503 results about "Automatic learning" patented technology

The invention discloses an intelligent power distribution network monitoring and fault detection system, and relates to the technical field of power distribution network detection. Comprising a power distribution network region division module, a periodic data acquisition and monitoring module, a data preprocessing and feature extraction module, an intelligent load state evaluation module, a sub-region state division module, a conventional monitoring module and a dynamic adjustment and anomaly response module, therefore, finer monitoring and management can be realized. According to the method, the power distribution network is divided into the sub-regions, and periodic state scanning is combined, so that the system can accurately identify local overload and abnormal load and respond to abnormal fluctuation in time. LSTM is combined to evaluate load fluctuation, automatic learning adapts to load change, the intelligence, flexibility and stability of the system are improved, and safety and reliability of power supply are ensured.

The invention relates to the technical field of power equipment fault diagnosis, in particular to a power equipment fault intelligent diagnosis method and system based on deep learning. The method comprises the following steps: automatically learning high-dimensional space-time correlation features in original time series data through a deep feature extraction network, and generating feature vectors representing potential abnormal modes of equipment; performing adaptive weight distribution on the high-dimensional space-time correlation features by using an attention enhancement mechanism, and marking a fault sensitive area to form enhanced fault features; inputting the enhanced fault features into a multi-level classifier for joint fault mode recognition and severity evaluation, and outputting a diagnosis result tensor containing a fault type and confidence; and an equipment maintenance decision signal is triggered based on the diagnosis result tensor, and the feature extraction network and classifier parameters are iteratively optimized according to feedback data, so that the intelligent level of operation and maintenance of the power equipment can be comprehensively improved.

The invention discloses a massive multi-source and multi-modal data fusion method, which is a technology for efficiently fusing and processing various types of data from different sources, realizes effective integration, utilization and seamless integration of multi-source and multi-modal data, and improves the breadth and depth of data analysis. According to the multi-modal feature extraction and fusion model based on deep learning, the deep features of all modal data can be automatically learned and extracted, and efficient fusion is carried out in the model. Besides, a data quality evaluation and self-adaptive adjustment mechanism is introduced, parameters and strategies in the data fusion process are dynamically adjusted according to the quality and distribution condition of the data so as to ensure the stability and reliability of the fusion result, and the method can be widely applied to multiple fields such as big data analysis, artificial intelligence, social media analysis, medical health and smart cities.

The invention discloses a wind turbine generator operation and maintenance knowledge base construction method based on a large model and mechanism self-learning. The wind turbine generator operation and maintenance knowledge base construction method comprises the steps of wind turbine generator operation and maintenance domain knowledge Schema definition and large model cue word template design used for wind turbine generator operation and maintenance knowledge extraction; obtaining operation and maintenance multi-modal data of the wind turbine generator, performing preprocessing, and performing knowledge extraction through a large model based on a designed cue word template; a dynamic knowledge association and wind turbine generator operation and maintenance knowledge base fault mechanism self-learning updating mechanism is established, operation and maintenance data and a knowledge graph are associated in real time, and the knowledge base is automatically learned and updated through an exception triggering mechanism; and constructing and storing a wind turbine generator operation and maintenance knowledge graph based on a knowledge extraction result, generating a semantic association sub-graph through clustering, generating a sub-graph clustering report, and realizing efficient knowledge retrieval. Based on the above content, the wind turbine generator operation and maintenance knowledge base which is efficient, accurate and updated in real time is constructed.

The invention provides a video monitoring early warning method based on a multi-modal behavior mode, and the method comprises the following steps: extracting multi-modal data according to a video stream, carrying out the preprocessing, carrying out the alignment and fusion of the preprocessed multi-modal data, and enabling the multi-modal data to comprise RGB frames, an optical flow field, human skeleton key points, and a scene semantic segmentation map; performing short-time behavior pattern feature and behavior pattern library establishment and scene behavior baseline construction on the fused multi-modal data; updating a dynamic behavior pattern library based on online learning, reducing a historical data weight in combination with a time decay factor, and adjusting an anomaly judgment threshold according to the complexity of a current scene; performing multi-level early warning according to the abnormal judgment threshold value, and performing feedback optimization; according to the method, an intelligent video monitoring early warning scheme which can automatically learn a scene behavior mode, dynamically adjust an early warning threshold value and fuse multi-dimensional feature analysis is provided, and the problems of high false alarm rate and poor adaptability in the prior art are solved.

