Aiming at the shortcomings of Salp Swarm Algorithm (SSA: Salp Swarm Algorithm) that are easy to fall into local optimality and slow convergence when solving feature selection problems, Based on salp swarm optimization algorithm, its improved version, differential evolution salp swarm feature selection algorithm(DESSA: Differential Evolution Salp Swarm Algorithm) is proposed. Differential evolution strategy is applied to replace the ordinary operator as the new way of moving particles to enhance search capabilities. And evolutionary population dynamics ( EPD: Evolution Population Dynamics) is proposed to enhance convergence efficiency.Utilizing K-nearest neighbor (KNN: K-Nearest Neighbor) as classifier and eight datasets come from the UCI (University of California Irvine) machine learning repository, DESSA is compared with the SSA and other high performing approaches proposed recently. The experimental result confirms the efficiency of DESSA in improving the SSA in several respects and its ability to better solve the problem of feature selection compared with other approaches of feature selection.

DAS (Distributed Acoustic Sensor) technology is a novel high-precision acquisition method. And DAS is considered as a new generation of seismic data acquisition technology which has the potential to replace the traditional seismometer array. However, the processing ability for the DAS records needs to be further improved due to the influence of complex background noise. To solve the problems, the stationarity and Gaussianity of complex DAS background noise are investigated by using the time series stationarity testing method based on substituted series and Lilliefors test method respectively. The actual seismic data analyzed are the borehole DAS records collected under the requirements of actual exploration industry. The results show that DAS background noise is a non-stationary and non-Gaussian random process. The noise characteristics have been tested changing with the noise recorder length. The accurate understanding for the characteristic of DAS background is obtained, which has potential significance for improving processing capability of DAS data.

The energy limitation of sensor nodes is the key factor to restrict the performance of the oil and gas pipeline Internet of things. The end-to-end delay of the network determines the response time of the oil and gas pipeline Internet of things to pipeline accidents. For solving the problems of energy limitation and end-to-end delay, the pipeline internet of things wireless routing algorithm based on multi-object optimization is proposed. It can realize the energy balance of the Internet of things by taking the remaining energy of nodes and the distance between nodes and aggregation nodesas the index of selecting candidate forwarding nodes. The end-to-end delay of the network is reduced by decreasing the number of data transmission hops. The simulation results show that the routing algorithm can effectively improve the network performance by extending the network life and reducing the network delay compared to the classical opportunistic routing algorithm.

An image denoising algorithm based on BAS-2D-VMD ( Beetle Antennae Search Algorithm-2D-Variational Mode Decomposition ) is proposed for the defect that the 2D-VMD ( 2D-Variational Mode Decomposition) algorithm needs artificially preset K value. This method first uses the BAS algorithm to search for the optimal value of the 2D-VMD parameter K, and then decomposes the image on 2D-VMD to obtain K sub-modes. Mode screening is performed by calculating the correlation coefficient of each mode. Finally, the remaining modes to an image is reconstructed. The image quality is judged by calculating the peak signal-to-noise ratio, and the denoising effect is verified by simulation. This algorithm is applied in the field of image processing of overhead transmission lines, and denoising line images collected by the UAV(Unmanned Aerial Vehicle).Compared with the traditional median filtering and wavelet threshold denoising algorithms, a better denoising effect is obtained. The problem that the line image noise is difficult to remove is solved.

In view of problems such as the difficulty in selecting effective modal components and the unsatisfactory denoising effect after the decomposition of VMD(Variational Mode Decomposition) algorithm, an optimization algorithm is proposed to improve PSO ( Particle Swarm Optimization) by adjusting the inertia weight and the acceleration factor, which combines PSO with VMD algorithm. The improved PSO algorithm is used to optimize the decomposition mode number k and the punishment factor α of VMD and to conduct the decomposition of mode. Then the ED( Euclidean Distance) is calculated between the probability density function of each modal component and the probability density function of the signal, and the effective modal component is selected to reconstruct the signal. Experimental results show that compared with VMD-CORR ( Variational Mode Decomposition-Correlation Coeffificient ) algorithm and EMD-ED ( Empirical Mode Decomposition-Euclidean Distance) algorithm, the proposed algorithm achieves better denoising effect for both simulated signals and actual pipeline leakage signals, which verifies its effectiveness in pipeline leakage detection.

