最新刊期
- “介绍了其在钓鱼网站检测领域的研究进展，相关专家提出了一种基于钓鱼工具包指纹识别的检测方法，为解决现有检测方法依赖人工部署且准确率低的问题提供了高效解决方案。”摘要：A phishing toolkit is a collection of code resources widely used for the rapid, bulk deployment of phishing websites. By identifying the specific toolkit used by a phishing website, one can efficiently detect other sites deployed using the same toolkit, supporting precise detection and attribution analysis of phishing websites. However, existing methods mostly rely on manual deployment of toolkits and suffer from low accuracy. To address these issues, this paper proposes a phishing website detection method based on phishing toolkit fingerprinting. This method extracts fingerprints of the toolkit from its directory structure, text, and resources without the need for deployment, and employs different methods to calculate fingerprint similarity. Comparing the proposed method with three baseline methods, experiments demonstrate that without deploying the toolkit, the proposed method achieves a phishing website detection accuracy of 99.5%, which is comparable to the best baseline method. Furthermore, it achieves a phishing toolkit identification accuracy of 80.5% (an improvement of 26.2 percentage points over the best baseline) and a phishing toolkit family identification accuracy of 85.4% (an improvement of 8.6 percentage points). Ablation experiments show that the directory structure fingerprint is the core component, while text and resource fingerprints provide effective complementarity; fusing all three significantly improves the performance of toolkit and family identification.关键词：phishing website;phishing toolkit;fingerprint;similarity;tree edit distance;perceptual hash1|0|0更新时间：2026-02-05
- “三峡库区地质灾害防治研究取得新进展，专家构建了空天地内立体感知监测体系，建立基于数字孪生的智能决策支持体系，研发绿色韧性协同工程防治体系，为提升三峡库区地质灾害防治能力与推动长江经济带高质量发展提供有力支撑。”摘要：Geological hazards in the Three Gorges Reservoir Area are influenced by the combined effects of regional geological structures, topography and geomorphology, stratigraphic lithology, reservoir operation, and engineering disturbances, resulting in highly complex hazard types and posing long-term and systemic challenges to geological hazard prevention and risk control. Focusing on three typical geological hazards in the reservoir area—bank collapse, landslide, and rockfall—this study systematically summarizes the main controlling factors and corresponding engineering mitigation measures. It further elaborates on the technological advances of the regional prevention-control system in terms of monitoring, decision support, and engineering prevention and control. On this basis, future development directions are proposed, including the construction of a space-sky-ground-interior stereoscopic monitoring system, the establishment of a digital-twin-based intelligent decision-support system, and the development of green and resilient synergistic engineering prevention-control technologies, to jointly support the modernization of geological hazard prevention and control in the Three Gorges Reservoir Area and promote the high-quality development of the Yangtze River Economic Belt.关键词：Three Gorges Reservoir Area;disaster prevention and control;prevention–control system;green resilience;digital twin;artificial intelligence47|12|0更新时间：2026-02-05
- 摘要：The five Central Asian countries have long suffered from drought disasters, with severe water scarcity. In particular, droughts occurring during the growing season (April—June) pose severe threats to local ecosystems and economic development. Based on reanalysis data from 1961 to 2022, four typical growing-season drought events (in 1975, 1977, 2012, and 2021) in the five Central Asia countries were identified. By constructing a land–atmosphere water balance framework, tracing moisture transport using the HYSPLIT model, and applying a moisture budget decomposition method. This study systematically analyzes the anomalous moisture transport and source regions responsible for atmospheric moisture deficits during growing-season droughts, and evaluate the dominant physical driving processes. The results show that: 1) During growing-season droughts in the five Central Asian countries, the marked reduction in precipitation contributed by advected moisture emerges as the primary cause of rainfall deficiency, while diminished precipitation contributed by evaporative recycling further intensifies drought severity. 2) Moisture deficits associated with air parcels originating from the Mediterranean region (in 1975, 2012, and 2021) or from northern Europe and the North Atlantic (in 1977) are key factors leading to atmospheric moisture shortages during growing-season droughts. Specifically, in terms of moisture sources, the contributions from the Caspian Sea and Turkmenistan decreased substantially. 3) The moisture transport during growing-season droughts is mainly influenced by changes in stationary vortices. The dynamic effects induced by wind changes (especially divergent dynamic effects) lead to negative anomalies in the moisture budget and dominate the atmospheric moisture deficit over the five Central Asian countries.关键词：five Central Asian countries;growing-season drought;moisture budget;moisture transport;dynamic mechanism;thermodynamic mechanism11|2|0更新时间：2026-02-04
