Abstract:

Abstract:

Abstract: To address the sensing security issue of unauthorized radar station (URS) stealing target information in multi-radar scenarios, a secure wireless sensing system model based on intelligent reflecting surface (IRS) assistance was proposed. This system deployed an IRS with sensing capabilities on the target and adopted a two-phase sensing scheme. In the first phase, the IRS sensing unit estimated the angle information of all radars. In the second phase, the IRS reflection coefficients were designed based on the estimation results to minimize the perception probability of URS. Specifically, under the constraints of ensuring the signal-to-noise ratio of the legitimate radar station (LRS) and the IRS reflection phase shift modulus, an optimization problem was formulated to minimize the maximum signal-to-noise ratio of the URS. An iterative optimization algorithm based on the Dinkelbach method and semidefinite relaxation (SDR) technique was proposed. Simulation results showed that compared to the scheme without IRS, the signal-to-noise ratio of the LRS improved by approximately 3 dB, while the signal-to-noise ratio of the URS decreased by about 12 dB, demonstrating that the proposed scheme significantly enhanced system security performance.

Abstract: When unsupervised neural machine translation was trained with monolingual data, it inevitably brought a lot of noise information. The errors of the machine translation model accumulated continuously during the training iteration process, affecting the translation effect. To solve this problem, in this study an unsupervised neural machine translation method was proposed based on pseudo-siamese network on the basis of cross-lingual pre-training model (XLM). The model encoder was divided into two modules, in which the pseudo-Siamese network part introduced a noise filtering gate mechanism to filter the noise features in the encoding process, so that the model could better learn the mapping relationship between the source language and the target language. The experimental results showed that in the interactive translation task between English, German, French, and Romanian, the proposed method had an average improvement of 3.5 percentage points compared with the baseline system, which proved its superiority in translation effect. Ablation experiments were used to verify the effectiveness of each component of the model. At the same time, the performance test of the method with different noise conditions was simulated in the German-English translation task, and it also showed good noise resistance.

Abstract: It is found that multi-modal information such as images and text possesses semantic complementarity, which could effectively enhance the representation of entities in knowledge graphs, thereby improving the accuracy and interpretability of recommendations. A knowledge-aware recommendation method that could integrate multimodal information was proposed by analyzing the characteristics of semantically related multimodal data in recommendation systems. On the basis of knowledge graph propagation, multi-modal information that was semantically related to entities in the graph was integrated, and feature fusion was performed with the corresponding entities to enrich entity representation, aiming to explore users′ potential interest preferences. In this method, the dependency and interactivity between multimodal information was considered, intermodal attention was adopted to focus on important information of each modality, and semantically associated multimodal embedding features were obtained. Through gated attention, the multi-modal embedding features corresponding to entities were fused with entity representations, further enriching the multi-modal semantic information of entities, thereby enhancing the representation of users and items. In order to verify the effectiveness of the method, experiments were conducted on MovieLens-1M and Book-Crossing data sets, and comparative analysis was conducted with 9 methods including RippletNet, KGAT, CKAN, LKGR, COAT, CKE, KGCN, SKGCR and KGCL. The experimental results showed that it was better than the other two indicators in AUC and ACC. On the MovieLens-1M and Book-Crossing datasets, the AUC of the proposed method were 0.936 6 and 0.763 7, respectively, with an increase of 0.027 2 and 0.029 1 compared to the average values of other models. The ACC values of the proposed methods were 0.862 3 and 0.708 9, respectively, with an increase of 0.028 3 and 0.030 5 compared to the average values of other models.

Abstract: The high-dimensional characteristics of microbial data, the high zero-value rate, and the scarcity of minority-class samples, which led to class imbalance, had significantly weakened classifiers′ ability to identify minority class. Existing augmentation algorithms are sensitive to high imbalance ratios (IR) and struggle to effectively synthesize samples. In this study a microbial data augmentation algorithm based on feature transformation and minority class clustering (FTMC) was presenteed. Firstly, the feature transformation stage used the principal components analysis algorithm to down thescale high-dimensional data to alleviate the problem of strong data sparsity. Subsequently, in the minority class clustering stage, the K-Means algorithm was used to capture the local features of the minority classes and obtain multiple clusters. In the cluster screening stage, based on the density and difficulty of each cluster, combined with the IR and weight ratio, its weight value was calculated and used to screen a subset of core clusters for subsequent sample generation. Finally, in the sample augmentation and filtering stage, a linear interpolation algorithm was used augment to the samples for each core cluster, and a local anomaly factor algorithm was used to filter outliers to ensure the quality of the augmented samples. The experiments were conducted on 12 microbial datasets and the performance was compared with 8 sampling algorithms of the same type with 3 classifiers.Results indicated that samples generated by FTMC were more diverse, with an average improvement of 26.42% in the Recall metric. This demonstrated that the algorithm could correctly identify more positive samples.

