To avoid the deficiency of excessive dependence on named entity recognition during the understanding perio, logic relationships among vital information in Chinese natural language question were understood by semantic relation. An algorithm for recognizing semantic relations based on dependency structures was proposed, which first recognized three kinds of valuable dependency structures that were vital for obtaining semantic relations, and then combined or transformed these dependency structures to obtain semantic relations. The effectiveness and scalability of the proposed method were verified by extensive experiments over Chinese benchmark question answering datasets, and the experiments results showed that this method could also understand Chinese natural language questions when recognition of named entity failed.

Aiming at automatic target recognition of aircrafts, a Chirplet neural network for joint feature extraction and target classification was proposed to realize recognition of one-dimensional high resolution range profiles. Based on the multilayer feedforward neural network structure, the Chirplet-atom transform was used to replace the conventional excitation function in the input layer for feature extraction, and the hidden layer and output layer constituted the classifier of the network. The network weights and the parameters of Chirplet-atom node were simultaneously adjusted and optimized to achieve joint feature extraction and target classification. The simulation results of the four types of aircrafts showed that the Chirplet neural network method with the four-feature-parameters had higher recognition rate and anti-noise performance than the time-frequency transformation and Gabor atoms network.

To investigate the clinical value of breast tumor magnetic resonance (MR) images in differentiating fibroadenoma of breast (FB), invasive ductal carcinoma (IDC) and invasive lobular carcinoma (ILC), the region of interest of MR image was selected and the MR image was decomposed by wavelet transform, and the region of tumor was segmented by K-means algorithm. Gabor wavelet was used to filter the region of interest from 8 directions and 5 scales, and the mean value of the tumor location was taken as the feature. The extracted features were analyzed and the key features were obtained. Different classification algorithms were compared in machine learning, such as support vector machine, Bayesian, and neural network, to classify and predict the key features, and calculate the accuracy, sensitivity and specificity of classification, so as to get the most suitable parameter settings for classification model. Texture analysis of breast MR images could distinguish three kinds of common breast tumors, and the prediction accuracy was 77.36%, which showed that MR image had important clinical value in differentiating FB, IDC and ILC.

To reduce the additional communication modules, complexity and costs of the different Internet of Things(IoT) devices communication, a multi-microcontroller communication method based on UART asynchronous serial ports was proposed. Based on the universal asynchronous receiver/transmitter(UART) serial communication method of the microcontroller, the control line was utilized to control the usage of the communication lines by the communication device, and a method of occupying the signal line by multiple machines in a time-sharing manner was realized. The master-slave control strategy was used to set the communication protocol. The master implements signal forwarding and identification, and the slave got signals from the master to achieve reliable and stable communication among multiple machines. By transplanting the μC/OS-Ⅱ operating system to the STM32 microcontroller, and using the real-time multitasking characteristics of μC/OS-Ⅱ, the signal reception, transmission and identification were designed into tasks of different priorities, and the master and the slave were realized. The functions of information receiving, sending and identification and the characteristics of multi-slave expansion were achieved through the communication protocol, solving the problem of multi-microcontroller communication that the traditional UART method could not achieve. The feasibility of the proposed method was verified through experiments, which provided a new solution for multi-microcontroller communication of edge devices in the Internet of Things.

A multi-model subspace learning semantic tag generation method was proposed, whic was based on the visual space and label space tag correlation modeling method separately. This method reconstructed the "image-tag" correlation in a non-linear manner by establishing a visual feature similarity map, thereby unifying the visual modal representation of the image and the text modal representation of the tag into a multi-model subspace, and ensuring space structure preservation before and after conversion. In this space, the text modal information of the label and the modal information of the visual content of the image were complementary to each other. The semantically related images and labels were mapped to similar sample points in the space, and the semantic label generation problem was then transformed into the nearest label-neighbors retrieval problem. The results showed that the performance of the proposed method was 36.88% on FLICKR-25K data set, and 44.17% on NUS-WIDE data set, which indicated that the proposed method could greatly improve the accuracy of label generation.

