Railway signal infrastructure data management method and system
Through multimodal sensor array and deep learning technology, combined with blockchain and multi-party security computing, the problem of low data utilization in the data management of railway signal basic equipment is solved, high-quality data acquisition and fault prediction are achieved, and equipment is improved Intelligent management level.
Patent Information
- Application Number
- CN202411248807.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-06
- Publication Date
- 2024-11-29
AI Technical Summary
The existing railway signal basic equipment data management methods cannot make full use of multi-source heterogeneous data, and there are shortcomings in data collection, quality assurance, fusion, knowledge representation and privacy protection, which cannot meet the equipment status monitoring, fault diagnosis and predictive maintenance. need.
Data is collected through multimodal sensor arrays, data compression and feature extraction algorithms are used to generate device operation feature vectors, combined with deep learning and causal reasoning technology to perform automatic semantic understanding and fault omen analysis, and data storage using blockchain and multi-party security computing technology and privacy protection, building a knowledge base of knowledge graphs and behavior patterns.
It realizes the acquisition and transmission of high-quality data, extracts key equipment information, improves the accuracy of fault prediction and the security of data storage, and improves the intelligent management level of railway signal equipment.
Smart Images

Figure CN119046409A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data processing, and in particular to a method and system for managing railway signal infrastructure equipment data. Background Art
[0002] With the continuous development and modernization of railway signal systems, the types and quantity of railway signal infrastructure equipment have continued to increase, generating a large amount of heterogeneous data. These data include equipment operating parameters, environmental information, status information, etc., which have the characteristics of multi-source heterogeneity, large data volume, and strong real-time performance.
[0003] Traditional data management methods for railway signal infrastructure equipment mainly rely on manual records and simple digital storage, with low data utilization, and cannot meet the needs of equipment status monitoring, fault diagnosis, predictive maintenance, etc. With the development of technologies such as big data and artificial intelligence, it has become a trend to use advanced data management and analysis technologies to improve the management level of railway signal infrastructure equipment;
[0004] At present, there have been some studies and applications in the data management of railway signal infrastructure equipment, which mainly use sensor technology to collect equipment operation data and realize equipment status monitoring and fault diagnosis through data analysis. However, the existing methods are still insufficient in data collection, data quality assurance, data fusion, knowledge representation, privacy protection, etc., and cannot give full play to the value of data.
[0005] Therefore, a solution is urgently needed to solve the problems existing in the prior art. Summary of the invention
[0006] The embodiments of the present invention provide a railway signal infrastructure equipment data management method and system, which can at least solve some of the problems existing in the prior art.
[0007] A first aspect of an embodiment of the present invention provides a railway signal infrastructure data management method, comprising:
[0008] The operation parameters, environmental information and status information of railway signal equipment are collected through a pre-set multimodal sensor array to form multi-source heterogeneous data, the multi-source heterogeneous data are adaptively compressed and feature extracted through a data compression algorithm and a feature extraction algorithm to obtain an equipment operation feature vector, the abnormal points in the equipment operation feature vector are identified and repaired through an abnormal point detection algorithm to obtain a high-quality data set, the network bandwidth is matched with the element form in the high-quality data set through a self-organizing data transmission network, and the high-quality data set is transmitted to a control center as unstructured data;
[0009] Based on the unstructured data, automatic semantic understanding is performed through a deep learning algorithm to generate a semantic feature library, key information corresponding to railway signal equipment is extracted, feature data enhancement is performed on the equipment operation feature vector through a small sample learning algorithm, and the fault precursor corresponding to the current equipment is determined through a spatiotemporal correlation representation learning algorithm, and a health status assessment report is generated. A railway signal knowledge graph is constructed based on the semantic feature library and the health status assessment report, and a full-cycle behavior pattern knowledge base is constructed through causal reasoning technology;
[0010] Based on the full-cycle behavior pattern knowledge base, the multi-source heterogeneous data are classified and stored through a distributed storage system, and the credibility is verified in combination with blockchain technology. For the multi-source heterogeneous data that have passed the verification, according to the access popularity, the multi-source heterogeneous data are hierarchically managed through multi-level intelligent storage and data sharing and fusion are performed through multi-party secure computing technology. For the initial fused data after fusion, the privacy score is determined through a pre-set privacy risk assessment model and the initial fused data is screened according to the privacy score to obtain high-quality fused data and encrypted in combination with blockchain technology.
[0011] In an optional embodiment,
[0012] The operation parameters, environmental information and status information of railway signal equipment are collected through a pre-set multimodal sensor array to form multi-source heterogeneous data, the multi-source heterogeneous data are adaptively compressed and feature extracted through a data compression algorithm and a feature extraction algorithm to obtain an equipment operation feature vector, the abnormal points in the equipment operation feature vector are identified and repaired through an abnormal point detection algorithm to obtain a high-quality data set, the network bandwidth is matched with the element form in the high-quality data set through a self-organizing data transmission network, and the high-quality data set is transmitted to a control center as unstructured data including:
[0013] Based on a pre-set multimodal sensor array, operating parameters, environmental information and status information of railway signal equipment are collected, wherein the multimodal sensor array determines the sampling rate of different sensors according to the pre-set importance of each device, performs data collection according to the sampling rate, obtains the operating parameters, environmental information and status information corresponding to each device, and combines them to obtain the multi-source heterogeneous data;
[0014] For the multi-source heterogeneous data, according to the edge computing nodes in the multimodal sensor array, the data distribution characteristics of the multi-source heterogeneous data are determined by an adaptive data compression algorithm, and the spatial correlation of the data corresponding to the coding dictionary and the railway signal equipment is combined to perform collaborative compression through distributed wavelet transform to generate initial compressed data, and the key features in the initial compressed data are determined by empirical mode decomposition and principal component analysis algorithms to obtain the equipment operation feature vector;
[0015] Initialize a pre-set outlier detection model based on an isolation forest and a data repair rule library based on the equipment operation feature vector, divide the equipment operation feature vector into multiple data blocks of the same size according to the model input requirements, and for each data block, vote and generate a voting result through the judgment module in the outlier detection model, and determine the detection result in combination with the spatiotemporal correlation of the data corresponding to the railway signal equipment and the outlier detection mechanism. If an outlier is detected, perform data verification on the current data block and locate the real outlier. For the real outlier, perform data repair according to the data repair rule library and random walk to obtain the high-quality data set;
[0016] The high-quality data set is sent to the control center through a self-organizing data transmission network, wherein the self-organizing data transmission network includes a network convergence layer, a network access layer and a network optimization layer. The network convergence layer builds a multi-channel heterogeneous fusion network based on wireless communication technology and sends the received high-quality data set to the network access layer. The network access layer accesses data through an adaptive multiple access mechanism according to the structure of the elements in the high-quality data set, and at the same time assigns attention weights to the elements in the high-quality data set based on the attention mechanism, assigns channels to each element according to the attention weight and sends it to the control center. Before sending, the network optimization layer determines the node energy consumption based on the network traffic prediction algorithm and self-optimizes the allocation of channel resources through cognitive radio technology and spectrum sensing to obtain the unstructured data.
[0017] In an optional embodiment,
[0018] For the real outliers, data repair is performed according to the data repair rule base in combination with random walks as shown in the following formula:
[0019]
[0020] in, represents the eigenvector of abnormal node a after the t+1th iteration, μ represents the random walk iteration coefficient, SI(f j , f a ) represents the similarity measurement function of the feature vectors between abnormal nodes a and j, ∑k represents the sum of normal nodes in the neighborhood, SI(f j , f k ) represents the similarity measurement function between node j and the normal node k in the neighborhood, represents the feature vector of abnormal node a after the tth iteration, Represents the initial feature vector of abnormal node a.
