Abstract:

Abstract: Knobs tuning is a key technology that affects the performance and adaptability of databases. However, traditional tuning methods have difficulty in finding the optimal configuration in high-dimensional continuous parameter spaces. The development of machine learning could bring new opportunities to solve this problem. By summarizing and analyzing relevant work, existing work was classified according to development time and characteristics, including expert decision-making, static rules, heuristic algorithms, traditional machine learning methods, and deep reinforcement learning methods. The database tuning problem was defined, and the limitations of heuristic algorithms in tuning problems were discussed. Traditional machine learning-based tuning methods were introduced, including random forest, support vector machine, decision tree, etc. The general process of using machine learning methods to solve tuning problems was described, and specific implementations were provided. The shortcomings of traditional machine learning models in adaptability and tuning capabilities were also discussed. The principles of deep reinforcement learning models were emphasized, and the mapping relationship between tuning problems and deep reinforcement learning models was defined. Recent relevant work on improving database performance, time consumption and model characteristics was introduced, and the process of building and training agents based on deep neural networks was described. Finally, the characteristics of existing work were summarized, and the research hotspots and development directions of machine learning in database tuning were outlined. Distributed scenarios, multi-granularity tuning, adaptive algorithms and self-maintenance capabilities were identified as future research trends

Abstract: Traditional fixed-parameter admittance models could not adjust the compliance of upper limb rehabilitation robots in real-time, existing variable-parameter admittance models required improvement based on actual rehabilitation needs. To address this issue, a novel variable-admittance control strategy was proposed for upper limb rehabilitation robots with a self-developed series-parallel hybrid end-effector traction structure, combining fuzzy and admittance control and tailored to actual rehabilitation needs. Four fuzzy rules that could benefit rehabilitation efficiency and safety were developed. This strategy proposed the use of interaction force error and its rate of change as inputs to fuzzy control, to adjust admittance model parameters and achieve autonomous compliance control in real time. Simulation and experimental results validated the feasibility of the proposed variable-admittance control strategy and the effectiveness of the developed four fuzzy rules. In scenarios where patients could adapt to training intensity, the variable-admittance model could reduce the redundant path generated during path tracking by up to 56. 13% thus improving rehabilitation training efficiency. When rehabilitation movements exceeded the patients′ tolerance limit, the variable-admittance model could change the tracking path half a second earlier, improving rehabilitation safety

Abstract: when solving problems in the grinding force model, most of the methods of segmental calculation or column equations were used to calculate each coefficient directly, which not only demanded a large amount of calculation but also could not guarantee its accuracy. In addition the traditional regression model was easy to fall into local optimal, difficult to describe the nonlinear relationship. Therefore, the genetic algorithm was introduced into the parameter optimization of the nonlinear fitting function, and the coefficient optimization method of the theoretical model of grinding force was studied based on the existing model data such as the model of cylindrical transverse grinding, the model of plane grinding and the model of cylindrical longitudinal grinding. Correlation analysis results showed that the predicted accuracy of grinding force of the three models was increased by 14. 69% -42. 54%. The average error of normal grinding force predicted by the three models was 5. 9%, 9. 13% and 3. 23%, respectively. The mean error of tangential force was 6. 78%, 8. 36% and 3. 69%, respectively. Through comparison, it could be concluded that the optimized model had a better fitting degree, and the prediction accuracy of the model was significantly improved. The nonlinear fitting function GA-LSQ algorithm optimized by genetic algorithm was more suitable for solving grinding force model and could provide reference for predicting grinding force and parameter optimization in actual production.

Abstract: The study was conducted to examine the durability issue occurred in front stabilizer bar bracket connected to sub-frame in full vehicle testing. Firstly, stabilizer bar system was plastered with strain gauges and calibrated, and drop link force and stabilizer bar twist displacement acquired on proving ground, sub-frame with stabilizer bar system physical test rig was designed and built, and rig tests in accordance with durability specifications were conducted. The test results showed that built physical test rig could greatly reappear crack location in full vehicle testing, the fatigue life of the physical bench had a deviation of 2. 5% compared to full vehicle testing. Based on this, a stabilizer bar and sub-frame multi-body virtual model was built with the same constraint boundary and the same loading method of the physical test rig. Then CAE fatigue simulation was used through quasi-static finite element fatigue life analysis method to reappear related area risk. The simulation results showed that phase and amplitude had a good coincidence in time domain, the PSD spectrum also had a good accuracy in frequency domain, the relative damage was almost closed to 1 with the comparison between the simulation and test in droplink force and stabilizer bar relative displacement. The deviation between the simulated fatigue life at the relevant risk position and the test life of the full vehicle was 6. 25%. A higher accuracy risk position load was obtained, and the reappearance of durability risk position was achieved. Finally, based on simulation fatigue load, the optimization risk structure was evaluated. Optimized proposal eventually passed the test rig and full vehicle testing successfully.

