A smart logistics scheduling method and system based on big data analysis

Abstract

The application discloses a kind of wisdom logistics scheduling method and system based on big data analysis, it is related to big data analysis technical field, including: executing multi-source data acquisition preprocessing, generates original logistics state data;Based on Hampel filtering and spatial consistency discrimination, generate reliable logistics state record;Carry out space-time alignment and task binding, generate multi-source reliable logistics state frame;Input improved FEDformer prediction, generate logistics situation prediction result;Execute risk feature extraction, generate scheduling risk state table;Carrying out candidate combination and execution state marking, generate scheduling constraint mask table;Based on candidate screening and range limitation, generate candidate scheduling feasible region;Execute scheduling plan generation, output wisdom logistics scheduling result.The application can improve multi-source logistics state identification accuracy, disturbance response stability and scheduling execution continuity.

Description

Technical Field

[0001] This invention relates to the field of big data analytics, and in particular to a smart logistics scheduling method and system based on big data analytics. Background Technology

[0002] As logistics order volume expands, delivery timeliness requirements increase, and transportation resource types diversify, intelligent logistics scheduling is gradually shifting from manual order dispatch and fixed-rule scheduling to data-driven scheduling. Existing logistics scheduling methods primarily rely on order management platforms, transportation management platforms, warehouse management platforms, and vehicle positioning platforms to collect operational data and generate delivery tasks based on route planning, vehicle matching, and timeliness rules, enabling them to complete order allocation, route arrangement, and departure plans in typical scenarios.

[0003] Existing technologies still have shortcomings in processing multi-source logistics operation data and making dynamic scheduling decisions. Order, vehicle, warehouse, road, environmental, and fulfillment data sources are scattered, with differences in temporal granularity, spatial granularity, and data reliability. Vehicle positioning drift, warehouse scanning delays, and road status lags can affect scheduling status judgments. Existing filtering methods often perform anomaly replacement on single time series, lacking spatial proximity consistency judgment, easily treating real road congestion, warehouse delays, and order surges as data noise. Existing predictive scheduling methods mostly focus on single-point prediction or static rule adjustments, lacking joint modeling of causal propagation between logistics objects, perturbation fidelity representation, and prediction reliability calibration. After dynamic perturbations occur, existing methods tend to use global rearrangement or local patching, making it difficult to form stable candidate scheduling feasible regions based on scheduling risk states and logistics execution constraints, resulting in insufficient continuity of scheduling results.

[0004] Therefore, how to provide a smart logistics scheduling method and system based on big data analysis is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a smart logistics scheduling method and system based on big data analysis. This invention utilizes the spatial proximity consistency Hampel filtering algorithm and the improved FEDformer model to perform reliable processing, logistics situation prediction, and scheduling constraint screening on multi-source logistics operation data, and completes vehicle allocation, delivery sequence, route execution plan, and rescheduling record generation. It has the advantages of accurate state identification, stable disturbance response, and high scheduling execution continuity.

[0006] According to an embodiment of the present invention, a smart logistics scheduling method based on big data analysis includes: Collect multi-source logistics operation data and perform preprocessing to generate raw logistics status data; Based on the original logistics status data, the Hampel filtering algorithm is executed, spatial proximity consistency constraints are introduced, window scale selection, credibility threshold correction and spatial proximity consistency discrimination are performed to generate a reliable logistics status record. Spatiotemporal alignment, spatial node mapping, and task status binding are performed on trusted logistics status records to generate multi-source trusted logistics status frames. An improved FEDformer model is constructed. Multi-source reliable logistics state frames are input into the improved FEDformer model. After processing by the perturbation-fidelity dual-track decomposition and reconstruction layer, the logistics heterogeneous causal spectrum interaction layer, and the joint feature reliable frequency band calibration layer, logistics situation prediction results are generated. Based on the logistics situation prediction results, extract logistics operation risk characteristics and generate a scheduling risk status table; Based on the scheduling risk status table, candidate combinations of order-vehicle-route are generated, and the execution status is marked in combination with logistics execution constraints to generate a scheduling constraint mask table. Based on the scheduling constraint mask table, candidate combination screening and adjustment range limitation are performed to generate candidate scheduling feasible domains; Based on the candidate scheduling feasible domain, vehicle allocation results, delivery sequence, route execution plan, departure plan and rescheduling record are generated, and the intelligent logistics scheduling result is output.