[0021] In an optional embodiment,
[0022] Based on the unstructured data, automatic semantic understanding is performed through a deep learning algorithm to generate a semantic feature library, key information corresponding to railway signal equipment is extracted, feature data enhancement is performed on the equipment operation feature vector through a small sample learning algorithm, and the fault precursor corresponding to the current equipment is determined through a spatiotemporal correlation representation learning algorithm, and a health status assessment report is generated. A railway signal knowledge graph is constructed based on the semantic feature library and the health status assessment report, and a full-cycle behavior pattern knowledge base is constructed through causal reasoning technology, including:
[0023] Based on the unstructured data, format conversion is performed through a deep learning algorithm and the text is converted into a word-unit identity sequence in a vocabulary through a word segmentation operation, the word-unit identity sequence is added to a pre-trained deep bidirectional processing model, and a semantic feature vector corresponding to each word-unit is output through an encoder;
[0024] Based on the semantic feature vector, a semantic understanding model is constructed and an ontology in the field of railway signal equipment is constructed, equipment type, component name, status description and failure mode are formally defined as key concepts, hierarchical relationships between the key concepts are defined and concept nodes in the ontology are embedded into the bidirectional processing feature space corresponding to the deep bidirectional processing model, the semantic association between the ontology and the word is determined in combination with the attention mechanism, and the key information of the railway signal equipment is automatically extracted in combination with the conditional random field sequence labeling model to obtain the key information;
[0025] Based on the key information, a small sample feature learning is performed on the equipment operation feature vector through a model-independent meta-learning algorithm, the equipment operation feature vector is divided into a meta-training set and a meta-testing set, two-level optimization is performed through the meta-training set and an initial feature extractor is determined, the initial feature extractor is optimized based on the meta-testing set, and feature data enhancement is performed in combination with a generative adversarial network, and for the enhanced equipment operation feature vector, spatiotemporal correlation modeling is performed through a spatiotemporal attention network, and the time dependency of railway signal equipment and the spatial correlation characteristics between adjacent equipment in the network topology are determined respectively, and the fault precursor is determined based on the time dependency and the spatial correlation characteristics, and a health status assessment report is generated;
[0026] A railway signal knowledge graph is constructed based on the semantic feature library and the health status assessment report. Entity connection and attribute matching are performed through a rule-based entity linking method, and pre-acquired unstructured data is linked to the railway signal knowledge graph. In combination with causal reasoning technology, entity relationships and logical associations are modeled through vector operations, and missing values of the graph are predicted and expanded to obtain the full-cycle behavior pattern knowledge base.
[0027] In an optional embodiment,
[0028] The objective function corresponding to the conditional random field sequence labeling model is shown in the following formula:
[0029]
[0030] Where X represents the input sequence, Y represents the predicted label sequence, Z(X) represents the partition function, and t h (y i-1 ,y i , X,i) is the transfer feature function, indicating the label y i-1 to label y i The transfer characteristics, λ h represents the transfer characteristic function t h The weight, μ k Represents the state characteristic function s k The weight of k (y i , X, i) is the state feature function, indicating that the label y at position i i and the features between the observation sequence X.
[0031] In an optional embodiment,
[0032] Based on the full-cycle behavior pattern knowledge base, the multi-source heterogeneous data is classified and stored through a distributed storage system, and the credibility is verified in combination with blockchain technology. For the multi-source heterogeneous data that has passed the verification, according to the access popularity, the multi-source heterogeneous data is hierarchically managed through multi-level intelligent storage and data sharing and fusion are performed through multi-party secure computing technology. For the initial fused data after fusion, the privacy score is determined through a pre-set privacy risk assessment model and the initial fused data is screened according to the privacy score to obtain high-quality fused data and encrypt it in combination with blockchain technology, including:
[0033] Based on the full-cycle behavior pattern knowledge base, determine the data category corresponding to the multi-source heterogeneous data, and classify and save the multi-source heterogeneous data according to the data category through the distributed storage system;
[0034] Based on blockchain technology, the multi-source heterogeneous data is uploaded to the blockchain network according to the predefined format and specification. For each piece of multi-source heterogeneous data, a corresponding transaction is generated, and a consensus is reached between different nodes in combination with the consensus mechanism and packaged and sent to the block. For the multi-source heterogeneous data uploaded to the block, the credibility is verified through the verification rules pre-defined in the smart contract, and a credibility score is generated and compared with a preset credibility threshold. If the credibility score is greater than the credibility threshold, it is considered that the verification is passed, and the multi-source heterogeneous data corresponding to the current credibility score is retained, otherwise the multi-source heterogeneous data is deleted;
[0035] For the verified multi-source heterogeneous data, a heat score is calculated based on the user's data request records and access time statistics. The multi-source heterogeneous data is divided into three levels: hot data, warm data, and cold data according to the heat score. A multi-level storage architecture is designed to store the verified multi-source heterogeneous data in combination with the levels. According to the change in heat score, data migration and update are performed through a heat-aware data migration strategy. The storage ratio is jointly calculated through a multi-party secure computing protocol, and the multi-source heterogeneous data is divided into multiple shards according to the storage ratio through secret sharing technology for data encryption. The encrypted data is combined with homomorphic encryption technology for fusion calculation to obtain initial fusion data.
[0036] For the initial fused data, a privacy scoring model based on information entropy is used to calculate the privacy score of each piece of initial fused data in combination with the sensitivity and associated risk of the data. The privacy score is screened according to a preset privacy risk threshold, and the initial fused data with a privacy score lower than the privacy risk threshold is removed to obtain the high-quality fused data, which is then saved in combination with blockchain encryption technology.
[0037] In an optional embodiment,
[0038] The privacy score of each piece of initial fused data is calculated by combining the sensitivity of the data and the associated risk as shown in the following formula:
[0039]
[0040] Among them, PS represents the privacy score, w s represents the sensitivity weight, P m represents the probability of the data taking the value of the mth attribute, w r represents the associated risk weight, R represents the associated risk of the data, and w t represents the timeliness weight, T represents the timeliness of the data, and w d represents the data volume weight, and D represents the amount of data.
[0041] A second aspect of an embodiment of the present invention provides a railway signal infrastructure data management system, including:
[0042] The first unit is used to collect railway signal equipment operation parameters, environmental information and status information through a pre-set multimodal sensor array and form multi-source heterogeneous data, adaptively compress and extract features of the multi-source heterogeneous data through a data compression algorithm and a feature extraction algorithm to obtain an equipment operation feature vector, identify and repair abnormal points in the equipment operation feature vector through an abnormal point detection algorithm to obtain a high-quality data set, match the network bandwidth with the element form in the high-quality data set through a self-organizing data transmission network, and transmit the high-quality data set to a control center as unstructured data;
[0043] The second unit is used to perform automatic semantic understanding and generate a semantic feature library based on the unstructured data through a deep learning algorithm, extract key information corresponding to railway signal equipment, perform feature data enhancement on the equipment operation feature vector through a small sample learning algorithm, and determine the fault omen corresponding to the current equipment through a spatiotemporal correlation representation learning algorithm, generate a health status assessment report, build a railway signal knowledge graph based on the semantic feature library and the health status assessment report, and build a full-cycle behavior pattern knowledge base through causal reasoning technology;
[0044] The third unit is used to classify and store the multi-source heterogeneous data through a distributed storage system based on the full-cycle behavior pattern knowledge base, and to verify the credibility in combination with blockchain technology. For the multi-source heterogeneous data that have passed the verification, according to the access popularity, the multi-source heterogeneous data are hierarchically managed through multi-level intelligent storage and data sharing and fusion are performed through multi-party secure computing technology. For the initial fused data after fusion, the privacy score is determined through a pre-set privacy risk assessment model and the initial fused data is screened according to the privacy score to obtain high-quality fused data and encrypt it in combination with blockchain technology.
[0045] According to a third aspect of the embodiments of the present invention,
[0046] An electronic device is provided, comprising:
[0047] processor;
[0048] a memory for storing processor-executable instructions;
[0049] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.