Abstract: In order to solve the disadvantages of traditional brake system logic threshold control, which was complicated in logic and difficult to make full use of road adhesion, and the problems of nonlinearity and uncertainty of ABS, a control strategy of ABS based on road identification on the basis of electronic mechanical braking system was proposed. Firstly, quarter vehicle braking model was established through Simulink. Next, the differences of adhesion coefficient on different road conditions and the variation rules of adhesion coefficient and wheel angle deceleration were analyzed. An efficient and accurate road identification algorithm was designed to estimate the optimal slip rate of the current road surface. Finally, an ABS control strategy based on integral sliding mode control was designed to track the optimal slip rate. The simulation results showed that the road identification algorithm had fast response and high accuracy. The designed ABS control strategy could stably track the optimal slip rate and had strong adaptability to different road conditions. Compared with the traditional braking system based on logical threshold control, on single road condition the braking time was reduced by 11. 89%, and the braking distance was shortened by 12. 7%. On variable road condition the braking time was reduced by 17. 8%, and the braking distance was shortened by 19. 9%.

Abstract: Taking the rail friction pair with surface texture in the pin insertion machine as the research object, the influence of surface texture on its friction performance was studied with the consideration of surface roughness, load fluctuation, speed fluctuation, and time-varying oil film extrusion effect. A rough surface with autocorrelation function was generated by computer simulation, and the mixed friction model was constructed by coupling the roughness contact model established by Greenwood and Tipp and the average oil film fluid lubrication model modified by Patir and Cheng. The oil film pressure, asperity contact pressure and oil film thickness at each moment in a cam rotation cycle were calculated by MATLAB software. The characteristics of the oil film pressure distribution of the friction pair in the mixed lubrication stage were analyzed, and the effects of texture quantity, surface roughness, and surface texture size parameters on the lubrication characteristics of the friction pair were obtained. The results showed that when the variance of the roughness of the friction pair of the base increased, the oil film pressure decreased. When the radius of the micro-dimples was 60 μm, the area occupation ratio was 40%, and the depth of the microdimples was 5 μm, the friction pair could achieve the best lubrication effect

Abstract: In order to solve the problem of stability and control accuracy of intelligent vehicle active steering tracking control on low adhesion road surface, an intelligent vehicle trajectory tracking control strategy based on adaptive time domain parameters was proposed. Based on the vehicle dynamics model and model predictive control algorithm (MPC) , a linear time-varying MPC controller was established, and dynamic constraints including tire side deflection constraints, centroid side deflection constraints and front wheel angle constraints were added to solve the optimal front wheel steering angle. The influence of time domain parameters in the controller on the control effect was analyzed, and an adaptive time domain parameter controller was designed. According to the acquired vehicle speed, the optimal predictive time domain and control time domain parameters were obtained and input to the controller, improving the control accuracy and stability of the controller at different speeds. By building the MATLAB / SimuLink and CarSim co-simulation platform, the fixed time domain controller and adaptive time domain controller were compared and simulated with the condition of low adhesion road surface. The results showed that the adaptive time-domain controller could effectively improve the performance of the controller, reduce the lateral deviation, and improve the control accuracy of trajectory tracking. At the same time, it also had strong adaptability to different speeds, and the lateral deflection angle of the vehicle center of mass was controlled within 0°-1. 5°, which effectively ensured the stability of the vehicle.

Abstract: Due to the limited block information storage capacity, the existing blockchain traceability system mostly adopted the combination of blockchain and cloud storage, which could not fundamentally solve the problem of blockchain traceability information storage and data leakage. The dual blockchain mode was constructed, the query chain was used to complete the uploading of traceability information and to realize the basic functions of the traceability system. The storage chain was combined with the interplanetary file system to ensure data integrity and security. On this basis, the improved capacity proof consensus algorithm was used to ensure the underlying security and meet the application requirements of blockchain traceability system while reducing energy consumption. The test showed that the average upload speed of the system could reach 80. 66 M / s, and the average download speed could reach 90. 75 M / s, with good upload and download performance. Blockchain transactions per second reached 125. 88, and the carrying capacity of a single transaction could meet the demand for block transactions of the traceability system