Abstract: Optimization algorithms often perform optimally on specific types of fitness terrains due to the varying nature of optimization problems. To address this limitation, in this study an optimization algorithm scheduling method grounded in fitness terrain analysis was introduced. This method characterizes the terrain features of an optimization problem by extracting the local peak cluster number features of the optimization objective function. Based on these terrain features, the method selected the most suitable algorithm to maximize the advantages of different algorithms through effective scheduling. In particular, this study considered the balance between exploration and exploitation in optimization problems by selecting the harris hawks optimization algorithm (HHO), known for its high development capability, and the differential evolution algorithm (DE), recognized for its strong exploration ability, as the scheduling algorithms. The choice of algorithm was tailored to the specific adaptability characteristics of the terrain. Experimental results show that the convergence performance of FL-AMAS was improved by 75% compared with that of HHO alone, and by 40% compared with that of DE algorithm. Further, FL-AMAS was compared with six advanced algorithms, and FL-AMAS outperformed these algorithms in convergence accuracy on 75% of the benchmark set. The effectiveness and scalability of the proposed scheduling method were further validated through comparisons with other types of scheduling optimization algorithms.

Abstract: To address limitations of traditional ensemble attack methods, which were constrained by high computational resource requirements, including training data and time, a low computational complexity ensemble attack method based on genetic recombination was proposed. This method aimed to enhance the transferability of existing adversarial attacks by generating a more diverse set of ensemble models. Firstly, the concept of genetic recombination was introduced into knowledge distillation. In this process, student models were treated as independent individuals, with their parameters considered as genes. Each round of distillation learning was viewed as a gene evolution. Randomly exchanging parameters among student models during the evolution process achieves artificial genetic recombination, resulting in superior offspring genes. By setting different distillation temperatures, multiple diversified student models were obtained. Next, these diverse student models were integrated with the source teacher model. Finally, the integrated model was used to generate adversarial examples with stronger transferability. Experimental results on a subset of the ImageNet validation set demonstrated that the proposed method significantly improved the transferability of adversarial samples compared to other baseline algorithms. Using ResNet152 as the source model and PGD as the attack method, the proposed method achieved the highest transfer attack success rate across 11 black-box models, outperforming the baseline PGD method by an average of 34.52 percentage point, the PGI method by an average of 5.30 percentage point, and the DGM method by an average of 2.12 percentage point.

Abstract: The traditional method of human activity recognition based on channel state information (CSI) suffers from issues such as input data redundancy and limited feature extraction. To address this, a human activity recognition approach based on CSI principal components and a dual-layer sliding window mechanism was proposed. Firstly, autlier removal and noise reduction were performed on the amplitude the use of a dual-layer sliding window mechanism based on principal component analysis enabled activity segmentation of preprocessed CSI data to eliminate irrelevant information and enhance model training efficiency. Subsequently, spatial and temporal analysis of the CSI data was conducted using convolutional neural network and bidirectional gated recurrent unit, with the integration of a multi-head attention mechanism to focus on key information for achieving high-precision recognition of human activities. Experimental validation was performed using the WiAR and BAHAR public datasets, demonstrating that the proposed method could effectively recognize various human activities in diverse environments, while reducing the data volume by 5%. The accuracy achieved on the WiAR dataset was 96.53%, indicating superior performance compared to existing methods.

