A safety risk assessment system based on railway yard information

Abstract

The application discloses a safety risk assessment system based on railway station yard information, relates to the technical field of railway transportation safety, and solves the technical problem that the existing technology is time-consuming and low in efficiency in constructing a risk prediction model for a newly-built station yard and a low-risk station yard, and a small amount of training data is often used, so that the deployed safety risk assessment system is time-consuming and low in efficiency; a railway safety knowledge graph is constructed according to station yard data, and a risk prediction model is generated based on the railway safety knowledge graph; after a prediction risk event and a prediction risk level are generated according to the risk prediction model, an alarm signal is generated, the railway safety knowledge graph and the risk prediction model are fused to enhance the risk identification precision and realize early warning; a multi-source heterogeneous data time sequence coding technology is adopted to realize comprehensive analysis of unified cross-frequency data; a target station yard prediction model is trained by using multi-source station yard risk data, the problem of insufficient training data of the target station yard is solved, and the precision and application efficiency of the safety risk assessment system are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of railway transportation safety, and specifically relates to a safety risk assessment system based on railway station yard information. BACKGROUND

[0002] A railway station yard is a core component of a railway transportation system, usually referring to a comprehensive railway facility for train arrival and departure, marshalling, maintenance, cargo loading and unloading, and passenger boarding and alighting; according to the classification of use, it can be divided into passenger station yards, freight station yards, and mixed station yards; safety risk assessment is a systematic analysis method aiming to identify, quantify, and manage potential risks to prevent accidents and improve system safety. In the context of a railway station yard, safety risk assessment is particularly important because it involves complex factors such as train operation, equipment status, and personnel operation.

[0003] In existing railway station yards, data such as equipment status, operation process, and environmental monitoring within the station yard are scattered in different systems, making it difficult to achieve unified analysis of multi-source heterogeneous data, resulting in low efficiency of risk assessment for the entire station yard; at the same time, in the construction of station yard risk prediction models for new station yards and low-risk station yards, there is often a lack of training data, which leads to a long time required for deploying a safety risk assessment system and poor accuracy of the safety risk assessment system; therefore, the safety risk assessment system for railway station yard information still needs further improvement. SUMMARY

[0004] The present application aims to at least solve one of the technical problems existing in the prior art; for this purpose, the present application proposes a safety risk assessment system based on railway station yard information to solve the technical problem that the prior art often has a small amount of training data when constructing a risk prediction model for a new station yard and a low-risk station yard, resulting in a long time required for deploying a safety risk assessment system and low efficiency.

[0005] To achieve the above-mentioned purpose, the first aspect of the present application provides a safety risk assessment system based on railway station yard information, comprising a data acquisition module, a data analysis module, an early warning module, and a database.

[0006] The data acquisition module acquires station yard data through a data acquisition device; the station yard data includes station yard ID, station yard purpose, equipment data, environmental data, and operation data;

[0007] The data analysis module constructs a railway safety knowledge graph according to the station yard data; generates a risk prediction model according to the railway safety knowledge graph; generates a predicted risk event and a predicted risk level according to the risk prediction model; and generates an alarm signal according to the predicted risk level;

[0008] The early warning module makes a prompt according to the alarm signal and contacts the management personnel;

[0009] The database is used to store the data of each module and store the historical data required for training the model.

[0010] The present application constructs the railway safety knowledge graph and the risk prediction model through the above steps, not only improves the accuracy of risk identification, but also realizes the pre-warning and active prevention of risk; by uniformly coding different frequency data, the unified analysis of multi-source heterogeneous data is realized; by using the cross-site data enhancement strategy under the transfer learning framework, the risk transmission law between similar sites is mined, so that the newly built site can still construct a high-precision prediction model under the condition of insufficient training data, and the scene adaptability and engineering landing efficiency of the safety evaluation system are improved.

[0011] Further, the railway safety knowledge graph is constructed according to the station data, comprising:

[0012] Obtaining station data; the station data includes station ID, equipment data, environment data and operation data;

[0013] Generating standard data according to the equipment data, the environment data and the operation data; the standard data refers to standard data used for constructing the railway safety knowledge graph, and the standard data includes a plurality of entities, a plurality of attributes and a plurality of relationships;

[0014] Using a graph database to construct a safety knowledge graph, and embedding spatio-temporal data storage and rule driving in the safety knowledge graph; the spatio-temporal data storage is used to store the spatio-temporal data corresponding to the station data; the rule driving is constructed by experts according to the station data and the corresponding risk degree;

[0015] Inputting the standard data into the safety knowledge graph to obtain the railway safety knowledge graph.