Abstract: A quadrature amplitude modulation (QAM) based cooperative physical layer authentication mechanism was proposed for the identity authentication in two-user cooperative communication systems. In the proposed mechanism, two single-antenna users transmitted their own messages and tags as well as those of the partner with the inphase component and orthogonal component, respectively. Thus, the messages and tags of two users could be uniquely decomposed. Maximal ratio combining was employed at the base station to detect the messages and tags, and physical layer identity authentication was performed through tag comparison. In a typical scenario with equal power distribution and symmetric user uplink channels, the closed expressions of symbol error rates of message and tag were derived. The simulation results showed that the authentication probability of the proposed mechanism was improved by 12% - 20% compared with the non-cooperative physical layer authentication when the signal-to-noise ratio ( SNR) was 6-15 dB. Meanwhile, the authentication probability of illegal attack remained only around 10 - 3 . Consequently, the proposed mechanism had better authentication performance

Abstract: A video frame prediction model based on gated spatio-temporal attention was proposed to address the issues of low accuracy, slow training, complex structure, and error accumulation in recurrent video frame prediction architectures. Firstly, high-level semantic information of the video frame sequence was extracted by a spatial encoder while preserving background features. Secondly, a gated spatio-temporal attention mechanism was established, utilizing multi-scale deep bar convolutions and channel attention to learn both intra-frame and inter-frame spatio-temporal features. A gate fusion mechanism was employed to balance the feature learning capability of spatiotemporal attention. Finally, a spatial decoder reconstructed the high-level features into predicted realistic images and complements background semantics to enhance the details. Experimental results on the Moving MNIST, TaxiBJ, WeatherBench, and KITTI datasets showed that compared to the multi-input multi-output model SimVP, the mean squared error (MSE) was reduced by 14. 7%, 6. 7%, 10. 5%, and 18. 5%, respectively. In ablation and expansion experiments, the proposed model achieved good overall performance, demonstrating advantages such as high prediction accuracy, low computational complexity, and efficient inference.

Abstract: Few-shot spoken language understanding ( SLU) is one of the urgent problems in dialogue artificial intelligence (DAI) . The relevant literature on SLU task, combining the latest research trends both domestic and foreign was systematically reviewed. The classic methods for SLU task modeling in non-few-shot scenarios were briefly introduced, including single modeling, implicit joint modeling, explicit joint modeling, and pre-trained paradigms. The latest studies in few-shot SLU were introduced, which included three kinds of few-shot learning methods based on model fine-tuning, data augmentation and metric learning. Representative models such as ULMFiT, prototypical network, and induction network were discussed. On this basis, the semantic understanding ability, interpretability, generalization ability and other performances of different methods were analyzed and compared. Finally, the challenges and future development directions of SLU tasks were discussed, it was pointed out that zero-shot SLU, Chinese SLU, open-domain SLU, and cross-lingual SLU would be the research difficulties in this field

Abstract: In this paper, the true triaxial testing machine was used to conduct experimental research on granite specimens with size of 100 mm×100 mm×100 mm with cyclic loading, and to explore the dynamic elastic modulus and dynamic damping ratio of granite under different confining pressure, dynamic loading frequency, dynamic stress amplitude, and number of cycle. The changing law of the data was compared and analyzed. The test found that the dynamic elastic modulus of granite increased linearly with the confining pressure, and the dynamic damping ratio decreased linearly with the confining pressure. The growth of dynamic elastic modulus gradually became larger. The dynamic elastic modulus of granite increased exponentially with the dynamic loading frequency, and the dynamic damping ratio increased as a power function with the dynamic loading frequency. At a dynamic load frequency of 20 Hz the internal activity of the rock was more intense, and the frequency had a greater impact on the rock. The dynamic elastic modulus increased as a quadratic function with the dynamic stress amplitude, and the dynamic damping ratio decreased as a power function with the dynamic stress amplitude. The composition of the rock had an influence on the dynamic damping ratio. The improved Duncan-Chang model reflected well the dynamic constitutive relationship of the granite under different actions of the rock. The model obtained in this test could be used as a reference for subsequent inversion of the dynamic properties.