The invention discloses an unmanned aerial vehicle intelligent inspection system based on AI vision and a detection switch cabinet, and relates to the technical field of unmanned aerial vehicle intelligent inspection, the unmanned aerial vehicle intelligent inspection system comprises an unmanned aerial vehicle inspection platform, and the unmanned aerial vehicle inspection platform is in communication connection with the following modules: an unmanned aerial vehicle end, which is used for collecting and preprocessing video stream data of an inspection area; extracting a key frame from the preprocessed video stream data; and the AI visual analysis module is used for analyzing the extracted key frame by using an AI visual algorithm and identifying key information and abnormal fragments in the key frame. According to the invention, through the AI vision algorithm based on the convolutional neural network model, abnormal features can be automatically learned and identified, the abnormal types and specific conditions can be rapidly determined through deep analysis of the key frames, accurate positioning of abnormal segments and comparison with the preset abnormal feature database, compared with manual detection, the accuracy is greatly improved, and the detection efficiency is improved. Tiny abnormal changes can be found in time, and potential faults can be warned in advance.

The invention discloses an entity digitization and link framework algorithm based on a heterogeneous graph attention network, and the algorithm comprises the following steps: S1, heterogeneous information network construction: carrying out the unified modeling of all multi-source heterogeneous data into a heterogeneous information network containing various types of nodes and various types of edges, s2, meta-path definition and guidance: defining "meta-paths" connecting different types of nodes to capture a complex deep semantic relationship, S3, heterogeneous graph attention network embedding: adopting an attention mechanism to enable a model to automatically learn importance of different neighbor nodes and different meta-paths, generating a final embedding vector of each entity, and establishing a heterogeneous graph attention network model; according to the method, the information fidelity is higher, modeling is directly conducted on different types of nodes and relations on a heterogeneous graph, more abundant and heterogeneous semantic information in data can be reserved compared with a multi-view method, and the end-to-end learning ability is higher; and the complexity of manually designing a fusion strategy is reduced.

The invention relates to a distributed early warning log-based automatic learning and fault prediction method and system. The method comprises the following steps: analyzing historical log traffic characteristics, generating a log rate prediction value, dynamically adjusting a sampling frequency according to the rate prediction value, and outputting a lightweight feature vector with a timestamp; generating a real-time contribution weight of each node through a dynamic weight calculation function to obtain a region-level model parameter; inputting the regional-level model parameters into a feature distiller to generate representation vectors, and performing network superposition on the representation vectors to obtain a network topological relation matrix; tensor splicing is carried out on the feature vectors and a real-time network topology relation matrix, a graph neural network model is input, and a fault probability prediction value is generated in combination with a historical feature sequence; and when the real-time contribution weight of the node continuously decreases for N times and exceeds a preset threshold value, triggering an anomaly detection instruction, injecting a feature compensation vector into the FPGA preprocessing unit, and synchronizing the model updating frequency of the feature distiller.

The invention belongs to the technical field of dry quenching, and provides a dry quenching boiler inlet temperature prediction method based on a DeepSeek large model. According to the method, the inlet temperature T6 of the coke dry quenching boiler is predicted by using the data, namely the process variable, changed in real time in the coke dry quenching process; the key control variable is a manually controlled variable, the change of the process variable is controlled / adjusted by changing the key control variable, and automatic control is realized by utilizing a prediction-control joint optimization method and by means of reinforcement learning. According to the prediction method provided by the invention, a complex nonlinear relationship and interaction among multiple variables can be captured, and strong coupling and nonlinear influence among multiple process parameters in the dry quenching process can be processed; through fine tuning on a large amount of historical data, the method can automatically learn and adapt to different working conditions, reduces the dependence on artificial feature engineering, and improves the generalization capability and prediction performance of the model.