[0016] Further, the standard data is generated according to the equipment data, the environment data and the operation data, comprising:

[0017] Obtaining equipment data, environment data and operation data; the equipment data includes equipment ID, equipment state and equipment parameters; the environment data includes meteorological data and geological data; the operation data includes operation ID, operation instruction, maintenance record and monitoring video;

[0018] Classifying the equipment data, the environment data and the operation data to obtain text data, view data, time series data and structured data; the structured data refers to data with fixed format and clear fields; the view data refers to video data and image data;

[0019] The text data, view data, time series data and structured data are standardized by a hierarchical fusion strategy.

[0020] Through the above steps, the device data, environment data and operation data in the station yard are unified, which provides strong data support for the risk assessment of the subsequent railway safety knowledge graph and the risk prediction model for future time risk prediction.

[0021] Further, the hierarchical fusion strategy is operated by a multi-modal fusion method, including the following steps:

[0022] Step 1: Obtain text data, view data, time series data and structured data;

[0023] Step 2: Extract a plurality of semantic vectors and spatial vectors from the text data, view data, time series data and structured data; the semantic vectors and spatial vectors include a plurality of entities, attributes and relationships;

[0024] Step 3: Add uniform time coding to the plurality of semantic vectors, and align different frequency data by sliding window aggregation to obtain aligned semantic vectors;

[0025] Step 4: Obtain standardized data by concatenating the aligned semantic vectors and spatial vectors.

[0026] Further, the risk prediction model is generated according to the railway safety knowledge graph, including:

[0027] Obtain a plurality of railway safety knowledge graphs corresponding to the station yard ID and the matching priority PYi, and a railway safety knowledge graph corresponding to the target station yard ID; the matching priority is calculated by a plurality of historical station yard data corresponding to the station yard ID; the target station yard ID refers to the railway station yard ID that needs to build a risk prediction model;

[0028] Sort a plurality of matching priorities from large to small to obtain a matching sequence PL;

[0029] Select the station yard ID and the railway safety knowledge graph corresponding to the first N matching priorities in PL, and name a plurality of the railway safety knowledge graphs as source railway safety knowledge graphs; wherein N is an integer, N>1;

[0030] Extract a plurality of common subgraph data from the target station yard ID corresponding to the railway safety knowledge graph and the source railway safety knowledge graph;

[0031] Merge a plurality of common subgraph data and station yard data corresponding to the target station yard ID to obtain target comprehensive data; the target comprehensive data includes a plurality of subgraph data;

[0032] The risk prediction model is obtained by training the image enhanced time sequence prediction model on the target comprehensive data.

[0033] Further, the matching priority is calculated by historical station data corresponding to the station IDs, including:

[0034] Obtain historical station data corresponding to the station IDs; the historical station data includes historical risk levels, historical risk times, and occurrence times;

[0035] Calculate comprehensive risk scores corresponding to the station IDs according to the historical risk levels, the historical risk times, and the occurrence times;

[0036] Obtain a plurality of railway safety knowledge graphs and a railway safety knowledge graph corresponding to the target station ID;

[0037] Input the plurality of railway safety knowledge graphs and the railway safety knowledge graph corresponding to the target station ID into a graph similarity estimation model to obtain a plurality of similarity levels XD i ; the graph similarity estimation model is constructed by an artificial intelligence model;

[0038] Calculate the matching priority PY i by the formula ; wherein i represents the number of the station IDs, g is a proportionality coefficient, g ∈ (0, π / 2); α1 and α2 are weight coefficients, α1 and α2 ∈ (0, 1); and the sum of α1 and α2 is 1; ZFP i represents the comprehensive risk score corresponding to the i-th station ID.

[0039] Further, the calculation of the comprehensive risk score corresponding to the station IDs according to the historical risk levels, the historical risk times, and the occurrence times includes:

[0040] Obtain the historical risk levels FD, the historical risk times, and the occurrence times Δt;

[0041] Calculate the comprehensive risk score ZFP i by the formula ; wherein j represents the number of the historical risk times, max() represents the maximum value operation, max(j) represents the historical risk occurrence times, β represents a time decay coefficient, β ∈ (0, 1); Δt i,j represents the time length from the i-th station ID j-th risk occurrence to the present time; DT represents a unit time length; the log() function is used to retain sensitivity to the total risk times, but suppress extreme values by the logarithmic function.