Abstract: To improve the accuracy of predicting the response of melanoma patients to immune checkpoint inhibitor (ICI) therapy, a new method integrating bulk RNA-seq and single-cell RNA-seq data was proposed. Firstly, a patient-cell correlation matrix was constructed through Pearson correlation analysis, and the Louvain algorithm was used to classify single-cell RNA-seq data into cell groups. The importance of cell groups in immune response related pathways was quantified using the CellChat tool. On this basis, a double group minimax concave penalty logistic regression model (DMCPLR) was proposed by introducing the cell group importance evaluation criterion constructed based on the cell-cell communication network and combining with the group minimax concave penalty. The experiments on the GSE35640 dataset showed that the prediction accuracy of the DMCPLR model reached 80.18%, with precision, recall, and F1 score of 82.24%, 89.71%, and 85.11%, respectively, significantly better than the performance of 14 comparison methods including Lasso regression and random forest, while reducing the fatal error rate to 8.30%. The ablation analysis experiment confirmed that the introduction of cell group weight mechanism and L2 regularization term can improve the performance of the model.

Abstract: In order to reduce the influence of wake disturbance on the total output power of wind farm, Informer neural network algorithm was proposed in the proposed wind farm yaw optimization control framework, and an intelligent equivalent model of power conversion for wind farm yaw control was established. Based on the present model, an optimization problem maximizing the power output of wind farm with yaw angles as decision variables was defined, and particle swarm optimization algorithm was used to obtain the optimal yaw angle of each wind turbine and reduce the wake interference. Firstly, a wind farm consisting of 14 wind turbines was built,and its layout was Penmanshiel wind farm. Secondly, wind data was used to model the wind farm equivalently, and the results of the Informer model were compared with LSTM, GRU, RNN, and Transformer. The results showed that the established Informer intelligent equivalent model could consist with the actual characteristics of the wind farms. Comparing the proposed algorithm with the mantis search algorithm, the proposed algorithm could increase the total power of wind farms by 1.94 MW with the wind speed of 10 m/s and the wind direction of 195°. With continuous wind conditions (measured wind data on a certain day), the total power of the wind farm was increased by 292.97 kW on average, and the improvement results were superior to the mantis search algorithm. The proposed algorithm could improve the overall output power of the wind farm well.

Abstract: To meet grid connection requirements and ensure robust operation of the power system, a two-layer optimization model was proposed with objectives of reducing PV prediction errors, smoothing grid-connected power fluctuations, and minimizing annual equivalent costs. The upper-layer planning aimed to minimize annual equivalent costs, which included system investment, equipment replacement, maintenance costs, and carbon emission benefit costs. To improve system economic efficiency, a fuzzy genetic particle swarm algorithm was developed to optimize and analyze the model. In the lower-layer planning, the model aimed to minimize prediction errors and grid connection volatility. Based on distinct characteristics of supercapacitors and batteries, a charging-discharging power allocation strategy was constructed to enhance system response speed and extend battery cycle life. A solver was employed for control implementation to achieve PV prediction error compensation and PV output fluctuation smoothing. Finally, a model evaluation index function was established based on the proposed framework, with a PV power plant serving as a case study. Results demonstrated that the proposed algorithm exhibited faster convergence speed and superior optimization capability in this model. The RMSE and MAPE of prediction errors were reduced by 99.95% and 99.97% respectively, while the maximum grid connection fluctuation rate decreased by 96.08% after optimization. These findings verified the effectiveness and practicality of the proposed strategy.

Abstract: Extreme disasters occur frequently around the world, and extreme weather events seriously threaten the safe operation of distribution networks. In order to improve the resilience of distribution network to extreme disasters, a multi-type extreme weather model and a multi-dimensional resilience assessment method were proposed in this study. Firstly, based on the impact mechanism of typhoons, heavy rains and ice storms, a unified distribution network component failure rate model was constructed, and Monte Carlo sampling and K-means clustering algorithms were used to screen typical fault scenarios. Secondly, based on a comprehensive weight method a multi-dimensional resilience assessment index system was used to cover defense, adaptability and resilience, and a multisource collaborative optimization disaster the post-recovery model was built to verify the effectiveness of the proposed resilience assessment method. Finally, taking the IEEE33 node and IEEE69 node systems as examples, the three measures of reinforcing lines, dynamic dispatching of distributed power sources, and increasing mobile power capacity were compared. The results showed that the proposed method was effective. Compared with the traditional one the score of the proposed method increased by 11.2%, and the evaluation results were more accurate and comprehensive.