- “深度神经网络在计算机视觉领域广泛应用，但对抗扰动使其鲁棒性不足。专家提出U-Purify框架，包含扰动擦除和特征恢复模块，设计3种通用对抗净化方法，有效剔除对抗扰动，保持图像良性特征，还提出净化效率指标，实验表明U-Purify显著提高分类模型准确率和净化效率，为模型安全性提供解决方案。”摘要：In recent years, deep neural networks have been widely applied in the field of computer vision, achieving remarkable performance in tasks such as image recognition and object detection. However, these models are highly susceptible to adversarial perturbations deliberately crafted to mislead predictions, revealing critical robustness vulnerabilities and posing serious security challenges. The paper focuses on adversarial purification and proposes a general purification framework named U-Purify, which comprises two key modules: perturbation erasure and feature reconstruction. Based on the complexity of real-world scenarios, three universal adversarial purification methods with different levels of granularity are designed: 1) U-Purify(M), based on random pixel dropping and matrix completion, which performs purification using local pixel information and requires relatively low computational resources; 2) U-Purify(D), based on wavelet filtering and deep image prior, which leverages texture-level features for purification with moderate computational cost; 3) U-Purify(S), based on downsampling–upsampling and stable diffusion models, which utilizes global semantic information to achieve high-quality purification at a higher computational cost. The proposed approach is model-agnostic and dataset-agnostic, capable of effectively eliminating adversarial perturbation from input images while preserving as much of the original benign features of the image as possible. Furthermore, a new performance evaluation metric—Purification Efficiency—is introduced to assess the method’s ability to defend against adversarial attacks while accurately recognizing benign examples. Experimental results demonstrate that U-Purify, even without prior knowledge, significantly improves standard accuracy, robust accuracy and purification efficiency of classification models. On the Flowers-102 dataset, against the strongest AutoAttack, U-Purify(S) achieved a robust accuracy 6.37 percentage points higher than GDMP and 8.09 percentage points higher than Diffpure. Compared to traditional image preprocessing methods, it achieves a 62.50 percentage point improvement. Its purification efficiency surpasses the best-performing comparison method (Diffpure) by 3.70 percentage points, demonstrating excellent overall performance.关键词：deep neural network;image classification;adversarial purification;adversarial example13|10|0更新时间：2026-01-30
- “在环境科学领域，专家构建了水-土-气三相耦合模型，探究了消落带中活性氧物种的时空动态特征和主控因素，为认识消落带中活性氧物种的环境效应提供了理论依据。”摘要：The hydro-fluctuation belt is characterized by dynamic aerobic/anaerobic alternations, and it acts as a hotspot for reactive oxygen species (ROS) generation. However, the mechanism underlying vertical ROS production induced by oxygen supply from the vadose zone remains unclear. To depict this pattern, we conducted a series of saturated-unsaturated soil column experiments and developed a coupled water-soil-air three-phase model to investigate the spatiotemporal dynamics of ROS. With vertical oxygen supply from atmosphere, ROS distribution exhibited the pattern of surface enrichment, sharp decrease with increasing depth, and gradual accumulation over time. When oxygen was vertically suppled by surface water infiltration, ROS distribution was synchronized with the infiltration front of the water flow. Model sensitivity analysis reveals that ROS production is primarily governed by two key factors: reaction rate of solid electron-donating capacity (EDC) with dissolved oxygen （DO） and DO supply intensity. The results of this study provide the basis for understanding the spatiotemporal dynamics of ROS and associated environmental impacts in riparian zones under different conditions.关键词：riparian zones;oxygen;reactive oxygen species;column experiment;numerical simulation31|13|0更新时间：2026-01-23
- “最新研究发现，PVFT/PWH/MoS2复合薄膜具有优异压电性能和抗菌功能，有望在感染性组织修复治疗中应用。”摘要：The introduction of biologically active high-temperature annealed whitlockite (PWH) nanoparticles and MoS2 nanosheets to modify poly(vinylidene fluoride-trifluoroethylene) (PVFT) films has led to the development of a PVFT/PWH/MoS2 composite film with excellent piezoelectric performance and antibacterial function. Composite films with different MoS2 contents were prepared by the casting method, and their structures were characterized by scanning electron microscopy (SEM), X-ray diffraction (XRD), etc. Their piezoelectric responses, mechanical properties and biocompatibility were systematically tested. The results show that appropriate doping of MoS2 significantly promotes formation of the β phase in PVFT and enhances its electrical performance, with the piezoelectric coefficient, output voltage and surface potential reaching 24.8 pC/N, 2.33 V and -87.42 mV respectively; meanwhile, it endows it with excellent antibacterial performance, with an antibacterial rate of over 75%. The composite films exhibited a tensile strength of 148.67 MPa and a hemolysis rate below 5%, meeting the requirements for tissue repair applications. This research provides a novel strategy for designing multifunctional tissue repair materials, demonstrating significant potential for infected tissue regeneration therapies.关键词：piezoelectric film;poly(vinylidene fluoride-trifluoroethylene) (PVFT);whitlockite (WH);molybdenum disulfide (MoS2)35|14|0更新时间：2026-01-21