In order to solve the problem of poor adaptability of SuBSENSE algorithm in updating distance threshold in real complex scenes, which resulted in poor detection effect, SuBSENSE algorithm is proposed based on adaptive distance threshold correction of background complexity. A measure of background complexity is defined based on temporal consistency and spatial consistency, and the distance threshold correction strategy to get the accurate distance threshold as a criterion to achieve better detection results. This algorithm was compared with PBAS and traditional SuBSENSE algorithm. Experiments showed that the prospects of the proposed algorithm were more accurate in dynamic scenarios. The precision of the proposed algorithm was 6.70% and 0.80% higher than that of the PBAS algorithm and the traditional SuBSENSE algorithm, and the recall was 9.37% and 1.24% higher than that of the PBAS algorithm and the traditional SuBSENSE algorithm, respectively. After a comprehensive study of the three indicators, it was found that the proposed algorithm was superior to the contrast algorithms, and had higher robustness and detection accuracy in dynamic scenarios.

With How to use influence of social individuals to expand the scope of information dissemination was an Influence Maximization problem, which had become an important research field. A new ant colony algorithm was propsed to solve the problem, in the initial node selection process, we introduced two heuristic information to measure node-influence: priority selected nodes that were less likely to be activated by the precursor node; considered the impact of successors, especially multi-level successors node on the influence of spread. Based on this, a new ant colony optimization algorithm was proposed. The experiments showed that our method improved the problem of initial node selection, which was easy to fall into the local optimum, the results were better than the greedy method and the traditional ant colony optimization algorithm in the efficiency(raise 25%) and range of initial node dissemination(add 150 nodes).

Due to the interference of various factors in the complex situation of reality, the trackers had some problems such as model drift and tracking failure. An adaptive fusion target tracking based on joint detection algorithm was proposed to improve the robustness and accuracy of the tracker. The deep and shallow convolutional features acted on the correlation filters separately to obtain response scores according to their respective advantages, and adaptively fused the response scores of different convolutional features by minimizing the loss. Then it combined with the location detection method to judge the validity and authenticity of the predicted location, so as to get the optimal target tracking results. A large number of tests were done in two open databases: OTB-2015 and VOT-2017. The experimental results showed that the proposed method was 10% more robust and 3.9% more accurate than the LSART algorithm. It also had excellent performance for occlusion and scale variation.

In the label distribution learning framework, the example could be associated with the degree of description of the label. However, most of the algorithms were designed with complete data, and didn′t consider the noise in the data. Therefore, combined the noise reduction characteristics of the auto-encoder and the stability of the kernel extreme learning machine, the Label Distribution Learning Algorithm based on Kernel Extreme Learning Machine with auto-encoder was proposed in this paper. Firstly, we used the auto-encoder in kernel extreme learning machine to map the original feature space to obtain more robust feature representation. Secondly, we constructed the extreme learning machine model that adapted to the label distribution learning as a classifier to improve the classification efficiency and performance. Finally, the experimental results showed the proposed algorithm had certain advantages over other label distribution learning algorithms, and the hypothesis test method further illustrated the effectiveness of the algorithm.

GDEM software was used to analyze the mechanical properties of parallel layered jointed rock masses from different angles, and explore the influence of the change of the angle between the joint angle and the loading direction on the failure mode of the specimen, the parallel layered joint model was established by using different interlayer rock materials commonly in engineering. Through three loading methods, such as static load uniaxial compression, biaxial compression and pure shear, failure forms of such rock mass model at different inclined angles, the stress-strain relationship in the loading process and the variation trend of peak load were analyzed. This study found that the mechanical properties and peak strength of the parallel-level jointed rock mass were directly related to the joint inclination angle. It was found by simulation that the parallel-level jointed rock mass obvious elastic-brittle mechanics under three loading conditions.