Abstract: To address the multi-stage characteristics of rolling bearing degradation with random change points, in this paper a novel method was proposed to predict the remaining useful life (RUL) of multi-stage degradation processes. Initially, the prior parameters of each stage model were estimated using offline historical data. Then, for a single online device, real-time change point detection was performed using the Bayesian change point detection method. The Bayesian updating approach was adopted to update the parameters of the first stage before the change point occurs and the second stage after the change point. Subsequently, the multi-stage model was utilized for RUL prediction. Numerical simulations and case studies showed that the rolling bearing life prediction method based on Bayesian change point detection could improve change point detection accuracy by 85%, thereby achieving highprecision multi-stage RUL prediction.

Abstract: Considering the supply dynamics of crowdsourced personnel, an optimization method was proposed for last-mile delivery in scenarios involving both self-operated personnel and crowdsourced personnel in this study. A hybrid delivery optimization model was developed to minimize delivery costs, incorporating decision variables such as crowdsourced personnel compensation, customer assignment schemes, and multi-trip vehicle routing. The constraints was constructed based on a spatio-temporal network, including customer service requirements, load capacities, time windows, and spatio-temporal coordination between vehicles and crowdsourced personnel. To address the problem′s characteristics, the minimal cost insertion algorithm was improved, and ten types of destruction operators were designed. An adaptive large neighborhood search algorithm was then improved by integrating the simulated annealing algorithm′s concept of accepting inferior solutions to optimize delivery plans.Case studies with 100 customers demonstrated that the improved adaptive large neighborhood search algorithm achieved solutions with 24% lower in delivery cost on average compared with those obtained by GUROBI with 1 hour computation time, while taking only 48.5 seconds on average. The proposed algorithm also outperformed simulated annealing, achieving a maximum cost reduction of 5.5%. The hybrid delivery mode combining self-operated and crowdsourced personnel significantly reduced costs compared to the self-operated-personnel only mode. The hybrid mode proved particularly suitable for scenarios with tight time windows, high vehicle travel costs, and limited job opportunities. The supply dynamics of crowdsourced personnel exhibited significant and stochastic impacts on delivery costs.

Abstract: The measurement of stator roundness in hydro-generators traditionally relied on manually pushing the measurement arm to multiple test points for evaluation. This process was time-consuming and labor-intensive. To improve measurement efficiency, a stator roundness measurement system was designed, utilizing a smooth input shaper (SIS) to suppress residual vibrations during the stopping phase of the measurement arm. Firstly, simulations were performed to compare the SIS shaper with existing zero-vibration (ZV), zero-vibration-derivative (ZVD), and extra-insensitive (EI) shapers in terms of residual vibration, time delay, and sensitivity. Secondly, an experimental test platform was used to evaluate the effectiveness of these shapers. Finally, field tests were conducted at the Yangqu Hydropower Station in Qinghai Province. Results demonstrated that the SIS exhibited the strongest robustness, with minimal impact on vibration suppression performance when system parameters were imprecise. Even with inaccuracies in damping ratio and natural frequency, the SIS achieved effective vibration suppression, reducing the peak residual vibration at the measurement arm tip by 91.6%. The SIS-integrated measurement system maintained measurement accuracy while reducing vibration decay time by 87.9% during stator roundness assessments.

Abstract: During the excavating operation of a tunnel boring machine (TBM), the disc cutters on the cutterhead work together to break up rock. To investigate the influence of cutter breaking on TBM breaking efficiency, in this study a finite element numerical simulation approach was employed to compare the effects of three cutters linear cutting breaking and rotary cutting breaking.According to the number of crucial surfaces, the three cutters breaking was classifed into, sequential breaking, inward-outward breaking, and outward-inward breaking. The results showed that outward-inward rock breaking with two critical surfaces had the least overall trend of rolling force, normal force, and specific energy, indicating that it was the most efficient of the four rock breaking modes. When the blade spacing to penetration ratio was 30, the roller cutter had the lowest specific energy for breaking rocks and the maximum efficiency. When the cutter′s linear cutting test data was converted to rotary cutting test data, the conversion coefficients for rolling force and specific energy were 1.35, while the conversion coefficient for normal force was 0.87. This study casted extra light on a theoretical basis for determining the operating parameters and designing the cutter for the TBM′s operation procedure.