The invention relates to the technical field of rainfall type loess side slope instability early warning methods, and discloses a rainfall type loess side slope instability early warning method and system based on multi-source data fusion, and the method comprises the steps: S1, arranging a sensor network on a side slope, and collecting rainfall capacity, surface displacement and soil pressure data in real time; s2, preprocessing the collected multi-source data, including data cleaning, missing value filling and data standardization; and S3, extracting characteristic indexes capable of reflecting the stability state of the slope from the preprocessed data. Environmental data of the loess slope, such as rainfall, evaporation capacity, temperature and humidity, are acquired by using a sensor network; the deformation data comprises earth surface displacement, deep displacement and crack width; the mechanical data comprises soil pressure, pore water pressure and soil moisture content; image data: a slope surface image and a crack development image are collected, an early warning model is constructed based on a machine learning algorithm, and data features can be automatically learned.

The invention relates to the technical field of enterprise service intelligent recommendation, and discloses an enterprise service matching method and system based on a large language model, and the method comprises the following steps: S1, receiving an enterprise service demand text input by a user, carrying out the semantic analysis of the service demand text through the large language model, and obtaining a semantic analysis result; constructing a service semantic graph containing service nodes and semantic relationships; s2, establishing a plurality of agents representing different roles in the enterprise, wherein each agent corresponds to one service preference; s3, based on the service semantic graph and the service preference, node embedding is carried out by using a graph neural network, and the service combination potential is estimated; and S4, performing multi-target matching optimization according to the service preferences and the combined potentials of the multiple agents, and outputting a service matching result. A graph neural network is adopted to model service node embedding, so that a combination relationship among services is efficiently captured in a graph structure, and automatic learning of non-uniform importance among the service nodes is realized by introducing a multi-layer graph attention mechanism.

The invention discloses an embroidery stitching method backstepping and dynamic reproduction method based on deep learning, and belongs to the technical field of embroidery stitching method identification, and the method comprises the steps: collecting stitch data; fusing the three-dimensional point cloud data acquired from multiple view angles by using a point cloud registration algorithm, and constructing a multi-modal data set; constructing a dynamic graph space-time interaction network, and extracting feature vectors in the multi-modal data based on the dynamic graph space-time interaction network; constructing a stitch inverse model to obtain a stitch space layout instruction, dynamic process parameters and an optimized array sequence; and constructing a dynamic reproduction model to obtain an actual embroidering action sequence. According to the invention, the deep features of the embroidery method in the embroidery image can be automatically learned, and the accuracy of the embroidery method backstepping and the authenticity of the dynamic reproduction are obviously improved; the method can achieve the precise reverse deduction of a plurality of embroidery methods under the conditions of complex textures, different light conditions and various embroidery angles.

The invention discloses a multi-warehouse logistics path optimization method and related equipment. The method comprises the following steps: acquiring warehouse data and customer data; constructing a distribution graph model according to the obtained data, and defining a Markov decision process; extracting homogeneous features and heterogeneous features of nodes in the distribution graph model, and fusing the extracted features and initial features of the nodes to form super-relation features; inputting the super-relation features into an encoder for feature extraction, and outputting a high-dimensional vector of each node; and the decoder gradually constructs a path decision sequence of vehicle distribution according to the vector provided by the encoder. According to the method, fine description of a heterogeneous relationship is introduced into a neural network model, and end-to-end automatic solution is realized in combination with reinforcement learning. According to the method, the warehouse distribution and path optimization strategy can be automatically learned, the warehouse distribution and path planning strategy can be automatically learned without manual parameter adjustment, a better transportation cost result can be obtained in a large-scale multi-constraint distribution scene, and the method has intelligent and practical values.