Direct construction of roads and railways on coastal soft soil foundation leads to subgrade instability and other problems, and foundation treatment must be carried out. Based on the actual project, the pile arrangement simulation analysis was carried out and the pile arrangement parameters of cement-soil mixing piles were optimized. Through indoor mix proportion test and on-site pile forming test, the influencing factors of cement-soil strength and pile forming quality of cement-soil mixing pile were analyzed. The results showed that the proposed pile arrangement scheme was safe and feasible and cost saving. The optimal range of cement content in cement soil was 16%-18%, and the unconfined compressive strength at short-term age could reach 60%-70% of the standard age, which shortened the construction period; During the construction, the pile-forming technology of four stirring and four spraying was of the best quality. Preservatives were very important to the quality and durability of pile body. Through pile quality inspection, it was comprehensively judged that the reinforcement effect of cement-soil mixing pile in this project meeted the requirements. The research results had certain reference value for similar projects and provide on-site basis for the formulation of technical standards and construction methods.

A large number of studies showed that high ground stress had a certain influence on the wave velocity of deep buried rock, based on the wave equation, a mathematical model of the relationship between longitudinal wave velocity of sandstone and hydrostatic confining pressure was proposed. Based on the conventional triaxial test of rock, the static elastic modulus, static Poisson's ratio and longitudinal wave velocity of sandstone under different hydrostatic confining pressures were obtained, and the fitting curves and fitting formulas of static elastic modulus-hydrostatic confining pressure and static Poisson's ratio-hydrostatic confining pressure were obtained respectively. The test results showed that the static elastic modulus and static Poisson's ratio of sandstone increased with the increase of hydrostatic pressure, and the rate of increase of static elastic modulus decreased slowly. Based on the wave equation, the mathematical model of the longitudinal wave velocity-hydrostatic confining pressure was obtained, the longitudinal wave velocity calculated by the mathematical model showed that the longitudinal wave velocity of the sandstone increased with the increase of the hydrostatic pressure, and the increasing rate gradually became slower. The calculated longitudinal wave velocity was compared with the measured, the error range was 7.0%-8.3%. Therefore, the mathematical model of sandstone longitudinal wave velocity-hydrostatic confining pressure based on wave equation was reliable and accurate, it was of guiding significance to analyze and judge the physical and mechanical parameters of rock under high ground stress and the variation law of wave velocity.

The investigation was focused on the effects of using a row of holes for the reduction of nearby vibration response generated by the motion of a machine foundation on saturated soil. The analysis was accomplished with the aid of a series of field experiments and numerical simulation. 3D semi-analytical BE models were established to use a row of holes as active wave barrier to isolate the ground vibrations generated by the machine foundation laid on the surface of saturated soil foundation, and the effects of the model parameters on screening effectiveness were investigated and discussed in detail. The results showed that a row of holes could isolate the ground vibrations successfully. Increasing the radius and the depth, decreasing the net spacing between two successive holes increase the screening effectiveness. According to the results, it was suggested in the design that the hole radius should take the value of 0.1-0.15λ_R, and the hole depth and the net spacing between two successive holes should take the value no more than 1.0λ_R and 0.1λ_R, respectively. Moreover, the distance between the machine foundation and wave barriers affect the screening effectiveness, and the the distance was, the the screening effectiveness was. The number of holes in a row had less effect on the screening effectiveness. However, the bigger the number of holes in a row was, the larger the screening zone behind barriers was.

In order to improve the "Strong alternating current and weak direct current" characteristics of multi-infeed high voltage direct current systems, the synchronous condenser gradually gained attention.For the problem of multi-infeed high voltage direct current with synchronous condenser simultaneous commutation failure, a risk evaluation method considering the synchronous condenser reactive voltage characteristics was proposed. Firstly, the reactive voltage characteristics of the synchronous condenser were analyzed, and the multi-infeed interaction factor was calculated by reactive voltage sensitivity of the synchronous condenser, high voltage direct current system and static reactive power compensation device. Then, based on the nature of the commutation failure, the commutation failure evaluation factor was defined, and the simultaneous commutation failure risk of the multi-infeed high voltage direct current system with synchronous condenser was evaluated. Finally, the Shandong power grid was taken as an example to verify the results that the proposed method could effectively evaluate the multi-infeed high voltage direct current simultaneous commutation failure, and it was important in the early planning of the high voltage direct current transmission system and ensuring the stability and security of the power system.