[0007] Optionally, generating the raw logistics status data includes: The system reads order operation records, vehicle operation records, warehousing operation records, road traffic records, environmental impact records, and historical performance records from the logistics business platform to generate multi-source logistics operation data. Standardized logistics operation data is generated by unifying field names, units of measurement, and time formats for multi-source logistics operation data. Perform duplicate record removal, missing field marking, and abnormal format removal on standard logistics operation data to generate clean logistics operation data; Write data source markers, collection time markers, and business object markers into the cleaning logistics operation data to generate raw logistics status data.

[0008] Optionally, generating a reliable logistics status record includes: Read the field types, field values, data source markers, collection time markers, and business object markers from the raw logistics status data, configure the Hampel filter window scale according to the field types, and generate a field-level window sequence. The absolute deviation of the window median and the middle median is calculated based on the field-level window sequence. The data source credibility is generated by combining the data source label and the data source credibility is used to correct the anomaly judgment threshold and generate the field-level anomaly threshold. Anomaly detection is performed on field values ​​based on field-level anomaly thresholds to generate initial anomaly markers; For the business object corresponding to the initial anomaly label, retrieve the spatial neighboring objects, calculate the same-direction change status of the spatial neighboring objects, and generate a spatial consistency score; The initial anomaly markers are constrained and verified based on the spatial consistency score, which distinguishes between data noise and real logistics disturbances. Correction values ​​are generated for data noise, and disturbance markers are generated for real logistics disturbances. The correction values, disturbance markers, and spatial consistency scores are written into the trusted logistics status record.

[0009] Optionally, generating a multi-source trusted logistics status frame includes: Read the collection time stamp, business object stamp, correction value, disturbance stamp and spatial consistency score from the trusted logistics status record, perform time merging according to the unified scheduling time window, and generate time-aligned logistics status record; Based on time-aligned logistics status records, business object tags and spatial location fields are read, and order objects, vehicle objects, warehouse objects, road objects, and delivery area objects are mapped to unified logistics spatial nodes to generate spatial node mapping records. Based on the disturbance markers and spatial consistency scores in the spatial node mapping records and trusted logistics status records, establish the status association between nodes and generate logistics node association records; Based on the logistics node association records, read the task execution stage, vehicle carrying status and order fulfillment status, bind the task status, and generate task-bound logistics status records; The time-aligned logistics status record, spatial node mapping record, logistics node association record, and task-bound logistics status record are fused together to generate a multi-source trusted logistics status frame.

[0010] Optionally, the generation of logistics situation prediction results includes: An improved FEDformer model is constructed, which includes a perturbation-fidelity dual-track decomposition and reconstruction layer, a logistics heterogeneous causal spectrum graph interaction layer, and a joint feature reliable frequency band calibration layer. Read the time-aligned logistics status record, spatial node mapping record, logistics node association record and task-bound logistics status record from the multi-source trusted logistics status frame, and generate the model input sequence according to the prediction time window; The model input sequence is input into the perturbation-fidelity dual-track decomposition and reconstruction layer. The normal state input track is constructed based on the correction value, and the perturbation dynamic input track is constructed based on the perturbation label and spatial consistency score, generating normal state logistics representation and perturbation dynamic logistics representation. Normal logistics representation and disturbed logistics representation are input into the interaction layer of the heterogeneous causal spectrum graph of logistics. A spectrum interaction graph is constructed based on order nodes, vehicle nodes, warehouse nodes, road nodes and delivery area nodes. Causal spectrum interaction representation is generated according to the logistics node association records. The causal spectrum interaction representation is input into the joint feature trusted frequency band calibration layer, and trusted frequency band representation is generated by combining data source labeling, spatial consistency scoring and frequency band residual records. Predictive decoding is performed based on reliable frequency band representation to generate logistics situation prediction results; The improved FEDformer model was trained by using a combination of logistics situation prediction error, disturbance reconstruction error, causal spectrum interaction error, and reliable frequency band calibration error as the joint optimization objective. The network parameters of the disturbance fidelity dual-track decomposition reconstruction layer, the logistics heterogeneous causal spectrum interaction layer, and the joint feature reliable frequency band calibration layer were continuously optimized. When the change of the joint loss value in five consecutive training rounds was less than 0.001, the improved FEDformer model was considered to have completed convergence training.