The invention belongs to the technical field of hydroelectric generating set fault diagnosis, and particularly discloses a hydroelectric generating set cavitation fault diagnosis method based on multi-channel acoustic emission signal fusion. Through the technical means of acquiring the acoustic emission signals at multiple positions, a more comprehensive cavitation phenomenon data basis is provided; the representation of time-frequency characteristics is optimized through the Mel time-frequency diagram, and the problem that the dynamic characteristics of acoustic emission signals cannot be effectively captured through a traditional method is solved; multi-channel feature extraction is performed on the Mel time-frequency graph through a preset deep convolutional neural network model, so that automatic learning of deep cavitation working conditions is realized; and multi-channel fusion and classification identification are carried out on a feature extraction result through the model, so that effective integration and fault classification of features are realized. Compared with the prior art, a more accurate and comprehensive cavitation fault diagnosis effect is realized, and the diagnosis precision is improved.

The invention relates to the technical field of underground structure management, in particular to an underground structure full life cycle management method and system based on fusion sensing data. According to the technical scheme, the system comprises a sensing layer, a data transmission layer, a data processing and analysis layer and a decision support and application layer. Through cooperation of the distributed optical fiber sensor and the multiple sensors, all-around and multi-parameter monitoring of the underground structure is achieved, the monitoring accuracy and reliability are improved, historical data and operation rules of the structure are automatically learned through a data driving prediction model, different geological environments and load conditions are adapted, a scientific basis is provided for maintenance decision making, and meanwhile, the system has good application prospects. The self-adjusting capability is realized; monitoring system application and data accumulation and utilization in each stage are considered, seamless connection and cooperative work are achieved, full-life-cycle integrated management is achieved, the overall performance of the structure is improved, the service life of the structure is prolonged, the maintenance cost is reduced, and sustainable development is promoted.

The invention discloses a protein palmitoyl transferase prediction method and system based on a multi-branch deep convolutional neural network, and belongs to the technical field of bioinformatics and artificial intelligence. The method comprises the following steps: S1, obtaining a to-be-detected protein sequence; s2, inputting the protein sequence into a pre-trained iPalmT model; and S3, judging whether the target protein is palmitoyl transferase or not according to a model output result. The iPalmT model comprises a coding module, two paths of parallel convolution branches, a feature fusion module and a classification module; and after the convolution layers of each convolution branch are stacked, an SE module is arranged and is used for channel weighting and feature re-calibration. The model extracts multi-level sequence features through convolution kernels of different scales, realizes high-precision prediction through feature fusion and a residual structure, can automatically learn multi-scale features from large-scale data, realizes end-to-end palmitoyl transferase recognition, and has high accuracy and good universality.

The present application provides an endogenous-security network method and architecture, a medium, and a device. The method comprises: extracting forwarding characteristic information of normal service data flows; delivering the forwarding characteristic information of the normal service data flows to a transport network element, so as to form a forwarding table entry, a flow table, and a forwarding white list, wherein forwarding modes in a method for binding service data flow forwarding characteristics to forwarding table entries of transport network elements comprises: an authentication and authorization mode and an automatic learning mode; the transport network element performing packet forwarding according to the forwarding characteristic information of the normal service data flows; and discarding packets of abnormal service data flows, wherein packet identification is performed for forwarding operations on the basis of a whitelist automatically generated from configuration files of switches and routers. The present application solves the technical issue of an IP network having low inherent security protection capabilities due to the openness thereof and thus requiring the deployment of a large number of external security protection facilities. The issue results in poor effectiveness, high costs, and difficulty in establishing low-cost security protection capabilities, and ultimately makes the IP network easy to attack but difficult to defend.

