An intelligent attendance prediction system for locomotive crew members

Abstract

This invention discloses an intelligent attendance prediction system for locomotive crew members, relating to the field of artificial intelligence technology. It addresses the problems of high inefficiency in obtaining attendance predictions, low attendance efficiency and emergency response capabilities, and low accuracy in attendance predictions caused by significant deficiencies in data security. The intelligent attendance prediction system includes an attendance server, several departure and arrival terminals, and several attendance management terminals. The attendance management terminals include an information collection module, an information processing module, and a management and transmission module. The attendance server includes a prediction scheduling module and an information maintenance module. The departure and arrival terminals include an identity verification module, a departure and arrival management module, and a key operation verification module. The system updates crew attendance predictions in real time using streaming data processing technology. It improves the accuracy and practicality of attendance predictions through self-learning and optimization using machine learning models, providing personalized attendance predictions for each employee. A microservice architecture ensures the scalability and reliability of the attendance prediction system, improving multi-source data security and privacy protection.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an intelligent attendance prediction system for locomotive crew members. Background Art

[0002] In recent years, with the continuous commissioning of railway lines and the rapid development of high-speed rail, higher requirements have been put forward for the number of EMUs, operating speed, driving safety, etc. Locomotive crews are the last guarantee for train driving safety, and it is crucial to fully understand the prompts of key operating points of trains in the working section.

[0003] Attendance plans are a must-have document for crew members. At present, the attendance plans of locomotive crew members are still manually printed and delivered by the locomotive dispatcher in the dispatch room. The attendance reminders are brief, simple, and not targeted, and lack key operation reminders at key points in the section. Due to the lack of appropriate card control methods, managers cannot know whether the crew members have carefully read the delivered attendance plans. At the same time, there are more and more new drivers in recent years, and the safety risks brought by new drivers when operating trains are relatively large.

[0004] Patent No. CN2022108793595 discloses an attendance rate self-feedback intelligent class scheduling system, which belongs to the field of intelligent class scheduling systems. It is an attendance rate self-feedback intelligent class scheduling system. The teacher attendance feedback unit includes a teaching data processing module and a teacher teaching IC card recognition module and a teacher teaching portrait recognition module respectively connected to the teaching data processing module by signal. Through the cooperation of the class scheduling data processing unit, the blockchain transmission unit and the teacher attendance feedback unit, the actual teaching attendance of the teacher can be effectively monitored through the dual data of the teacher card and the portrait after the class scheduling is completed, which effectively avoids the phenomenon of graduate students substituting for tutors in university classrooms, and effectively regulates and restricts the teaching behavior of teachers, which can not only improve the teaching quality of the university and improve the learning effect of students, but also improve the fairness and justice of the subsequent efficient teacher selection, which is conducive to the healthy development of university teaching and improves the practicality of the intelligent class scheduling system.

[0005] Patent No. CN201510175138X discloses an intelligent attendance verification system and an intelligent attendance verification method. The system includes a student mobile terminal, which is configured to obtain an attendance response, forcibly execute the Bluetooth module, and forcibly turn off the Bluetooth module when the attendance verification time is pre-set; an attendance verification server, which is configured to pre-store student mobile terminal information, and when the attendance verification starts and the Bluetooth module is executed, a notification is pushed to the student mobile terminal; and a professor mobile terminal, which is configured to link to the attendance verification server, collect and display student mobile terminal information, press the attendance verification start button, send a command to the attendance verification server, and during the attendance verification, execute each student mobile terminal and Bluetooth docking, count the number of student mobile terminals that successfully dock with Bluetooth, generate attendance verification information and transmit the attendance verification information to the attendance verification server. The above invention uses a mobile phone to automatically and quickly conduct attendance verification in a short time to prevent wasting professors' lecture time.

[0006] However, the above patents have made great improvements in the attendance assessment method, but due to major defects in data security, the attendance expectations obtained by flight attendants are inefficient, the flight attendants' attendance efficiency and emergency response prediction capabilities are low, and the attendance prediction accuracy is low. Summary of the invention

[0007] The purpose of the present invention is to provide an intelligent attendance prediction system for locomotive crew members, which can update the attendance prediction of crew members in real time through streaming data processing technology; self-learn and optimize through machine learning models to improve the accuracy and practicality of crew members' attendance prediction, and provide personalized attendance prediction for each employee; ensure the scalability and reliability of the attendance prediction system through a microservice architecture, while paying attention to multi-source data security and privacy protection.

