Accurate forecasting of iron ore futures prices and related commodities is critical for mining investors, yet complex due to multifactorial influences. To address this, an enhanced Crossformer-based prediction model is proposed, integrating transfer entropy for feature selection to identify key drivers. The model innovatively combines a sequence stabilization module and a de-stationary attention mechanism, overcoming data over-stationarization limitations in the original Crossformer architecture. Experiments on iron ore, rebar steel, and coke futures datasets confirm the model’s superiority, achieving a 22.7% reduction in MAE and 47.6% reduction in MSE compared to the original Crossformer, and outperforming FEDformer by 15.2% in MAE for iron ore futures. By improving dynamic trend capture and noise reduction, this framework provides stakeholders with robust tools to mitigate market uncertainties and optimize decision-making.

Link prediction plays a crucial role in improving the knowledge graph in the field of military equipment, but the hit rate of link prediction is closely related to the structure of entities and relationships in the knowledge graph. The objective of this study is to address the deficiencies in current link prediction models, particularly the weak interaction between entity and relation vectors, and the challenges in effectively leveraging the interaction characteristics of entities and relations. In this paper, we first construct a knowledge graph of equipment with complex entity relationships (Military), and propose a multiscale feature fusion enhancement selection network, FusionE, using the Knowledge Graph Representation Learning (KGRL) technique. This network achieves dynamic interaction of entity relationships. Considering the characteristics of data dispersion in knowledge graph structures, a multi-scale feature fusion module is designed to perform feature fusion through spatial and channel dimensions. Then, the obtained messy features are subjected to feature cleaning and enhancement. Through a dynamic feature selection mechanism, weights are assigned to the features to improve the generalization ability of the model’s knowledge reasoning. Finally, a series of benchmark models are compared and experimented on the public datasets WN18RR, FB15K-237 and Military, and the results are better than the benchmark models. The experimental results show that the method proposed in this paper effectively solves the pain point problem of missing knowledge graph in the military equipment domain, and tests on public datasets reflect the generalization ability of the model.

China Railway Express assembly center occupies an important position in China Railway Express operation network. In order to promote the development of China Railway Express, this study takes the assembly centers as the research object. Firstly, it constructs a competitiveness index system based on relevant policy documents, and uses the critic-entropy weight combination weighting method to measure and analyze logistics competitiveness. Secondly, it combines the Gini coefficient and offset-share method were used to clarify its competitive evolution characteristics from 2011 to 2023. Finally, targeted strategies to enhance competitiveness were proposed to provide reference for the construction and development of China Railway Express.

In the context of the digital age, studying the specific impact mechanism of digital transformation on enterprises has become one of the research hotspots. This article takes multi-source data of A-share listed companies on the Shanghai and Shenzhen stock markets from 2012 to 2022 as the research sample, scientifically quantifying digital transformation, and using a fixed effects panel model to empirically study the impact mechanism of digital transformation on enterprise innovation and further on enterprise development. Research has shown that digital transformation has a significant positive impact on enterprise innovation. Digital transformation can enhance the research and development intensity of enterprises, thereby affecting enterprise innovation, and transparency can positively regulate the impact of digital transformation on enterprise innovation; Digital transformation will have a promoting effect on enterprise innovation, which in turn will promote the development of enterprises. Based on the results of mechanism research, corresponding suggestions have been proposed for enterprise decision-making for reference.