[0011] Optionally, generating the scheduling risk status table includes: Read the prediction time window, node prediction value, disturbance impact marker and prediction confidence from the logistics situation prediction results, and generate prediction result analysis record; Based on the prediction results, the analysis records are used to extract logistics operation risk characteristics associated with orders, vehicles, warehouses, roads, and delivery areas, generating object-level risk characteristic records; Risk propagation associations are established based on object-level risk feature records, disturbance impact markers, and logistics node association records to generate logistics operation risk association records. Based on the risk association records and prediction reliability of logistics operation, the risk characteristics of logistics operation are labeled with risk levels, and risk label records are generated. By integrating the fields of object-level risk characteristic records, logistics operation risk association records, and risk label records, a scheduling risk status table is generated.

[0012] Optionally, generating the scheduling constraint mask table includes: Read the object-level risk characteristic records, logistics operation risk association records, and risk marker records from the scheduling risk status table to generate the basis for candidate combination generation; Based on the candidate combination generation criteria and the spatial node mapping records in the multi-source trusted logistics status frames, order objects, vehicle objects and route objects are matched to generate order-vehicle-route candidate combinations. Based on the order-vehicle-route candidate combination and the task-bound logistics status record, extract logistics execution constraints and generate execution constraint records; Based on the execution constraint record, the candidate combinations of order-vehicle-route are marked as executable, semi-locked and non-executable, and an execution status record of the candidate combinations is generated. Write the execution status record of the candidate combination into the order-vehicle-route candidate combination to generate a scheduling constraint mask table.

[0013] Optionally, the generation of candidate scheduling feasible domains includes: Read the order-vehicle-route candidate combinations, candidate combination execution status records, and risk marker records from the scheduling constraint mask table, and generate candidate combination screening records; Candidate combination screening records are classified according to executable status, semi-locked status and non-executable status. Order-vehicle-route candidate combinations corresponding to non-executable status are removed, and a set of executable candidate combinations is generated. Read the task-bound logistics status record and logistics operation risk association record corresponding to the semi-locked state, determine the vehicle holding range, order holding range and route adjustment range of the semi-locked candidate combination, and generate the adjustment range limitation record; The executable candidate combination set and the adjustment range limit record are associated, and the schedulable object flag, restricted adjustment flag and risk level flag are written to generate the candidate scheduling feasible domain.

[0014] Optionally, the output of the intelligent logistics scheduling result includes: Read the schedulable object markers, restricted adjustment markers, risk level markers, and order-vehicle-route candidate combinations from the candidate scheduling feasible domain to generate the basis for scheduling execution; Based on the scheduling execution criteria, matching order objects, vehicle objects, and route objects generates vehicle allocation results; Order delivery priorities are ranked based on vehicle allocation results and risk level markings to generate a delivery sequence; Generate route execution plans and departure plans based on delivery sequence, route objects, and restricted adjustment tags; Based on vehicle allocation results, delivery sequence, route execution plan, departure plan, and restricted adjustment flags, a rescheduling record is generated, and the intelligent logistics scheduling result is output.

[0015] According to an embodiment of the present invention, a smart logistics scheduling system based on big data analysis includes the following modules: The data acquisition and preprocessing module is used to collect multi-source logistics operation data and perform standardized processing to generate raw logistics status data. The Hampel filtering module is used to execute the Hampel filtering algorithm with spatial proximity consistency constraints based on the original logistics status data to generate reliable logistics status records. The state frame construction module is used to perform spatiotemporal alignment, spatial node mapping, and task state binding on trusted logistics state records to generate multi-source trusted logistics state frames. The model prediction module is used to build an improved FEDformer model and generate logistics situation prediction results based on multi-source reliable logistics status frames. The risk status generation module is used to extract the risk characteristics of the stream operation and generate a scheduling risk status table; The constraint mask generation module is used to generate a scheduling constraint mask table by combining the scheduling risk status table and logistics execution constraints; The feasible region generation module is used to filter candidate combinations based on the scheduling constraint mask table and generate candidate scheduling feasible regions. The scheduling output module is used to generate vehicle allocation results, delivery sequence, route execution plan, departure plan and rescheduling record based on candidate scheduling feasible domains, and output intelligent logistics scheduling results.

