Desert seismic recordings contain lots of complex noise which reduces signal-to-noise ratio. To solve this problem, an adaptive rank convergence denoising algorithm combining VMD ( Variational Mode Decomposition) with MoG-RPCA (Mixture of Gauss-Robust Principal Component Analysis) is proposed. The desert seismic data is firstly decomposed by VMD. All the decomposed modalities are rearranged into a new signal matrix, and then the matrix is subjected to low-rank decomposition by MoG-RPCA. When the error of decomposition satisfies the pre-determined requirement, the efficient low-rank component is extracted. Finally superimpose all the modalities of each channel signal in the low-rank matrix and substract from the original seismic data to achieve denoising. This method avoids choosing the modes of VMD and performs an adaptive rank convergence to the traditional low-rank decomposition. Simulation experiment and actual data processing show that the algorithm can effectively suppress low-frequency noise while maintaining more than 85% amplitude of the effective signal.

Aiming at the problem that the common performance monitoring method for engine is not applicable under high power microwave environment, a novel performance monitoring system combining field-wire with electromagnetic pulse protection and passive optical network access technology is proposed. An electromagnetic simulation model of monitoring node coupling response is built, and the electromagnetic protection performance of CAN (Controller Area Network) bus diagnosis node under typical high power microwave environment is analyzed quantitatively. The test results indicate that the proposed performance monitoring system has strong electromagnetic pulse immunity under broadband high power microwave with a peak field intensity of 50 kV/ m, monitoring data and control command have been transmitted between test area and monitoring area in a real time.

Aiming at the characteristic that it is difficult to accurately extract useful signals when the signal is interfered by strong noise while there is a leaking at the natural gas pipeline, an effective signal denoising method combining variational modal decomposition and generalized morphological filtering is proposed. Firstly by using VMD (Variational Mode Decomposition) several modal component is decomposed. Then the mean absolute value for autocorrelations function of the model component is calculated. Using SCT (Statistical Change-point) modal and effective modal noise are distinguished. And reconstruction after effective modal component as the denoising signal. Finally, the GMF (Generalized Morphological Filtering) is used to give further filtering to the denoising signal. The experimental results show that compared with the method based on Hausdorff distance VMD, VMD combined with correlation number and wavelet, and VMD based on mutual information, the proposed method has better denoising effect.

Aiming at the problem that the GCC(Generalized Cross-Correlation) time delay estimation method will produce large errors under the condition of low signal-to-noise ratio, a method of time delay estimation based on VMD( Variational Mode Decomposition ) combined with GSCC ( Generalized Second Cross-Correlation ) is proposed. This method first performs variational modal decomposition of the two signals separately, separates the effective mode and the noise mode, uses the HD (Hausdorff Distance) to optimize the mode and reconstructs the signal. Then uses the generalized second cross-correlation to analyze the processed signal and perform delay estimation. Theoretical analysis and simulation experiment results show that compared with the generalized second cross-correlation method, wavelet denoising combined with the generalized second cross-correlation (WT-GSCC: Wavelet-GSCC) method, this method can effectively improve the estimation accuracy and has good anti-noise performance.

 Due to the uncertainty of renewable energy and load, power flow analysis of the power system needs effective tools. Many existing literatures assume a given set of PDF ( Probability Density Functions) to model uncertainties and develop parametric probabilistic power flow tools. A nonparametric probabilistic power flow analysis method is proposed to determine the partial differential equation of power flow output. The method is based on the first order saddle point approximation of the mean value. For system with N random variables, the first order Taylor series expansion is established by power flow calculation, and then the probability characteristics of the expected output variables are determined by saddle point approximation. The proposed nonparametric estimator can provide accurate results while requiring reasonable computation. And the probability distribution function and cumulative distribution function of power flow output are directly established without using integral or differential operators. The test results on IEEE 14 bus and IEEE 118 bus test systems show that compared with other methods, mvfospa(Mean Value First Order Saddle Point Approximation) reduces the running time of MCS (Monte Carlo Simulation )algorithm by 12% . The effectiveness of MVFOSPA method is verified.