[0008] The present invention utilizes the following technical solutions:

[0009] A locomotive crew intelligent attendance prediction system includes an attendance server, a plurality of attendance terminals and a plurality of attendance management terminals; wherein,

[0010] The attendance management terminal includes an information collection module, an information processing module, and a management transmission module;

[0011] The attendance management terminal is used to collect crew information, weather information, train information and historical road condition information of different railway sections through the information collection module; and to clean, associate and store the crew information, weather information, train information and historical road condition information through the information processing module to build a full-course train map; and to manage the stored crew information, weather information, train information, historical road condition information and full-course train map through the management transmission module, and transmit them to the attendance server at the same time;

[0012] The attendance server includes an expected scheduling module and an information maintenance module;

[0013] The attendance server is used to update the attendance event prediction for all crew members of various functional departments of the railway section in real time based on the train full-course map by using the multimodal prediction model through the expected arrangement module, so as to generate the attendance event prediction table and transmit it to the attendance terminal at the same time; and encrypt and maintain the attendance event prediction table and crew information through the information maintenance module;

[0014] The attendance terminal includes an identity authentication module, an attendance management module and a key operation verification module;

[0015] The attendance terminal is used to verify the crew information according to the daily attendance schedule through the identity authentication module; and to adjust the daily attendance schedule according to the crew's attendance application form or attendance change application form through the attendance management module to dispatch the crew; and to verify the communication matters and train key operation points in the attendance event prediction table in the form of questions through the key operation verification module.

[0016] Preferably, the information collection module uses input and output devices to collect crew information, weather information, train information and historical road condition information; crew information includes function, name, gender, age, education, height and weight; weather information includes temperature, humidity, air pressure, wind speed and precipitation; train information includes train number, running direction, speed level, line type, train model and internal facilities; historical road condition information includes train running position, train delay, track circuit status, switch location, station track, section blockade, temporary speed limit, safety monitoring and equipment failure.

[0017] Preferably, the information processing module uses a data cleaning algorithm to clean the crew information, weather information, train information and historical road condition information to remove duplicate information data; and uses a data association algorithm to associate the crew information, weather information, train information and historical road condition information that have completed data cleaning according to the train number and railway section to construct a train data block; also uses a knowledge graph to construct several train data blocks into a full-route train map according to the railway section, and uses a metadata database to store them; the management and transmission module manages the stored crew information, weather information, train information, historical road condition information and full-route train map, and uses the CANopen communication protocol and the Profibus communication protocol to transmit them to the attendance server.

[0018] Preferably, the anticipated scheduling module uses a multimodal prediction model to update the attendance event prediction for all crew members of each functional department of the railway section in real time according to the full train map to generate an attendance event prediction table; the information maintenance module uses a lossless encryption algorithm through the B\S architecture to encrypt and regularly maintain the attendance event prediction table and crew information; the attendance event prediction table includes one car and one card, train timetable, three fixed items, section operation related items and communication matters.

[0019] Preferably, the multimodal prediction model first uses the information separation layer to separate the historical road condition information of each train number in the full train map; then uses the natural language branch of the feature extraction layer to extract the keywords in the historical road condition information, and converts the keywords into feature vectors to form a keyword vector matrix; at the same time, the information extraction branch of the feature extraction layer is used to extract the eigenvalues ​​of the full train map to form a map vector matrix; then the iterative training layer is used to perform several iterative training on the map vector matrix to form a comprehensive weight matrix; then the recognition update layer is used in combination with the keyword vector matrix and the global loss function to update and optimize the comprehensive weight matrix to obtain the spatiotemporal weight matrix of the precise time series; finally, the attendance prediction layer sorts out the attendance events of each crew member according to the spatiotemporal weight matrix and crew information to generate an attendance event prediction table.

