A logistics intelligent management optimization method fusing multi-source data analysis

Abstract

The present application relates to the technical field of railway logistics operation and maintenance, and discloses a logistics intelligent management optimization method fusing multi-source data analysis, which comprises the following steps: obtaining business text and structured equipment monitoring data, combining train working diagram and passenger flow density parameters to extract fault action vocabulary; using train arrival and departure time alignment data to lock effective physical time window containing fault evolution; performing passenger flow load decoupling processing on monitoring data, using a baseline model to remove basic load to generate standard residual waveform; fusing building information model and logical network, constructing spatial logical index based on comprehensive proximity; performing cross-station similar equipment matching based on multi-dimensional working condition feature vector and generating maintenance report. Through effectively stripping the interference of passenger flow environmental noise on equipment monitoring data, combining spatial logical correlation and working condition similarity matching, the accuracy of fault feature extraction and the reuse efficiency of cross-site operation and maintenance knowledge are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of railway logistics operation and maintenance technology, specifically to a logistics intelligent management optimization method that integrates multi-source data analysis. Background Technology

[0002] With the rapid development of high-speed railway networks, railway logistics operation and maintenance management has gradually transformed towards digitalization and intelligence. Existing integrated logistics management platforms typically integrate building equipment monitoring systems and operation and maintenance management databases, enabling real-time status monitoring and repair management of key facilities within stations, such as elevators, HVAC systems, water supply and drainage systems, and lighting. Maintenance personnel primarily rely on time-series data collected by sensors to determine the current operating status of equipment, combined with regular manual inspection records, to perform equipment maintenance. This model plays a fundamental role in ensuring the daily operation of station facilities and responding to unexpected failures.

[0003] In the specific scenario of railway stations, equipment operating parameters are significantly constrained by the dynamic tidal effects of passenger flow density and train arrival and departure times. Existing monitoring data analysis methods often focus on determining absolute thresholds for individual equipment physical quantities or simple trend analysis, lacking a mechanism to deeply correlate equipment monitoring data with external environmental background parameters. During peak train arrival and departure times, the drastic fluctuations in equipment load caused by high-intensity passenger flow can easily be misinterpreted by monitoring algorithms as equipment performance anomalies, or genuine early, subtle fault characteristics can be masked by the background noise of fluctuating passenger flow loads. This limits the accuracy of fault diagnosis and makes it difficult to achieve precise anomaly identification in complex dynamic environments.

[0004] Existing fault tracing and maintenance decision support methods primarily rely on physical topology and simple text keyword retrieval. In highly integrated station systems, equipment not only exists in physical adjacency but also in implicit logical connections such as electrical circuits, signal control, and pipeline networks. Relying solely on physical spatial models makes it difficult to quickly locate cascading fault sources across regions. Furthermore, traditional maintenance knowledge base retrieval lacks consideration of equipment operating conditions, making it difficult to perform cross-site experience matching based on the similarity of actual scenarios such as passenger flow pressure, temperature and humidity environments, and operating duration. Consequently, when facing equipment faults of the same type but with significantly different operating conditions, the reference value and reuse efficiency of historical maintenance solutions cannot be effectively guaranteed. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a logistics intelligent management optimization method that integrates multi-source data analysis. This method solves the technical problems in existing railway logistics operation and maintenance, such as the difficulty in effectively integrating multi-source heterogeneous data, the susceptibility of equipment fault characteristics to interference from passenger flow and environmental noise leading to false alarms, and the difficulty in effectively reusing cross-site operation and maintenance experience based on the similarity of operating conditions.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a logistics intelligent management optimization method integrating multi-source data analysis, implemented on a railway logistics intelligent integrated management platform, comprising the following steps:

[0007] Acquire business text data, structured equipment monitoring data, and environmental background parameters including train operation diagrams and passenger flow density; perform semantic parsing on the business text data to extract equipment entities and fault action words.

[0008] The search domain is determined by using a rule base for mapping action and timeliness features, and an effective physical time window containing the fault evolution process is locked by an event alignment algorithm based on train arrival and departure times.

[0009] The structured equipment monitoring data within the effective physical time window is subjected to passenger flow load decoupling processing, and the basic load is removed using the passenger flow and equipment load response baseline model to generate residual waveforms;

[0010] The building information model is analyzed to establish a physical spatial tree structure and overlay a logical association network. The comprehensive proximity is calculated based on physical distance and logical hop count to generate a spatial inverted index.

[0011] In response to external query requests, the system performs cross-site matching of similar equipment based on multi-dimensional operating condition feature vectors, and generates maintenance auxiliary reports by retrieving historical maintenance records based on residual waveforms.

[0012] Specifically, for the acquisition and semantic parsing of multi-source data, the method utilizes natural language processing (NLP) techniques to parse business text data, extract target device entities, and map them to unique device codes in the device ledger. Simultaneously, it converts action descriptions in the business text data into standardized fault terms using a thesaurus and extracts degree adverbs modifying the actions. Subsequently, based on the base scores of the fault terms and the correction coefficients of the degree adverbs, it calculates fault level weight values, outputting a structured fault vector containing unique device codes, standardized fault terms, and fault level weight values.

[0013] To clarify the operational context of the equipment, this method performs operational scenario labeling on the structured equipment monitoring data. First, a passenger flow density threshold is calculated based on station architectural design specifications. This threshold is calculated as the product of the design limit capacity density and the congestion saturation coefficient. If the real-time passenger flow density monitoring value is greater than or equal to the passenger flow density threshold, and the current time falls within the peak train arrival and departure period defined in the train timetable, the structured equipment monitoring data for that period is marked as a high passenger flow load mode. Based on this, a scenario context vector is generated, containing the data sampling timestamp, station operation stage code, and normalized passenger flow density value.

