A laboratory calibration and a downhole online calibration method for the JLU-FG03A downhole fluxgate magnetometer are designed for the application needs of geomagnetic field observation under high pressure and humid environment, and the construction of the experimental test system and the related test methods are presented. The laboratory calibration method tests the range, bandwidth, noise and sensitivity of temperature drift of the magnetometer, while the downhole online calibration method tests the reliability of the magnetometer data and noise levels when working in the well. Laboratory calibration results show that the instrument has a range of +- 100 000 nT, a bandwidth of DC-10 Hz, RMS ( Root Mean Square ) noise less than 0. 01 nT (DC-0. 3 Hz) and a temperature drift of 23. 39 ppm when the adaptive feedback function is on, which meets the application requirements for downhole geomagnetic observation. The downhole online calibration experiments show that the instrument is in normal working condition downhole and the noise measurement results indicate a good downhole magnetic environment, which is conducive to the best performance of the magnetometer.

Feature extraction is the basic work for data mining. The quality will largely affect the results of the excavation. The algorithm for LLE ( Locally Linear Embedding) does not consider the correlation between different characteristics of the same data, and can not well retain the main form trend of timing signals. We proposed CC-LLE ( Local Linear Embedding Algorithm Based on Characteristic Correlation) which is used to diagnosis of bearings. In response to the periodic characteristics of the bearing fault signal, the algorithm combines the data segmentation during the feature extraction stage. The standard deviations on each segment are selected as a characteristic to construct the characteristic sample set of the original data to effectively extract the identification characteristics. The experiments on the bearing data set proved the effectiveness of the algorithm in the feature extraction.

In order to solve the problem that the existing scheduling algorithm can not adapt to the high retransmission probability caused by the rapid decline of the air port signal quality in the handover scenario, a new handover command transmission scheme optimized for the handover scenario applied to the B-TrunC is proposed to reduce the bit error rate of handover command. In the new transmission scheme, the PDCP( Packet Data Convergence Protoco1) module identifies the handover command sent to the terminal, and distinguishes the handover command from the ordinary signaling and data transmission processing, using the specified low-key mode and coding rate transmission. The simulation results of switching scene show that the new transmission scheme can effectively reduce the re-transmission and failure probability of handover command, and significantly improve the handover delay performance of B-TrunC.

Since various unpredictable disturbances are found in the actual system, it is difficult for the system to accurately achieve the expected performance. Therefore, the probability constraint H_∞ performance is used to make the system more suitable for practical engineering applications. In order to avoid network congestion and resource occupation, RR protocol is introduced to schedule data transmission between network nodes. The filter of probability-guaranteed and non-fragile is constructed. The uncertain parameters of the system are controlled by random variables, which are uniformly distributed and independent of each other. The filter that can guarantee performance requirements under probability constraints is designed by seeking a parameter box. The probability-guaranteed H∞ filtering problem is solved by recursive matrix inequality. Finally, the effectiveness of the filtering scheme is verified by a simulation example. 

In the two-phase hybrid stepping motor position control system, the sliding mode algorithm is adopted. However, the problem of large chattering exists. The sliding mode control algorithm based on the concept of boundary layer is improved. And the quasi-sliding mode sliding control algorithm is optimized. A two-phase hybrid stepping motor model is built in SIMULINK, and simulation verification is carried out. The results show that the use of quasi-sliding mode control reduces the litter amplitude of the system by 50% compared to the general sliding mode control. Dynamic performance is greatly improved.

In order to optimize the magnetic coupling mechanism of underground radio energy transmission system in oil field, an improved solenoid type magnetic coupling mechanism is proposed for the magnetic coupling mechanism of the underground wireless power transmission system in the oil field. And an optimization design method is proposed by using the Maxwell electromagnetic field simulation tool. The characteristics of the LCC-S high-order compensation topology are studied. The compensation design formula and main circuit characteristics of LCC-S are obtained. Finally, the feasibility and correctness of the proposed coupling structure and compensation topology are verified by closed-loop PI ( Proportion Integration) control through PSIM ( Power Simulation) simulation software.

In order to improve the magnetic flux leakage caused by the traditional Halbach magnetizing structure in the non-air gap side of the linear generator, which leads to the reduction of the main flux content, power density and load capacity, the traditional Halbach magnetizing structure and the magnetic field distribution of the linear generator is improved. Firstly, the structural dimensions of traditional rectangular permanent magnets are optimized, and the variants of trapezoidal, T-shaped and U-shaped Halbach magnetizing structures are proposed. Then, the fundamental wave content of magnetic flux density under different sizes is analyzed, so as to find the optimal design size of each variant under the condition of maximum magnetic flux density content. Through the simulation results, it is found that the U-shaped Halbach magnetizing structure is more obvious than other structures in improving the magnetic flux leakage of the non-air gap side, and the output power and efficiency are improved.