[0020] Preferably, the information separation layer uses a separation extraction function to separate the historical road condition information of each train number in the train full-route map;

[0021] A,B,C,D=Split[TFJM] (1)

[0022] Among them, TFJM represents the train full-route map, A represents the crew information set, B represents the weather information set, C represents the train information set, D represents the historical road condition information set, Split[] represents the separation extraction function, i represents the train sequence number, d i Indicates the historical road condition information of each train number;

[0023] The natural language branch includes 2 CBM blocks, 3 InceptionV2 blocks, and 2 Transformer encoding blocks; the CBM block includes 1 3×3 convolution layer with a stride of 2, 2 5×5 convolution layers with a stride of 1, 2 adaptive pooling layers, and MISH activation function;

[0024] The information extraction branch includes three 3×3 convolutional layers, four 5×5 convolutional layers, one convolutional block consisting of a 3×1 convolutional layer and a 1×3 convolutional layer, and one 5×5 maximum pooling layer;

[0025] The iterative training layer includes the first training branch and the second training branch; the first training branch includes 3 CBR blocks, 2 residual blocks, 3 InceptionV4 blocks, 2 Transformer encoding blocks, 1 batch standard layer and Silu activation function; the CBR block includes 1 5X5 convolution layer with a stride of 2, 4 batch standard layers and Relu activation function;

[0026] The second training branch includes 3 CBS blocks, 2 inverted residual blocks, 2 InceptionV3 blocks and 2 Softmax activation functions; the CBS block includes 3 5×5 convolutional layers with a stride of 2, 2 3×3 convolutional layers with a stride of 1, 1 7×7 dilated convolutional layer with a stride of 2 and 2 Silu activation functions;

[0027] The recognition update layer includes 3 normalization layers, 2 fully connected layers, 2 random dropout layers, and 2 Softmax activation functions;

[0028] The attendance prediction layer uses the content generator according to the crew information to match the corresponding attendance events from the spatiotemporal weight matrix in sequence, and integrates all the matched attendance events in time series to form an attendance event prediction table for each crew member.

[0029] Preferably, the identity authentication module uses an alcohol detector to detect the alcohol concentration of the flight attendant, uses an IC card detector to verify the flight attendant's identification document based on the flight attendant's information, and uses a facial recognizer to match and authenticate the flight attendant's facial features based on the ID card photo.

[0030] Preferably, after the flight attendant passes the alcohol concentration test, identity document verification and facial matching, the key operation verification module uses the question generator to test the attendance event prediction table in the form of question answering, and uses the posture sensor to evaluate the flight attendant's action behavior for special standard actions. After passing the question test and action evaluation, the flight attendant's status is shown as normal attendance in the daily attendance schedule; the attendance and leave management module uses the automatic approval algorithm to approve the flight attendant's leave application form or change of attendance application form, and then generates a daily attendance schedule.

[0031] The present invention uses streaming data processing technology to update the attendance prediction of flight attendants in real time; performs self-learning and optimization through machine learning models to improve the accuracy and practicality of the attendance prediction of flight attendants, and provides personalized attendance prediction for each employee; ensures the scalability and reliability of the attendance prediction system through a microservice architecture, and improves multi-source data security and privacy protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the drawings required for use in the embodiments or related technical descriptions are briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0033] Figure 1 Envision a system flow chart for smart attendance;

[0034] Figure 2 This is a schematic diagram of the principle of the intelligent attendance envisioning system. DETAILED DESCRIPTION

[0035] The present invention is described in detail below with reference to the accompanying drawings and embodiments:

[0036] like Figure 1 to Figure 2 As shown, the locomotive crew intelligent attendance prediction system described in the present invention includes an attendance server, a plurality of attendance terminals and a plurality of attendance management terminals; wherein,

[0037] The attendance management terminal includes an information collection module, an information processing module, and a management transmission module;

[0038] The attendance management terminal is used to collect crew information, weather information, train information and historical road condition information of different railway sections through the information collection module; and to clean, associate and store the crew information, weather information, train information and historical road condition information through the information processing module to build a full-course train map; and to manage the stored crew information, weather information, train information, historical road condition information and full-course train map through the management transmission module, and transmit them to the attendance server at the same time;

[0039] The attendance server includes an expected scheduling module and an information maintenance module;

[0040] The attendance server is used to update the attendance event prediction for all crew members of various functional departments of the railway section in real time based on the train full-course map by using the multimodal prediction model through the expected arrangement module, so as to generate the attendance event prediction table and transmit it to the attendance terminal at the same time; and encrypt and maintain the attendance event prediction table and crew information through the information maintenance module;