[0050] A fourth aspect of the embodiments of the present invention is:
[0051] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.
[0052] In the present invention, by using a self-organizing data transmission network, the network bandwidth is matched with the element form in the high-quality data set, so that data transmission can be optimized. According to the importance of the data and the network status, the transmission strategy is dynamically adjusted to improve the network utilization efficiency and ensure the timely transmission of key data. The deep learning algorithm is used to automatically understand the semantics of unstructured data and generate a semantic feature library, which can extract the key information of railway signal equipment. The small sample learning algorithm is used to enhance the feature data of the equipment operation feature vector, expand the training sample, and improve the generalization ability of the model. Based on the semantic feature library and the health status assessment report, a railway signal knowledge graph is constructed to form a structured representation of knowledge such as equipment status, failure mode, and maintenance strategy. Causal reasoning technology builds a full-cycle behavior pattern knowledge base to reveal the causal relationship and behavior pattern of equipment state changes. The knowledge graph and behavior pattern knowledge base can provide intelligent decision support for equipment management, fault diagnosis, predictive maintenance, etc. Based on the full-cycle behavior pattern knowledge base, multi-source heterogeneous data are classified and stored through a distributed storage system to improve data storage and access efficiency. For multi-source heterogeneous data that has passed verification, hierarchical management is performed through multi-level intelligent storage according to the access popularity, and frequently accessed hot data is stored on high-speed storage media to improve data access performance. In summary, the present invention can effectively improve the level of intelligent management of railway signal equipment data, improve equipment reliability and safety, and has important application value. BRIEF DESCRIPTION OF THE DRAWINGS
[0053] Figure 1 A schematic diagram of a flow chart of a railway signal infrastructure equipment data management method according to an embodiment of the present invention;
[0054] Figure 2 The present invention is a schematic diagram of the structure of a railway signal infrastructure data management system according to an embodiment of the present invention. DETAILED DESCRIPTION
[0055] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0056] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.
[0057] Figure 1 FIG. 1 is a flow chart of a method for managing data of railway signal infrastructure equipment according to an embodiment of the present invention. Figure 1 As shown, the method includes:
[0058] S1. Collect railway signal equipment operation parameters, environmental information and status information through a pre-set multimodal sensor array and form multi-source heterogeneous data, adaptively compress and extract features of the multi-source heterogeneous data through a data compression algorithm and a feature extraction algorithm to obtain an equipment operation feature vector, identify and repair the abnormal points in the equipment operation feature vector through an abnormal point detection algorithm to obtain a high-quality data set, match the network bandwidth with the element form in the high-quality data set through a self-organizing data transmission network, and transmit the high-quality data set to the control center as unstructured data;
[0059] The multimodal sensor array is a system that integrates multiple sensors and can simultaneously collect multiple types of data, such as vision, audio, temperature, pressure, etc. The data compression algorithm is used to reduce the size of data storage and transmission, thereby saving storage space and bandwidth. The feature extraction algorithm is used to extract representative information from the original data and simplify the complexity of data processing. It is often used in pattern recognition and machine learning. The adaptive compression is a technology that dynamically adjusts the compression strategy according to data characteristics, aiming to optimize compression efficiency and compression quality. The outlier detection algorithm is used to identify data points in a data set that are significantly different from the normal pattern. It is often used in fault detection, financial fraud detection and other fields. The self-organizing data transmission network is a network that does not require a fixed infrastructure. Nodes can automatically configure and manage network connections to achieve efficient data transmission.
[0060] In an optional embodiment,
[0061] The operation parameters, environmental information and status information of railway signal equipment are collected through a pre-set multimodal sensor array to form multi-source heterogeneous data, the multi-source heterogeneous data are adaptively compressed and feature extracted through a data compression algorithm and a feature extraction algorithm to obtain an equipment operation feature vector, the abnormal points in the equipment operation feature vector are identified and repaired through an abnormal point detection algorithm to obtain a high-quality data set, the network bandwidth is matched with the element form in the high-quality data set through a self-organizing data transmission network, and the high-quality data set is transmitted to a control center as unstructured data including:
[0062] Based on a pre-set multimodal sensor array, operating parameters, environmental information and status information of railway signal equipment are collected, wherein the multimodal sensor array determines the sampling rate of different sensors according to the pre-set importance of each device, performs data collection according to the sampling rate, obtains the operating parameters, environmental information and status information corresponding to each device, and combines them to obtain the multi-source heterogeneous data;
[0063] For the multi-source heterogeneous data, according to the edge computing nodes in the multimodal sensor array, the data distribution characteristics of the multi-source heterogeneous data are determined by an adaptive data compression algorithm, and the spatial correlation of the data corresponding to the coding dictionary and the railway signal equipment is combined to perform collaborative compression through distributed wavelet transform to generate initial compressed data, and the key features in the initial compressed data are determined by empirical mode decomposition and principal component analysis algorithms to obtain the equipment operation feature vector;
[0064] Initialize a pre-set outlier detection model based on an isolation forest and a data repair rule library based on the equipment operation feature vector, divide the equipment operation feature vector into multiple data blocks of the same size according to the model input requirements, and for each data block, vote and generate a voting result through the judgment module in the outlier detection model, and determine the detection result in combination with the spatiotemporal correlation of the data corresponding to the railway signal equipment and the outlier detection mechanism. If an outlier is detected, perform data verification on the current data block and locate the real outlier. For the real outlier, perform data repair according to the data repair rule library and random walk to obtain the high-quality data set;
[0065] The high-quality data set is sent to the control center through a self-organizing data transmission network, wherein the self-organizing data transmission network includes a network convergence layer, a network access layer and a network optimization layer. The network convergence layer builds a multi-channel heterogeneous fusion network based on wireless communication technology and sends the received high-quality data set to the network access layer. The network access layer accesses data through an adaptive multiple access mechanism according to the structure of the elements in the high-quality data set, and at the same time assigns attention weights to the elements in the high-quality data set based on the attention mechanism, assigns channels to each element according to the attention weight and sends it to the control center. Before sending, the network optimization layer determines the node energy consumption based on the network traffic prediction algorithm and self-optimizes the allocation of channel resources through cognitive radio technology and spectrum sensing to obtain the unstructured data.
[0066] The sampling rate refers to the number of times samples are extracted from a continuous signal per second, usually in Hertz (Hz). The edge computing node is a device or system that processes data close to the data source. The data distribution characteristics describe the distribution of data in the feature space, such as statistical properties such as mean, variance, skewness and kurtosis. The distributed wavelet transform is a technology used for signal processing. It decomposes the signal at multiple scales through a distributed computing architecture to extract different frequency components. The principal component analysis is a dimensionality reduction technology that projects high-dimensional data into a low-dimensional space through linear transformation and retains the component with the largest variance in the data. The random walk is a mathematical model that describes the process of random movement of an object in space. The multi-channel heterogeneous fusion network is a network architecture that fuses data from different types of sensors. The adaptive multiple access mechanism is a technology that dynamically adjusts resource allocation and access strategies to optimize network performance and user experience. The network traffic prediction algorithm is used to predict future network traffic trends and help optimize network resource management and load balancing. The spectrum sensing is a technology for detecting and identifying available frequency bands in the radio spectrum, which is used for dynamic spectrum access and cognitive radio networks.
[0067] A multimodal sensor array is pre-set to collect operating parameters, environmental information and status information of railway signal equipment. The sampling rates of different sensors are determined according to the importance of each device. The operating parameters, environmental information and status information of each device are collected using the sensor array according to the set sampling rate. The collected data are combined to form multi-source heterogeneous data. The edge computing nodes in the multimodal sensor array are used to determine the data distribution characteristics of the multi-source heterogeneous data through an adaptive data compression algorithm. The data is collaboratively compressed through distributed wavelet transform in combination with the coding dictionary and the spatial correlation of the railway signal equipment data to generate initial compressed data. The key features are extracted from the initial compressed data through empirical mode decomposition and principal component analysis algorithms to obtain the equipment operation feature vector.