[0014] To lock the effective physical time window for sensor data, the maximum delayed response time is first searched in the action and timeliness feature mapping rule base according to standardized fault terminology, and a fault backtracking search domain is constructed with the repair reporting timestamp as the endpoint. Next, the absolute value of the time difference between the event time and the repair reporting time for each train in the set of train numbers within the fault backtracking search domain is calculated. During the calculation, a train formation passenger capacity weight is introduced, determined by the ratio of the train's rated passenger capacity to the maximum allowed single train formation passenger capacity at the station. Finally, the train time with the smallest weighted time distance is selected as the baseline train event anchor point, and an effective physical time window is generated by combining the preceding and following offsets determined by the equipment's physical attributes.

[0015] In the passenger flow load decoupling and residual waveform generation stage, passenger flow density data is first interpolated to construct a synchronous passenger flow density sequence aligned with the timestamps of structured equipment monitoring data. A baseline model for passenger flow and equipment load response is pre-trained using historical operation and maintenance data, and a multinomial regression algorithm is used to fit the nonlinear mapping relationship between passenger flow density and the physical quantities monitored by the equipment. The synchronous passenger flow density sequence is input into the baseline model to calculate the theoretical background load at each time point. Subsequently, point-to-point difference operations are performed between the structured equipment monitoring data within the effective physical time window and the calculated theoretical background load to generate the original residual sequence. The baseline arithmetic mean and standard deviation of the residual sequence under historical normal operating conditions are used to perform a standardization transformation on the original residual sequence, converting the original residual amplitude into a deviation value relative to the normal noise level, thereby generating a dimensionless standard fault characteristic waveform.

[0016] Regarding the construction of the hierarchical and partitioned spatial logical index, the method analyzes the 3D geometric center coordinates of equipment from the Building Information Model (BIM) and constructs a physical spatial tree-like topology structure containing sites, functional areas, and equipment entities based on physical zoning standards. Simultaneously, electrical wiring diagram data and signal flow diagram data are read, and logical connections are established between equipment nodes connected by direct electrical loops, signal control links, or physical pipelines, forming a logical association network with an undirected graph structure. Based on this structure, the physical Euclidean distance between two equipment nodes in 3D space is calculated, and the physical proximity component is calculated based on the distance attenuation constant. The shortest path hop count between two equipment nodes in the logical association network is calculated, and the logical association component is calculated based on the reciprocal of the hop count. The physical proximity weight coefficient and logical association weight coefficient are obtained by looking up a table according to the fault type. The physical proximity component and the logical association component are then weighted and summed to obtain the comprehensive association weight and generate a weighted adjacency matrix.

[0017] Finally, in response to external query requests, a multi-dimensional operating condition feature vector is constructed. This vector includes static inherent attributes describing the station's geographical and climatic zones and equipment models, as well as dynamic operating environment attributes describing passenger congestion levels, temperature and humidity values, and continuous equipment operating time. Historical fault cases of similar equipment are filtered by equipment type. The weighted similarity between the current multi-dimensional operating condition feature vector and the operating condition feature vectors of historical fault cases is calculated, where the weight coefficients of the feature dimensions are assigned based on a preset equipment type mapping relationship. Based on the calculation results, standardized fault terminology is used as a semantic index, combined with the selected historical fault cases, to retrieve the corresponding fault cause entities and handling solution entities in the operation and maintenance knowledge graph. A comprehensive recommendation score for each handling solution is calculated, which is weighted by the weighted similarity and the historical repair success rate of the handling solution. Maintenance auxiliary reports, including troubleshooting steps and supporting similar historical cases, are generated in descending order of comprehensive recommendation scores.

[0018] This invention provides a method for optimizing intelligent logistics management by integrating multi-source data analysis. It has the following beneficial effects:

[0019] 1. This invention introduces train timetables and real-time station passenger flow density data as environmental background parameters, utilizes an event alignment algorithm based on train arrival and departure times to pinpoint the fault evolution process, and combines passenger flow load decoupling processing to eliminate basic load. This mechanism effectively isolates environmental noise from interfering with equipment monitoring data, restoring the true operating state of the equipment, thereby significantly improving the accuracy of fault feature extraction in complex passenger flow fluctuation environments and preventing equipment parameter fluctuations caused by peak passenger flow from being misjudged as equipment faults.

[0020] 2. This invention constructs a hierarchical, partitioned spatial logical index by integrating Building Information Modeling (BIM) with electrical and signal logic networks, and calculates comprehensive proximity based on physical distance and logical hop count. This dual indexing mechanism considers not only the physical adjacency of devices but also the logical connections between devices in electrical circuits and signal control. During fault diagnosis, this mechanism helps maintenance personnel quickly locate potential cascading fault points or hidden related fault sources, optimizing the fault tracing path in complex systems.

[0021] 3. This invention constructs a multi-dimensional operating condition feature vector that includes static inherent attributes and dynamic operating environment, enabling cross-site matching of operating conditions for similar equipment. This method changes the traditional approach of relying solely on equipment model or simple keywords for retrieval, and can accurately push historical maintenance experience based on the similarity of actual operating conditions such as passenger flow congestion, temperature, and humidity. This allows maintenance personnel to directly refer to handling solutions validated under similar environmental pressures, effectively improving the relevance of cross-site maintenance knowledge reuse and the efficiency of resolving actual faults. Attached Figure Description

[0022] Figure 1 This is a flowchart of the method of the present invention;

[0023] Figure 2 This is a schematic diagram illustrating the process of acquiring multi-source railway logistics data and semantic parsing based on operational scenarios in accordance with the present invention.

[0024] Figure 3 This is a flowchart illustrating the process of dynamically determining the effective physical time window of sensor data based on action vocabulary and railway passenger transport patterns according to the present invention.

[0025] Figure 4 This is a schematic diagram of the process for nonlinear decoupling of passenger flow load and generation of residual waveforms in this invention;

[0026] Figure 5 This is a schematic diagram illustrating the process of constructing a hierarchical partitioned spatial logical index according to the present invention;

[0027] Figure 6 This is a schematic diagram of the cross-site working condition matching and experience retrieval process of the present invention. Detailed Implementation

[0028] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] See attached document Figure 1 , Figure 1 This is a flowchart of a logistics intelligent management optimization method integrating multi-source data analysis according to an embodiment of the present invention. The present invention provides a logistics intelligent management optimization method integrating multi-source data analysis, which is executed based on a railway logistics intelligent integrated management platform and includes the following steps:

[0030] S100 implements the acquisition of multi-source railway logistics data and semantic parsing based on operational scenarios: real-time acquisition of business text data entered by maintenance personnel and time-series data collected by sensors, and access to train timetables and real-time passenger flow density data of stations as environmental background parameters, while extracting target equipment and fault action words from the text.