This study aims to address the problem of named entity recognition (NER) in wind turbine fault texts. Currently, most wind turbine fault texts are characterized by an abundance of technical terms, significant long-distance dependencies, and unclear entity boundaries. Therefore, we propose an improved WC-LSTM model (GR-WCM). First, we use a self-attention mechanism to encode lexical information, capturing all lexical information for word-character fusion. Second, we introduce a Global Context Mechanism (GCM) that integrates the representation of the entire future and past sentences into each unit of the Bi-LSTM framework to obtain semantic features of long sequences. We also employ the R-drop training algorithm during model training to further enhance the model’s generalization ability. Finally, we use Conditional Random Fields (CRF) for sequence decoding to obtain the globally optimal sequence. Experimental results on a self-built wind turbine fault dataset and the Resume dataset show that, compared with the baseline WC-LSTM model, the proposed method improves the F1 score by 2.12

In response to the issues of sample imbalance, complex backgrounds, and insufficient detection accuracy of small targets in the images of insulators collected by unmanned aerial vehicles (UAVs) during power grid inspection, a method based on multiple improvement strategies of YOLOv5 is proposed. Firstly, large-resolution data is sliced to increase the proportion of small targets in the overall image. Secondly, the kmeans++ _CIoU method is employed to optimize anchor box selection, enabling the model to better adapt to insulator bounding boxes of different sizes within the dataset. Additionally, the CBAM attention mechanism is incorporated to enhance focus on insulator features from both channel and spatial perspectives. Finally, Focal Loss is incorporated to encourage the model to prioritize more challenging data, alleviating the impact of sample imbalance. The experimental results indicate that the detection method exhibits varying degrees of improvement in average precision as well as in the accuracy of detecting large, medium, and small targets.

Wind energy is a renewable resource with abundant reserves, and its sustainable development and utilization are crucial. The components of wind turbines, particularly the blades and various surfaces, require meticulous defect detection and maintenance due to their significance. The operational status of wind turbine generators directly impacts the efficiency and safe operation of wind farms. Traditional surface defect detection methods for wind turbines often involve manual operations, which suffer from issues such as high subjectivity, elevated risks, low accuracy, and inefficiency. The emergence of computer vision technologies based on deep learning has provided a novel approach to surface defect detection in wind turbines. However, existing datasets designed for wind turbine surface defects exhibit overall category scarcity and an imbalance in samples between categories. The algorithms designed face challenges, with low detection rates for small samples. Hence, this study first constructs a benchmark dataset for wind turbine surface defects comprising seven categories that encompass all common surface defects. Simultaneously, a wind turbine surface defect detection algorithm based on improved YOLOv5 is designed. Initially, a multi-scale copy-paste data augmentation method is proposed, introducing scale factors to randomly resize the bounding boxes before copy-pasting. This alleviates sample imbalances and significantly enhances the algorithm’s detection capabilities for targets of different sizes. Subsequently, a dynamic label assignment strategy based on the Hungarian algorithm is introduced that calculates the matching costs by weighing different losses, enhancing the network’s ability to learn positive and negative samples. To address overfitting and misrecognition resulting from strong data augmentation, a two-stage progressive training method is proposed, aiding the model’s natural convergence and improving generalization performance. Furthermore, a multi-scenario negative-sample-guided learning method is introduced that involves incorporating unlabeled background images from various scenarios into training, guiding the model to learn negative samples and reducing misrecognition. Finally, slicing-aided hyper inference is introduced, facilitating large-scale inference for wind turbine surface defects in actual industrial scenarios. The improved algorithm demonstrates a 3.1% increase in the mean average precision (mAP) on the custom dataset, achieving 95.7% accuracy in mAP_50 (the IoU threshold is half of the mAP). Notably, the mAPs for small, medium, and large targets increase by 18.6%, 16.4%, and 6.8%, respectively. The experimental results indicate that the enhanced algorithm exhibits high detection accuracy, providing a new and more efficient solution for the field of wind turbine surface defect detection.