[0042] Further, the graph similarity estimation model is constructed by an artificial intelligence model, including:

[0043] Obtain several historical railway safety knowledge graphs and railway safety knowledge graphs corresponding to historical target station IDs and their corresponding historical similarity levels;

[0044] Divide the several historical railway safety knowledge graphs and railway safety knowledge graphs corresponding to historical target station IDs and their corresponding historical similarity levels into training data, validation data, and test data; and perform data preprocessing on the training data, validation data, and test data to obtain a training set, a validation set, and a test set;

[0045] Select an artificial intelligence model as a base model;

[0046] Train the base model through the training set, and adjust the learning rate and hyperparameters on the validation set to obtain a pre-trained model;

[0047] Verify the pre-trained model on the test set, and finally obtain a graph similarity prediction model with several railway safety knowledge graphs and railway safety knowledge graphs corresponding to target station IDs as input and several similarity levels as output.

[0048] Further, the risk prediction model obtained by training the graph-enhanced time series prediction model on the target comprehensive data comprises:

[0049] Obtain target comprehensive data; the target comprehensive data comprises several subgraph data;

[0050] Extract several historical subgraph data and their corresponding several historical time series data, historical risk levels, and risk events;

[0051] Integrate the several historical subgraph data and their corresponding several historical time series data into several historical prediction data;

[0052] Divide the several historical prediction data and their corresponding historical risk levels and risk events into training data, validation data, and test data; and perform data preprocessing on the training data, validation data, and test data to obtain a training set, a validation set, and a test set;

[0053] Select a graph-enhanced time series prediction model as a base model;

[0054] Train the base model through the training set, and adjust the learning rate and hyperparameters on the validation set to obtain a pre-trained model;

[0055] Verify the pre-trained model on the test set, and finally obtain a risk prediction model with prediction data as input and predicted risk levels and predicted risk events as output.

[0056] Further, the generation of the predicted risk events and the predicted risk levels according to the risk prediction model comprises:

[0057] obtain the time sequence data corresponding to the target station yard ID and the risk prediction model and the railway safety knowledge graph in real time; the railway safety knowledge graph comprises a plurality of subgraph data;

[0058] extract historical time sequence data and corresponding historical subgraph data, and integrate the current time sequence data and subgraph data to obtain prediction data;

[0059] input the prediction data into the risk prediction model to obtain a predicted risk level and a predicted risk event.

[0060] Further, the generation of the alarm signal according to the predicted risk level comprises:

[0061] obtain the predicted risk level and the risk level interval; the risk level interval comprises a low risk interval, a medium risk interval and a high risk interval; the risk level interval is set by experts according to historical risk levels and corresponding historical risk events;

[0062] when the predicted risk level belongs to the low risk interval, a low risk event warning signal is generated;

[0063] when the predicted risk level belongs to the medium risk interval, a medium risk event warning signal is generated;

[0064] when the predicted risk level belongs to the high risk interval, an emergency risk warning signal is generated.

[0065] Through the above steps, when the predicted risk level occurs, the relevant departments can be notified in time to make advance prevention arrangements, thereby reducing safety hazards and improving the safety of the safety risk assessment system.

[0066] Compared with the prior art, the application has the following beneficial effects:

[0067] 1. The application constructs a railway safety knowledge graph according to station yard data, generates a risk prediction model according to the railway safety knowledge graph, generates a predicted risk event and a predicted risk level according to the risk prediction model, and generates an alarm signal according to the predicted risk level. The construction of the railway safety knowledge graph and the risk prediction model not only improves the accuracy of the prediction of the coming risk, but also accurately predicts the risk and avoids it in advance. By uniformly coding different frequency data, the unified analysis of multi-source heterogeneous data is realized. The risk prediction model of the target station yard is trained by using the risk data of other source stations, so that the target station yard can obtain a risk prediction model with high accuracy even with a small amount of training data, thereby improving the application efficiency of the safety risk assessment system.

[0068] 2、The application obtains the matching priority of other railway stations and the target station, uses the data of several stations with higher matching priority and similar to the target station to train the risk prediction model of the target station, improves the deployment efficiency of the risk prediction model of the target station and the accuracy of model prediction in the case of less training data of the target station, and further improves the application efficiency of the safety risk assessment system.