The invention relates to the technical field of industrial equipment state monitoring and fault diagnosis, in particular to a carrier roller fault monitoring method based on sound multi-feature fusion for a carrier roller of a belt conveyor. The method comprises the following steps: firstly, collecting a sound signal when the carrier roller runs, and segmenting the sound signal into segments with fixed duration; then, for each sound segment, a logarithmic Mel spectrum, a Mel frequency cepstral coefficient graph and a spectral contrast graph are extracted in parallel. The feature images are superposed and fused into a three-channel feature image after being subjected to independent channel normalization processing and size unification. And finally, inputting the fused three-channel feature image into a convolutional neural network for training and reasoning of a fault classification model, and realizing identification of various carrier roller fault types. According to the method, the defects of an existing carrier roller fault monitoring method in the aspects of recognition accuracy, noise immunity, fault type subdivision capability, intelligent degree and the like are overcome, and the accuracy and robustness of fault monitoring can be remarkably improved by fusing multiple complementary acoustic features and utilizing the powerful automatic learning capability of the deep learning model.

The invention relates to the technical field of recharging abnormity monitoring, in particular to a big data and artificial intelligence-based recharging behavior abnormity monitoring system, which is characterized in that historical recharging behavior data and real-time user behavior data are acquired, a user static portrait and an equipment I P portrait are constructed, deviation measurement is performed on real-time behavior characteristics, and the difference between the real-time behavior characteristics and a historical baseline is quantified, so that the recharging behavior abnormity can be monitored. The method is used for identifying abnormal operation modes. A time sequence neural network model is trained based on a historical operation sequence, typical operation links and behavioral rhythms of a user are automatically learned, and time sequence structure abnormity is accurately identified. And the abnormity monitoring module fuses static offset and behavior offset results, adopts a reinforcement learning model, dynamically adjusts a response strategy according to a risk interception effect, a false alarm condition and user feedback, realizes adaptive optimization of risk control measures, and improves the real-time intelligent risk control capability of the system in a complex recharging scene.

An object of a system according to an embodiment is to automatically learn a behavior pattern of a target person, detect an abnormality, and issue an alert.SOLUTION: A system according to an embodiment includes a GPS device, a AI learning unit, an abnormality detection unit, and an alert generation unit. The GPS device keeps track of the current location of the subject. The AI learning unit automatically learns a daily behavior pattern based on the position information of the target person acquired by the GPS device. The abnormality detection unit detects an abnormality by comparing the behavior pattern learned by the AI learning unit with the current position information. The alert issuing unit issues an alert based on the abnormality detected by the abnormality detection unit.SELECTED DRAWING: Figure 1

The invention discloses an image recognition system for constructing osteoporotic vertebral fracture healing based on artificial intelligence, and particularly relates to the technical field of orthopedic treatment, which comprises a data preprocessing module for extracting and optimizing data from an original medical image. According to the method, the advantage of discipline crossing is developed, a recognition system on medical imaging is developed based on deep learning, the advantage that the recognition system can discover detail features and hidden rules which cannot be found by human beings is utilized, and then a large amount of structured data is utilized to automatically learn and train visual feature expression of abstract data, so that the clinical missed diagnosis rate is reduced, and the clinical diagnosis efficiency is improved. The artificial intelligence architecture is trained by a rigorous and scientific means, so that the artificial intelligence architecture has clinical experience and knowledge of high-age dominant doctors and is popularized in local hospitals to achieve an auxiliary diagnosis effect, the difference of medical differentiation is reduced, the technical and experience defects of local medical personnel are made up, and the technical and experience defects of the local medical personnel are overcome. The aim of improving the definite diagnosis rate of diseases and reducing the missed diagnosis rate is one of the purposes of the invention.

The invention relates to the technical field of robots, in particular to a self-adaptive manipulator which comprises a sensing system and a control system. A multi-degree-of-freedom motion joint assembly is arranged at the lower end of the base, a driving device for providing power is installed at the lower end of the top of the base, the joint assembly comprises a lifting sleeve and a first connecting base, a lifting plate is fixedly connected to the lower end of the lifting sleeve, and second connecting bases annularly distributed at equal intervals are fixedly connected to the lower end of the lifting plate; a first rotating joint arm is rotatably connected in the first connecting seat, a third connecting seat is fixedly connected to the inner side of the first rotating joint arm, a second rotating joint arm is rotatably connected between the third connecting seat and the second connecting seat, and the control system is arranged in the base; the shape and hardness of an object can be sensed in real time, the grabbing force and posture are dynamically adjusted, it is ensured that different objects are stably grabbed, the grabbing strategy can be automatically learned and optimized, and high precision and reliability of grabbing are ensured through a high-precision sensor and a real-time feedback control mechanism.