The introduction of deep neural network technology greatly improves the extraction accuracy of text semantic features of relation classification. The common sense knowledge graph is used to construct the contextual semantics other than the text′s own semantics, and the pre-trained model is used to obtain the contextual semantic features. Aiming at the semantic features of text, context and marked entity, a multiple semantic fusion mechanism is established to realize the relation classification model, which is named MSF-RC. The model is tested on two different datasets, SemEval-2010 task and TARCED. The experimental results show that the introduction of contextual information helps to strengthen the semantic understanding of labeled entities, and the hierarchical fusion of multiple semantics can further improve the performance of relation classification model.

In the previous alarm correlation analysis algorithms, the alarm importance is regarded as the same. In order to distinguish the difference in importance of different alarms and the difference in the amount of information contained in the alarms, an alarm correlation analysis algorithm with improved weighting method is proposed. First, the attributes related to alarm importance in the alarm information are quantified, and the XGBoost(eXtreme Gradient Boosting) integrated learning model is used to train them to obtain the weight value of the alarm attribute, and the weight assigned to the alarm data. Then, the network topology data is added to the sliding window to improve the problems in the division of transactions by the sliding window. The improved transaction set divided by the sliding window is more realistic and reliable. Finally, the weighted alarm transaction set is used to mine frequent alarms and association rules by using the weighted FP-Growth(Frequent
Pattern Growth ) algorithm. Experiments show that the alarm correlation analysis algorithm with improved weighting method has good performance in mining frequent alarms, important association rules and time.

At present, the integration models in various fields mainly focus on data integration and information integration, but these integration models lack the integration of standards. Ontology and linked data technology can fundamentally solve the “knowledge island冶 phenomenon, and link semantics at the pattern level, reducing data redundancy, and providing an integrated knowledge model for knowledge elements. Firstly, the traditional knowledge organization methods are compared and analyzed, and then a standard driven knowledge integration model is proposed and constructed according to the knowledge organization methods of ontology and linked data. The model realizes the representation of domain knowledge ontology by re-using the domain thesaurus and the vocabularies in the relevant international fact standard ontology. Finally, the RDF ( Resource Description Framework) is serialized through the linked data technology, and linked with entities in the external knowledge base to publish unified and integrated linked data to provide knowledge services for the field. In order to verify the feasibility of the proposed model, taking the petroleum field as the background, the standard vocabularies such as
petroleum subject words is re-used, and the standardized petroleum field ontology is constructed, which integrates the petroleum knowledge in the petroleum field with the knowledge of GeoNames and DBpedia encyclopedia. The rationality and availability of the proposed model are verified. The proposed standardized integration model is also suitable for the integration of knowledge in other fields.

At present, Face sketch synthesis has a series of problems, such as generateing fuzzy outline, lacking of detail texture and so on. Therefore, using CycleGAN(Cycle-Generative Adversarial Networks) as a solution to build multi-scale cyclegan is proposed. Method innovation is mainly reflected in: The generator adopts the deep supervised U-net++ structure as the basis, and performs down sampling dense jump connection at its decoder; The encoder end of the generator designs the channel attention and spatial attention mechanism to form a feature enhancement module; a pixel attention module is added to the generator. Compared with some existing classical algorithms, from the subjective visual evaluation and using the existing four image quality evaluation algorithms for quality evaluation, the experimental results show that this algorithm can better synthesize the geometric edge and facial detail information of sketch image, and improve the quality of sketch image.

Traditional segmentation algorithms have problems such as low accuracy, low precision and time- consuming. Therefore, an adaptive segmentation algorithm for 3D breast ultrasound images using improved deep learning is proposed. First, the images are pre-processed, and the deep multiple example learning method is used to detect the lesion image blocks and remove the normal image blocks. Second, the breast ultrasound image data set is augmented and processed for neural network training. Then, a residual convolutional neural network model is constructed, a residual learning unit is designed, a feature mapping is formed by combining the augmented dataset. A softmax function is used to train the network and perform feature block judgment. Finally, combined with threshold settings, the achieves adaptive segmentation of 3D breast ultrasound images is realized. The results show that the proposed algorithm can complete image segmentation more carefully, and the average running time of the algorithm is 52. 3 s, the image segmentation accuracy is 95. 5% , the F1 score value is high, and the overall performance is good, which provides a reference for the application of convolutional neural network segmentation.

In order to solve the problem of strong forward security in digital signature schemes, a class of forward security digital signature scheme based on elliptic curve is analyzed. It is proved that the scheme does not satisfy the backward security. An elliptic curve digital signature scheme with strong forward security is proposed by introducing the double private key evolution method. After security analysis, the improved scheme has strong forward security and anti-forgery. It is also a backward-secure digital signature scheme.