[0069] 3、The application considers the historical risk level, historical risk times and occurrence time of the source station, calculates the comprehensive risk score of the station by using multi-source data, improves the representativeness of the comprehensive risk score, and provides data support for finding suitable source station data. BRIEF DESCRIPTION OF DRAWINGS

[0070] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0071] Figure 1 A safety risk assessment system based on railway station information according to the present application;

[0072] Figure 2 A safety risk assessment method based on railway station information according to the present application;

[0073] Figure 3 A hierarchical fusion strategy flowchart according to the present application. DETAILED DESCRIPTION

[0074] The technical solutions of the present application will be described in detail below with reference to the embodiments. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0075] Please refer to Figures 1-2 The first aspect of the present application provides a safety risk assessment system based on railway station information, which comprises a data acquisition module, a data analysis module, a warning module and a database.

[0076] Data acquisition module: acquire station data through data acquisition equipment; station data includes station ID, station purpose, equipment data, environment data and operation data; data acquisition equipment includes a plurality of sensors, etc.

[0077] The data analysis module: constructing a railway safety knowledge graph according to station data, the railway safety knowledge graph is a graph representation of the entire station information; generating a risk prediction model according to the railway safety knowledge graph, the risk prediction model is a model for predicting risks; generating a predicted risk event and a predicted risk level according to the risk prediction model; generating an alarm signal according to the predicted risk level;

[0078] The early warning module: making a prompt according to the alarm signal and contacting the management personnel; the alarm signal includes a low-risk event occurrence early warning signal and an emergency risk alarm signal, etc.

[0079] The database is used to store the data of each module and store the historical data required for training the model.

[0080] In the embodiment, the railway safety knowledge graph is constructed according to the station data, which includes:

[0081] Obtaining station data; the station data includes station ID, equipment data, environment data and operation data;

[0082] Generating standardized data according to the equipment data, the environment data and the operation data, the standardized data refers to standard data used for constructing the railway safety knowledge graph, and the standardized data includes a plurality of entities, a plurality of attributes and a plurality of relationships;

[0083] Using a graph database to construct the safety knowledge graph, and embedding spatiotemporal data storage and rule driving in the safety knowledge graph; the spatiotemporal data storage is used to store the spatiotemporal data corresponding to the station data; the rule driving is constructed by experts according to the station data and the corresponding risk degree; the graph database selects Neo4j; the spatiotemporal data storage selects PostGIS extension;

[0084] Inputting the standardized data into the safety knowledge graph to obtain the railway safety knowledge graph.

[0085] In the embodiment, the standardized data is generated according to the equipment data, the environment data and the operation data, which includes:

[0086] Obtaining equipment data, environment data and operation data; the equipment data includes equipment ID, equipment state and equipment parameters, etc.; the environment data includes meteorological data and geological data; the operation data includes operation ID, operation instruction, maintenance record and monitoring video, etc.

[0087] Classifying the equipment data, the environment data and the operation data to obtain text data, view data, time series data and structured data; the structured data refers to data with fixed format and clear fields; the view data refers to video data and image data.

[0088] The text data, view data, time series data and structured data are obtained by a hierarchical fusion strategy; the hierarchical fusion strategy is operated by a multi-modal fusion manner; since the station yard data has multiple forms and the collection frequency of each data is not the same, it is essential to unify the multi-modal data in order to obtain the risk assessment state of the entire station yard through a single system.

[0089] Referring to Figure 3 The hierarchical fusion strategy in the embodiment is operated by a multi-modal fusion manner, including the following steps:

[0090] Step one: obtaining text data, view data, time series data and structured data;

[0091] Step two: obtaining a plurality of semantic vectors and spatial vectors from the text data, view data, time series data and structured data by feature extraction; the semantic vectors and spatial vectors include a plurality of entities, attributes and relationships; a pre-trained language model such as a BERT model is used to extract semantic vectors from the text data; for view data, features are extracted by a ResNet network, and spatial positions are located by target detection and SLAM technology;

[0092] Step three: adding a unified time coding to the plurality of semantic vectors, the unified time coding includes Time2Vec, and the alignment of different frequency data is realized by a sliding window aggregation manner to obtain aligned semantic vectors; the sliding window aggregation manner can be set to fuse multi-modal data in every M-minute window, M can be set according to experience, which can be set to fuse multi-modal data in every 5-minute window, or can be set to fuse multi-modal data in every 10-minute window; when the semantic vectors are aligned at the same timestamp, each modal feature can be directly spliced by directly splicing each modal feature, or can be directly spliced by a cross-modal attention mechanism dynamically assigning weights to each modal feature;

[0093] Step four: obtaining standardized data by splicing the aligned semantic vectors and spatial vectors; when splicing the aligned semantic vectors and spatial vectors, a direct splicing manner is used for splicing.