[0016] The beneficial effects of this invention are: This invention proposes a smart logistics scheduling method and system based on big data analytics, constructing a complete data processing chain from multi-source logistics operation data collection, Hampel filtering for reliable processing, construction of multi-source reliable logistics status frames, to improved FEDformer model prediction. By introducing spatial proximity consistency constraints into the Hampel filtering algorithm, combined with window scale selection and reliability threshold correction, reliable identification of vehicle positioning drift, warehouse scanning delay, and road status lag is achieved, while retaining the real disturbance information corresponding to road congestion, warehouse delay, and order surges, thereby improving the reliability of logistics status data and the accuracy of scheduling status identification.

[0017] This invention, based on an improved FEDformer model, sets up a perturbation-fidelity dual-track decomposition and reconstruction layer, a logistics heterogeneous causal spectrum graph interaction layer, and a joint feature-based reliable frequency band calibration layer. It performs perturbation characterization, causal spectrum interaction, and predictive reliability calibration on multi-source reliable logistics state frames, generating logistics situation prediction results and a scheduling risk state table. Through the construction of a scheduling constraint mask table and candidate scheduling feasible regions, it completes the generation of vehicle allocation results, delivery order, route execution plan, departure plan, and rescheduling records, reducing execution fluctuations caused by global rescheduling and improving scheduling stability, execution continuity, and logistics operation risk response capabilities under dynamic perturbation scenarios. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a smart logistics scheduling method based on big data analysis proposed in this invention; Figure 2 This is a schematic diagram of the improved FEDformer model structure in the intelligent logistics scheduling method based on big data analysis proposed in this invention. Figure 3 This is a system architecture diagram of a smart logistics scheduling system based on big data analysis proposed in this invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figure 1 and Figure 2 A smart logistics scheduling method based on big data analysis includes: Collect multi-source logistics operation data and perform preprocessing to generate raw logistics status data; Based on the original logistics status data, the Hampel filtering algorithm is executed, spatial proximity consistency constraints are introduced, window scale selection, credibility threshold correction and spatial proximity consistency discrimination are performed to generate a reliable logistics status record. Spatiotemporal alignment, spatial node mapping, and task status binding are performed on trusted logistics status records to generate multi-source trusted logistics status frames. An improved FEDformer model is constructed. Multi-source reliable logistics state frames are input into the improved FEDformer model. After processing by the perturbation-fidelity dual-track decomposition and reconstruction layer, the logistics heterogeneous causal spectrum interaction layer, and the joint feature reliable frequency band calibration layer, logistics situation prediction results are generated. Based on the logistics situation prediction results, extract logistics operation risk characteristics and generate a scheduling risk status table; Based on the scheduling risk status table, candidate combinations of order-vehicle-route are generated, and the execution status is marked in combination with logistics execution constraints to generate a scheduling constraint mask table. Based on the scheduling constraint mask table, candidate combination screening and adjustment range limitation are performed to generate candidate scheduling feasible domains; Based on the candidate scheduling feasible domain, vehicle allocation results, delivery sequence, route execution plan, departure plan and rescheduling record are generated, and the intelligent logistics scheduling result is output.