Accurately predicting agricultural product prices is crucial for the stability of the agricultural market. However, due to the collection intervals of agricultural price data, which are typically daily, weekly, or monthly, the data volume is relatively small. This limitation affects the accuracy of deep learning models in price prediction. Data augmentation is an effective method to improve the accuracy of agricultural price predictions. This research proposes a novel data augmentation model, EMD-TCN-TimeGAN. Based on TimeGAN, this model has been significantly improved by replacing the RNN-based components with TCN, which enhances the model’s ability to capture long-term dependencies. Additionally, the model incorporates an EMD module to decompose the original price series, thereby better capturing the intrinsic structural information of the time series and further improving the quality of the generated data. Experiments conducted on a collected corn dataset show that adding the augmented data to the training set of the prediction model and retraining it significantly improves the prediction accuracy. The coefficient of determination (R2) increased from 0.928831 to 0.968742, and the mean absolute error (MAE) decreased from 0.026023 to 0.016628. Comparisons with the TimeGAN model using RNN components validate the effectiveness of replacing the components with TCN; comparisons with TCN-TimeGAN validate the effectiveness of the EMD module in further improving prediction performance. The EMD-TCN-TimeGAN proposed in this study generates sequences that better preserve the temporal dynamics of the time series data, maintaining the temporal relationships between the original variables. This improves the accuracy of small-sample agricultural price predictions, which is of significant importance and benefit to agricultural production and related activities.

针对钢板表面缺陷检测中存在的边缘分割粗糙、漏检和误检率高等问题,提出了一种引入注意力机制的多尺度特征融合的DeepLabv3+检测方法.在DeepLabv3+网络的解码区中,充分利用多尺度特征信息,对跃层特征融合进行优化,保留浅层特征并对深层特征进行了细化的上采样操作,获得更精细的缺陷边缘;在编码区主干网络ResNet101中引入坐标注意力机制,增强特征提取能力,提高分割准确率.设计了加权Dice损失和二元交叉熵损失(BCEloss)结合的优化损失函数来缓解样本不均衡的问题,提高分割精度.改进DeepLabv3+网络的Dice系数和mIoU值分别提高了6.0%和7.92%,刮痕缺陷边缘分割更准确,对凹坑、边缘裂纹与氧化铁皮缺陷的分割效果提升明显,实验结果验证了该方法处理钢板表面缺陷问题的有效性.

Long-term forecasting of wheat prices is important for agricultural production, market policy making and food security assurance. In this study, an improved long time series forecasting model for wheat price, DWT-TimesNet, is proposed, which is deeply improved on the basis of the original TimesNet model. The main improvement lies in the use of Discrete Wavelet Transformation (DWT) instead of the original Fast Fourier Transformation (FFT) for time series decomposition. Since the FFT has certain limitations in dealing with non-smooth signals, such as fixed frequency resolution and the inability to provide time and frequency information at the same time, the DWT overcomes these shortcomings through the characteristics of multi-resolution analysis, and is able to localize signals at different scales and provide a more refined time-frequency representation. In the improved DWT-TimesNet model, we use the mean plus standard deviation method to dynamically determine a threshold that is used to filter valid cycles, and cycles corresponding to amplitudes above this threshold are considered valid cycles. This improvement allows the model to be more flexible in adapting to signal variations, and improves the model’s ability to capture the periodic properties of the signal compared to the fixed cycle selection mechanism in the original method. Furthermore, the improved method returns an additional parameter k indicating the number of selected cycles, which provides more information for subsequent processing and increases the interpretability of the method. To validate the effectiveness of the DWT-TimesNet model, we compare it with the original TimesNet model and other mainstream time series forecasting models. The experimental results show that the DWT-TimesNet model achieves remarkable results in wheat price forecasting, with results far ahead of other models. This research is not only academically significant, but also has important implications and benefits for practical production and life. The accurate prediction of wheat prices can help farmers rationalize their planting plans and reduce production costs; the government can formulate reasonable market policies based on the prediction results and stabilize market prices; and it also helps to achieve food security guarantee.

针对金融文本语义特征难以准确提取的问题,提出一种面向金融文本的实体关系抽取方法.使用预训练模型Fin-BERT对输入金融文本进行字、词粒度特征提取;在编码层引入时序格子网络将词粒度特征动态融合到字粒度特征中;采用字级别注意力机制合并所有字粒度特征,获取更准确的句子表示,进行关系分类.实验结果表明,该方法在金融文本数据集上F1值达到91.8％,较目前主流方法效果更好.