[0094] The risk prediction model generated according to the railway safety knowledge graph in the embodiment includes:

[0095] obtaining a railway safety knowledge graph corresponding to a plurality of station yard IDs and a matching priority PY i and a railway safety knowledge graph corresponding to a target station yard ID; the matching priority is calculated by a plurality of station yard ID corresponding historical station yard data; the target station yard ID refers to a railway station yard ID for which a risk prediction model needs to be constructed;

[0096] The matching priorities are sorted in descending order to obtain a matching sequence PL;

[0097] The station yard IDs and the railway safety knowledge graphs corresponding to the first N matching priorities in PL are selected, and the first N railway safety knowledge graphs are named as source railway safety knowledge graphs; where N is an integer, N > 1, and the specific value is set according to experience, and N is set to 3 in this embodiment;

[0098] Extracting a plurality of common subgraph data from the railway safety knowledge graph corresponding to the target station yard ID and the source railway safety knowledge graphs; This step mainly searches for common subgraph data related to entities, attributes and relationships in the railway safety knowledge graph corresponding to the target ID in the source railway safety knowledge graph. Using these common subgraph data can help train the risk prediction model of the target station yard ID and provide a large amount of relevant training data;

[0099] Merging the plurality of common subgraph data and the station yard data corresponding to the target station yard ID to obtain target comprehensive data; the target comprehensive data includes a plurality of subgraph data, i.e., all data required for training the risk prediction model for the target station yard;

[0100] Training the graph-enhanced time series prediction model on the target comprehensive data to obtain the risk prediction model.

[0101] The embodiment selects a plurality of station yards with high matching priorities from other railway station yards and the target station yard by obtaining the matching priorities of the other railway station yards and the target station yard, and then applies the data similar to the target station yard in these station yards to train the risk prediction model of the target station yard. When the training data of the target station yard is limited, this method can significantly improve the deployment efficiency of the risk prediction model of the target station yard and greatly improve the prediction accuracy of the model, thereby effectively enhancing the application efficiency of the safety risk assessment system.

[0102] The matching priorities in this embodiment are calculated based on the historical station yard data corresponding to a plurality of station yard IDs, including:

[0103] Obtaining historical station yard data corresponding to a plurality of station yard IDs; the historical station yard data includes historical risk levels, historical risk frequencies and occurrence times;

[0104] Calculating the comprehensive risk scores of the plurality of station yard IDs based on the historical risk levels, the historical risk frequencies and the occurrence times;

[0105] Obtaining a plurality of railway safety knowledge graphs and the railway safety knowledge graph corresponding to the target station yard ID;

[0106] Inputting the plurality of railway safety knowledge graphs and the railway safety knowledge graph corresponding to the target station yard ID into a graph similarity estimation model to obtain a plurality of similarity levels XD iThe atlas similarity prediction model is constructed by an artificial intelligence model.

[0107] The matching priority PY is calculated by the formula i ; wherein, i represents the number of station ID, g is a proportional coefficient, g ∈ (0, π / 2), and g is set to make the matching priority PY i ∈ (0, 1), and the specific value is set according to experience; α1 and α2 are weight coefficients, α1 and α2 ∈ (0, 1), and the sum of α1 and α2 is 1, and the specific value is set according to experience, when considering the importance of the similarity level is higher than the comprehensive risk score in calculating the matching priority, α1 < α2, when considering the importance of the comprehensive risk score is higher than the similarity level, α1 > α2, and in the embodiment, the weight coefficient is set to α1 < α2; ZFP i represents the comprehensive risk score corresponding to the i-th station ID; as the comprehensive risk score and the similarity level corresponding to the station increase, the matching priority corresponding thereto also increases.

[0108] In the embodiment, by considering other stations from two dimensions of the comprehensive risk score and the similarity level, the station with high comprehensive risk score and high similarity level is selected as the source station of the training data, more data meeting the demand of the target station and being helpful to the training of the risk prediction model can be obtained, which not only improves the accuracy of the prediction result of the target station risk prediction model, but also optimizes the deployment effect of the model, so that the risk prediction model can more efficiently and accurately serve the actual scene.

[0109] In the embodiment, the comprehensive risk score corresponding to the station ID is calculated according to the historical risk level, the historical risk number and the occurrence time, which includes:

[0110] The historical risk level FD, the historical risk number and the occurrence time Δt are obtained.