[0031] S200 dynamically determines the effective physical time window of sensor data based on action vocabulary and railway passenger transport patterns: a pre-built action-time feature mapping rule base is constructed, and an event alignment algorithm based on train arrival and departure times is used to search for high-load periods before and after train arrival as the core analysis window to lock in data segments containing the complete fault evolution process.

[0032] S300 performs passenger flow load decoupling processing on the intercepted sensor data to generate waveform fingerprints that reflect the inherent health of the equipment: using the energy consumption-passenger flow normalization elimination algorithm, the basic load caused by passenger flow density is deducted from the actual operating data to obtain the residual waveform that is purely determined by the equipment performance, and then it is encoded to generate feature fingerprints.

[0033] S400 integrates equipment attributes and spatial logical relationships to construct a hierarchical and partitioned spatial logical index: it parses the building information model to establish a physical spatial tree structure, overlays a cross-regional electrical and signal logical relationship network, calculates the comprehensive proximity based on physical distance and logical hop count, and thus generates an inverted index library containing equipment spatial neighborhood relationships.

[0034] The S500 responds to external query requests, generates query fingerprints, and matches cross-site maintenance experience based on the index: it receives real-time data to generate fingerprints, performs cross-site matching of similar equipment operating conditions, retrieves historical maintenance records of equipment that have exhibited the same residual waveform under similar passenger flow and temperature conditions, and provides feedback.

[0035] The above steps will be explained in detail below with reference to specific embodiments.

[0036] See attached document Figure 2 , Figure 2 This is a flowchart illustrating the acquisition of multi-source railway logistics data and semantic parsing based on operational scenarios, according to an embodiment of the present invention.

[0037] Step S100 primarily relies on the algorithm processing layer of the railway logistics intelligent integrated management platform. This algorithm processing layer is deployed in a central server cluster and logically comprises: a data acquisition server for data access and distribution, a natural language processing unit for text parsing, a scene analysis unit for spatiotemporal data correlation, and a logical verification unit for rule determination. Each unit interacts with data through an internal message queue, specifically executing the following sub-steps:

[0038] S101, construct a multi-source heterogeneous data acquisition channel to acquire unstructured business text data, structured equipment monitoring data, and environmental background parameter data respectively.

[0039] The data acquisition server connects to the database of the railway logistics integrated management platform and the on-site IoT gateway via a data bus. For unstructured business text data, the data acquisition server retrieves maintenance records submitted by maintenance personnel through mobile work terminals. These records include natural language descriptions of fault phenomena, maintenance measures, and operation timestamps. For structured equipment monitoring data, the data acquisition server reads sensor values ​​deployed in various functional areas of the station at a preset sampling frequency. These sensor values ​​specifically include the voltage and current of the power supply circuit, the temperature and air volume of the HVAC equipment, and the pressure and flow of the water supply and drainage system.

[0040] As a key technical feature of this embodiment, the data acquisition server synchronously accesses railway-specific environmental background parameter data. This environmental background parameter data specifically includes train timetable data and real-time station passenger flow density data. The train timetable data includes train numbers, arrival and departure times, and station shutdown and maintenance windows; the real-time station passenger flow density data is provided by station entrance and exit gate counting or video surveillance analysis systems, and is used to characterize the real-time passenger load status of various areas of the station. The data acquisition server integrates these three types of data into a data preprocessing buffer, providing basic input for subsequent analysis.

[0041] S102, preprocesses unstructured business text data and extracts key semantic elements to establish a structured mapping between equipment entities and fault behaviors.

[0042] For the collected maintenance record text, the natural language processing unit performs text cleaning to remove meaningless stop words and special symbols. For routine processing such as word segmentation and part-of-speech tagging, those skilled in the art can use existing mature algorithm libraries. Based on this, the natural language processing unit utilizes a pre-built railway logistics professional dictionary to perform named entity recognition and relation extraction.

[0043] The natural language processing unit focuses on extracting two types of semantic elements and constructing structured fault vectors. The first type of element is the target equipment entity, that is, the specific physical object where the fault occurred. The natural language processing unit maps it to a unique equipment code in the equipment ledger. ;

[0044] The second category of elements consists of core action vocabulary, which are combinations of verbs or adjectives describing changes in equipment status. The natural language processing unit uses a thesaurus to standardize the descriptions given by maintenance personnel into standardized fault terms and extracts adverbs of degree that modify the actions.

[0045] To quantify the severity of faults, the natural language processing unit uses a weighted calculation formula to determine the fault level weight value. :

[0046] ;

[0047] In the formula, The base score represents the core action vocabulary. The base score is predefined in the fault terminology library (e.g., 1.0 for shutdown and 0.6 for abnormal noise).

[0048] The correction factor for adverbs of degree (e.g., 1.5 for severe, 0.8 for slight, and 1.0 for no adverb).

[0049] and These are the basic weighting coefficient for the action and the weighting coefficient for the degree correction, respectively. Finally, the natural language processing unit outputs a structured fault vector containing device codes, standard fault terms, and fault level weights, defined as follows:

[0050] ;

[0051] In the formula, This represents a unique device code extracted from text and mapped to the device ledger using entity recognition technology.

[0052] Standardized fault terminology after synonym standardization;

[0053] This represents the fault level weight value calculated in the preceding steps. Structured fault vector. As standardized semantic feature inputs, they are used for subsequent data fusion and retrieval matching.

[0054] S103, based on train timetables and passenger flow density data, performs operational scenario labeling on structured equipment monitoring data.

[0055] The scenario analysis unit reads the time-series data collected by the sensors and correlates and matches it with the train timetable and passenger flow density data based on the timestamp. The scenario analysis unit has preset a variety of operating scenario modes, including high passenger flow and high load mode, daily off-peak mode, nighttime low-peak mode, and shutdown and maintenance mode.