The invention discloses a protein mass spectrum coding method based on self-supervised learning, and the method comprises the steps: guaranteeing the consistency of model input through the preprocessing steps of data enhancement, normalization, sequence length standardization and the like; by constructing a self-supervised learning task and utilizing a multi-head attention mechanism of a Transform architecture, efficient extraction of global and local features of mass spectrum data is realized; through an autoregressive encoder training framework, the model can automatically learn the internal structure of mass spectrum data without a large amount of labeled data, and high-dimensional coding representation with robustness and generalization ability is generated. According to the method, the workload of manual feature design is remarkably reduced, and the generated coding representation can be applied to downstream tasks such as mass spectrum data quality evaluation, protein identification and quantification, post-translational modification identification and biomarker discovery, and has wide applicability and high efficiency; and an innovative intelligent solution is provided for proteomics research.

The invention relates to the technical field of deep learning, in particular to a radiotherapy plan dose distribution verification method based on deep learning, and the method comprises the following steps: collecting historical radiotherapy plan data, generating a physical reference dose field through a Monte Carlo algorithm, unifying the voxel resolution of an anatomical structure to 1 cubic millimeter, and normalizing the dose according to a prescription, data enhancement is carried out only by adopting translation and mirror transformation, trace Gaussian noise is added, and a physical information enhanced three-dimensional training data set is constructed. According to the method, a three-dimensional convolutional network is utilized to automatically learn a dose distribution rule of a historical high-quality plan, a physical constraint module is embedded to ensure that a prediction result accords with a radiology principle, a real-time clinical rule engine is combined to instantly identify and correct a violation hot spot cold region, and an uncertainty quantification technology is assisted to position a high-risk region, so that the accuracy of a prediction result is improved. Finally, minute-level full-automatic verification is achieved, executable optimization suggestions are output, and efficiency is improved by dozens of times while safety is improved.

The invention discloses an AI and GIS combined natural resource space-time mapping knowledge domain construction method, which is suitable for intelligent mapping and dynamic management of natural resource space units. GIS data are acquired, multi-scale spatial range automatic identification is carried out, and map nodes are constructed; then, introducing crowd LBS behavior data, and establishing an edge relationship between dynamic nodes; a spatial topological structure between nodes is automatically learned by adopting a graph convolutional network model, and intelligent classification and correlation prediction between spatial units are realized; extracting node attribute codes, service type codes and state label codes, constructing a two-dimensional code data string, generating a standard two-dimensional code image through error correction codes and formatted information, and embedding the standard two-dimensional code image into node data; and a terminal user can access the atlas data by scanning a code and trigger permission verification and call log records. According to the method, automatic space structure identification, dynamic map updating and lightweight calling are realized, and the intelligence, integration and safety of natural resource data management are improved.

The invention discloses a multi-modal data collaborative analysis risk quantitative evaluation system, and the system comprises a multi-source data collection unit which is used for collecting different modal data; the data integration unit adopts a data cleaning technology and is used for integrating data of different formats and sources into a unified database; the fusion module is used for applying a deep neural network model based on an attention mechanism to automatically learn importance degrees of different modal data during risk assessment and dynamically distribute weights; the visual interaction unit is used for generating a visual interface according to the multi-modal data; the dynamic evaluation unit is used for constructing a personalized initial evaluation model by utilizing a machine learning algorithm; the method has the beneficial effects that the importance of different modal data in risk assessment is automatically identified by building an omnibearing acquisition system, collecting physiological, behavior, language and text multi-modal information and mining associated features through cross-modal contrast learning, and potential risk signals of a target are more accurately captured.