[0111] The comprehensive risk score ZFP is calculated by the formula i ; wherein, j represents the number of historical risk, max() represents the maximum value operation, max(j) represents the number of historical risk, β represents the time attenuation coefficient, β ∈ (0, 1), and the specific value is set according to experience, and in the embodiment, β is set to 0.2, and the time attenuation coefficient is set to consider that the historical risk event occurring a long time ago has less influence on the comprehensive risk score; Δt i,j ​represents the time length from the i-th station yard ID j-th risk occurrence to the present time; DT represents a unit time length, and the specific value is set according to experience, and in the embodiment, DT is set to 10 days; the log() function is used to retain the sensitivity to the total number of risks, but the extreme value case is suppressed by the logarithmic function; the shorter the risk event occurrence time, and the higher the risk occurrence times and risk levels, the greater the corresponding comprehensive risk score.

[0112] In the embodiment, when calculating the comprehensive risk score of the station yard, the historical risk levels, the historical risk times and the occurrence time and other multi-source data of the station yard are fully taken into account, so that the representativeness of the comprehensive risk score is improved, the root of the source station yard data meeting the needs is laid, and the subsequent search for suitable source station yard data is effectively supported.

[0113] The graph similarity estimation model in the embodiment is constructed by an artificial intelligence model, including:

[0114] Obtain a plurality of historical railway safety knowledge graphs and railway safety knowledge graphs corresponding to the historical target station yard ID and the corresponding historical similarity levels;

[0115] The plurality of historical railway safety knowledge graphs and railway safety knowledge graphs corresponding to the historical target station yard ID and the corresponding historical similarity levels are divided into training data, verification data and test data; the training data, the verification data and the test data are preprocessed to obtain a training set, a verification set and a test set; the ratio between the training set, the verification set and the test set is set to 7:2:1;

[0116] Select an artificial intelligence model as a basic model; the artificial intelligence model includes a neural network model and the like;

[0117] The basic model is trained by the training set, and the learning rate and the hyperparameter are adjusted on the verification set to obtain a pre-trained model;

[0118] The pre-trained model is verified on the test set, and finally a graph similarity estimation model with a plurality of railway safety knowledge graphs and railway safety knowledge graphs corresponding to the target station yard ID as input and a plurality of similarity levels as output is obtained.

[0119] In the embodiment, the risk prediction model is obtained by training the graph-enhanced time series prediction model on the target comprehensive data, including:

[0120] Obtain target comprehensive data; the target comprehensive data includes a plurality of subgraph data;

[0121] Extract a plurality of historical subgraph data and a plurality of historical time series data, historical risk levels and risk events corresponding thereto;

[0122] Integrate the historical subgraph data and the corresponding historical time series data into historical prediction data;

[0123] Divide the historical prediction data, the corresponding historical risk levels and risk events into training data, validation data and test data; perform data preprocessing on the training data, the validation data and the test data to obtain a training set, a validation set and a test set; the ratio between the training set, the validation set and the test set is set to 7:2:1;

[0124] Select a graph-enhanced time series prediction model as a base model; the graph-enhanced time series prediction model includes a GETM model and the like;

[0125] Train the base model through the training set, and adjust the learning rate and the hyperparameters on the validation set to obtain a pre-trained model;

[0126] Finally, obtain a risk prediction model with the prediction data as the input and the predicted risk level and the predicted risk event as the output by verifying the pre-trained model on the test set.

[0127] The generation of the predicted risk event and the predicted risk level according to the risk prediction model in the embodiment includes:

[0128] Obtain the time series data corresponding to the target station ID, the risk prediction model and the railway safety knowledge graph in real time; the railway safety knowledge graph includes a plurality of subgraph data;

[0129] Integrate the historical time series data and the corresponding historical subgraph data to obtain prediction data;

[0130] Input the prediction data into the risk prediction model to obtain the predicted risk level and the predicted risk event.

[0131] The generation of the alarm signal according to the predicted risk level in the embodiment includes:

[0132] Obtain the predicted risk level and the level interval; the level interval includes a low-risk interval, a medium-risk interval and a high-risk interval; the level interval is set by experts according to the historical risk level and the corresponding historical risk event;

[0133] When the predicted risk level belongs to the low-risk interval, a low-risk event warning signal is generated;

[0134] When the predicted risk level belongs to the medium-risk interval, a medium-risk event warning signal is generated;

[0135] When the predicted risk level belongs to the high-risk interval, an emergency risk warning signal is generated.