- “在量子信息领域，研究者基于腔辅助光力交叉科尔非线性效应，提出了两量子比特Werner态并发度的直接测量方案，为声子纠缠态度量问题提供新方法。”摘要：To address the problem that phononic entanglement is difficult to quantify, this paper, based on the cavity-assisted optomechanical cross-Kerr nonlinearity, by correlating the single-photon detection probabilities obtained from multiple measurements at the output ports of an interferometer with the concurrence of the phononic entangled state, proposes direct measurement scheme for the concurrence of two-qubit Werner state. The existing single-copy direct measurement scheme requires only a single copy of the initial Werner state per measurement, yet it relies on prior knowledge of the coefficients. In this work, we propose a multi-copy direct measurement scheme that does not require knowledge of the coefficients α and β. Although this scheme requires multiple copies of the initial Werner state per measurement, it offers better universality. The scheme proposed in this paper do not involve complex Controlled-NOT (CNOT) gate operations. Theoretical analysis shows that the proposed scheme exhibit reliable performance under experimentally achievable conditions in the current research field.关键词：entanglement measurement;Werner state;phonon;optomechanical system25|22|0更新时间：2026-01-19
-
Graph Incremental Learning Fault Diagnosis Method Based On Horizontal Visibility Adjacency Matrix AI导读
“在非欧氏工业数据建模领域，专家提出了基于水平可见邻接矩阵的图增量学习故障诊断方法，有效提升了诊断精度和模型适应性。”摘要：Graph Neural Networks (GNNs) have been widely adopted for non-Euclidean industrial data modeling. However, traditional methods relying on global adjacency matrices often fail to capture localized topological features of individual samples. Moreover, their offline operation neglects valuable information from newly collected data, limiting diagnostic accuracy. To address these challenges, this paper proposes an incremental graph learning framework for fault diagnosis based on horizontal visibility adjacency matrices (HVAMs). First, the visibility criterion is applied to preserve intrinsic fluctuation patterns of raw signals, generating HVAMs that precisely encode local topological structures. Second, a Q-learning-based incremental learning framework is designed, integrating category-aware elastic loss weighting to dynamically select high-value incremental samples. This strategy balances class distributions and optimizes training sequences. Finally, a collaborative update mechanism is established, combining gradient sensitivity propagation with topology-preserving regularization, which enhances model adaptability to online scenarios. Experiments on the TE benchmark and coal mill datasets demonstrate that our method significantly improves incremental learning performance with streaming data, achieving higher diagnostic accuracy. The results validate its robustness in complex industrial environments.关键词：fault diagnosis;horizontal visibility adjacency matrix;incremental reinforcement learning;GCN（Graph Neural Network）17|9|0更新时间：2026-01-16 - “在新能源接入电网领域，专家提出了改进SOGI-PLL结构，有效提升电网同步检测精度，为电网稳定性研究提供新方向。”摘要：The high proportion of new energy integration complicates the working conditions of the power grid, leading to insufficient output accuracy of the second-order generalized integrator (SOGI) phase-locked loop. This article proposes an improved frequency adaptive phase-locked loop structure for SOGI. Firstly, an amplitude compensation branch is introduced to compensate for the tracking of the fundamental frequency signal by SOGI, ensuring the accuracy of the orthogonal signal output by SOGI under frequency fluctuation conditions. Secondly, by connecting two adaptive SOGI modules in series, the insufficient ability to suppress low order harmonics and DC components has been improved. Finally, the positive and negative sequence separation method of orthogonal signals is introduced to accurately extract the positive sequence synchronization signal of the power grid voltage and eliminate the influence of unbalanced components on the phase-locked loop. The comparative experiment with the classic SOGI-PLL method shows that the proposed phase-locked loop structure effectively improves the accuracy of power grid synchronization detection under complex working conditions such as harmonic components, unbalanced components, and frequency fluctuations. It can complete frequency detection within 30 ms, and the detected fundamental frequency positive sequence voltage has higher sinuosity and harmonic distortion rate within 0.5%.关键词：phase-locked loop;second-order generalized integrator;pre filtering;frequency adaptation;harmonic distortion19|17|0更新时间：2026-01-14