[0041] The attendance terminal includes an identity authentication module, an attendance management module and a key operation verification module;

[0042] The departure and return terminal is used to verify the crew information according to the daily attendance schedule through the identity authentication module; and adjust the daily attendance schedule according to the crew's departure application form or change of attendance application form through the departure and return management module to dispatch the crew; and also verify the communication items and train key operation points in the attendance event prediction table in the form of questions through the key operation verification module;

[0043] In the present invention, the information collection module uses input and output devices to collect crew information, weather information, train information and historical road condition information; crew information includes function, name, gender, age, education, height and weight; weather information includes temperature, humidity, air pressure, wind speed and precipitation; train information includes train number, running direction, speed level, line type, train model and internal facilities; historical road condition information includes train running position, train early or late, track circuit status, turnout position, station track, section blockade, temporary speed limit, safety monitoring and equipment failure;

[0044] In this embodiment, the functions include train conductor, train attendant, announcer, train inspector, train baggage attendant, food and water supply personnel and train police;

[0045] In the present invention, the information processing module uses a data cleaning algorithm to clean crew information, weather information, train information and historical road condition information to remove duplicate information data; and uses a data association algorithm to associate the crew information, weather information, train information and historical road condition information that have completed data cleaning according to the train number and railway section to construct a train data block; also uses a knowledge graph to construct a number of train data blocks into a full-course train map according to the railway section, and uses a metadata database to store them; the management and transmission module manages the stored crew information, weather information, train information, historical road condition information and full-course train map, and uses the CANopen communication protocol and the Profibus communication protocol to transmit them to the attendance server;

[0046] In the present invention, the pre-arrangement module uses a multimodal prediction model to update the attendance event prediction for all crew members of various functional departments of the railway section in real time according to the train full-course map to generate an attendance event prediction table; the information maintenance module uses a lossless encryption algorithm through the B\S architecture to encrypt and regularly maintain the attendance event prediction table and crew information; the attendance event prediction table includes one car and one card, train timetable, three fixed items, section operation related items and communication matters;

[0047] In this embodiment, three fixed means fixed track, fixed platform and fixed parking position; one car one card means each car corresponds to a train operation card; section operation related items include passenger business, freight business, operation operation, locomotive operation, vehicle business, vehicle operation, engineering operation, electrical operation, communication operation and signal operation;

[0048] In the present invention, the multimodal prediction model first uses the information separation layer to separate the historical road condition information of each train number in the full train map;

[0049] In this embodiment, the information separation layer uses a separation extraction function to separate the historical road condition information of each train number in the train full-route map;

[0050] A,B,C,D=Split[TFJM](1)

[0051] D=∑ i＝1 d i (2)

[0052] Among them, TFJM represents the train full-route map, A represents the crew information set, B represents the weather information set, C represents the train information set, D represents the historical road condition information set, Split[] represents the separation extraction function, i represents the train sequence number, d i Indicates the historical road condition information of each train number;

[0053] Then, the natural language branch of the feature extraction layer is used to extract keywords from the historical traffic information and convert the keywords into feature vectors to form a keyword vector matrix;

[0054] In this embodiment, the natural language branch includes 2 CBM blocks, 3 InceptionV2 blocks and 2 Transformer encoding blocks; the CBM block includes 1 3×3 convolution layer with a stride of 2, 2 5×5 convolution layers with a stride of 1, 2 adaptive pooling layers and a MISH activation function;

[0055] At the same time, the information extraction branch of the feature extraction layer is used to extract the feature values ​​of the train's entire map to form a map vector matrix;

[0056] In this embodiment, the information extraction branch includes three 3×3 convolutional layers, four 5×5 convolutional layers, one convolutional block consisting of a 3×1 convolutional layer and a 1×3 convolutional layer, and one 5×5 maximum pooling layer;

[0057] Then, the graph vector matrix is ​​iteratively trained several times using the iterative training layer to form a comprehensive weight matrix;

[0058] In this embodiment, the iterative training layer includes a first training branch and a second training branch; the first training branch includes 3 CBR blocks, 2 residual blocks, 3 InceptionV4 blocks, 2 Transformer encoding blocks, 1 batch standard layer and Silu activation function; the CBR block includes 1 5X5 convolution layer with a stride of 2, 4 batch standard layers and Relu activation function;