随着新冠感染患者数量的增多,产生了大量与之相关的流调数据.以流调数据为基础,通过分析患者间的语义关联特征可以在个体层面表达疾病的传播过程,深入探讨患者染病的特征分布、患者之间的传播路径等问题.基于此,研究以患者为中心并兼顾语义关联特征,借助知识图谱技术完成对患者流调数据的建模.首先在解析流调数据的基础上定义患者语义关系,据此设计患者关系图谱的模式层.然后,通过识别患者、地点实体,抽取"患者-关系-患者"及"患者-居住-地点"三元组等任务完成数据层构建.最后,利用Neo4j图数据库实现患者关系图谱的可视化并加以分析.结果表明,通过对超级传播源分析和传播路径追溯等层面进行验证,患者关系图谱可以挖掘患者的内在关联、有效整合患者语义关系,表达疾病在患者间的传播过程.

对大宗商品企业类交易商进行信用风险评估,能为交易市场精准服务和监管机构分级分类监管提供依据.为有效全面地评估交易商企业信用风险,基于违约相依理论,构建了财务指标和违约传染变量相结合的信用风险评估指标体系.在此基础上,文章采用二元Logistics逐步回归方法对指标进行筛选,建立了基于BP神经网络的企业信用风险评估模型,并采用284家交易商企业相关数据进行实证分析.结果表明,相对于传统的财务指标模型,引入传染变量的模型整体预测准确率提高了6.6％,高风险企业的预测准确率提高了28.3％.

Price forecasting is important for the stability of bulk agricultural commodity markets, but bulk agricultural commodity prices have complex correlations with multiple factors. In order to address the current problems of strong dependence on data integrity and the difficulty of single models to fully utilize multiple data features in price forecasting,a boosting ensemble learning method that combines the attention mechanism-based convolutional bi-directional long short-term memory neural network（CNN-BiLSTM-Attention）, support vector regression（SVR）, and LightGBM is proposed, and experiments are conducted on the datasets containing historical trades, weather, exchange rate, oil price,and other features data, respectively. The experiment takes the price forecasting of wheat and cotton as the target task,uses the mutual information method for feature selection, selects the CNN-BiLSTM-Attention model with low error as the base model, and performs boosting ensemble learning with the machine learning model through linear regression. The experimental results show that the root mean square error（RMSE） of the ensemble learning method is 12.812 and 74.356for wheat and cotton datasets, which are 11.00%, 0.94%, 4.44%, 1.99%, 13.03%, and 4.39% lower than the three base models, respectively. The method can effectively improve the accuracy of price forecasting.

棉花价格受多种因素影响而复杂多变,通过选择合适的数据特征和预测模型可提高棉花价格预测精度.本文以棉花日现货价格数据为研究目标,采集了供需关系、国际市场、宏观经济、产业链这4个方面的9项影响因素作为特征,使用极限梯度提升(XGBoost)算法对棉花价格影响因素进行特征评估筛选,选取其中5项特征后,采用引入注意力机制(Attention)的时间卷积网络(TCN)TCN-Attention、TCN、LSTM、GRU等模型对棉花价格进行预测.通过消融实验和对比实验,结果表明:(1)经过XGBoost特征筛选后,TCN-Attention价格预测的平均绝对误差(MAE)和均方根误差(RMSE)为41.47和58.76,与未筛选相比分别降低了 77.57％和76.49％.(2)与TCN、LSTM、GRU相比,本文提出的TCN-Attention模型预测结果更准确,MAE和RMSE均降低50％以上,运行时间较 LSTM、GRU 缩短 60％.