[0136] Part of the data in the above formula is calculated by removing the dimension, and the formula is obtained by software simulation of a large amount of collected data to obtain a formula closest to the actual situation; the preset parameters and the preset threshold in the formula are set by the person skilled in the art according to the actual situation or obtained by a large amount of data simulation.

[0137] The working principle of the present application is: acquiring station data through a data acquisition device; constructing a railway safety knowledge graph according to the station data; generating a risk prediction model according to the railway safety knowledge graph; generating a predicted risk event and a predicted risk level according to the risk prediction model; generating an alarm signal according to the predicted risk level; making a prompt according to the alarm signal, and contacting the management personnel; enhancing the risk identification accuracy by fusing the constructed railway safety knowledge graph and the risk prediction model, and realizing the early warning; using the multi-source heterogeneous data time sequence coding technology to realize the comprehensive analysis of the unified cross-frequency data; training the target station prediction model by using the multi-source station risk data, solving the problem of insufficient training data of the target station, and improving the accuracy and application efficiency of the safety risk assessment system, which avoids the problem that the existing technology often has a small amount of training data when constructing the risk prediction model for the newly-built station and the low-risk station, so that the deployed safety risk assessment system has a long time consumption and low efficiency.

[0138] The above embodiments are only used to illustrate the technical method of the present application and not to limit it. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical method of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical method of the present application.

Claims

1. A safety risk assessment system based on railway yard information, characterized by, include: Data acquisition module, data analysis module, early warning module, and database; The data acquisition module acquires station data through data acquisition equipment; the station data includes station ID, station purpose, equipment data, environmental data, and operational data. The data analysis module: constructs a railway safety knowledge graph based on station data; A risk prediction model is generated based on the railway safety knowledge graph; predictable risk events and risk levels are generated based on the risk prediction model. An alert signal is generated based on the predicted risk level; The risk prediction model generated based on the railway safety knowledge graph includes: Obtain the railway safety knowledge graph and matching priority PYi corresponding to several station IDs, as well as the railway safety knowledge graph corresponding to the target station ID; the matching priority is calculated using historical station data and the railway safety knowledge graph corresponding to several station IDs; the target station ID refers to the railway station ID for which a risk prediction model needs to be constructed. The matching sequence PL is obtained by sorting several matching priorities from largest to smallest. Select the station IDs and railway safety knowledge graphs corresponding to the top N matching priorities in PL, and name several of the railway safety knowledge graphs as source railway safety knowledge graphs; where N is an integer, N>1; Extract several common subgraph data from the railway safety knowledge graph corresponding to the target station ID and the source railway safety knowledge graph; The target comprehensive data is obtained by merging several common sub-map data and the station data corresponding to the target station ID; the target comprehensive data includes several sub-map data. A risk prediction model is obtained by training a graph-augmented time-series prediction model on target integrated data; The matching priority is calculated using historical station data corresponding to several station IDs and a railway safety knowledge graph, including: Obtain historical station data corresponding to several station IDs; the historical station data includes historical risk level, historical risk frequency, and occurrence time. Calculate the comprehensive risk score corresponding to several station IDs based on historical risk level, historical risk frequency, and occurrence time; Obtain several railway safety knowledge graphs and railway safety knowledge graphs corresponding to the target station ID; inputting a plurality of railway safety knowledge graphs and a railway safety knowledge graph corresponding to a target station yard ID into a graph similarity estimation model to obtain a plurality of similarity levels XD i ; the graph similarity estimation model is constructed by an artificial intelligence model; The matching priority PY is calculated by the formula i ; wherein i represents the number of station yard IDs, g is a proportional coefficient, g ∈ (0, π / 2); α1 and α2 are weight coefficients, α1 and α2 ∈ (0, 1); and the sum of α1 and α2 is 1; ZFP i represents the comprehensive risk score corresponding to the i-th station yard ID.​ 2. The system for safety risk assessment based on railway yard information according to claim 1, wherein, The construction of a railway safety knowledge graph based on station data includes: Acquire station data; the station data includes station ID, equipment data, environmental data, and operational data; Standardized data is generated based on equipment data, environmental data, and operational data. The standardized data refers to the standard data used to construct a railway safety knowledge graph. The standardized data includes several entities, several attributes, and several relationships. A security knowledge graph is constructed using a graph database, and spatiotemporal data storage and rule-driven mechanisms are embedded within it. The spatiotemporal data storage is used to store the spatiotemporal data corresponding to the site data. The rule-driven mechanism is constructed by experts based on the site data and its corresponding risk level. The railway safety knowledge graph is obtained by inputting standardized data into the safety knowledge graph.