[0059] The second training branch includes 3 CBS blocks, 2 inverted residual blocks, 2 InceptionV3 blocks and 2 Softmax activation functions; the CBS block includes 3 5×5 convolutional layers with a stride of 2, 2 3×3 convolutional layers with a stride of 1, 1 7×7 dilated convolutional layer with a stride of 2 and 2 Silu activation functions;

[0060] Then, the recognition update layer is used to combine the keyword vector matrix and the global loss function to update and optimize the comprehensive weight matrix to obtain the spatiotemporal weight matrix of the precise time series;

[0061] In this embodiment, the recognition update layer includes 3 normalization layers, 2 fully connected layers, 2 random dropout layers and 2 Softmax activation functions;

[0062] Global loss function Loss:

[0063] Among them, σ represents the number of iterative training layers, μ1 represents the calculation accuracy coefficient, and Model 1 represents the information extraction layer, μ2 represents the preset optimization weight coefficient, and Model 2 represents the iterative training layer, G represents the quantization function, B σ Represents the iterative training loss function for each layer, E σ Indicates the optimization efficiency of each layer’s iterative training;

[0064] Finally, the attendance prediction layer sorts out the attendance events of each crew member according to the spatiotemporal weight matrix and crew information to generate an attendance event prediction table;

[0065] In this embodiment, the attendance prediction layer uses the content generator according to the crew information to sequentially match the corresponding attendance events from the spatiotemporal weight matrix, and integrates all the matched attendance events according to the time series to form the attendance event prediction table of each crew member;

[0066] In the present invention, the identity verification module uses an alcohol detector to detect the alcohol concentration of the crew member, uses an IC card detector to verify the crew member's identity document based on the crew member's information, and uses a facial recognition device to match and authenticate the crew member's facial features based on the ID card photo;

[0067] In the present invention, after the crew member passes the alcohol concentration test, identity document verification and face matching, the operation key verification module uses the question generator to test the attendance event prediction table in the form of question answering, and uses the posture sensor to evaluate the action behavior of the crew member for special standard actions. After passing the question test and action evaluation, the crew member's status is displayed as normal attendance in the daily attendance schedule; the attendance and leave management module uses the automatic approval algorithm to approve the crew member's leave application form or change application form, and then generates a daily attendance schedule;

[0068] Example:

[0069] Train managers collect crew information, weather information, train information and historical road condition information through the information collection module and input and output devices at the attendance management end; crew information includes function, name, gender, age, education, height and weight; weather information includes temperature, humidity, air pressure, wind speed and precipitation; train information includes train number, running direction, speed level, line type, train model and internal facilities; historical road condition information includes train running position, train early or late, track circuit status, switch location, station track, section blockade, temporary speed limit, safety monitoring and equipment failure; the attendance management end collects crew information, weather information, train information and historical road condition information through the information processing module using data cleaning algorithm The weather information, train information and historical road condition information are cleaned to remove duplicate information data; the crew information, weather information, train information and historical road condition information that have completed data cleaning are associated according to the train number and railway section using the data association algorithm to construct the train data block; the knowledge graph is also used to construct several train data blocks into a full-course train map according to the railway section, and the metadata database is used for storage; the attendance management end manages the stored crew information, weather information, train information, historical road condition information and full-course train map through the management transmission module, and transmits them to the attendance server using the CANopen communication protocol and the Profibus communication protocol;

[0070] The attendance server uses the information separation layer of the multimodal prediction model to separate the historical road condition information of each train in the train map through the expected arrangement module; then, through the natural language branch of the feature extraction layer, two CBM blocks, three InceptionV2 blocks and two Transformer encoding blocks are used to extract the keywords in the historical road condition information, and the keywords are converted into feature vectors to form a keyword vector matrix; at the same time, through the information extraction branch of the feature extraction layer, three 3×3 convolutional layers, four 5×5 convolutional layers, one convolutional block consisting of a 3×1 convolutional layer and a 1×3 convolutional layer, and one 5×5 maximum pooling layer are used to extract the feature values ​​of the train map to form a map vector matrix; then, through the first training layer of the iterative training layer, the feature vector matrix is ​​obtained. The training branch includes 3 CBR blocks, 2 residual blocks, 3 InceptionV4 blocks, 2 Transformer encoding blocks, 1 batch standard layer and Silu activation function, and the graph vector matrix is ​​iteratively trained several times to form a comprehensive weight matrix; then the recognition update layer uses 3 normalization layers, 2 fully connected layers, 2 random dropout layers and 2 Softmax activation functions, combined with the keyword vector matrix and the global loss function, to update and optimize the comprehensive weight matrix to obtain the spatiotemporal weight matrix of the precise time series; finally, the attendance prediction layer uses the content generator according to the crew information to match the corresponding attendance events from the spatiotemporal weight matrix in sequence, and integrates all the matched attendance events according to the time series to form the attendance event prediction table of each crew member;