[0068] Use the equipment operation feature vector to initialize the pre-set outlier detection model based on isolation forest and the data repair rule library, divide the equipment operation feature vector into multiple data blocks of the same size as the input of the model, and vote for each data block through the judgment module in the outlier detection model to generate a voting result. Combine the spatiotemporal correlation of railway signal equipment data and the outlier detection mechanism to determine the final outlier detection result. If an outlier is detected, perform data verification on the current data block and locate the real outlier. For the determined real outlier, perform data repair according to the data repair rule library combined with the random walk algorithm, and combine the repaired data with the original normal data to form a high-quality data set.
[0069] High-quality data sets are sent to the control center through a self-organizing data transmission network. The self-organizing data transmission network includes a network convergence layer, a network access layer, and a network optimization layer. The network convergence layer builds a multi-channel heterogeneous fusion network based on wireless communication technology, receives high-quality data sets and sends them to the network access layer. The network access layer accesses data through an adaptive multiple access mechanism based on the structure of the elements in the high-quality data set. At the same time, the network access layer assigns attention weights to the elements in the high-quality data set based on the attention mechanism, assigns channels to each element according to the attention weight, and sends the data to the control center. Before sending, the network optimization layer determines the node energy consumption based on the network traffic prediction algorithm, and self-optimizes the allocation of channel resources through cognitive radio technology and spectrum sensing. The optimized data transmission network sends the high-quality data set to the control center to form unstructured data.
[0070] In this embodiment, by using a preset multimodal sensor array and adjusting the sampling rate according to the importance of the equipment, the data of key equipment can be collected in a targeted manner, the collection of redundant data can be reduced, and the efficiency of data collection can be improved. By using an adaptive data compression algorithm and a distributed wavelet transform, combined with the spatial correlation of the equipment data, the data can be collaboratively compressed, which can significantly reduce the data transmission and storage overhead, and reduce the bandwidth and storage requirements of the system. By using an anomaly detection model and a data repair rule library, the equipment operation feature vector is detected and repaired. By combining the spatiotemporal correlation and the random walk algorithm, the anomalies in the data can be effectively identified and repaired, and the quality and reliability of the data can be improved. By using a self-organizing data transmission network, the transmission of data in the network is optimized through multi-channel heterogeneous fusion, adaptive multiple access, attention mechanism and other technologies. By using network traffic prediction, cognitive radio and spectrum sensing and other technologies, channel resources are self-optimized and allocated to improve the transmission efficiency and reliability of the network. In summary, this embodiment realizes the intelligent and efficient processing of railway signal equipment data, which is of great significance to improving the safety, reliability and operation efficiency of the railway system.
[0071] In an optional embodiment,
[0072] For the real outliers, data repair is performed according to the data repair rule base in combination with random walks as shown in the following formula:
[0073]
[0074] in, represents the eigenvector of abnormal node a after the t+1th iteration, μ represents the random walk iteration coefficient, SI(f j , f a ) represents the similarity measurement function of the feature vectors between abnormal nodes a and j, ∑ krepresents the sum of normal nodes in the neighborhood, SI(f j , f k ) represents the similarity measurement function between node j and the normal node k in the neighborhood, represents the feature vector of abnormal node a after the tth iteration, Represents the initial feature vector of abnormal node a.
[0075] In this embodiment, by considering the feature similarity between the abnormal node and its normal neighboring nodes and using the information in the local topological structure for data repair, the normal data information around the abnormal point can be better utilized to improve the accuracy of the repair. The random walk mechanism can improve the robustness of the repair process, reduce the impact of local noise and abnormal points, and make the repair result more stable and reliable. The weighted method based on similarity can adaptively adjust the contribution of different nodes to the repair value, and improve the adaptability and accuracy of the repair. In summary, this embodiment effectively repairs the real abnormal points in the railway signal equipment data, has high accuracy, robustness and adaptability, can significantly improve the data quality, and provide a reliable data basis for the intelligent operation and maintenance and decision-making optimization of the railway system.
[0076] S2. Based on the unstructured data, automatic semantic understanding is performed through a deep learning algorithm to generate a semantic feature library, key information corresponding to railway signal equipment is extracted, feature data enhancement is performed on the equipment operation feature vector through a small sample learning algorithm, and the fault precursor corresponding to the current equipment is determined through a spatiotemporal correlation representation learning algorithm, and a health status assessment report is generated. A railway signal knowledge graph is constructed based on the semantic feature library and the health status assessment report, and a full-cycle behavior pattern knowledge base is constructed through causal reasoning technology;
[0077] The small sample learning algorithm aims to learn an effective model from a small number of training samples, so that accurate prediction and classification can be performed even when the samples are limited. The spatiotemporal correlation representation learning algorithm is used to capture the spatial and temporal correlations in the data and is widely used in video analysis, traffic forecasting and other fields. The fault precursor refers to identifying potential signs of faults by monitoring and analyzing relevant data before a system failure occurs. The railway signal knowledge graph is a structured representation of railway signal-related knowledge and data to form a knowledge network to support intelligent management and analysis of railway signal systems. The causal reasoning technology is used to infer causal relationships between variables from data, rather than just correlation relationships. The full-cycle behavior pattern knowledge base is used to record and analyze the behavior patterns of objects throughout their life cycle.
[0078] In an optional embodiment,
[0079] Based on the unstructured data, automatic semantic understanding is performed through a deep learning algorithm to generate a semantic feature library, key information corresponding to railway signal equipment is extracted, feature data enhancement is performed on the equipment operation feature vector through a small sample learning algorithm, and the fault precursor corresponding to the current equipment is determined through a spatiotemporal correlation representation learning algorithm, and a health status assessment report is generated. A railway signal knowledge graph is constructed based on the semantic feature library and the health status assessment report, and a full-cycle behavior pattern knowledge base is constructed through causal reasoning technology, including:
[0080] Based on the unstructured data, format conversion is performed through a deep learning algorithm and the text is converted into a word-unit identity sequence in a vocabulary through a word segmentation operation, the word-unit identity sequence is added to a pre-trained deep bidirectional processing model, and a semantic feature vector corresponding to each word-unit is output through an encoder;
[0081] Based on the semantic feature vector, a semantic understanding model is constructed and an ontology in the field of railway signal equipment is constructed, equipment type, component name, status description and failure mode are formally defined as key concepts, hierarchical relationships between the key concepts are defined and concept nodes in the ontology are embedded into the bidirectional processing feature space corresponding to the deep bidirectional processing model, the semantic association between the ontology and the word is determined in combination with the attention mechanism, and the key information of the railway signal equipment is automatically extracted in combination with the conditional random field sequence labeling model to obtain the key information;
[0082] Based on the key information, a small sample feature learning is performed on the equipment operation feature vector through a model-independent meta-learning algorithm, the equipment operation feature vector is divided into a meta-training set and a meta-testing set, two-level optimization is performed through the meta-training set and an initial feature extractor is determined, the initial feature extractor is optimized based on the meta-testing set, and feature data enhancement is performed in combination with a generative adversarial network, and for the enhanced equipment operation feature vector, spatiotemporal correlation modeling is performed through a spatiotemporal attention network, and the time dependency of railway signal equipment and the spatial correlation characteristics between adjacent equipment in the network topology are determined respectively, and the fault precursor is determined based on the time dependency and the spatial correlation characteristics, and a health status assessment report is generated;
[0083] A railway signal knowledge graph is constructed based on the semantic feature library and the health status assessment report. Entity connection and attribute matching are performed through a rule-based entity linking method, and pre-acquired unstructured data is linked to the railway signal knowledge graph. In combination with causal reasoning technology, entity relationships and logical associations are modeled through vector operations, and missing values of the graph are predicted and expanded to obtain the full-cycle behavior pattern knowledge base.