Aiming at the problems of low recognition accuracy and high miss-detection and misdetection rates encountered in the task of helmet wear detection for site operations, this study proposes an improved YOLOv7-tiny target detection algorithm for enhancing the performance of helmet wear detection. The algorithm employs the Mosaic method to randomly splice the input images, and subsequently realizes efficient enhancement of the data by performing random rotation, scaling and cropping operations on the spliced images. To enhance the stability of the model, we replace the coordinate loss in the YOLOv7-tiny network with WIoU (Wise-IoU) loss. To enhance the fusion effect of the feature layer and to solve the problem of gradient vanishing in the deep network, we introduce the RepBlock layer in the original network. Experiments on the publicly available helmet detection dataset show that the improved YOLOv7-tiny target detection algorithm not only makes the convergence of the loss function more stable compared to the original network, but also improves the detection accuracy to 94.2%, which is an improvement of 18.4 percentage points compared to the original network; at the same time, the mAP is also significantly improved to 92.8%. In addition, the leakage false detection is also effectively reduced, thus proving the significant advantage of the improved algorithm over the original network in terms of detection performance.

针对大宗农产品价格波动研究中对新闻事件影响分析不足的问题,以生猪市场的新闻数据为例,提出了一种大宗农产品价格趋势事理图谱构建方法.首先,利用因果触发词匹配的方法,对领域新闻进行因果三元组抽取;随后结合事件规则模板匹配与k-means聚类方法,将三元组中的事件描述划分为12类,并根据实际价格波动情况,分别构建了上升、平稳、下降三种价格趋势下的事理图谱;最后,利用贝叶斯理论对新发生的事件进行事件推理.实验结果表明,通过该方法所构建出的事理图谱在上升、平稳、下降三种价格趋势下的事件推理平均倒数排名(mean reciprocal rank,MRR)分别为0.541、0.571、0.510,整体效果均优于传统的TransE方法,证明了该方法能够有效地辅助分析不同价格趋势下的事件关系.

供应链金融是银行将核心企业和上下游企业的资信捆绑起来提供金融产品和服务的一种融资模式,可以有效缓解中小型企业融资难的问题.基于演化博弈理论并借助Matlab软件探讨了大宗商品交易平台中农产品供应链应收账款融资问题.研究发现供应链企业为银行带来的额外收益、双方都违约时平台给予银行的补偿款、银行的放款额度及支付的货款订金会影响银行的演化策略;支付的货款订金、供应链违约惩罚及缴纳的交易保证金会影响农产品买方的演化策略;供应链企业带给银行的额外收益、双方都违约时平台给予银行的补偿款、银行的放款额度、缴纳的交易保证金及供应链违约惩罚会影响农产品卖方的演化策略.最后结合数值仿真结果为大宗商品交易平台及银行的管理人员提供了参考性建议.

For the problem of short-term power load forecasting, a numerical algorithm is presented based on Temporal Convolutional Network (TCN) and XGBoost. Firstly, correlations between load and influencing factors are analyzed so as to extract important features by XGBoost. Secondly, residual blocks in TCN mainly deal with gradient disappearance or explosion for the long time series. Combined with attention mechanism, important feature vectors may be benefit to advance the accuracy of forecasting results. Short-term power load forecasting is carried out with multiple scales by two groups of public data. Compared with TCN, Gate Recurrent Unit (GRU) and Long Short-Term Memory (LSTM), the presented algorithm needs less time when high-frequency or high-dimensional data appears.

近年来,随着有关大宗商品电子交易市场由恶意主体导致的信用风险事件的媒体报道的披露,市场和行业对企业类交易商信用违约事件的关注度持续高涨.评估企业类交易商信用风险大小对预防市场内企业违约事件的发生具有重要意义.本文在研究文献和市场风险案例的基础上,试图找出目前企业信用风险评估指标体系中存在的问题,并遵照评估指标体系的构建原则,梳理信用风险成因,选取多维度、细粒度的指标,运用层次分析法,计算各级指标合成权重,进而构建了一套更系统、更有针对性的大宗商品企业类交易商信用风险评估指标体系,以供业内人士参考.