[0071] The crew uses an alcohol detector to test the crew's alcohol concentration through the identity verification module of the departure and arrival terminal, uses an IC card detector to verify the crew's identity document based on the crew's information, and uses a facial recognition device to match and authenticate the crew's facial features based on the identity card photo; the departure and arrival management module uses an automatic approval algorithm to approve the crew's departure application form or attendance change application form, and then generates a daily attendance schedule; through the key verification module of the operation, after the crew passes the alcohol concentration test, identity document verification and facial matching, the attendance event prediction table is tested in the form of questions using a question generator, and the crew's action behavior is evaluated using a posture sensor for special standard actions. After passing the question test and action evaluation, the crew's status is shown as normal attendance in the daily attendance schedule;

[0072] The locomotive crew intelligent attendance prediction system also reads the transportation safety system data, combines with professional managers to organize and import various types of data in advance, realizes intelligent retrieval based on key information such as different train numbers, sections, and vehicle models, and combines them to obtain a reminder sheet in the style of a driver's manual.

Claims

1. An intelligent attendance prediction system for locomotive crew, characterized in that: It includes attendance server, several attendance terminals and several attendance management terminals; among which, The attendance management terminal includes an information collection module, an information processing module, and a management transmission module; The attendance management terminal is used to collect crew information, weather information, train information and historical road condition information of different railway sections through the information collection module; and to clean, associate and store the crew information, weather information, train information and historical road condition information through the information processing module to build a full-course train map; and to manage the stored crew information, weather information, train information, historical road condition information and full-course train map through the management transmission module, and transmit them to the attendance server at the same time; The attendance server includes an expected scheduling module and an information maintenance module; The attendance server is used to update the attendance event prediction for all crew members of various functional departments of the railway section in real time based on the train full-course map by using the multimodal prediction model through the expected arrangement module, so as to generate the attendance event prediction table and transmit it to the attendance terminal at the same time; and encrypt and maintain the attendance event prediction table and crew information through the information maintenance module; The attendance terminal includes an identity authentication module, an attendance management module and a key operation verification module; The attendance terminal is used to verify the crew information according to the daily attendance schedule through the identity authentication module; and to adjust the daily attendance schedule according to the crew's attendance application form or attendance change application form through the attendance management module to dispatch the crew; and to verify the communication matters and train key operation points in the attendance event prediction table in the form of questions through the key operation verification module.

2. The intelligent attendance prediction system for locomotive crew members according to claim 1 is characterized in that: The information collection module uses input and output devices to collect crew information, weather information, train information and historical road condition information; crew information includes function, name, gender, age, education, height and weight; weather information includes temperature, humidity, air pressure, wind speed and precipitation; train information includes train number, running direction, speed level, line type, train model and internal facilities; historical road condition information includes train running position, train delay, track circuit status, switch location, station track, section blockade, temporary speed limit, safety monitoring and equipment failure.

3. The intelligent attendance prediction system for locomotive crew members according to claim 1 is characterized in that: The information processing module uses a data cleaning algorithm to clean crew information, weather information, train information and historical road condition information to remove duplicate information data; and uses a data association algorithm to associate the crew information, weather information, train information and historical road condition information that have completed data cleaning according to the train number and railway section to construct a train data block; The knowledge graph is also used to construct a full train map of several train data blocks according to the railway section, and the metadata database is used for storage; The management and transmission module manages the stored crew information, weather information, train information, historical road condition information and train full-route maps, and transmits them to the attendance server using the CANopen communication protocol and the Profibus communication protocol.