[0084] The word-gram is the basic unit in natural language processing, usually a word, subword or character in a text. The word-gram identity sequence is a list of word-grams after the text is lemmatized. The deep bidirectional processing model can simultaneously use the previous and next context information of the sequence for processing. The bidirectional processing feature space refers to the feature space generated by the model using the bidirectional context information when extracting features. The conditional random field sequence labeling model is a probabilistic graph model for sequence labeling, which is used to predict tags in a sequence. The model-independent meta-learning algorithm is a meta-learning algorithm that aims to quickly adapt to new tasks through training data of a small amount of tasks. The generative adversarial network is a generative model composed of a generator and a discriminator. The generator is responsible for generating realistic data samples, while the discriminator is used to distinguish between generated data and real data. Through adversarial training, the generator continuously improves its generation ability and the discriminator continuously improves its discrimination ability, thereby achieving high-quality data generation. The network topology describes the network nodes and their connection methods. The rule-based entity linking method matches and associates entities in the text with entities in the knowledge base through predefined rules.
[0085] Preprocess the unstructured data and convert it into a format suitable for subsequent processing through a deep learning algorithm. Perform word segmentation on the converted text data and convert the text into a sequence of word identities in the vocabulary. Add the obtained word identity sequence to the pre-trained deep bidirectional processing model and output the semantic feature vector corresponding to each word through the encoder as the basis for subsequent semantic understanding.
[0086] Based on the obtained semantic feature vectors, a semantic understanding model is constructed to understand the semantic information in the field of railway signal equipment, and an ontology in the field of railway signal equipment is constructed. Key concepts such as equipment type, component name, status description and failure mode are formally defined. The hierarchical relationship between key concepts in the ontology is defined to represent the hierarchical and associative relationships between concepts. The concept nodes in the ontology are embedded into the bidirectional processing feature space corresponding to the deep bidirectional processing model, and a mapping between concepts and semantic features is established. Combined with the attention mechanism, the semantic association between ontology concepts and word units is determined to achieve semantic level association and understanding. The constructed semantic understanding model and ontology knowledge are combined with the conditional random field sequence labeling model to automatically extract key information from the unstructured data of railway signal equipment. Through the sequence labeling model, key information such as equipment type, component name, status description, failure mode, etc. is identified and extracted, and the extracted key information is structured to obtain a key information set of railway signal equipment.
[0087] Based on the key information obtained, small sample feature learning is performed on the device operation feature vector to improve the effectiveness of feature representation. The device operation feature vector is divided into a meta-training set and a meta-test set through a model-independent meta-learning algorithm. The meta-training set is used for two-level optimization to determine the parameters and hyperparameters of the initial feature extractor. The initial feature extractor is optimized based on the meta-test set to improve the generalization ability of feature extraction. Combined with the generative adversarial network, the optimized feature vector is data enhanced to generate more synthetic feature samples.
[0088] For the enhanced equipment operation feature vector, the spatiotemporal correlation modeling is performed through the spatiotemporal attention network. The time dependency of railway signal equipment is determined by the time attention mechanism, and the temporal evolution law of the equipment operation status is captured. The spatial attention mechanism is used to determine the spatial correlation characteristics between adjacent devices in the network topology and capture the interaction between devices. Based on the temporal dependency and spatial correlation characteristics obtained by modeling, the operation trend and abnormal conditions of the equipment are analyzed, potential fault precursors are determined, and a health status assessment report of the railway signal equipment is generated to reflect the current status and future trends of the equipment.
[0089] Based on the obtained semantic feature library and health status assessment report, a railway signal knowledge graph is constructed. Through the rule-based entity linking method, the entities in the unstructured data are linked and the attributes are matched with the nodes in the knowledge graph. Using causal reasoning technology, the relationship and logical association between entities are modeled through vector operations. Based on the reasoning results, the missing values in the knowledge graph are predicted, and the graph is expanded and improved. Based on the constructed railway signal knowledge graph, a full-cycle behavior pattern knowledge base is generated. The knowledge base contains key information such as the full-cycle operation behavior pattern, fault signs, and health status of the equipment, which is used to support application scenarios such as fault diagnosis, predictive maintenance, and intelligent decision-making of railway signal equipment.
[0090] In this embodiment, through deep learning algorithms and lemmatization technology, unstructured data is converted into semantic feature vectors, and a semantic understanding model is constructed. Combined with the ontological knowledge in the field of railway signal equipment, semantic-level understanding of equipment data is achieved, laying the foundation for subsequent key information extraction and knowledge graph construction. The semantic understanding model and conditional random field sequence labeling model are used to automatically extract key information from unstructured data of railway signal equipment, greatly improving the utilization efficiency of data and reducing the workload of manual processing. Through the model-independent meta-learning algorithm and generative adversarial network, small sample feature learning and data enhancement are performed on the equipment operation feature vector. It can capture subtle changes in the operating status of equipment and improve the accuracy of fault prediction. Through the spatiotemporal attention network, the time dependency and spatial correlation characteristics of railway signal equipment are modeled, which fully considers the temporal evolution law of the equipment operating status and the interaction between equipment, and realizes a comprehensive analysis of the equipment status, which helps to timely discover potential fault signs. The knowledge graph represents and associates key information such as the equipment's operating status, fault signs, and maintenance strategies in a structured manner, providing rich knowledge support for intelligent decision-making. In summary, this embodiment realizes the deep mining and utilization of unstructured data of railway signal equipment, and provides key technical support for intelligent operation and maintenance.
[0091] In an optional embodiment,
[0092] The objective function corresponding to the conditional random field sequence labeling model is shown in the following formula:
[0093]
[0094] Where X represents the input sequence, Y represents the predicted label sequence, Z(X) represents the partition function, and t h (y i-1 ,y i , X,i) is the transfer feature function, indicating the label y i-1 to label y i The transfer characteristics, λ h represents the transfer characteristic function t h The weight, μ k Represents the state characteristic function s k The weight of k (y i , X, i) is the state feature function, indicating that the label y at position i i and the features between the observation sequence X.
[0095] In this embodiment, the conditional random field model characterizes the transfer dependency between adjacent labels in the label sequence by introducing a transfer feature function, which can capture the contextual dependency and sequential relationship of the key information of railway signal equipment in the sequence, making the label prediction more accurate. The conditional random field model fuses the observation sequence with the features of the label through the state feature function, which can capture the intrinsic connection between the key information of railway signal equipment and the observation data, so that the model can make full use of the observation information for label prediction. By defining a global feature function, the transfer features and the state features are weighted and combined, and normalized through the distribution function, the predicted label sequence can have the maximum probability in the global scope, thereby improving the accuracy of key information extraction. In summary, this embodiment can effectively handle long-distance dependencies, reduce label bias, and integrate domain knowledge, thereby significantly improving the accuracy and generalization ability of key information extraction, laying an important data foundation for subsequent knowledge graph construction and equipment status analysis, and helping to improve the level of intelligent operation and maintenance of the railway system.
[0096] S3. Based on the full-cycle behavior pattern knowledge base, the multi-source heterogeneous data are classified and stored through a distributed storage system, and the credibility is verified in combination with blockchain technology. For the multi-source heterogeneous data that have passed the verification, according to the access popularity, the multi-source heterogeneous data are hierarchically managed through multi-level intelligent storage and data sharing and fusion are performed through multi-party secure computing technology. For the initial fused data after fusion, the privacy score is determined through a pre-set privacy risk assessment model and the initial fused data is screened according to the privacy score to obtain high-quality fused data and encrypted in combination with blockchain technology.