Aiming at the problem of voltage over-limit in urban distribution network, this paper proposed a voltage control method for electric vehicles participating in charging and discharging of street charging piles. The method utilized the transformer capacity released by the traditional high-pressure sodium light replaced by using the LED street light, and built a street light charging pile to participate in the voltage control of the urban distribution network as a controllable resource for the electric vehicle charged and discharged by the street light charging pile. Based on the analysis of the load characteristics of electric vehicle charging and discharging of street light charging pile, according to the characteristics of various voltage regulating resources in the distribution network, the multi-level voltage control strategy of urban distribution network was studied, and the optimal control cost of each voltage regulating measure was taken as the objective function. The voltage regulation model was established and the model was solved by particle swarm optimization. According to the characteristics of urban street lighting load, the daytime and nighttime scenarios were simulated. The simulation results verified the effectiveness of electric vehicle′s use of streetlight pile charging and discharging in urban distribution network voltage control. The control effect was verified by comparative analysis. It was better than traditional voltage control methods.

An energy scheduling method of smart home integrated with photovoltaic (PV) generation units under time-of-use pricing (TOUP) was proposed. Based on TOUP and PV generation output, the mathematical model of day-ahead energy scheduling to obtain maximal gains for smart home was established considering the appliances with adjustable power. The bilateral gains equations of smart home and the grid were incorporated in the proposed model. Based on linear transformation method, the proposed mixed integer programming model was reformulated as a linear programming one, which made it easy to obtain optimal solution. The influence of maximal exchange power and controllable appliances on gains of smart home was also studied. Test results verified the effectiveness and practicability of the proposed model and method.

The improved elastic gradient descent algorithm of back propagation was used, and 14 rainfall runoff processes from 1997 to 2014 in the upper reaches of Chongyang River were selected. The back propagation (BP) neural network rainfall-runoff forecasting model of the elastic gradient descent algorithm was established, which took the measured rainfall of six rainfall stations in Yangzhuang, Wubian, Da′an, Kengkou, Lingyang, and Langu in the basin and the preliminary flow data of Wuyishan Hydrological Station as inputs, and selected the corresponding flow of Wuyishan Hydrological Station as output. The 7-rainfall runoff process was used to test the model, the test results showed that the proposed method required fewer parameters and had higher operation speed than the traditional back propagation algorithm. The prediction accuracy of the model could meet the requirements, and provide the basis for flood control and disaster reduction.

To reduce the payload pendulation of shipboard cranes, a mechanical anti-swing device based on triple-cables limiting-location was proposed. Three cables were used to pull the hook, which limited the spatial position of the payload to prevent the payload pendulation. According to the established kinematic model of the shipboard crane, the effect of length on the anti-swing was analyzed. The dynamic model of the payload system was established, and the effect of tension value on the anti-swing was analyzed. The models were verified by physical experiment based on a self-built test platform. The experimental results proved that the proposed mechanical anti-swing device based on triple-cables limiting-location had good anti-swing effect in practical applications. The overall anti-swing effect could reach more than 61%.

Chemical oxygen demand and suspended solid were important monitoring indices of effluent discharge in paper-making industry. An effective model of effluent quality of wastewater treatment processes was of key importance to monitoring and controlling pollution emission. Concerning the strong correlations among the input variables and the complicated characteristics of wastewater treatment processes in paper-making industry, partial least squares (PLS) method was applied to extract information of variables importance in projection (VIP) for variable selection (VS). Then the optimal variables were chosen as new input variables for soft sensor models to predict the effluent qualities of a papermaking wastewater treatment process. Compared to the LSSVM model, the root mean square error (RMSE) of VS-based LSSVM model was reduced by 15.2%, and the correlation coefficient (r) was increased by 14.4%. For the effluent SS, the value of RMSE was decreased by 20.5%, and the value of r was increased by 16.1%. The results showed that the proposed method not only reduced the model complexity, but also enhanced the model generalization capacity.