Abstract: In order to study the influence of mud content on salt swelling and mechanical properties of sand sulfate soils, sand sulfate soils with different fine-grained soil content were manually configured. Based on the unidirectional freezing salt swelling test with 1% and 3% salt content, sand samples with fine-grained soil mass fraction of 5%, 15%, and 30% were selected to conduct triaxial shear tests at constant and low temperature. The results showed that with the conditions of this test, the freezing temperature of sand sulfate saline soil of different levels was within the range of -0. 7--0. 1 ℃ . When the pore solution concentration of the soil samples reached saturation above the freezing temperature, salt crystals could first be generated in the cooling process. With the condition of 1% salt content, the initial swelling temperature of sand samples with high fine-grained soil content (≥30%) was in the range of 4-9 ℃ , while the initial swelling temperature of sand samples with low fine-grained soil content was near 0 ℃ . The swelling temperature of 3% salt content sand sample was 20-23 ℃ . The process of salt frost heave was inseparable from the participation of free water. The test water content and fine-grained soil content had a significant effect on salt frost heave by influencing the content of free water in the soil. In terms of mechanical properties, with the increase of the content of fine-grained soil, the shear strength of sand sulfate soils increased first and then decreased, and the fine-grained soil changed from enhancing friction to " lubrication " between particles. In addition, after freezing, the sand was transformed into a " soil-salt-ice skeleton structure " with a stronger bearing capacity, and the shear strength increased greatly, showing obvious brittle failure characteristics. Affected by relative temperature, the failure stress of frozen sand decreased first and then increased with the increase of salt content.

Abstract: Taking the tunnel project from Fengsha Station to Creative Park Station of Foshan Metro Line 3 as background, the inherent variation tendency of boring parameters of shield when the EPB shield tunnel crossed the soft soil strata was analyzed in detail through the on-site measured data, and the different prediction models of driving speed were built. Firstly, the shield tunneling parameters were analyzed by mathematical statistics, and the distribution of each tunneling parameter was tested. Secondly, the Pearson correlation analysis was performed to find out the variation law between the parameters with strong linear correlation. Then using the feature selection algorithm based on mutual information, the parameter variables with high nonlinear correlation with the driving speed were screened. Finally, the random forest regression prediction and the BP neural network prediction model based on genetic algorithm optimization were established respectively to predict the driving speed. The research results showed that in shield tunnel projects in soft formations, lower cutterhead speed, cutterhead torque, higher tunneling speed, penetration, total shield thrust and soil silo pressure were usually used. The parameters such as the excavation speed passed the normality test using the K-S test method. There was a strong correlation between the speed of excavation and the degree of penetration. The average absolute error, root mean square error and goodness of fit of the random forest regression prediction model in the test set were 4. 055, 5. 038 and 0. 871, respectively, while the optimization of the BP neural network prediction model based on genetic algorithm was 0. 822, 1. 244 and 0. 991, respectively

Abstract: There were still some research gaps in reliability evaluation of distribution network considering centralized feeder automation, and most studies only focused on the impact of power failure. Considering the pre-arranged maintenance and capacity constraints, the influence of load transfer, and combined with the type and operation logic of feeder automation, according to the related technical indexes of feeder automation, the load nodes appearing in the process of power restoration were classified in detail. And the calculation formulas of expected power restoration time and power supply reliability indexes of different types of loads were derived. Combined with an example, the average outage duration SAIDI of feeder system was reduced by 0. 95-1. 08 h / ( user·a) with different terminal configurations in the example, which showed that optimized the terminal configuration could effectively improve the power supply reliability of distribution network, which proved the accuracy and practicability of the evaluation method in this study. The influences of different terminal configurations on reliability were compared.

Abstract: Aiming at the requirement of real-time detection of various electrical equipment in substation, an electrical equipment identification method based on improved YOLOv5s was proposed, and an electrical equipment identification APP based on Android was designed to recognize and learn electrical equipment. Six common electrical equipments of substation, such as power transformer and insulator string, were taken as examples to construct image data set. After image preprocessing of data set, YOLOv5s algorithm was improved, introducing C2f module to improve the detection accuracy of small targets, and using Soft-NMS to improve the screening ability of detection frame, so as to reduce the phenomenon of missing and false detection. The improved algorithm was used to train the model of data set. The trained identification network model was deployed through the TensorFlow Lite framework, and the electrical equipment identification APP was designed. It was verified that the mAP value of the improved substation electrical equipment identification network model was stable at 91. 6%, which was 3. 3 percentage points higher than that of the original model. After deployment, the APP had the interface of equipment recognition and equipment introduction, and the recognition time of each image was less than 1 s when using mobile terminal, which had a fast recognition speed and high recognition accuracy, and could effectively realize the real-time detection and equipment learning of electrical equipment in substation.