoring.

Abstract:At present, there are few researches on the somatosensory function of stroke patients, and the two-point discriminative test disc used in hospitals is primitive. This article discusses a quantitative test system for the sensory function of stroke patients, which makes up for the lack of digitization and modernization of the traditional two-point discrimination test in rehabilitation hospitals at this stage. Through the software and hardware design of the system, this system integrates multiple functions such as testing, electronic recording, chart analysis, etc. After field tests and comparative experiments in hospital, the results of quantitative test are almost the same as those of traditional test method, and the test time can be reduced about 90%. The system achieves such high accuracy and efficiency that it can be further popularized and applied.

Abstract:In order to solve the deficiencies in bandwidth, flexibility, transmission distance and reliability of traditional buses in high-speed data transmission between FPGA and FPGA, a Serial RapidIO (SRIO) interconnection architecture and photoelectric conversion module are proposed. The long-distance high-speed data transmission program. This solution uses Xilinx's ZYNQ7000 series FPGA and Kintex7 series FPGA as the interconnection equipment of RapidIO. It implements the SRIO transmission protocol by calling Xilinx's IP core and completes the communication by programming the user interface of the logic layer. The data exchange interface adopts a photoelectric conversion module and replaces it with an optical cable. The cable realizes long-distance data exchange. After a large number of tests, the solution achieved a transmission speed of 444Mbps by connecting a 10km optical cable at both ends of the communication, zero error and zero frame loss during the transmission process, and the system is stable and reliable.

Abstract:On the basis of analyzing on localization error of DV-Hop algorithm in WSN, an improved DV-Hop algorithm (ISSA-DH) combining hop distance improvement and sparrow search algorithm is proposed. Firstly, the accurate hop number is refined by communication radius, and the weighted correction factor is added to reduce the error of average hop distance; Then the hop distance is further weighted by the deviation between the estimated distance and the actual distance; Then the improved sparrow algorithm is used to replace the least square method, and the localization problem of wireless sensor is transformed into an optimal problem to obtain the position of the unknown node. Simulation experiments show that under different conditions, the localization error of ISSA-DH algorithm was reduced by 50.3% and 34.3% ,compared to the DV hop algorithm and improved algorithm (DCAI DV-Hop).

Abstract:Continuous phase modulation is a kind of modulation with constant envelope, which has high frequency spectrum utilization and power utilization. Its constant envelope property can be well adapted to nonlinear amplifier modulation. So continuous phase modulation can be used in satellite communication, deep space communication and other communication fields. In order to improve the system performance, a system model combining serial cascade code and continuous phase modulation is used. Due to the high complexity of the traditional decoding algorithm, the improved Max-Log-APP algorithm is used for decoding. In the additive Gaussian white noise channel, a parameter optimization method is designed to solve the problem of the correlation between decoding iterations and CPM parameters. The results show that under the condition of low SNR, the system performance can be improved about 1dB by optimizing the bit error rate.

Abstract:In order to solve the problem of high peak average power ratio (PAPR) in optical OFDM system effectively, a design scheme of optical OFDM baseband system is proposed, which combines optical OFDM baseband system and constant-enveloping OFDM modulation.Constant envelope optical OFDM baseband system modulates the OFDM signal to the phase of the constant envelope carrier signal on the basis of ordinary OFDM modulation, and finally modulates to the optical carrier signal. The PAPR of the optical OFDM signal obtained is reduced to 0dB, which solves the nonlinear problem in optical fiber transmission.In this paper, the mathematical model of constant envelopment OFDM baseband system is built in Matlab for simulation, which is implemented on FPGA hardware, and the optimal modulation index is found in the built system.The experimental results show that the PAPR can be reduced to nearly 0dB in the constant envelopment OFDM base-band system in practice, and the system obtains the highest spectral utilization when the modulation index is less than or equal to 1, and the system obtains the lowest bit error rate when the modulation index is 1.3.

Abstract:Linear regression classification is a fast and effective method in face recognition. However, linear regression classification is based on image vector recognition, which leads to the fact that the original matrix image is often high-dimensional data, and the face image is often contaminated. In order to solve this problem, a robust linear regression classification algorithm based on PCA and IGG weight function is proposed in this paper. Firstly, PCA is used to reduce the dimensionality of the face image, then the IGG weight function is adopted to classify the contaminated face image. Linear regression classification, robust linear regression classification based on IGG weight function and robust linear regression classification based on PCA and IGG weight function methods are compared with the public ORL and Yale databases. The experimental results show that the average recognition rate of the proposed method is above 92.07% without noise and with salt and pepper noise and speckle noise, which are higher than the other two methods in the ORL and Yale databases.

Abstract:In the contour measurement system based on machine vision, optical distortion exists in the imaging due to the uneven lens transmittance of the camera, which affects the measurement accuracy. In order to correct the optical distortion, the distortion mathematical model is firstly established, and then the distorted contour data is obtained by measuring the square calibration plate. According to the geometric characteristics of the distorted contour data and the linear constraint conditions, the parameters in the distortion model are calibrated by using LM optimization algorithm. The same contour measurement system was calibrated several times by experiments and the calibration results were compared with the existing methods. The experimental results show that the standard deviation of the proposed method is smaller than that of the existing algorithms, which indicates that the proposed method has better robustness.

Abstract:The millimeter-wave is an electromagnetic wave without ionizing radiation. It can penetrate the insulating cloth and is harmless to the human body. These characteristics make the millimeter wave have a wide range of application prospects in the field of public safety. Apply deep learning to the field of millimeter-wave image object detection, a millimeter-wave image object detection method based on improved YOLOv3-Tiny is proposed. Firstly, add convolutional layers to the feature extraction network to increase the depth of the network and increases to 3 different scale prediction layers to enhance the detection ability of millimeter-wave image object. Then, the Convolutional block attention module is introduced in the Feature pyramid network to make the network pay more attention to the features of targets and ignore the characteristics of redundant background noise. The results show that the improved network has mean average accuracy up to 93.4%, single frame detection speed is 15 ms, model parameters are only 38.7M, which provides a reference value for the research of high precision and miniaturization of millimeter wave security system.

Abstract:In remote sensing and scene text images, the target has the characteristics of directional diversity and large scale change, which makes the common target detection methods have poor detection effect in these two scenes. Aiming at this problem, many specially designed detection methods have been born. Integrating the orientation angle information into the candidate area generation network or designing a special orientation angle prediction network is the mainstream method of orientation target detection, which is of great significance to remote sensing and scene text image detection. This paper summarizes the research status of rotating target detection in remote sensing and scene text. According to whether there is an anchor box or not, the current rotation detection methods based on deep learning are divided into three types: one-stage method based on anchor, two-stage method based on anchor and anchor free method, and compared from the aspects of advantages and disadvantages, backbone network and applicable scene. Finally, the development prospect and research direction of rotating target detection methods are prospected.