3. The system for safety risk assessment based on railway yard information according to claim 2, characterized in that, The process of generating standardized data based on equipment data, environmental data, and operational data includes: Acquire equipment data, environmental data, and operational data; the equipment data includes equipment ID, equipment status, and equipment parameters; the environmental data includes meteorological data and geological data; the operational data includes operational ID, operational instructions, maintenance records, and monitoring videos; Equipment data, environmental data, and operational data are categorized into text data, view data, time-series data, and structured data; the structured data refers to data with a fixed format and defined fields; the view data refers to video data and image data. Standardized data is obtained by using a hierarchical fusion strategy to combine text data, view data, time-series data, and structured data; the hierarchical fusion strategy is operated through a multimodal fusion method.

4. The system for safety risk assessment based on railway yard information according to claim 3, wherein, The layered fusion strategy operates through multimodal fusion, including the following steps: Step 1: Obtain text data, view data, time series data, and structured data; Step 2: Extract several semantic vectors and spatial vectors from text data, view data, time-series data, and structured data through feature extraction; the semantic vectors and spatial vectors include several entities, attributes, and relationships; Step 3: Add a uniform time code to several semantic vectors, and align data of different frequencies using a sliding window aggregation method to obtain aligned semantic vectors; Step 4: Obtain the canonical data by concatenating the aligned semantic vector and the spatial vector.

5. A safety risk assessment system based on railway station information according to claim 1, characterized in that, The calculation of comprehensive risk scores corresponding to several station IDs based on historical risk levels, historical risk frequency, and occurrence time includes: Obtain historical risk level FD, historical risk frequency, and occurrence time Δt; Through formula Calculate the Comprehensive Risk Score (ZFP) i Where j represents the historical risk occurrence number, max() represents the maximum value operation, max(j) represents the number of historical risk occurrences, and β represents the time decay coefficient, β∈(0,1); Δt i,j DT represents the time elapsed since the occurrence of the j-th risk at the i-th station ID, where DT represents the unit of time.

6. A safety risk assessment system based on railway station information according to claim 1, characterized in that, The map similarity prediction model is constructed using an artificial intelligence model, including: Obtain several historical railway safety knowledge graphs and railway safety knowledge graphs corresponding to historical target station IDs, as well as their corresponding historical similarity levels; Several historical railway safety knowledge graphs and railway safety knowledge graphs corresponding to historical target station IDs, along with their corresponding historical similarity levels, are divided into training data, validation data, and test data. Data preprocessing is performed on the training data, validation data, and test data to obtain the training set, validation set, and test set. Choose an artificial intelligence model as the base model; The base model is trained on the training set, and the learning rate and hyperparameters are adjusted on the validation set to obtain the pre-trained model. By validating the pre-trained model on the test set, the final result is a railway safety knowledge graph with several railway safety knowledge graphs and the railway safety knowledge graph corresponding to the target station ID as inputs, and several similarity prediction models of the graphs with similarity levels as outputs.

7. A safety risk assessment system based on railway station information according to claim 1, characterized in that, The method of obtaining a risk prediction model by training a graph-enhanced time-series prediction model on target aggregated data includes: Acquire comprehensive target data; the comprehensive target data includes several sub-map data. Extract several historical subgraph data and their corresponding historical time series data, historical risk levels, and risk events; Several historical subgraph data and their corresponding historical time series data are integrated into several historical prediction data. Several historical prediction data and their corresponding historical risk levels and risk events are divided into training data, validation data and test data; data preprocessing is performed on the training data, validation data and test data to obtain the training set, validation set and test set. The graph-augmented time series prediction model was selected as the base model. The base model is trained on the training set, and the learning rate and hyperparameters are adjusted on the validation set to obtain the pre-trained model. By validating the pre-trained model on the test set, a risk prediction model is finally obtained, which takes the prediction data as input and outputs the predicted risk level and the predicted risk event.

8. A safety risk assessment system based on railway station information according to claim 1, characterized in that, The process of generating predicted risk events and predicted risk levels based on the risk prediction model includes: Real-time acquisition of time-series data and risk prediction models corresponding to target station IDs, as well as railway safety knowledge graphs; the railway safety knowledge graphs include several subgraph data. Extract several historical time series data and their corresponding historical subgraph data, and integrate them with the current time series data and subgraph data to obtain the prediction data; The predicted data is input into the risk prediction model to obtain the predicted risk level and predicted risk events.

Images