[0097] The distributed storage system is a storage system that stores data in multiple physical locations in a dispersed manner, aiming to improve the availability, reliability and access speed of data. The credibility verification is a method of evaluating whether data or a system is reliable and trustworthy. The access popularity refers to the frequency with which data is accessed. Data with high access popularity is usually given priority in storage and caching strategies. The multi-level intelligent storage achieves the optimal configuration of storage resources by storing data on storage media with different performance and cost. The data sharing and fusion is to integrate data from different sources to achieve unified access and utilization of data. The privacy risk assessment model is used to assess the risk of data leakage or abuse.
[0098] In an optional embodiment,
[0099] Based on the full-cycle behavior pattern knowledge base, the multi-source heterogeneous data is classified and stored through a distributed storage system, and the credibility is verified in combination with blockchain technology. For the multi-source heterogeneous data that has passed the verification, according to the access popularity, the multi-source heterogeneous data is hierarchically managed through multi-level intelligent storage and data sharing and fusion are performed through multi-party secure computing technology. For the initial fused data after fusion, the privacy score is determined through a pre-set privacy risk assessment model and the initial fused data is screened according to the privacy score to obtain high-quality fused data and encrypt it in combination with blockchain technology, including:
[0100] Based on the full-cycle behavior pattern knowledge base, determine the data category corresponding to the multi-source heterogeneous data, and classify and save the multi-source heterogeneous data according to the data category through the distributed storage system;
[0101] Based on blockchain technology, the multi-source heterogeneous data is uploaded to the blockchain network according to the predefined format and specification. For each piece of multi-source heterogeneous data, a corresponding transaction is generated, and a consensus is reached between different nodes in combination with the consensus mechanism and packaged and sent to the block. For the multi-source heterogeneous data uploaded to the block, the credibility is verified through the verification rules pre-defined in the smart contract, and a credibility score is generated and compared with a preset credibility threshold. If the credibility score is greater than the credibility threshold, it is considered that the verification is passed, and the multi-source heterogeneous data corresponding to the current credibility score is retained, otherwise the multi-source heterogeneous data is deleted;
[0102] For the verified multi-source heterogeneous data, a heat score is calculated based on the user's data request records and access time statistics. The multi-source heterogeneous data is divided into three levels: hot data, warm data, and cold data according to the heat score. A multi-level storage architecture is designed to store the verified multi-source heterogeneous data in combination with the levels. According to the change in heat score, data migration and update are performed through a heat-aware data migration strategy. The storage ratio is jointly calculated through a multi-party secure computing protocol, and the multi-source heterogeneous data is divided into multiple shards according to the storage ratio through secret sharing technology for data encryption. The encrypted data is combined with homomorphic encryption technology for fusion calculation to obtain initial fusion data.
[0103] For the initial fused data, a privacy scoring model based on information entropy is used to calculate the privacy score of each piece of initial fused data in combination with the sensitivity and associated risk of the data. The privacy score is screened according to a preset privacy risk threshold, and the initial fused data with a privacy score lower than the privacy risk threshold is removed to obtain the high-quality fused data, which is then saved in combination with blockchain encryption technology.
[0104] The blockchain network is a decentralized distributed ledger technology that ensures the immutability and transparency of data through a consensus mechanism. The smart contract is a code that is automatically executed on the blockchain and is used to execute and verify contract terms. The verification rule is a standard used by the blockchain network to verify transactions and blocks. The credibility verification ensures the authenticity and reliability of data and transactions in the blockchain network. The heat score is used to evaluate the attention paid to data or transactions in the blockchain network. The multi-party secure computing protocol is a protocol that allows multiple parties to jointly calculate the value of a function without disclosing their respective inputs. The secret sharing technology divides the secret into several parts, each participant holds a part, and only a specific number of participants can recover the secret together. The sharding is a method to improve the scalability of the blockchain by dividing the blockchain into multiple sub-chains (shards) to process transactions in parallel. The privacy risk threshold is a standard for measuring the acceptable degree of data privacy risk. The blockchain encryption technology is used to ensure the security of data and transactions, including asymmetric encryption, hash functions and zero-knowledge proofs.
[0105] Based on the full-cycle behavior pattern knowledge base, the data categories of multi-source heterogeneous data are determined, and the distributed storage system is used to classify and save the multi-source heterogeneous data according to the data categories. Through the division and classified storage of data categories, orderly management and efficient access to data are achieved;
[0106] Upload multi-source heterogeneous data to the blockchain network in a predefined format and specification. For each piece of uploaded data, generate corresponding transaction information, reach consensus among different nodes through a consensus mechanism, package the transaction and send it to the block, and use the pre-defined verification rules in the smart contract to verify the credibility of the data uploaded to the block, generate a credibility score for each piece of data, and compare it with the preset credibility threshold. If the credibility score is greater than the threshold, the verification is considered to be passed and the data is retained. Otherwise, the data is deleted. The authenticity and credibility of the data are guaranteed through the decentralization and tamper-proof characteristics of the blockchain.
[0107] For verified multi-source heterogeneous data, the heat score is calculated based on the user's data request records and access time statistics. The data is divided into three levels: hot data, warm data, and cold data according to the heat score. A multi-level storage architecture is designed to store data according to the heat level of the data. As the data heat score changes, the data is migrated and updated through a heat-aware data migration strategy. The storage ratio of the data is jointly calculated through a multi-party secure computing protocol. The multi-source heterogeneous data is divided into multiple shards according to the storage ratio using secret sharing technology. The shard data is encrypted to protect the confidentiality of the data. The homomorphic encryption technology is used to perform fusion calculations on the encrypted data to obtain the initial fusion data.
[0108] For the initial fused data, privacy assessment is performed through a privacy scoring model based on information entropy. The privacy score of each piece of initial fused data is calculated based on the sensitivity and associated risk of the data. The privacy score is screened according to the preset privacy risk threshold, and data with a privacy score below the threshold is removed to obtain high-quality fused data. The privacy security of the fused data is ensured through privacy assessment and screening.
[0109] The selected high-quality fusion data is saved using blockchain encryption technology, uploaded to the blockchain network, and consensus is reached through a consensus mechanism and packaged for storage.
[0110] In this embodiment, through data classification and distributed storage based on a full-cycle behavior pattern knowledge base, orderly management and efficient access to multi-source heterogeneous data are achieved, blockchain technology is used to verify the uploaded multi-source heterogeneous data, and the authenticity and credibility of the data are ensured through the consensus mechanism and the verification rules of the smart contract, which provides a guarantee for the trusted storage of data. Through the multi-level storage architecture design and the data migration strategy based on heat perception, the optimal configuration of storage resources is achieved. Through the homomorphic encryption technology, fusion calculations are performed on encrypted data to achieve data privacy protection and secure fusion analysis. Through the privacy scoring model based on information entropy, the initial fused data is evaluated for privacy, and the privacy score of each data is calculated, which effectively balances the needs of data utilization and privacy protection. In summary, this embodiment realizes the secure and reliable storage and fusion management of multi-source heterogeneous data, lays a solid foundation for subsequent data applications, and helps to maximize the value of data.
[0111] In an optional embodiment,
[0112] The privacy score of each piece of initial fused data is calculated by combining the sensitivity of the data and the associated risk as shown in the following formula:
[0113]
[0114] Among them, PS represents the privacy score, w s represents the sensitivity weight, P m represents the probability of the data taking the value of the mth attribute, w r represents the associated risk weight, R represents the associated risk of the data, and w t represents the timeliness weight, T represents the timeliness of the data, and w d represents the data volume weight, and D represents the amount of data.
[0115] In this embodiment, by calculating the privacy score, the privacy risk of the data can be intuitively measured to provide a basis for subsequent privacy protection decisions. The model uses the concept of information entropy to measure the probability distribution of data values on different attributes, thereby evaluating the sensitivity of the data. By calculating the information entropy, the sensitivity of the data can be objectively evaluated. The two weight factors of timeliness and data volume are introduced, and the time sensitivity of the data and the impact of the data scale on privacy are considered, making the privacy assessment more comprehensive and dynamic. In summary, this embodiment can adaptively evaluate the privacy risk of the data, provide a basis for privacy protection decisions, and balance the needs of data utilization and privacy protection.