The invention provides an automatic learning engine device based on a packaging large model training platform, and the device comprises a unified access specification module which carries out the standardized training and reasoning parameter configuration through a YAML configuration template file; the multi-device support and parallel computing framework module is compatible with three hardware devices including a CPU (Central Processing Unit), a GPU (Graphic Processing Unit) and an NPU (Network Processing Unit), and supports two parallel computing frameworks including Accelerate and DeepSpeed; the training engine module adopts a three-layer architecture to realize task allocation, state monitoring and exception handling; and the training algorithm framework module is used for analyzing the running configuration file, carrying out data set splitting and data format conversion, supporting various fine tuning training methods and evaluating the model. According to the overall scheme, the technical threshold of large model fine adjustment is lowered, non-professional personnel can complete complex model training tasks through simple configuration, and popularization and application of the large model technology are promoted.

The invention discloses a VR scene automatic generation method based on deep learning, and relates to the related technical field of computer vision, and the method comprises the following steps: collecting image, sound and smell multi-modal feature data in a scene, and carrying out the preprocessing and feature extraction; different visual angles in the scene are collected, and a user freely switches the visual angles for observation; collecting real-time weather data, and generating a VR scene to reflect the current weather condition; evaluating the emotional state of the user; and constructing a scene generation model of the generative adversarial network, and automatically generating a VR scene according to the scene data, the visual angle, the weather and the emotional state. According to the invention, the deep learning can automatically learn and generate a VR scene, the requirements for artificial modeling and texture mapping are reduced, the production efficiency is improved, the GANs can continuously optimize the generated scene, the quality and diversity of the scene are improved, various leading-edge technologies such as smell linkage, multi-view switching, emotional state and real-time weather system are fused together, and the method is suitable for popularization and application. And a unique VR scene generation system is formed.

The invention provides a multi-dimensional financial risk early warning and dynamic management and control system based on AI driving, and the system comprises a data collection and preprocessing module which is used for collecting internal and external multi-source heterogeneous financial data of an enterprise in real time, carrying out the cleaning, standardization and feature engineering processing, and generating a structured feature vector; the AI risk assessment module is used for constructing a multi-dimensional risk assessment model based on a federated learning framework, and the federated learning framework comprises a longitudinal federated learning sub-framework, a transverse federated learning sub-framework and a model optimization unit; according to the method, a self-adaptive feature alignment algorithm based on differential privacy is adopted, cross-mechanism feature distribution and semantic level alignment are automatically learned, a mapping relation does not need to be manually preset, the dynamic change of multi-source data can be quickly and accurately adapted, the feature alignment efficiency is greatly improved, and the accuracy of feature alignment is improved. Therefore, federal learning can more timely utilize multi-mechanism data to carry out risk assessment, and more timely financial risk early warning is provided for enterprises.

The invention discloses an unmarked steel rail surface defect screening method based on self-supervised learning, and relates to the technical field of steel rail maintenance. Comprising the following steps: S100, acquiring steel rail surface image data and carrying out data preprocessing to generate an enhanced image pair; s200, constructing a defect screening basic feature encoder through a multi-scale visual pre-training model, and generating a final multi-scale fusion feature vector based on the enhanced image pair; and S300, constructing a dynamic pseudo tag generation unit, and calculating the cosine similarity between the final multi-scale fusion feature vector and the nearest neighbor normal sample feature vector. According to the method, a multi-scale visual pre-training framework is constructed, deep visual features representing the normal state and the abnormal state of the surface of the steel rail are automatically learned from massive original steel rail images on the premise that manual labeling is not needed, and a dynamic pseudo-label generation mechanism and a cross-scene migration adaptation unit are combined, so that the real-time performance of the system is improved. High-precision automatic screening of steel rail surface defects is achieved, and the generalization ability of the model in a complex environment is improved.