In order to further analyze and design the control system, the dynamic output feedback controller is designed for discrete time-delay singularly perturbed uncertain control systems to make the closed-loop system asymptotically stable. A new quadratic summation type L-K function is constructed for both delay dependent and delay independent case. The functional difference process is amplified by cross-term defined method, and the uncertainty of the system is eliminated by using the appropriate lemmas, the sufficiently existent criterion of dynamic output feedback controller with time delay is derived to expand the perturbation control range of the controller. The effectiveness and feasibility of the proposed method are verified by the showed example. Based on comparing the corresponding literature, it is shown that the proposed controller has certain advantages and makes the closed-loop system asymptotically stable. It meets the design requirements, and can achieve the secondary control effect.

The traditional tracking differentiator will vibrate at high frequency after entering steady state, in order to overcome this phenomenon, an improved algorithm is introduced. The tracking differentiator is modeled and packaged by using S-Function in Matlab / Simulink software, then the simulation experiment analysis is done. Numerical simulation results show that the tracking differentiator using the new comprehensive control function can track the input signal without overshoot and eliminate the high frequency vibration of the differential signal.For the mixed noise signal, it can effectively suppress the noise, restore the initial signal, and has a goodfiltering effect.

In order to solve the problems of slow convergence speed and low precision of hybrid frog leaping algorithm for continuous function optimization problem, a cellular shuffle leapfrog algorithm is proposed. The neighborhood structure of cell is used to replace the grouping method of basic leapfrog algorithm to overcome the shortcomings of classical shuffle leapfrog algorithm. The algorithm reduces the selection pressure and maintains the population diversity through the neighborhood structure and evolution rules of cellular automata. The improved spiral evolution method and chaos mutation method are used to balance the relationship between local search and global optimization to improve the speed and accuracy of optimization. Comparing the proposed
algorithm with five improved leapfrog algorithm, it can be seen that the algorithm can get good results, for 10 typical benchmark function optimization problems and oilfield measure planning scheme to solve output input ratio.

In order to solve the problem of using voice signals to accurately identify animals so as to protect and research wild animals. We propose a FCNN (Fully Convolutional Neural Network) combining a fully connected algorithm and a sparse connection algorithm for automatic speech recognition. The fully connected algorithm is used to extract more combined features, and the sparse connection algorithm to select important features to speed up the convergence. The specific model structure and algorithm flow are given, and speech recognition experiments are carried out. The experimental results show that the fully convolutional neural network deep learning algorithm is an effective method for automatic speech recognition. It can solve the problem of frog sound recognition and provide a reference for animal speech recognition.

In view of the fact that the fault tree analysis method can not analyze the pipeline risk polymorphism, and can not realize two-way reasoning, a calculation method of pipeline failure probability based on Bayesian network is proposed. Firstly, the fault tree model of oil field pipeline failure risk is established, and the Bayesian network structure is determined by the transformation of fault tree and Bayesian network to complete the construction of the Bayesian network model structure of pipeline failure risk. Secondly, considering the large estimation error of network parameters determined by expert knowledge experience and expectation maximization algorithm, genetic algorithm is introduced to complete the Bayesian network structure. Finally, this method is applied to the actual risk problem of oil field pipeline, and the failure probability of oil field pipeline is calculated by using the genie Bayesian network simulation software. And each risk factor is analyzed and the cause chain affecting the pipeline failure is obtained. The experimental results show that the method proposed has significantly improved the evaluation accuracy.

In order to solve the problem of multiple and complex data sources for the credit degree of enterprise,a new method based on the traditional linear regression method is proposed to evaluate the credit degree of enterprise. Then, the obtained credit value of enterprise can effectively define the reliability of the enterprise. To calculate the average value of several important parameters from the enterprise, we can obtain the balance value of the credit rating for enterprise. Because the improved linear regression reduces low-impact parameters, it optimizes the errors generated by the traditional linear regression method and accurately calculates the creditworthiness of the enterprise. The experiment is shown that the improved linear regression is generally better than the traditional linear regression in the evaluation process of enterprise credit. In particular, it requires a small number of variables to achieve similar results, and produces the better classification efficiency.