The coverage of radar communication network gradually increases, the number of access users increases sharply, and the risk of terminal access expands. Radar communication terminal is one of the key equipment to block the access of bad users. Therefore, a security access control algorithm of radar communication terminal based on electronic signature technology is proposed. The security access control framework of radar communication terminal is build, applying CA authentication technology to authenticate the identity information of the user to enter the radar communication terminal, the user authentication information in the radar communication is stored terminal database, the user's personal electronic signature is made through electronic signature technology, and the electronic signature is matched through SIFT matching algorithm, It determines whether the electronic signature authentication information of the user of the radar communication terminal to be accessed is safe, to realize the control of the security access of the radar communication terminal. The experimental data shows that the security rate and security access efficiency of radar communication terminal obtained by the proposed algorithm are larger, and the security access control effect of radar communication terminal is better.

During the evolution of public opinion in group social networks, it is difficult for the current methods to obtain the data in key nodes, resulting in the inability to accurately obtain parameters such as the number of public opinion propagation, search index, time to reach the peak of public opinion, and the problem of low evolution accuracy. A simulation method of public opinion evolution in group social network based on clustering algorithm and SIS(Susceptible Infected Susceptible Model) model is proposed. The PageRank algorithm is used to obtain the key nodes, and the clustering algorithm is used to cluster the data in the key nodes. The SIS model is constructed, and the public opinion evolution simulation of the group social network is completed through the SIS model. The experimental results show that the proposed method can accurately obtain the parameters such as the number of public opinion propagation, search index and the time to reach the peak of public opinion, and the evolutionary simulation accuracy is high.

Even subgraph covers is an important subject in graph theory. Inorder to study the shortest cycle covers conjecture, using the connection between integer flows and even subgraph covers, a new upper boundary of shortest cycle covers of bridgeless graphs is obtained by means of the conclusion of expanding an integer 4-flow in a circuit of a graph. The result improves Fan's conclusion.

Aiming at the problems of unreasonable route selection and low distribution efficiency of distribution vehicles, a distribution vehicle route optimization method based on improved cuckoo algorithm is proposed. According to the principle of shortest route distribution, the objective function is built, the relevant constraints of route optimization is set in order to simplify the model structure, ensure that each demand point can only be distributed once, and the vehicle must drive within the maximum distance load range, and establish the route optimization model. The nest position update process of classical cuckoo algorithm is analyzed, adjustment factor is added and the dynamic inertia weight is introduced. The optimization model is solved by cuckoo search algorithm, and the global optimal solution is continuously found through the process of population initialization and nest location update. When the iteration stops condition is met, the optimal optimization scheme is output. Experimental results show that this method has strong searching ability, uniform distribution of solution set, and can ensure the shortest distribution path and improve distribution efficiency.

In the short-term power load forecasting, when the power load data is sufficient, the accuracy of load forecasting is usually high, but when the data is missing or the data quantity is small, the accuracy of load forecasting is often poor. Therefore, when the power load data in a certain region is small, the load prediction accuracy is difficult to meet the prediction accuracy requirements. A short-term load prediction method based on 1DCNN-LSTM ( 1D Convolutional Neural-Long Short-Term Memory Networks ) and parameter transfer is proposed. 1DCNN-LSTM combined with transfer learning is used to solve the problem of low prediction accuracy. The actual load data of a certain area in the United States are used for simulation analysis. Experimental results show that this method can effectively improve the accuracy of load prediction when regional power load data is missing.

Aiming at the problems of poor prediction effect in the existing power system short-term load forecasting, a power system short-term load forecasting algorithm based on deep data mining is proposed. Taking the normalized historical power system load data, fuzzy temperature data, weather conditions, precipitation probability and other data as the input of the prediction model, a power system short-term load prediction model based on fuzzy gbdt is constructed, and the boosting algorithm is introduced to solve the problems of slow training speed and large memory occupation in the prediction model. The experimental results show that the short-term load forecasting results of the proposed method are close to the actual load at different times on weekdays and weekends. The MAPE and rmspe values of power system short-term load forecasting in the next week are lower than 0. 2% .

Failure diagnosis of oil pumping machine has low identification accuracy due to various faults and complex system, which increases the difficulty of fault diagnosis. After clarifying the working principle of SVM, the ant colony algorithm is carefully studied to adjust the penalty coefficient of SVM(Support Vector Machine) and the kernel function parameters. The ant colony algorithm has the problem of easy to fall into the local optimal solution, which introduces the idle ant to update the pheromone again after the ant colony algorithm fails to enable the ant group to obtain new paths. In order to further reduce the problem of local optimal solution of ant colony algorithm and improve the search speed of ordinary ants in the early stage of optimization, idle ants are optimized by using chaotic initialization and chaotic perturbation. The test data of the pumping machine shows that the proposed fault diagnosis system has high fault identification accuracy.