[0056] The scenario analysis unit determines the current operational phase of the station based on the train arrival and departure timetable. To accurately determine high-passenger-flow, high-load patterns, the scenario analysis unit first calculates the passenger flow density threshold based on station architectural design specifications. The calculation formula is:

[0057] ;

[0058] In the formula, This indicates the design limit capacity density of the waiting area of ​​the station, which is derived from the station's architectural design blueprint data.

[0059] This represents the congestion saturation coefficient (recommended value is 0.75 to 0.85), used to define the critical point at which a high-load state is reached.

[0060] The scene analysis unit obtains the current passenger flow density monitoring value in real time. If the current time falls during a peak train arrival and departure period and meets the following conditions...

[0061] The scenario analysis unit marks the equipment monitoring data for that period as a high-passenger-flow, high-load mode label; if the current time is within the maintenance window period specified in the train timetable, the scenario analysis unit marks the data for that period as a shutdown and maintenance mode label.

[0062] To facilitate subsequent processing of the computer model, the scene analysis unit encapsulates the above-mentioned labeling results into a scene context vector. :

[0063] ;

[0064] In the formula, Indicates the specific timestamp of the data sampling;

[0065] This indicates the station's operational phase code as determined by the train timetable (e.g., 1 represents operation, 0 represents a shutdown window).

[0066] This represents the normalized real-time passenger flow density value, calculated as follows: By constructing this vector, the scenario analysis unit endows the simple physical quantity values ​​with structured semantic attributes of railway business scenarios, clarifies the external operating conditions of the equipment, and provides a calculation benchmark for distinguishing between normal high load and abnormal faults.

[0067] S104 performs compliance checks based on the three-dimensional logic of people, place, and event, and cleans up abnormal and non-compliant business records.

[0068] The logic verification unit extracts the operator's identity information, work location information, and work time information from the business text data. It then calls the data from the scheduling management system to verify whether the operator is on duty during the specified work hours. Simultaneously, the logic verification unit performs logic conflict detection in conjunction with train timetable data.

[0069] Specifically, the logic verification unit determines whether the maintenance activity shown in the business record occurred within the permitted operation time window. For equipment maintenance records involving train safety areas, if the operation time does not fall within the time window specified in the train timetable, or if the area is in a high-risk period for trains entering or leaving the station when the operation occurs, the logic verification unit determines that the business record violates physical and safety logic. For records that fail such logic verification, the logic verification unit marks them as invalid data and removes or isolates them to prevent invalid operation records from interfering with subsequent fault model analysis, thereby ensuring the authenticity and validity of the data entered into the database.

[0070] See attached document Figure 3 , Figure 3 This is a flowchart illustrating the process of dynamically determining the effective physical time window of sensor data based on action words and railway passenger transport patterns according to an embodiment of the present invention.

[0071] This step S200 relies on the time window locking unit in the algorithm processing layer. The time window locking unit maintains communication with the natural language processing unit in the preceding steps and the database storing the train timetable, and locks the physical time window through the following sub-steps:

[0072] S201, Load the action-time feature mapping rule base and determine the fault backtracking search domain based on the standard fault terms in the structured fault vector.

[0073] The time window locking unit is pre-configured with an action-time-effect feature mapping rule base. This rule base is a set of key-value pairs built based on statistical analysis of historical maintenance data and prior knowledge from railway equipment experts. It stores the time-series delay characteristic relationship between different equipment fault actions and passenger flow load. Specifically, the action-time-effect feature mapping rule base defines the maximum delayed response time for various fault actions. For example, for electrical faults, the maximum delayed response time. Set as a short time span value; for temperature- or pressure-related cumulative faults, its maximum delayed response time It is set as a long time span value.

[0074] The time window locking unit receives and parses the structured fault vector output in step S100. Extracting structured fault vectors Standard fault terminology The time window locking unit retrieves standard fault terms from the action-time feature mapping rule base. For the matched records, obtain the corresponding maximum latency response time. Based on maximum latency response time The time window locking unit uses the timestamp of the maintenance personnel submitting the repair request. As the endpoint, construct the fault backtracking search domain. Fault backtracking search domain The time range is defined as:

[0075] ;

[0076] In the formula, This indicates that the time window locking unit is based on the timestamp of the maintenance personnel submitting the repair request.

[0077] S202 executes an event alignment algorithm based on train arrival and departure times to locate the reference train event anchor point within the fault backtracking search domain.

[0078] To eliminate the randomness of manual repair reporting times, the time window locking unit uses train events that cause changes in equipment load as a physical benchmark. The time window locking unit reads train timetable data and traverses the fault backtracking search domain. .

[0079] The time window locking unit filters out the fault backtracking search domain. The set of all trains arriving or departing within the range. The time window locking unit calculates the time of each train event in the set. Repair time The absolute value of the time difference, combined with the weight of the train formation's passenger capacity. The train time with the smallest weighted time distance is selected as the benchmark train event anchor point. .

[0080] To clarify the calculation method of the weights, this embodiment uses the following preferred calculation formula to determine the benchmark train event anchor point. :

[0081] ;

[0082] And the weight of passenger capacity in train formation The calculation formula is:

[0083] ;

[0084] In the formula, Indicates the final determined anchor point time of the benchmark train event;

[0085] Indicates the fault backtracking search domain Inner The arrival and departure times of the train;

[0086] This indicates the timestamp when the maintenance personnel submitted the repair request;

[0087] This represents the fault backtracking search domain determined in step S201;

[0088] Indicates the first The passenger load weighting coefficient for each train;

[0089] Indicates the first The rated passenger capacity of a train;

[0090] This indicates the maximum passenger capacity of a single train formation allowed by the station's design. By introducing passenger capacity weighting, the time window locking unit can prioritize locking train operation events that are close to the repair request time and have a significant impact on equipment load.

[0091] S203 generates dynamic offsets based on the physical attributes of the equipment, and constructs an effective physical time window that includes the complete fault evolution process.