[0116] Figure 2 FIG. 1 is a schematic diagram of a railway signal infrastructure data management system according to an embodiment of the present invention. Figure 2 As shown, the system comprises:
[0117] The first unit is used to collect railway signal equipment operation parameters, environmental information and status information through a pre-set multimodal sensor array and form multi-source heterogeneous data, adaptively compress and extract features of the multi-source heterogeneous data through a data compression algorithm and a feature extraction algorithm to obtain an equipment operation feature vector, identify and repair abnormal points in the equipment operation feature vector through an abnormal point detection algorithm to obtain a high-quality data set, match the network bandwidth with the element form in the high-quality data set through a self-organizing data transmission network, and transmit the high-quality data set to a control center as unstructured data;
[0118] The second unit is used to perform automatic semantic understanding and generate a semantic feature library based on the unstructured data through a deep learning algorithm, extract key information corresponding to railway signal equipment, perform feature data enhancement on the equipment operation feature vector through a small sample learning algorithm, and determine the fault omen corresponding to the current equipment through a spatiotemporal correlation representation learning algorithm, generate a health status assessment report, build a railway signal knowledge graph based on the semantic feature library and the health status assessment report, and build a full-cycle behavior pattern knowledge base through causal reasoning technology;
[0119] The third unit is used to classify and store the multi-source heterogeneous data through a distributed storage system based on the full-cycle behavior pattern knowledge base, and to verify the credibility in combination with blockchain technology. For the multi-source heterogeneous data that have passed the verification, according to the access popularity, the multi-source heterogeneous data are hierarchically managed through multi-level intelligent storage and data sharing and fusion are performed through multi-party secure computing technology. For the initial fused data after fusion, the privacy score is determined through a pre-set privacy risk assessment model and the initial fused data is screened according to the privacy score to obtain high-quality fused data and encrypt it in combination with blockchain technology.
[0120] According to a third aspect of the embodiments of the present invention,
[0121] An electronic device is provided, comprising:
[0122] processor;
[0123] a memory for storing processor-executable instructions;
[0124] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.
[0125] According to a fourth aspect of the embodiments of the present invention,
[0126] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.
[0127] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.
[0128] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for managing data of railway signal infrastructure equipment, It is characterized in that include: The operation parameters, environmental information and status information of railway signal equipment are collected through a pre-set multimodal sensor array to form multi-source heterogeneous data, the multi-source heterogeneous data are adaptively compressed and feature extracted through a data compression algorithm and a feature extraction algorithm to obtain an equipment operation feature vector, the abnormal points in the equipment operation feature vector are identified and repaired through an abnormal point detection algorithm to obtain a high-quality data set, the network bandwidth is matched with the element form in the high-quality data set through a self-organizing data transmission network, and the high-quality data set is transmitted to a control center as unstructured data; Based on the unstructured data, automatic semantic understanding is performed through a deep learning algorithm to generate a semantic feature library, key information corresponding to railway signal equipment is extracted, feature data enhancement is performed on the equipment operation feature vector through a small sample learning algorithm, and the fault precursor corresponding to the current equipment is determined through a spatiotemporal correlation representation learning algorithm, and a health status assessment report is generated. A railway signal knowledge graph is constructed based on the semantic feature library and the health status assessment report, and a full-cycle behavior pattern knowledge base is constructed through causal reasoning technology; Based on the full-cycle behavior pattern knowledge base, the multi-source heterogeneous data are classified and stored through a distributed storage system, and the credibility is verified in combination with blockchain technology. For the multi-source heterogeneous data that have passed the verification, according to the access popularity, the multi-source heterogeneous data are hierarchically managed through multi-level intelligent storage and data sharing and fusion are performed through multi-party secure computing technology. For the initial fused data after fusion, the privacy score is determined through a pre-set privacy risk assessment model and the initial fused data is screened according to the privacy score to obtain high-quality fused data and encrypted in combination with blockchain technology.
2. The method according to claim 1, It is characterized in that The operation parameters, environmental information and status information of railway signal equipment are collected through a pre-set multimodal sensor array to form multi-source heterogeneous data, the multi-source heterogeneous data are adaptively compressed and feature extracted through a data compression algorithm and a feature extraction algorithm to obtain an equipment operation feature vector, the abnormal points in the equipment operation feature vector are identified and repaired through an abnormal point detection algorithm to obtain a high-quality data set, the network bandwidth is matched with the element form in the high-quality data set through a self-organizing data transmission network, and the high-quality data set is transmitted to a control center as unstructured data including: Based on a pre-set multimodal sensor array, operating parameters, environmental information and status information of railway signal equipment are collected, wherein the multimodal sensor array determines the sampling rate of different sensors according to the pre-set importance of each device, performs data collection according to the sampling rate, obtains the operating parameters, environmental information and status information corresponding to each device, and combines them to obtain the multi-source heterogeneous data; For the multi-source heterogeneous data, according to the edge computing nodes in the multimodal sensor array, the data distribution characteristics of the multi-source heterogeneous data are determined by an adaptive data compression algorithm, and the spatial correlation of the data corresponding to the coding dictionary and the railway signal equipment is combined to perform collaborative compression through distributed wavelet transform to generate initial compressed data, and the key features in the initial compressed data are determined by empirical mode decomposition and principal component analysis algorithms to obtain the equipment operation feature vector; Initialize a pre-set outlier detection model based on an isolation forest and a data repair rule library based on the equipment operation feature vector, divide the equipment operation feature vector into multiple data blocks of the same size according to the model input requirements, and for each data block, vote and generate a voting result through the judgment module in the outlier detection model, and determine the detection result in combination with the spatiotemporal correlation of the data corresponding to the railway signal equipment and the outlier detection mechanism. If an outlier is detected, perform data verification on the current data block and locate the real outlier. For the real outlier, perform data repair according to the data repair rule library and random walk to obtain the high-quality data set; The high-quality data set is sent to the control center through a self-organizing data transmission network, wherein the self-organizing data transmission network includes a network convergence layer, a network access layer and a network optimization layer. The network convergence layer builds a multi-channel heterogeneous fusion network based on wireless communication technology and sends the received high-quality data set to the network access layer. The network access layer accesses data through an adaptive multiple access mechanism according to the structure of the elements in the high-quality data set, and at the same time assigns attention weights to the elements in the high-quality data set based on the attention mechanism, assigns channels to each element according to the attention weight and sends it to the control center. Before sending, the network optimization layer determines the node energy consumption based on the network traffic prediction algorithm and self-optimizes the allocation of channel resources through cognitive radio technology and spectrum sensing to obtain the unstructured data.
3. The method according to claim 2, It is characterized in that For the real outliers, data repair is performed according to the data repair rule base in combination with random walks as shown in the following formula: in, represents the eigenvector of abnormal node a after the t+1th iteration, μ represents the random walk iteration coefficient, SI(f j , f a ) represents the similarity measurement function of the feature vectors between abnormal nodes a and j, ∑ k represents the sum of normal nodes in the neighborhood, SI(f j , f k ) represents the similarity measurement function between node j and the normal node k in the neighborhood, represents the feature vector of abnormal node a after the tth iteration, Represents the initial feature vector of abnormal node a.