In order to realize the fusion of infrared image and visible image, a neural network structure of infrared image and visible image fusion based on the visual imaging mechanism of rattlesnake is designed. Firstly, according to the six response modes of dual-mode cells, six response results of infrared and visible image are obtained. Then, based on the mathematical model of visual receptive field, the neural network structure of infrared image and visible image fusion is designed, and six kinds of dual-mode cell responses are input into a two-layer network structure composed of on countermeasure system and off countermeasure system. Finally, the mapping values of R, G and B channels and the pseudo color image enhancement results are output. Four groups of registered infrared and visible images are fused respectively. And the fusion results are compared with the classical Waxman method. The experimental results show that the fusion image effect of the designed network structure is better, and the information entropy and average gradient are better than the classical Waxman method.

In the traditional vibration measurement, the laser monitoring device measures the single parameter of the motor only. To measure the mechanical vibration of the motor whose shaft center height is 56 mm and above, the dual channel independent measurement system is established, which can be used to measure the speed and vibration of the motor shaft at the same time. The vibration signal and velocity signal are obtained based on the principle of Fabry-Perot interference and speckle interferometry. After the synthetic signal is decomposed, the vibration signal is demodulated by Hilbert transform and the velocity signal is processed by autocorrelation function. The experimental results show that the accuracy of vibration measurement is less than 255 nm, and the relative error of velocity measurement is less than 0. 11% . The advantages of this experiment are non-contact, anti-electromagnetic interference and high precision. It is simple in structure and easy to operate and implement.

In order to solve the problems such as low recognition rate and poor anti-interference ability of speech signal feature extraction, a method of feature extraction based on improved empirical modal decomposition algorithm is presented. Classification by the method including noise speech signal decomposition, two types of modal component processing, reconstruction and feature extraction, respectively, to solve present most speech signal feature extraction process will filter out part of the original signal, on the basis of effectively eliminate the noise signal, as much as possible to save the original signal. And the recognition performance of system is improved obviously. Experimental results show that the proposed algorithm can achieve a 95. 5% recognition rate without adding noise. Compared with several traditional algorithms, this algorithm maintains a high recognition rate when adding different proportion of noise.

When multiple UAVs (Unmanned Aaerial Vehicle) fight in coordination, cooperative path planning is needed to improve mission success rate. After transforming constraints of cooperative path planning into multiple targets, the fusion design of NSGA(Non-Dominated Sorting Genetic Algorithm)-Ⅲ algorithm and potential field ant colony algorithm are carried out. Firstly, the potential field of the map is constructed to make the nodes close to the obstacles difficult to be selected, and to guide the search direction. Then, the path cost, spatial cooperative constraint and temporal cooperative constraint are modeled and converted into numerical indicators, and are set as multiple targets of NSGA-Ⅲ algorithm. For NSGA-Ⅲ algorithm, critical layer selection method and evolutionary algorithm are designed. Finally, in two-dimensional and three-dimensional grid map, the improved NSGA-Ⅲ algorithm uses each population to search the desired path for each UAV. Simulation results show that the UAV paths obtained by planning are safe and cost less.

In order to improve the accuracy of gesture recognition algorithm using sEMG ( Surface Electromyography) signals and solve the problem of accuracy affected by various features extracted, a sEMG's recognition method based on MB-CNN (Multi-Branch Convolutional Neural Networks) is proposed. Firstly, a MYO armband is utilized to sample sEMG signals of 8 different gestures. Secondly, the sliding window method is used to detect active segment of sEMG signals and the original training samples with the size of 64×8 are obtained. Thirdly, as a comparative experiment, seven different time-domain and frequency-domain features are extracted from original samples and machine learning algorithms are used to achieve the gesture recognition. Finally, in the case of avoiding conventional feature extraction, a MB-CNN model is constructed to achieve the gesture recognition and the accuracy of test set gains 97.89% . Experiment shows the proposed method is efficient and feasible for gesture recognition.

Aiming at the problems of serious chattering and slow control speed of traditional sliding-mode observers, the ant colony algorithm is used to optimize the sliding mode observation controller. In order to improve the search speed and the global optimal solution optimizing ability of ant colony algorithm, the chaos idea is combined with ant colony algorithm. And the improved chaotic ant colony algorithm is used to optimize the gain of the sliding-mode observer to advance dynamic response characteristics and speed estimation accuracy of IPMSM(Interior Permanent Magnet Synchronous Motor) speed control system. Finally, the method proposed is verified by simulation. The simulation results show that this speed control system has faster response speed, lower speed estimation error and better steady-state accuracy in speed control. The effectiveness of this method is fully demonstrated.