With the emergence of semantic web technologies and knowledge maps, most of the current query models require semantic matching between query results and user queries. The simple query process can not meet the user’s query requirements. Therefore, the rewriting techniques and implementation methods involved in knowledge graph query are studied. On the basis of defining the rewrite rule set of SPARQL(SPARQL Protocol and RDF Query Language) query mode, the SPARQL is rewritten by Prolog. In the distributed data storage environment, through the test analysis of the experimental data of LUBM(Lehigh University Benchmark), it is found that the rewritten query can mine more semantic information in the knowledge map than the original query.

So as to ensure that the learners can understand learning progress of the learner, group and peer learners accurately and induce the learners' meta-cognitive learning experience and social comparative learning experience, we use social comparative theory to design and develop an open social learner model based on the table which is called as TableOSLM. TableOSLM presents learning progress of the learner, group and peer learners in tabular form, In the empirical research, evaluation data is collected through questionnaire survey, experimental research method and face-to-face interviews. The results show that the TableOSLM can well induces meta-cognitive learning experience on learning planning and self reflection, TableOSLM can well induces social comparative learning experience on comparison method and self-improvement. Learners have shown strong acceptance to usability, usefulness, and satisfaction with TableOSLM. Usefulness, and satisfaction with MindOLM.

In order to measure the resonant frequency and the quality factor of quartz-enhanced photoacoustic spectroscopy gas sensing system, to serve the teaching requirements of R-L-C frequency selection circuit in the ourse of “ high frequency electronic circuit冶, to and promote the full integration of teaching and scientific research, a frequency sweep and performance test experimental system for quartz tuning fork is designed and implemented. The principle of R-L-C electrical model of quartz tuning fork is firstly analyzed, and the model is simulated by Matlab. In order to amplify the output signal of a quartz tuning fork under sinusoidal excitation or external sound field, a transimpedance amplifier circuit is designed. The frequency sweep and performance test system of quartz tuning fork is established by using a lock-in amplifier, a trans resistance amplifier, a signal generator and a LabVIEW signal platform. Using the given quartz tuning fork, the equivalent circuit model parameters, resonant frequency and quality factor are tested under vacuum packaging and shell removal. This system integrates the basic knowledge of R-L-C frequency selection circuit and expands the scientific research application cases based on quartz tuning fork. The practical application shows that the system can meet the experimental teaching requirements of R-L-C frequency selection circuit and scientific research needs of quartz enhanced photoacoustic spectroscopy technology, and has achieved good teaching and research application results.

In the process of iOS application development, due to the insufficient functions of the system's own methods, some business requirements can not be effectively realized. In order to solve this problem, the main API interface usage of the Runtime library is studied and the available interfaces are found. The message forwarding mechanism of Runtime is studied to prove that the essence of the function call is the transmission of messages. And through actual cases, it proves that the application of Runtime can solve the problems caused by the lack of system methods. Therefore, it can dynamically add or modify member variables and member methods of the system's own classes through the Runtime library. And finally it is proved that the use of Runtime library can solve the problem of insufficient functions of the system, which can provide reference for iOS developers.

In view of the low correlation of the variables in the traditional method, which leads to the big error in the prediction results, a method based on logistic algorithm is proposed to predict the performance variables of the postgraduate entrance examination. The data of postgraduate entrance examination and student achievement over the years is collected and processed as the initial data of achievement variable prediction. The predictive variables of the performance of the postgraduate entrance examination is set up, the logistic regression classification algorithm model is established, and the correlation between the performance variables through the operation of the model is improved. Based on the analysis of the development law of the data and the influencing factors of the variables, the prediction results of the variables are obtained. Through the comparative experimental analysis, it is concluded that the prediction error rate of the method based on logistic algorithm is low and the prediction accuracy is high.

In order to achieve the intelligent control of home system, a solution based on brain computer interface, recognition and classification of brain electrical signals and AR (Augmented Reality) is proposed.EEG (Electro Encephalo Gram) signals are collected and extracted from the brain by wearing a device, and the data is de-noised by wavelet transform and further processed by STFT ( Short Time Fourier Transform). Then,PCA(Principal Component Analysis) and CNN(Convolutional Neural Network) are used for classification to form a classification model. According to the classification results, the instructions is obtained from the brain to control the home. Combined with AR technology, the control process can be visualized and is more interactive,which is in line with the development trend of the control method of the future smart home.