Abstract: To address the influence of salt freeze-thaw environment on low-exothermic polymer materials, an investigation was conducted to explore the water absorption, mass, and compressive strength of these materials following exposure to freeze-thaw cycles in salt solutions (CaCl2 and CH3 COOK) and pure water. Scanning electron microscopy (SEM) analysis was employed to elucidate the mechanisms underlying the loss of mass and compressive strength in the materials from the microscopic point of view. Finally, based on the principles of damage mechanics, the damage variable D was determined using the mass loss rate and compressive strength loss rate as parameters. Subsequently, an evolution equation for freeze-thaw damage was established. The results indicated that as the number of freeze-thaw cycles increased, the water absorption rate of the material initially rose and then decreased slightly, while its mass and compressive strength gradually declined, suggesting that the material was subjected to certain freeze-thaw damage, and the freeze-thaw damage originated from the deformation and rupture of the cellular structure within the material. The materials suffered more severe freeze-thaw damage in salt solutions, especially in CaCl2 solutions. After 200 freeze-thaw cycles, the average mass loss and average strength loss of samples with different densities in CaCl2 solution were about 1.7 times and 1.5 times of those in pure water, respectively, furthermore, the higher the density of the material, the stronger its freeze-melt resistance.

Abstract: The wetland area between Huayuankou and Jiahetan in the lower reaches of the Yellow River is a vital component of the Yellow River corridor ecosystem. Its formation, evolution, and degradation are closely linked to the unique variations in the river′s water and sediment dynamics.To clarify the evolution characteristics of different types of wetland landscape patterns in this section and their response mechanisms to changes in water and sediment elements such as runoff, sediment transport volume and water level, in this study the section of the Yellow River from Huayuankou to Jiahetan was examined. Utilizing 12 phases of Landsat satellite imagery from 1994 to 2023, a wetland landscape database was constructed. The study employed land use transfer matrices and landscape pattern index methods to quantitatively analyze the spatiotemporal changes in wetland types, patch characteristics, and landscape structure.Subsequently, by integrating the flow, water level, sediment concentration and other data from the Huayuankou station, the response relationship between the landscape pattern characteristics of different types of wetlands and water and sediment elements was highlighted.The results indicated that: ①From 1994 to 2023, the runoff at the Huayuankou Station showed an upward trend, while sediment discharge and water levels exhibited an overall downward trend.②The areas of wetland types such as rivers, ponds, mudflats, and wild grasslands in the study area showed an overall shrinking trend, with an average annual reduction of 5.90 km2. The most significant transformation was the conversion of wetlands into farmland. ③At the landscape level, following the implementation of the water and sediment regulation policy, the overall connectivity of wetland landscapes improved, while landscape richness and evenness increased and tended to stabilize. ④In pond wetlands, mudflat wetlands, and wild grassland wetlands, indices such as the largest patch index, mean patch area, and contagion index showed significant correlations with sediment discharge and water level at the Huayuankou Station, with a particularly stronger correlation with sediment discharge. The findings of this study provided a theoretical basis for wetland ecological protection and restoration in the Yellow River Basin.

Abstract: Due to the lack of systematic research, the formula of the shear bearing capacity of the planar gate slot was absent for a long time, meanwhile the Code for Design of Hydraulic Concrete Structures required the review of the shear bearing capacity of the planar gate slots subjected to large gate force. As a special study on the revision of the design code, an experimental study on the shear bearing capacity of the downstream side gate slots with free side and end was carried out. Taking the emergency gate slots in the diversion tunnel of Laxiwa Hydropower Station as the prototype, the static experiments of the shear bearing capacity of two batches of total 14 gate slots were conducted. The process of shear failure and the failure mode was studied, the influencing factors of the shear bearing capacity of gate slots were discussed. Based on the analysis of influencing factors and the regression of the test results, the calculation formula for the shear bearing capacity of the gate slot and the lower limit for the size and concrete strength were proposed, which were applied in practical engineering. The study showed that the shear failure of the gate slots was brittle, the failure mode of the shear failure was that the crack starting from the inner edge of the gate slot inclined and extended into the downstream part of the pier with the action of the shear force, and the failure occurred while the concrete of the remain section was not enough to bear the gate force. The tensile strength of concrete, the amount of transverse reinforcement, and the width of the neck of the pier were positively correlated with the shear bearing capacity of the gate slot, and the influence of transverse reinforcement had an upper limit. The proposed calculation formula could be used to review the shear capacity of the gate slot structures.