In order to solve the effective refractive index measurement problem of the ridged waveguide phase modulation, a three-port MZI(Mach-Zehnder Interferometer) structure is proposed. It can quantitatively measure and analyze the relative changes of the real and imaginary parts of the effective refractive index with the voltage varies of the PN ridged silicon optical waveguide and the polynomial fitting equation is derived. The experimental results are in good agreement with the fitting results, and then finally the characteristics of the ridged waveguide in the phase modulation process are obtained. This measurement method is simple and feasible, and can be used in silicon based optoelectronic integrated chips as a quantitative device for the detection of carrier modulation characteristics.

In order to achieve the rapid test of the optical phase control performance of the optical phased array (OPA: Optical Phased Array), i. e. to quickly test the configuration voltage corresponding to the OPA beam scanning angle, a closed-loop test system with self-feedback function is designed. The 64-channel drive power supply is controlled by the host computer to provide the configuration voltage for the OPA. The far-infrared camera detects the light field distribution of the OPA's emitted light on the Imaging screen, and the beam scanning angle is determined according to the main lobe position. Then we compare the deflection angle with the preset deflection angle by cosine similarity, and further adjust the configuration voltage of each OPA until the best actual configuration voltage corresponding to the set beam scanning angle is obtained. Experiments have proved that the system can achieve a rapid test of 64 configuration voltages corresponding to the OPA specified beam scanning angle within 16 minutes, and the peak sidelobe level obtained is about -10 dB. The system has high stability, strong anti-interference ability and fast speed.

The marine environment is complex and changeable, and the target recognition accuracy of a single underwater sensor can not meet the performance requirements of the system. Multi-source sensor fusion is an effective method that can improve the target recognition rate. It has received extensive attention and research. Underwater data is noisy and has high dimensions, and direct data fusion can not get better results. For multi- scene underwater multi-source test data, denoising autoencoder and multiple dimensionality reduction methods is used for multi-angle feature-level fusion. First, the denoising autoencoder is used to remove noise and reduce the data dimension, and extract new features from the source data. Then, the data cascade method is used for multisource data fusion for new features. The fusion methods include principal component analysis, independent component analysis and isometric mapping. Comparative experiment results in different scenarios show that the proposed method gets better classification results, and principal component analysis can achieve a higher target recognition rate.

In order to understand the relationship between social media usage and information cocoon, this research takes Sina Weibo as an example, to analyze information cocoon accompanied by Weibo usage, activity, and impact. We use Word2vec, one of accessible NLP ( Natural Language Processing) technology of word embedding, and k-means, a kind of clustering method, to explore the information cocoon and narrowing scope. The result of statistical paired T test shows that, as the development of users' level in social media, there is a remarkable trend of rising in semantic similarity of UGC(User Generated Content). The distribution and richness of content categories will also decrease accordingly. The result inspires us to rethink the relation between social media usage and information cocoon. The classification of users does not bring more flexible discourse space. Rather, deeper, and higher users suffer more from similar content.

In order to detect the duplicate of Chinese data, we propose a duplicate record detection algorithm based on SNM (Sorted-Neighborhood Method) algorithm, which integrates the extended version of synonym word forest and Chinese word segmentation. Using the extended version of synonym word forest and Jaccard algorithm to calculate the similarity of words, the Chinese word segmentation in Python is used to segment sentences, to optimize cosine similarity algorithm and to calculate the similarity of sentences. The improved algorithm can effectively detect duplicate records of fields and sentences recorded in Chinese. The experiment on the test data set of students in a counseling institution shows that the recall ratio of the new algorithm is much higher than that of the traditional SNM algorithm.

In order to provide scientific and technological resources, and improve the quality and level of information services, the data mining technology and user portrait modeling method are applied to the information service platform of science and technology literature in Jilin Province ( referred to as “ platform”), and the correlation between data can be found and mined according to the historical data of user downloaded literature. The upgrade of platform function greatly improves the retrieval performance and information acquisition performance of platform users, improves the scientific research performance of users, and improves the support and guarantee function of platform. The user behavior analysis has been well applied in platform V2. 0.