- “最新研究提出了基于知识图谱的迭代改写RAG方法，有效提升了复杂多跳问答的推理能力，为解决幻觉问题提供新方案。”摘要：Retrieval-Augmented Generation (RAG) methods have attracted widespread attention since their proposal, with the core goal of enhancing large language models’ (LLMs) reasoning capabilities by incorporating external information to effectively mitigate hallucination. However, RAG approaches face challenges in complex multi-hop question answering, including incomplete information retrieval in a single pass and error propagation caused by redundant interference during iteration. To address these issues, we propose a Knowledge Graph-based Iterative Rewriting RAG method (KGIR-RAG), which constructs evidence subgraphs to filter structured knowledge from retrieved passages and utilizes a maximum spanning tree algorithm to eliminate noisy triplets. It iteratively rewrites queries by replacing multi-hop dependencies in questions with factual relations from the knowledge graph, progressively reducing reasoning complexity. Finally, a reflection-based reasoning module evaluates answer completeness to drive iteration termination. We conducted experiments on the HotpotQA and MuSiQue datasets to validate the effectiveness of each module. The results demonstrate that KGIR-RAG outperforms several recently proposed methods in both generation and retrieval performance.关键词：large language model;retrieval augmented generation;knowledge graph;iterative rewriting41|32|0更新时间：2026-01-09
- “在数据异构应用场景中，pFedCS算法加快联邦学习训练速度，提高模型准确度。”摘要：In the context of data heterogeneity, existing federated learning methods face challenges such as low local training speed and poor model stability after aggregation. To address these issues, this paper proposes a Personalized Federated Learning with Clustered Sampling for non-IID datasets (pFedCS) to accelerate training and improve model accuracy. The algorithm introduces a regularization loss function during local training to prevent significant deviations between the local model and global model parameters. Additionally, a similarity-based clustering method is proposed to cluster clients, determine sampling weights based on the number of samples in each cluster, and select representative clients from each cluster to participate in model aggregation. When a cluster has fewer samples, sampling is still performed to enhance sample diversity, allowing better capture of the global data distribution’s characteristics. Experimental results demonstrate that, on the MNIST and Synthetic datasets, pFedCS achieves higher accuracy and faster convergence compared to FedAvg, Per-FedAvg, FedProx, and FedTC.关键词：personalized federated learning;data heterogeneity;cluster sampling119|155|0更新时间：2025-12-17
- “在人工智能领域，专家建立了深度学习体系，为智能系统发展提供新方向。”
- “在云数据中心资源消耗预测领域，专家提出了CNN-TCAM-BiGRU预测方法，有效提高了预测准确性，为资源管理和节能优化提供技术方案。”摘要：Cloud data center resource consumption prediction involves predicting energy consumption, computing resource requirements, and other factors of the data center. In order to improve the accuracy of resource consumption prediction in cloud data centers, so as to better manage resources and optimize energy saving effect, a novel prediction method CNN-TCCAM-BIGRU is proposed, which integrates multi-channel convolutional neural network with bidirectional gated cycle unit and time-channel attention mechanism. The method first uses signal decomposition technology to preprocess the original data, aiming at eliminating noise and highlighting key features in the data. Subsequently, two focused modules are constructed along the channel and time dimensions, and then an intermediate feature map is then built using these two modules.This intermediate feature map focuses on capturing key frequency bands and semantically relevant time periods. Compared with existing benchmark algorithms, it is verified that the proposed method shows remarkable performance advantages in three main evaluation indexes. CNN-TCAM-BiGRU effectively enhances the accuracy of resource consumption prediction in cloud data centers, offering an effective technical plan for resource management and energy-saving optimization in cloud data centers.关键词：temporal attention mechanism;variational mode decomposition;bidirectional gated recurrent unit;cloud data center93|79|0更新时间：2025-12-05