[0092] Determine the benchmark train event anchor point Then, the time window locking unit calculates the preceding offset based on the physical properties of the target device. and post offset The time window locking unit accesses the device attribute database and reads the preset load response parameters corresponding to the target device type. (Previous offset) The preset duration corresponds to the period of increased equipment load, used to cover the stage of personnel gathering or equipment pre-start before train arrival; Post-offset amount A preset duration corresponding to the equipment load recovery period is used to cover the passenger evacuation or equipment cooling recovery phase. The time window locking unit utilizes a preceding offset. and post offset Calculate and generate the final effective physical time window. This effective physical time window The formula for calculating the start and end times is:

[0093] ;

[0094] In the formula, Indicates the final effective physical time window; The reference train event anchor point determined in step S202; The preset load increase period duration is based on the equipment type; The preset load recovery period duration based on the equipment type.

[0095] S204: Extract raw sensor data segments based on the effective physical time window and complete data integrity verification.

[0096] The time window locking unit sends a data extraction request to the time series database. The data extraction request includes the target device code and the calculated start and end times of the valid physical time window. The time series database returns the data sequence of all sampling points within that time period.

[0097] The time window locking unit performs integrity checks on the returned data sequence, checking for any data loss or discontinuities. Specifically, the preset threshold is configured as the minimum effective sampling rate standard to ensure that the time series data can accurately reflect the equipment's operating trend. In practical applications, this threshold is typically set to 15% to 20% of the theoretical total number of sampling points within the window. This value is based on the fact that when the amount of missing data is controlled within this range, the data supplemented by linear interpolation can still maintain the original signal characteristics without significant distortion; however, once this ratio is exceeded, the supplemented data will lose statistical reliability and cannot accurately reflect the actual response of the equipment during train operations.

[0098] If the proportion of missing data in the data sequence is lower than the preset threshold, the time window locking unit uses a linear interpolation algorithm to fill in the missing data. If the proportion of missing data is higher than the preset threshold, the time window locking unit marks the valid physical time window as invalid and generates a data anomaly log. Finally, the time window locking unit outputs the locked original sensor data segment, which retains the equipment operating status information associated with the specific train operation event.

[0099] See attached document Figure 4 , Figure 4 This is a schematic diagram of the process for nonlinear decoupling of passenger flow load and generation of residual waveforms according to an embodiment of the present invention.

[0100] This step S300 primarily relies on the load decoupling unit in the algorithm processing layer. The load decoupling unit is connected to both the time-series database and the time window locking unit from the preceding step, aiming to separate characteristic waveforms reflecting the internal state of the equipment from sensor data superimposed with environmental background noise. By eliminating the interference of passenger flow fluctuations on the equipment operating baseline, this step S300 can extract fault characteristics from the background load. Specifically, the following sub-steps are executed:

[0101] S301 performs frequency alignment and interpolation synchronization processing for heterogeneous time-series data.

[0102] The load decoupling unit reads the raw sensor data segment output in step S200. In addition, there is passenger flow density data for the same period. Because the sampling frequency of sensor data is higher than the update frequency of passenger flow density data, there is a granularity mismatch between the two in the time dimension. The load decoupling unit uses a cubic spline interpolation algorithm to upsample the low-frequency passenger flow density data, constructing a synchronized passenger flow density sequence that is strictly aligned with the timestamps of the sensor data. .

[0103] To address boundary effects during the interpolation process, the load decoupling unit utilizes historical data points before and after the time window as boundary constraints to ensure the generated synchronous passenger flow density sequence is optimized. The first derivative is kept continuous at the beginning and end of the time window. This step ensures that each sensor sampling point has a corresponding, continuously changing background value of passenger flow load in subsequent calculations, providing a data foundation for point-to-point load calculation.

[0104] S302, Construct a baseline model of passenger flow-equipment load response based on multinomial regression, and calculate the theoretical background load.

[0105] The load decoupling unit pre-builds and stores a passenger flow-equipment load response baseline model for the target equipment. To ensure the model's effectiveness, the load response baseline model is trained offline. Specifically, the load decoupling unit selects historical operation and maintenance data from the historical database that are marked as "normal" and cover the entire range from zero load to full load as the training sample set. The load decoupling unit uses historical passenger flow density as the input feature and the corresponding historical equipment monitoring physical quantities (such as current and temperature) as the output label, and uses the least squares algorithm to fit the nonlinear mapping relationship between the two to determine the regression coefficients.

[0106] Considering that equipment load in railway scenarios typically exhibits a non-linear, saturated growth trend with increasing passenger flow, this embodiment employs a high-order polynomial function to characterize this physical law. During the real-time processing phase, the load decoupling unit processes the synchronous passenger flow density sequence generated in step S301. Input the data into the trained passenger flow-equipment load response baseline model to calculate the theoretical background load at each time step. Its calculation formula is defined as:

[0107] ;

[0108] In the formula, Indicates in The theoretical health load value of the equipment is predicted based on the current passenger flow density at all times;

[0109] express Real-time passenger flow density values;

[0110] This embodiment preferably represents the order of the polynomial. To fit the nonlinear characteristics of passenger flow from sparse to crowded;

[0111] Indicates the index of the current term in the polynomial;

[0112] Indicates the first The regression coefficients of order are obtained by fitting the data through the aforementioned offline training process, where... This is a constant bias term, representing the basic operating load of the equipment under no-load conditions.

[0113] S303 performs differential operations to generate the original residual sequence and filters out environmental load interference.

[0114] The load decoupling unit locks the original sensor data segment in step S204. The theoretical background load calculated in step S302 Point-to-point differential operations are performed. The physical meaning of this operation is that the raw sensor data is considered a superposition signal of equipment fault components and passenger flow environment components, while the theoretical background load characterizes the passenger flow environment component. Through the subtraction operation, the system can filter out baseline drift caused by normal passenger flow fluctuations from the total signal. The load decoupling unit outputs the original residual sequence. The calculation formula is as follows:

[0115] ;

[0116] In the formula, Indicates at a point in time The residual amplitude;

[0117] This indicates the actual physical quantity value collected by the sensor;

[0118] This represents the theoretical health value estimated based on the current passenger flow. If the equipment is in a healthy state, this original residual sequence... It exhibits random fluctuations around zero; if the equipment has a potential fault, this original residual sequence... This will include outlier components that deviate from zero.