[0021] In this embodiment, generating the original logistics status data includes: The system reads order operation records, vehicle operation records, warehousing operation records, road traffic records, environmental impact records, and historical performance records from the logistics business platform to generate multi-source logistics operation data. Standardized logistics operation data is generated by unifying field names, units of measurement, and time formats for multi-source logistics operation data. Perform duplicate record removal, missing field marking, and abnormal format removal on standard logistics operation data to generate clean logistics operation data; The cleaning logistics operation data is written with data source markers, collection time markers, and business object markers to generate raw logistics status data, including: Generate raw logistics status data, specifically: The data set formed by writing data source markers, collection time markers, and business object markers into the cleaning logistics operation data; Raw logistics status data includes field type, field value, data source marker, collection time marker, and business object marker; Field types include spatial location fields, order operation fields, vehicle operation fields, warehouse operation fields, road access fields, environmental impact fields, and historical fulfillment fields; The field values ​​record the operational values ​​corresponding to the order status, vehicle status, warehouse status, road status, environmental status, and fulfillment status; The data source marker records the corresponding logistics business platform and data access interface from which the data source field originates. The data collection time stamp records the business occurrence time and system collection time corresponding to the data collection time field. The business object tags record fields corresponding to the order object, vehicle object, warehouse object, road object, delivery area object, and fulfillment object; The field type, field value, data source marker, collection time marker, and business object marker are associated with the cleaning logistics operation data and stored to generate the original logistics status data.