- “在自动驾驶领域，研究者提出了基于区域注意力的轻量级多任务全景驾驶感知网络APDP-Net，显著降低了计算开销，保持了出色的预测性能，为端侧自动驾驶感知提供了切实可行的解决方案。”摘要：Panoptic driving perception aims to simultaneously accomplish three perception tasks—traffic object detection, drivable area segmentation, and lane line detection—to meet the requirements of drivability prediction in autonomous driving. Existing methods usually adopt multi-task learning networks to process multiple tasks simultaneously, which have achieved certain progress in performance but generally suffer from high computational complexity as well as an imbalance between prediction accuracy and real-time performance. To address these problems, a lightweight multi-task panoramic driving perception network based on area attention, named APDP-Net, is proposed. First, the Residual-Efficient Layer Aggregation Network (R-ELAN) is introduced as the backbone network, in which the area attention module, characterized by a local and continuous attention mechanism, maintains the receptive field while reducing computational cost. Second, a GhostFPN-ELAN neck network with a feature pyramid structure is designed, introducing Ghost Convolution to reduce computation by reasonably compressing the feature space. Finally, for the lane line detection task, a Polarized Self-Attention mechanism is introduced to establish long-range connections between pixel-level features, enhancing the spatial feature representation capability of the network. Experimental results show that APDP-Net, through lightweight network design, significantly reduces computational overhead while maintaining excellent prediction performance, providing a feasible solution for edge-side autonomous driving perception.关键词：multi-task learning;panoptic driving perception;area attention;autonomous driving62|80|0更新时间：2025-12-01
- “在社会治理领域，专家构建了社区治理情景知识图谱，显著提升了治理效率和智能化水平。”摘要：Knowledge graphs, due to their openness, flexibility, and strong interoperability, have significant application value in solving various social governance problems. Currently, community governance faces frequent resident demands and low processing efficiency, primarily due to the lack of systematic analysis and effective cross-departmental coordination mechanisms. To address this issue, this paper proposes a new method for constructing contextual knowledge graphs, aimed at improving community governance efficiency. First, by fine-tuning a pre-trained model, key governance elements are efficiently extracted from a small amount of labeled data, significantly improving data processing efficiency. Secondly, by combining risk-cause chain analysis with spatial-temporal dimensions, a new community governance ontology (CGO) model is constructed, which provides a structured semantic framework for the contextual knowledge graph. Based on this, the constructed contextual knowledge graph (CKG) integrates spatial-temporal information and inference rules, enabling precise risk-cause analysis and emergency severity prediction, while supporting intelligent decision-making inference and visualization queries. Experimental results show that the constructed contextual knowledge graph has strong spatial-temporal interoperability and inference capabilities, significantly enhancing the intelligence level and governance efficiency of community governance.关键词：risk causal chain;model fine-tuning;spatiotemporal interoperability;inference rules;contextual knowledge graph77|82|0更新时间：2025-11-10
- “遥感技术在极端干旱区植被生物量估算领域取得新进展，专家利用Sentinel-2影像和机器学习算法，为干旱区植被生物量估测提供新方法。”摘要：Using remote sensing technology to estimate the aboveground biomass of vegetation in extremely arid areas is of great significance for regional desertification monitoring and sustainable development. In this study, the core area of oasis in Turpan City was taken as the research area. Based on Sentinel-2 image and field sample data, vegetation index and band variables were extracted, and three sets of variable combinations were constructed. The traditional statistical model and four machine learning algorithm models of random forest (RF), support vector machine (SVM), extreme gradient boosting (XGBoost) and K nearest neighbor (KNN) were compared to realize the estimation and spatial inversion of vegetation aboveground biomass in the core area of oasis in Turpan City. The results showed that in the traditional statistical model, the cubic curve equation established by the modified soil adjusted vegetation index (MSAVI) had the best fitting effect. The estimation accuracy of the machine learning model is better than that of the traditional statistical model as a whole. In the vegetation aboveground biomass estimation model constructed by the three variable combinations, the full variable combination composed of vegetation index and band variables has achieved the best estimation effect. The RF model has the highest estimation accuracy, with a coefficient of determination (R2) of 0.72, an average absolute error (MAE) of 105.64 g·m-2, and a root mean square error (RMSE) of 127.73 g·m-2. The total value of vegetation aboveground biomass in the oasis core area of Turpan City is about 1.599×106 kg, which has obvious spatial heterogeneity and shows a “fan” shape distribution. It shows that the oasis is larger than the desert, and the oasis gradually decreases to the hinterland of the desert. The RF model combined with Sentinel-2 image showed high accuracy and stability in the inversion of vegetation aboveground biomass in extremely arid areas, and could be used for the estimation of vegetation aboveground biomass in arid areas关键词：aboveground biomass of vegetation;machine learning;extremely arid areas;Sentinel-255|75|0更新时间：2025-11-10