[0119] S304 performs a standardized Z-Score transform on the original residual sequence to generate standard fault characteristic waveforms.

[0120] Because different types of equipment have vastly different physical quantities and absolute values, directly using the original residual amplitude is not conducive to subsequent unified fault classification. The load decoupling unit uses the original residual sequence... Standardization is performed to eliminate the influence of dimensions. To prevent outliers within the current window from affecting the statistical distribution, the load decoupling unit retrieves historical normal operating baseline statistics of the equipment stored in the database and performs the following transformation to generate standard fault characteristic waveforms. :

[0121] ;

[0122] In the formula, This represents the dimensionless standard fault characteristic waveform of the final output;

[0123] This represents the original residual sequence calculated in step S303;

[0124] This represents the baseline arithmetic mean of the residual sequence of the device under historical normal operating conditions;

[0125] This represents the baseline standard deviation of the residual sequence of the device under historical normal operating conditions;

[0126] For a preset minimum regularization constant (e.g.) ), used to prevent calculation overflow when the denominator approaches zero.

[0127] Through the above processing, the load decoupling unit outputs... The abnormal state of the equipment is converted into a deviation multiple relative to the normal noise level. This standard fault characteristic waveform eliminates baseline interference caused by passenger flow fluctuations and maps the fault characteristics of equipment with different power levels to the same scale space, providing standardized input data for subsequent fault diagnosis.

[0128] See attached document Figure 5 , Figure 5 This is a schematic diagram of the process for constructing a hierarchical partitioned spatial logical index according to an embodiment of the present invention.

[0129] In one embodiment of the present invention, step S400 is primarily executed by the spatial relationship construction unit in the algorithm processing layer. The spatial relationship construction unit is connected to both the basic geographic information database and the equipment asset management system. By establishing an index system that maps physical space to system logical relationships, it achieves precise location of the fault impact domain. Specifically, the following sub-steps are executed:

[0130] S401, analyze the building information model and establish a multi-level physical space tree topology.

[0131] The spatial relationship construction unit reads the station's Building Information Model (BIM) data file. It then parses the IFC standard format data from the BIM model, extracting spatial attribute information of building components. Based on the physical zoning standards of railway passenger stations, the spatial relationship construction unit constructs a physical spatial tree topology. This tree topology defines three levels of logical nodes: the root node corresponds to the entire station; second-level nodes correspond to functional areas, including waiting halls, ticket offices, security checkpoints, platform levels, and equipment rooms; and leaf nodes correspond to specific end-point monitoring equipment.

[0132] The spatial relationship building unit traverses all leaf nodes, calculates and extracts the coordinates of each device entity at the 3D geometric center of the BIM model. Simultaneously, the spatial relationship construction unit maps each device to a unique secondary functional area node based on its installation location attribute, forming a physical affiliation. This hierarchical structure discretizes the physical location of devices, ensuring that any device can be traced back to its functional area through a tree-like path.

[0133] S402 is a system logic association network that is superimposed across regions based on the functional coupling relationship of devices.

[0134] Since a single physical partition cannot represent the system linkage characteristics across regions, the spatial relationship building unit is based on the tree-like topology of physical space and superimposed with a logically related network.

[0135] The spatial relationship construction unit reads electrical wiring diagrams and signal flow diagrams from the equipment asset management system. For two equipment nodes connected by direct electrical loops, signal control links, or physical conduits, regardless of whether they belong to the same physical functional area, the spatial relationship construction unit establishes a logical connection edge between them. This logical connection edge is stored as an undirected graph data structure, enabling the spatial index to include system-level topological attributes between devices, thus covering cross-regional cascading reactions caused by power outages or network failures.

[0136] S403 quantizes the spatial-logical proximity between devices and generates a weighted adjacency matrix.

[0137] To provide quantified search priorities during fault backtracking, the spatial relationship building unit calculates the comprehensive proximity between each pair of device nodes and constructs a weighted adjacency matrix.

[0138] The spatial relationship construction unit utilizes a pre-defined computational model, combining physical Euclidean distance and logical connection depth, to calculate device nodes. With device nodes Comprehensive correlation weight between The calculation formula is as follows:

[0139] ;

[0140] In the formula, Indicates equipment With equipment The overall correlation weight between them;

[0141] This represents the physical Euclidean distance between the two devices in three-dimensional space, calculated based on the coordinates extracted in step S401.

[0142] The base of the natural logarithm;

[0143] This represents the distance attenuation constant, which is set to 15 meters in this embodiment to control the range of influence of physical distance.

[0144] This represents the shortest path hop count in the logical association network constructed in step S402 between the two devices. If the two devices are not logically connected, then... Take infinity;

[0145] and These are the physical proximity weight coefficient and the logical association weight coefficient, respectively. .

[0146] To achieve precise location of different fault types, the spatial relationship construction unit is pre-configured with a fault type-weight mapping table. When dealing with physical diffusion faults such as fires or water immersion, the spatial relationship construction unit looks up the table to set the parameters. , When handling logic propagation faults such as network communication or circuit interruptions, the spatial relationship construction unit is configured by looking up a table. , .

[0147] S404 generates a hierarchical partitioned spatial inverted index and completes data storage.

[0148] The spatial relationship construction unit constructs a spatial inverted index table based on the weighted adjacency matrix generated in step S403. This spatial inverted index table uses the unique identifier of each device as the key and the list of adjacent devices of that device as the value. The devices in the adjacent device list are arranged according to a comprehensive association weight. Sort from highest to lowest.

[0149] The spatial relationship construction unit serializes and stores the constructed physical spatial tree topology, system logical relationship network, and spatial inverted index table into the graph database. When the system subsequently receives a faulty device entity extracted by the natural language processing unit, the spatial relationship construction unit queries the graph database to return the physical area information where the device is located, as well as a list of affected devices sorted by correlation, thereby providing clear spatial logical guidance for fault location and troubleshooting.