4. The intelligent attendance prediction system for locomotive crew members according to claim 1 is characterized in that: The anticipated scheduling module uses a multimodal prediction model to update the attendance event prediction for all crew members of each functional department of the railway section in real time according to the full train map to generate an attendance event prediction table; the information maintenance module uses a lossless encryption algorithm through the B\S architecture to encrypt and regularly maintain the attendance event prediction table and crew information; the attendance event prediction table includes one car and one card, train timetable, three fixed items, section operation related items and communication matters.

5. The intelligent attendance prediction system for locomotive crew members according to claim 4 is characterized in that: The multimodal prediction model first uses the information separation layer to separate the historical road condition information of each train number in the train full-route map; then uses the natural language branch of the feature extraction layer to extract keywords in the historical road condition information, and converts the keywords into feature vectors to form a keyword vector matrix; at the same time, uses the information extraction branch of the feature extraction layer to extract the feature values ​​of the train full-route map to form a map vector matrix; then uses the iterative training layer to perform several iterative training on the map vector matrix to form a comprehensive weight matrix; Subsequently, the recognition update layer is used in combination with the keyword vector matrix and the global loss function to update and optimize the comprehensive weight matrix to obtain the spatiotemporal weight matrix of the accurate time series; finally, the attendance prediction layer sorts out the attendance events of each crew member according to the spatiotemporal weight matrix and crew information to generate an attendance event prediction table.

6. The intelligent attendance prediction system for locomotive crew members according to claim 5 is characterized in that: The information separation layer uses a separation extraction function to separate the historical road condition information of each train number in the train full-route map; A,B,C,D=Split[TFJM] (1) D=∑ i＝1 d i (2) Among them, TFJM represents the train full-route map, A represents the crew information set, B represents the weather information set, C represents the train information set, D represents the historical road condition information set, Split[] represents the separation extraction function, i represents the train sequence number, d i Indicates the historical road condition information of each train number; The natural language branch includes 2 CBM blocks, 3 InceptionV2 blocks, and 2 Transformer encoding blocks; the CBM block includes 1 3×3 convolution layer with a stride of 2, 2 5×5 convolution layers with a stride of 1, 2 adaptive pooling layers, and MISH activation function; The information extraction branch includes three 3×3 convolutional layers, four 5×5 convolutional layers, one convolutional block consisting of a 3×1 convolutional layer and a 1×3 convolutional layer, and one 5×5 maximum pooling layer; The iterative training layer includes the first training branch and the second training branch; the first training branch includes 3 CBR blocks, 2 residual blocks, 3 InceptionV4 blocks, 2 Transformer encoding blocks, 1 batch standard layer and Silu activation function; the CBR block includes 1 5X5 convolution layer with a stride of 2, 4 batch standard layers and Relu activation function; The second training branch includes 3 CBS blocks, 2 inverted residual blocks, 2 InceptionV3 blocks and 2 Softmax activation functions; the CBS block includes 3 5×5 convolutional layers with a stride of 2, 2 3×3 convolutional layers with a stride of 1, 1 7×7 dilated convolutional layer with a stride of 2 and 2 Silu activation functions; The recognition update layer includes 3 normalization layers, 2 fully connected layers, 2 random dropout layers, and 2 Softmax activation functions; The attendance prediction layer uses the content generator according to the crew information to match the corresponding attendance events from the spatiotemporal weight matrix in sequence, and integrates all the matched attendance events in time series to form an attendance event prediction table for each crew member.

7. The intelligent attendance prediction system for locomotive crew members according to claim 1 is characterized in that: The identity verification module uses an alcohol detector to detect the alcohol concentration of the crew, uses an IC card detector to verify the crew's identity document based on the crew's information, and uses a facial recognizer to match and authenticate the crew's facial features based on the identity card photo.

8. The intelligent attendance prediction system for locomotive crew members according to claim 1 is characterized in that: After the flight attendant passes the alcohol concentration test, identity document verification and facial matching, the key operation verification module uses the question generator to test the attendance event prediction table in the form of question answering, and uses the posture sensor to evaluate the flight attendant's action behavior for special standard actions. After passing the question test and action evaluation, the flight attendant's status is shown as normal attendance in the daily attendance schedule; the attendance and leave management module uses the automatic approval algorithm to approve the flight attendant's leave application form or change of attendance application form, and then generates a daily attendance schedule.

Images