- “在跨链身份互认领域，研究者设计了CAMRF机制，通过改进PageRank算法和BLS签名技术，有效提高了跨域身份认证的安全性和可靠性。”摘要：To address the challenge of identity authentication between entities across different administrative domains, a cross-blockchain authenticating mechanism based on random ferrying (CAMRF) was proposed in the present study. The mechanism first selects high-reputation candidate ferry nodes from ordinary nodes through the improved PageRank algorithm. Then, the consensus random number is generated based on the verifiable random function, and the ferry node group is dynamically elected as the notary group. Finally, the group nodes forward, sign and authenticate cross-domain messages, and use BLS(Boneh-Lynn-Shacham) lightweight aggregate signature technology to verify the authenticity and validity of the messages, while addressing issues related to centralization and security. Both theoretical analysis and experimental results indicate that CAMRF markedly enhances interoperability and strengthens the security and reliability of cross-domain authentication, thereby providing some theoretical significance and high practical value.关键词：cross-blockchain technology;identity authentication;verifiable random function(VRF);BLS signature;aggregate signatures46|78|0更新时间：2025-11-05
- 摘要：This study investigates the I-optimal design problem for two-response Scheffé canonical first-order and second-order mixture models. We derive the minimum trace of the average prediction variance matrix and provide the I-optimal measures for models with specific numbers of components. The results demonstrate that the I-optimal design for these two-response Scheffé mixture models is independent of the response covariance matrix.关键词：mixture experiments;multi-response;Scheffé model;I-optimal design65|80|0更新时间：2025-10-30
- “在乡村振兴和健康中国政策背景下，成都和上海8个农村新型社区的研究表明，物质环境设施、交通条件、社会生活环境满意度显著正向影响居民自评健康，为相关政策制定和资源配置提供参考。”摘要：Under the background of rural revitalization and healthy China policy, the research on the subjective perception of rural environment and residents’ self-rated health is increasing. Existing studies mostly focus on single indicators, and lack of discussion on the joint effect of group indicators. Taking 8 typical new rural communities in Chengdu and Shanghai as examples, this study conducted a questionnaire survey on individual villagers and obtained 493 valid sample data. Factor analysis method and stepwise regression model were used to explore the common factors affecting residents’ subjective environmental perception in new rural communities and their mechanism. The results showed that the comprehensive common factor of satisfaction with physical and environmental facilities and health management (F1), the comprehensive common factor of satisfaction with traffic conditions and interpersonal activities (F2), and the comprehensive common factor of satisfaction with social life environment (F3) all significantly positively affected the self-rated health of residents, and the F1 had the greatest influence. Based on the three comprehensive common factors, the promotion strategy of planning and design of new rural communities for health was proposed, which could provide reference for relevant policy making and resource allocation of new rural communities.关键词：subjective perception;self-rated health;factor analysis method;design strategy50|88|0更新时间：2025-10-11
- “在因果推断领域，专家提出了一种三阶段加权因果效应估计算法，有效解决了协变量缺失问题，为因果效应估计提供了新方案。”摘要：Faced with the challenges of obtaining randomized controlled trial data, causal inference based on observational data emerges as an alternative research paradigm. However, missing covariates in observational data often introduces bias, leading to distorted causal effect estimation. To solve this problem, this paper proposed a three-stage weighted causal effect estimation algorithm with missing covariates. First, covariates were categorized into confounders, exposure predictors and outcome predictors based on the causal graph, with multiple imputation implemented for missing confounders to preserve the integrity of causal structures. Second, a missing pattern was introduced, constructing propensity score models through an improved covariate balancing strategy. Finally, causal effects were estimated by integrating inverse probability weighting across missing patterns. Simulation results demonstrate that compared with gradient boosting machines, the proposed method reduces root mean square error in most scenarios, validating its effectiveness in estimating causal effects under missing data conditions.关键词：causal inference;missing data for covariates;treatment effect;propensity score104|86|0更新时间：2025-10-09
- Tel：027-68756952 Email：whdz@whu.edu.cn
- Technical support is provided by Beijing Founder electronics co., LTD 京ICP备09064830号-19
京公网安备11010802024621 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
- Total Visits：1339283 Visits Today：173
Bulk Citation
0