4. The method according to claim 1, It is characterized in that Based on the unstructured data, automatic semantic understanding is performed through a deep learning algorithm to generate a semantic feature library, key information corresponding to railway signal equipment is extracted, feature data enhancement is performed on the equipment operation feature vector through a small sample learning algorithm, and the fault precursor corresponding to the current equipment is determined through a spatiotemporal correlation representation learning algorithm, and a health status assessment report is generated. A railway signal knowledge graph is constructed based on the semantic feature library and the health status assessment report, and a full-cycle behavior pattern knowledge base is constructed through causal reasoning technology, including: Based on the unstructured data, format conversion is performed through a deep learning algorithm and the text is converted into a word-unit identity sequence in a vocabulary through a word segmentation operation, the word-unit identity sequence is added to a pre-trained deep bidirectional processing model, and a semantic feature vector corresponding to each word-unit is output through an encoder; Based on the semantic feature vector, a semantic understanding model is constructed and an ontology in the field of railway signal equipment is constructed, equipment type, component name, status description and failure mode are formally defined as key concepts, hierarchical relationships between the key concepts are defined and concept nodes in the ontology are embedded into the bidirectional processing feature space corresponding to the deep bidirectional processing model, the semantic association between the ontology and the word is determined in combination with the attention mechanism, and the key information of the railway signal equipment is automatically extracted in combination with the conditional random field sequence labeling model to obtain the key information; Based on the key information, a small sample feature learning is performed on the equipment operation feature vector through a model-independent meta-learning algorithm, the equipment operation feature vector is divided into a meta-training set and a meta-testing set, two-level optimization is performed through the meta-training set and an initial feature extractor is determined, the initial feature extractor is optimized based on the meta-testing set, and feature data enhancement is performed in combination with a generative adversarial network, and for the enhanced equipment operation feature vector, spatiotemporal correlation modeling is performed through a spatiotemporal attention network, and the time dependency of railway signal equipment and the spatial correlation characteristics between adjacent equipment in the network topology are determined respectively, and the fault precursor is determined based on the time dependency and the spatial correlation characteristics, and a health status assessment report is generated; A railway signal knowledge graph is constructed based on the semantic feature library and the health status assessment report. Entity connection and attribute matching are performed through a rule-based entity linking method, and pre-acquired unstructured data is linked to the railway signal knowledge graph. In combination with causal reasoning technology, entity relationships and logical associations are modeled through vector operations, and missing values of the graph are predicted and expanded to obtain the full-cycle behavior pattern knowledge base.
5. The method according to claim 4, It is characterized in that The objective function corresponding to the conditional random field sequence labeling model is shown in the following formula: Where X represents the input sequence, Y represents the predicted label sequence, Z(X) represents the partition function, and t h (y i-1 ,y i , X,i) is the transfer feature function, indicating the label y i-1 to label y i The transfer characteristics, λ h represents the transfer characteristic function t h The weight, μ k Represents the state characteristic function s k The weight of k (y i , X, i) is the state feature function, indicating that the label y at position i i and the features between the observation sequence X.
6. The method according to claim 1, It is characterized in that Based on the full-cycle behavior pattern knowledge base, the multi-source heterogeneous data is classified and stored through a distributed storage system, and the credibility is verified in combination with blockchain technology. For the multi-source heterogeneous data that has passed the verification, according to the access popularity, the multi-source heterogeneous data is hierarchically managed through multi-level intelligent storage and data sharing and fusion are performed through multi-party secure computing technology. For the initial fused data after fusion, the privacy score is determined through a pre-set privacy risk assessment model and the initial fused data is screened according to the privacy score to obtain high-quality fused data and encrypt it in combination with blockchain technology, including: Based on the full-cycle behavior pattern knowledge base, determine the data category corresponding to the multi-source heterogeneous data, and classify and save the multi-source heterogeneous data according to the data category through the distributed storage system; Based on blockchain technology, the multi-source heterogeneous data is uploaded to the blockchain network according to the predefined format and specification. For each piece of multi-source heterogeneous data, a corresponding transaction is generated, and a consensus is reached between different nodes in combination with the consensus mechanism and packaged and sent to the block. For the multi-source heterogeneous data uploaded to the block, the credibility is verified through the verification rules pre-defined in the smart contract, and a credibility score is generated and compared with a preset credibility threshold. If the credibility score is greater than the credibility threshold, it is considered that the verification is passed, and the multi-source heterogeneous data corresponding to the current credibility score is retained, otherwise the multi-source heterogeneous data is deleted; For the verified multi-source heterogeneous data, a heat score is calculated based on the user's data request records and access time statistics. The multi-source heterogeneous data is divided into three levels: hot data, warm data, and cold data according to the heat score. A multi-level storage architecture is designed to store the verified multi-source heterogeneous data in combination with the levels. According to the change in heat score, data migration and update are performed through a heat-aware data migration strategy. The storage ratio is jointly calculated through a multi-party secure computing protocol, and the multi-source heterogeneous data is divided into multiple shards according to the storage ratio through secret sharing technology for data encryption. The encrypted data is combined with homomorphic encryption technology for fusion calculation to obtain initial fusion data. For the initial fused data, a privacy scoring model based on information entropy is used to calculate the privacy score of each piece of initial fused data in combination with the sensitivity and associated risk of the data. The privacy score is screened according to a preset privacy risk threshold, and the initial fused data with a privacy score lower than the privacy risk threshold is removed to obtain the high-quality fused data, which is then saved in combination with blockchain encryption technology.
7. The method according to claim 6, It is characterized in that The privacy score of each piece of initial fused data is calculated by combining the sensitivity of the data and the associated risk as shown in the following formula: Among them, PS represents the privacy score, w s represents the sensitivity weight, P m represents the probability of the data taking the value of the mth attribute, w r represents the associated risk weight, R represents the associated risk of the data, and w t represents the timeliness weight, T represents the timeliness of the data, and w d represents the data volume weight, and D represents the amount of data.
8. A railway signal infrastructure data management system, used to implement the method described in any one of claims 1 to 7. It is characterized in that include: The first unit is used to collect railway signal equipment operation parameters, environmental information and status information through a pre-set multimodal sensor array and form multi-source heterogeneous data, adaptively compress and extract features of the multi-source heterogeneous data through a data compression algorithm and a feature extraction algorithm to obtain an equipment operation feature vector, identify and repair abnormal points in the equipment operation feature vector through an abnormal point detection algorithm to obtain a high-quality data set, match the network bandwidth with the element form in the high-quality data set through a self-organizing data transmission network, and transmit the high-quality data set to a control center as unstructured data; The second unit is used to perform automatic semantic understanding and generate a semantic feature library based on the unstructured data through a deep learning algorithm, extract key information corresponding to railway signal equipment, perform feature data enhancement on the equipment operation feature vector through a small sample learning algorithm, and determine the fault omen corresponding to the current equipment through a spatiotemporal correlation representation learning algorithm, generate a health status assessment report, build a railway signal knowledge graph based on the semantic feature library and the health status assessment report, and build a full-cycle behavior pattern knowledge base through causal reasoning technology; The third unit is used to classify and store the multi-source heterogeneous data through a distributed storage system based on the full-cycle behavior pattern knowledge base, and to verify the credibility in combination with blockchain technology. For the multi-source heterogeneous data that have passed the verification, according to the access popularity, the multi-source heterogeneous data are hierarchically managed through multi-level intelligent storage and data sharing and fusion are performed through multi-party secure computing technology. For the initial fused data after fusion, the privacy score is determined through a pre-set privacy risk assessment model and the initial fused data is screened according to the privacy score to obtain high-quality fused data and encrypt it in combination with blockchain technology.
9. An electronic device, It is characterized in that include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.
10. A computer-readable storage medium having computer program instructions stored thereon, It is characterized in that When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.
Citation Information
Cited By
Storage control method and device based on big data
CN119645322A
Generative adversarial network dual-data representation-based natural gas pipeline audio anomaly detection method
CN119993201A
Audio anomaly detection method based on dual data representation of generative adversarial network
CN119993201B
General data end-to-end reliable transmission method and system
CN120017228A
Subway operation and maintenance management method and system for intelligent rail transit and electronic equipment
CN120031416A