[0150] See attached document Figure 6 , Figure 6 This is a schematic diagram of the cross-site working condition matching and experience retrieval process according to an embodiment of the present invention.

[0151] This step, S500, relies on the knowledge retrieval unit in the application service layer. The knowledge retrieval unit maintains bidirectional communication with the operation and maintenance knowledge graph database and the real-time monitoring system, aiming to leverage historical operation and maintenance experience across the entire road network to resolve current sudden faults. This step achieves cross-regional and cross-environmental analogical reasoning by constructing a multi-dimensional operating condition feature space. Specifically, the following sub-steps are executed:

[0152] S501, construct a multi-dimensional working condition feature vector that includes static attributes and dynamic environment.

[0153] To ensure comparability between different stations, the knowledge retrieval unit extracts features from the station and its equipment environment where a fault has occurred, generating a multi-dimensional operating condition feature vector. This multi-dimensional operating condition feature vector consists of both static inherent attribute components and dynamic operating environment components.

[0154] The static inherent attributes section describes the basic configuration information of the station and equipment, specifically including: the geographical and climatic zone where the station is located, the line grade, and the specific brand, model, and firmware version of the faulty equipment. The geographical and climatic zone determines the aging baseline of the equipment, and the line grade determines the maintenance standards of the equipment.

[0155] The dynamic operating environment section describes the transient external conditions at the time of the failure, specifically including: the current passenger congestion level, the current indoor and outdoor temperature and humidity values, and the continuous operating time of the equipment before the failure. The knowledge retrieval unit performs numerical encoding or one-hot encoding on the above heterogeneous information and integrates it into a standardized feature description sequence, which serves as the query benchmark for subsequent similarity calculations.

[0156] S502, calculate the working condition similarity of historical cases based on a weighted attribute matching strategy.

[0157] The knowledge retrieval unit traverses the historical fault cases stored in the operations and maintenance knowledge graph database. In order to filter out cases with reference value from massive historical data, the knowledge retrieval unit performs a similarity calculation process based on a weighted attribute matching strategy.

[0158] The knowledge retrieval unit uses equipment type and core component model as hard filtering conditions to eliminate all historical records of non-similar equipment. For the selected similar equipment cases, the knowledge retrieval unit calculates the weighted similarity between the operating condition features in its historical records and the current multi-dimensional operating condition feature vector. The calculation formula is as follows:

[0159] ;

[0160] In the formula, This represents the similarity score between the current fault scenario and a historical case, with a value range of [value missing]. ;

[0161] This represents the total number of dimensions of the feature vector;

[0162] Indicates the first The weighting coefficients for each feature dimension are dynamically set based on the device type (e.g., for outdoor devices, the weights for temperature and humidity dimensions are...). Set a larger value; for high-load equipment, the weight of the passenger flow level dimension. (Set a larger value);

[0163] Represents the th element in the current operating condition feature vector. The values ​​of each dimension;

[0164] Indicates the first historical case The values ​​of each dimension;

[0165] Indicates the first The maximum possible range of values ​​for each feature dimension is used for normalization.

[0166] S503, retrieve the associated repair solutions and perform path traversal in the knowledge graph.

[0167] The knowledge retrieval unit uses the standard fault terms extracted in step S100 as semantic indexes, and combines them with the highly similar historical case IDs selected in step S502 to perform retrieval in the operation and maintenance knowledge graph.

[0168] The operations and maintenance knowledge graph is stored with the core link of fault phenomenon-operating condition background-fault cause-handling solution. The knowledge retrieval unit performs path walking operations in the graph: the knowledge retrieval unit locates the entity node that is consistent with the current standard fault terminology, filters out the intermediate nodes with highly similar operating condition backgrounds along the edge relationships, and traces to the corresponding fault cause entity and handling solution entity.

[0169] The knowledge retrieval unit extracts detailed information about the found solutions, including: the sequence of troubleshooting steps, a list of required spare parts, special tools needed for repair, and unstructured notes left by technical experts when handling the case.

[0170] S504 generates a recommendation list based on case similarity and the effectiveness of historical handling.

[0171] To assist maintenance personnel in making rapid decisions, the knowledge retrieval unit ranks and optimizes multiple potential solutions retrieved. The knowledge retrieval unit comprehensively considers the operational similarity score calculated in step S502 and the feedback on the effectiveness of solutions from historical case records to calculate a recommendation score for each solution. The calculation formula is as follows:

[0172] ;

[0173] In the formula, This indicates the overall recommended score for the repair plan;

[0174] The similarity of working conditions calculated in step 5502;

[0175] This indicates the number of times the solution has been marked as a "one-time fix" in historical applications;

[0176] This indicates the total number of times the scheme has been cited throughout history;

[0177] This characterizes the historical success rate of the repair scheme;

[0178] This is a balancing factor used to adjust the weight ratio between similarity and success rate. In this embodiment, the default value is 0.6.

[0179] Knowledge retrieval units are based on recommendation scores A structured maintenance assistance report is generated from highest to lowest priority. This report lists possible causes of failure and their corresponding troubleshooting methods in order of priority, and presents similar historical cases as supporting evidence. If the search results are empty or the highest recommended score is below the preset reference lower limit, the knowledge retrieval unit generates a request for human expert intervention and stores the current operating condition feature vector as the initial data for the new case in the processing queue, enabling dynamic expansion of the knowledge base.

Claims

1. A method for optimizing intelligent logistics management by integrating multi-source data analysis, characterized in that, This process, implemented using the railway logistics intelligent integrated management platform, includes the following steps: Acquire business text data, structured equipment monitoring data, and environmental background parameters including train operation diagrams and passenger flow density; perform semantic parsing on the business text data to extract equipment entities and fault action words. The search domain is determined by using a rule base for mapping action and timeliness features, and an effective physical time window containing the fault evolution process is locked by an event alignment algorithm based on train arrival and departure times. The structured equipment monitoring data within the effective physical time window is subjected to passenger flow load decoupling processing, and the basic load is removed using the passenger flow and equipment load response baseline model to generate residual waveforms; The building information model is analyzed to establish a physical spatial tree structure and overlay a logical association network. The comprehensive proximity is calculated based on physical distance and logical hop count to generate a spatial inverted index. In response to external query requests, cross-site matching of similar equipment is performed based on multi-dimensional operating condition feature vectors, and historical maintenance records are retrieved based on the residual waveform to generate a maintenance auxiliary report.