A multi-scale full-convolution timing model is proposed in order to detect the partial discharge phenomenon of high-voltage power lines in a timely manner. This method uses a multi-scale fully convolutional timing model to train the power signal data collected in high-voltage power lines. The trained model can be used to monitor the future continuous signal to detect whether it has a partial discharge phenomenon. The experimental results show that the model proposed has good accuracy on the used data set.

When the current method is used to detect the three-dimensional human motion image, it can not accurately obtain the target area in the image, resulting in the problems of low detection integrity, low detection accuracy, high false detection rate and low detection efficiency. Therefore, an intelligent detection algorithm of 3D human motion image based on adaptive projection is proposed. Firstly, the background rough extraction model is constructed, and the background region of 3D human motion image is extracted by using the model to obtain the human motion target region of the image. Secondly, the features of the target region are extracted by adaptive projection method, and the optimal classification function is constructed on the basis of support vector machine. The target region feature is input into the optimal classification function to complete the intelligent detection of three-dimensional human motion image. Experimental results show that the proposed algorithm has high detection integrity, high detection accuracy, low false detection rate and high detection efficiency.

Because the existing methods can not reduce the dimension of vehicle license plate, which leads to inaccurate recognition results, an intelligent vehicle license plate fuzzy recognition method based on Relief algorithm is proposed. The Relief algorithm is used to calculate the weight coefficients of different license plate image features and reduce the dimension of the feature set. The smart license plate is enhanced by sequence video images, the full convolution network is used to detect the significant areas of the license plate, roughly extract the significant areas in the image, the sliding window method is used to accurately detect the license plate in the candidate area, locate the exact position of the license plate, add the context information of the characters, accurately detect and recognize the characters, and finally achieve the fuzzy recognition of the smart license plate. The simulation results show that the proposed method can obtain high precision license plate recognition results.

In order to reduce the execution time of visual image target tracking and improve the accuracy of tracking track, a visual image target tracking method based on improved convolutional neural network algorithm is proposed. In order to obtain shorter target tracking execution time and better target tracking track, video image processing technology is used to extract the foreground of visual image, improved convolutional neural network algorithm is used to extract the features of visual image, MeanShift target tracking algorithm is used to track visual image targets on the basis of visual image features. And the tracking results of MeanShift target tracking algorithm are further optimized through Kalman filtering, realizing visual image target tracking. Experimental results show that the proposed method has short execution time and high tracking accuracy.

In the research of distributed multidimensional data stream frequent pattern mining algorithm, the non frequent items in multidimensional data stream are not deleted, and there is a problem of long average processing time. A distributed multidimensional data stream frequent pattern mining algorithm based on artificial neural network is proposed. According to the characteristics of artificial neural network, this method establishes an artificial neural network model and trains multi-dimensional data flow, so as to improve the mining efficiency; Based on the training results, a frequent pattern information tree, FR-tree ( Frequent Pattern tree ), is constructed. Because there are many expired multidimensional data streams in fr tree, it is necessary to prune fr tree and delete non frequent itemsets, so as to speed up the calculation of frequent patterns. Then, the distributed mining algorithm is used to mine the global fr tree to obtain the complete set of frequent itemsets of multidimensional data streams, so as to realize the mining of frequent patterns of distributed multidimensional data streams. The experimental results show that the average processing time of the method is tested to verify the practicability of the method.

In order to realize eye movement interaction with high frame rate, a deep-learning based single-camera dual-light source identification method is proposed. This method uses correlation between the reflected spot and the gaze landing point to obtain the mapping law from the eye image to the gaze. Human-computer eye movement interaction device is constructed, which results in a high-quality dataset, obtains the gaze estimating model with high precision and speed by training, solves the problems of complex mathematical model and large amount of calculation of gaze estimation. It realizes the function of real-time recognition and interaction of users蒺 gaze, supports the development and application of psychological experimental research and virtual reality application technology.

There is a deficiency of information available on waste classification, and many municipalities and educational institutions struggle with this issue. We address this challenge by utilizing the efficiency and accuracy of the neural networks to classify items and implement waste image classification with a deep residual shrinkage network built on the ResNet network and SENet network. By filtering the Garbage dataset to obtain the data set necessary for the experiment, and by enhancing ResNet, SENet and soft threshold processes are incorporated into the ResNet structure. And by training the network and optimizing its hyperparameters, a greater recognition rate and recognition effect are achieved for the classification of campus waste. The experimental findings indicate that the proposed approach is feasible to a certain extent.