2. The logistics intelligent management optimization method integrating multi-source data analysis according to claim 1, characterized in that, The steps of acquiring business text data and performing semantic parsing specifically include: Natural language processing techniques are used to extract the target device entity from the business text data and map it to a unique device code. The action descriptions in the business text data are converted into standardized fault terms using a thesaurus, and degree adverbs modifying the actions are extracted. The fault level weight value is calculated based on the base score of the fault term and the correction coefficient of the degree adverb, and a structured fault vector is output.

3. The logistics intelligent management optimization method integrating multi-source data analysis according to claim 1, characterized in that, The specific steps for tagging the structured equipment monitoring data with work scenarios include: The passenger flow density threshold is calculated based on the station building design code. The passenger flow density threshold is the product of the design limit capacity density and the crowding saturation coefficient. If the real-time passenger flow density monitoring value is greater than or equal to the passenger flow density judgment threshold and the current time is within the dense train arrival and departure period defined in the train operation diagram, the structured equipment monitoring data will be marked as a high passenger flow load mode. Generate a scene context vector that includes data sampling timestamps, station operation stage codes, and normalized passenger flow density values.

4. The logistics intelligent management optimization method integrating multi-source data analysis according to claim 1, characterized in that, In the step of locking the effective physical time window, locating the reference train event anchor point specifically includes: Based on the standardized fault terminology, the maximum delayed response time is searched in the action and timeliness feature mapping rule base, and a fault backtracking search domain is constructed with the repair timestamp as the endpoint. Calculate the absolute value of the time difference between the event time and the repair reporting time for each train in the set of train numbers within the fault backtracking search domain; The effective physical time window is generated by introducing the passenger capacity weight of train formations, selecting the train time with the smallest weighted time distance as the benchmark train event anchor point, and combining the preceding offset and following offset determined by the physical attributes of the equipment.

5. The logistics intelligent management optimization method integrating multi-source data analysis according to claim 1, characterized in that, In the process of decoupling passenger flow load, calculating the theoretical background load specifically includes: Interpolate the passenger flow density data to construct a synchronized passenger flow density sequence aligned with the timestamps of the structured equipment monitoring data; The baseline model of passenger flow and equipment load response is trained in advance using historical operation and maintenance data, and a multinomial regression algorithm is used to fit the nonlinear mapping relationship between passenger flow density and equipment monitoring physical quantities; The synchronous passenger flow density sequence is input into the passenger flow and equipment load response baseline model to calculate the theoretical background load at each time point.

6. The logistics intelligent management optimization method integrating multi-source data analysis according to claim 5, characterized in that, In the process of decoupling passenger flow load, generating the residual waveform specifically includes: The structured equipment monitoring data within the effective physical time window is compared with the theoretical background load using a point-to-point differential operation to generate the original residual sequence. Retrieve the baseline arithmetic mean and baseline standard deviation of the residual sequence from the device's historical normal operating conditions; The original residual sequence is subjected to a standardization transformation, which converts the original residual amplitude into a deviation value relative to the normal noise level, generating a dimensionless standard fault characteristic waveform.

7. The logistics intelligent management optimization method integrating multi-source data analysis according to claim 1, characterized in that, In the step of constructing a hierarchical partitioned spatial logical index, establishing a physical spatial tree structure and an overlay logical association network specifically includes: The three-dimensional geometric center coordinates of the equipment are extracted by parsing the building information model, and a physical space tree-like topology structure containing the site, functional area and equipment entity is constructed according to the physical zoning standard. Read electrical wiring diagram data and signal flow diagram data, establish logical connections between device nodes that have direct electrical connection loops, signal control links, or physical pipe connections, and form a logical association network with an undirected graph structure.

8. The logistics intelligent management optimization method integrating multi-source data analysis according to claim 7, characterized in that, In the step of constructing a hierarchical spatial logical index, calculating the comprehensive proximity specifically includes: Calculate the physical Euclidean distance between two device nodes in three-dimensional space, and calculate the physical proximity component based on the distance decay constant; Calculate the shortest path hop count between the two device nodes in the logical association network, and calculate the logical association component based on the reciprocal of the hop count; Based on the fault type, the physical proximity weight coefficient and logical association weight coefficient are obtained by looking up the table. The physical proximity component and the logical association component are then weighted and summed to obtain the comprehensive association weight and generate a weighted adjacency matrix.

9. The logistics intelligent management optimization method integrating multi-source data analysis according to claim 1, characterized in that, In the steps of responding to external query requests, performing cross-site matching of similar devices specifically includes: The multidimensional operating condition feature vector is constructed, which includes a static inherent attribute part describing the station's geographical and climatic zones and equipment models, and a dynamic operating environment part describing the passenger flow congestion level, temperature and humidity values, and continuous operating time of the equipment. Historical failure cases of similar equipment are filtered by equipment type, and the weighted similarity between the current multidimensional operating condition feature vector and the operating condition feature vector of historical failure cases is calculated.

10. The logistics intelligent management optimization method integrating multi-source data analysis according to claim 9, characterized in that, In the step of responding to external query requests, generating a maintenance support report specifically includes: By using standardized fault terminology as a semantic index and combining it with selected historical fault cases, the corresponding fault cause entities and handling solution entities are retrieved in the operation and maintenance knowledge graph. Calculate the comprehensive recommendation score for each treatment plan, which is composed of the weighted similarity and the historical repair success rate of the treatment plan. The maintenance assistance report is generated in descending order of the comprehensive recommendation score, and includes troubleshooting steps and supporting historical cases.

Images