[0022] In this embodiment, generating a reliable logistics status record includes: Read the field types, field values, data source markers, collection time markers, and business object markers from the raw logistics status data. Configure the Hampel filter window scale according to the field type to generate a field-level window sequence, where: Generate a field-level window sequence, specifically: Read the field types, field values, data source markers, collection time markers, and business object markers from the raw logistics status data; Based on the business object tag, field values ​​belonging to the same order object, vehicle object, warehouse object, road object, delivery area object, or fulfillment object are grouped into the same object sequence; Based on field type, the field values ​​in the same object sequence are divided into order operation field sequence, vehicle operation field sequence, warehouse operation field sequence, road access field sequence, environmental impact field sequence, and historical performance field sequence; Based on the time stamp of the data collection, the field values ​​in each field sequence are sorted by time to generate a field time series; Configure the Hampel filter window scale according to the field type to generate field-level window scale records corresponding to the field's time series. Using the value of each field in the field time series as the center field value, the values ​​of adjacent fields before and after the center field value are read according to the field-level window scale to generate field window records; Write the central field value, adjacent field values, field type, data source marker, collection time marker, and business object marker into the field window record to generate a field-level window sequence; The absolute deviation of the window median and mean is calculated based on the field-level window sequence. Data source credibility is generated by combining data source tags, and the anomaly detection threshold is adjusted using this credibility to generate a field-level anomaly threshold. Generate field-level exception thresholds, specifically as follows: Read field window records from the field-level window sequence. The field window records include the center field value, adjacent field values, field type, data source marker, collection time marker, and business object marker. Sort the values ​​in the center field and adjacent fields according to their numerical values ​​to generate a set of sorted values ​​for the window; Read the value of the field located in the middle position in the sorted set of values ​​in the window, and generate the value in the window; Calculate the absolute value of the difference between each field value in the window sorted value set and the middle value in the window, and generate a window deviation value set; Sort the set of window deviation values ​​according to their numerical values, read the deviation value in the middle position after sorting, and generate the median absolute deviation. Based on the data source marker and collection time marker in the field window records, we can count the actual number of records in the field window records under the same data source marker, the expected number of records corresponding to the field-level window scale, and the adjacent collection time interval. A continuous access score is generated based on the ratio of the actual number of records to the required number of records. A time stability score is generated based on the difference between adjacent acquisition time intervals; The access continuity score and time stability score are normalized to between 0 and 1 and then weighted to generate a source stability score. The source stability score is used as the data source credibility, and it is stored in association with data source tags, field types, and business object tags. Based on the field type, the anomaly judgment coefficient is configured, and the order operation field, vehicle operation field, warehouse operation field, road access field, environmental impact field and historical performance field are respectively matched with the corresponding anomaly judgment coefficient; The anomaly determination coefficient is multiplied by the absolute deviation of the median to generate the basic anomaly threshold for the field. A threshold correction coefficient is generated based on the credibility of the data source. The threshold correction coefficient is equal to 1 plus 1 minus the difference after the credibility of the data source. Multiply the basic anomaly threshold of the field by the threshold correction coefficient to generate the corrected anomaly threshold; The error threshold, window median, median absolute deviation, data source credibility, field type, data source marker, collection time marker, and business object marker are associated and stored to generate field-level error thresholds; The exception determination coefficient is configured based on the field type, specifically as follows: Read field window records from the field-level window sequence. Each field window record includes the field type, the value of the center field, the values ​​of adjacent fields, and the collection time stamp. Based on field type, the field window records are divided into the following sets: order operation field window set, vehicle operation field window set, warehouse operation field window set, road access field window set, environmental impact field window set, and historical performance field window set. For each field window set, read the value of the center field and the values ​​of adjacent fields in each field window record, calculate the difference between the maximum and minimum field values ​​in the field window record, and generate the window value span; Read the middle value of the corresponding window record in the field window, calculate the ratio of the window value span to the absolute value of the middle value plus 1, and generate the window value fluctuation ratio. Calculate the average of the window value fluctuation ratios within the same field window set to generate a field-type value fluctuation score; For each field window set, read the acquisition timestamps and arrange them in chronological order, calculate the time interval between adjacent acquisition timestamps, and generate an acquisition time interval sequence. Calculate the difference between the maximum and minimum time intervals in the data collection time interval sequence, and divide the difference by the value of the maximum time interval plus 1 to generate a field-type time fluctuation score; Multiply the field type numerical fluctuation score by 0.8, multiply the field type time fluctuation score by 0.2, and add the two products to generate the field type comprehensive fluctuation score; When the overall volatility score of a field type is greater than 1, the overall volatility score of the field type will be corrected to 1. When the overall volatility score of a field type is less than 0, the overall volatility score of the field type will be corrected to 0. Add 3 to the product of the field type comprehensive fluctuation score and 2 to generate the anomaly determination coefficient; The anomaly determination coefficient is associated with the corresponding field type and stored to generate an anomaly determination coefficient record for the field type. Anomaly detection is performed on field values ​​based on field-level anomaly thresholds, generating initial anomaly markers, where: Generate initial exception markers, specifically: Read the value of the central field, field type, data source marker, collection time marker, and business object marker from the field-level window sequence; Read the window value, corrected exception threshold, field type, data source flag, collection time flag, and business object flag from the field-level exception threshold; Based on field type, data source marker, collection time marker, and business object marker, the value of the central field is matched with the corresponding window value and the correction anomaly threshold; Calculate the difference between the center field value and the window center value to generate the field deviation value; Calculate the absolute value of the field deviation value to generate the field deviation range. When the field deviation range is less than or equal to the correction anomaly threshold, determine that the center field value is a normal field value and generate a normal mark. When the deviation of a field exceeds the correction threshold and the field deviation value is greater than 0, the value of the central field is determined to be a high-value anomaly, and a high-value anomaly marker is generated. When the deviation of a field is greater than the correction threshold and the field deviation value is less than 0, the value of the center field is determined to be a low value anomaly, and a low value anomaly marker is generated. The normal flag, high-value anomaly flag, or low-value anomaly flag are associated and stored with the central field value, field type, data source flag, collection time flag, and business object flag to generate the initial anomaly flag; For each business object corresponding to the initial anomaly marker, spatial neighbor objects are retrieved, the same-direction change status of these neighbor objects is calculated, and a spatial consistency score is generated, where: The spatial consistency score is generated as follows: Read the exception type, central field value, field type, collection time marker, and business object marker from the initial exception marker; Filter records with high-value anomaly markers or low-value anomaly markers to generate anomaly records to be verified; Based on the business object tags in the abnormal records to be verified, vehicle location, warehouse node, road segment, delivery area and order association area are extracted from the field values ​​of the original logistics status data to generate object spatial location records; Based on the object spatial location record, calculate the spatial distance between the business object corresponding to the abnormal record to be verified and other business objects, and take the business objects whose spatial distance is less than or equal to the proximity distance threshold of 3km as spatial proximity objects. When there is no spatial distance value between business objects, spatially adjacent objects are determined based on the direct connection relationship between warehouse nodes, road segments, delivery areas and order-related areas. Associate spatial neighbor objects with the abnormal records to be verified and store them to generate a set of spatial neighbor objects; Read the field values ​​of each spatial neighbor object in the spatial neighbor object set within a time matching threshold of 15 minutes and under the same field type, and read the window midpoint in the field-level anomaly threshold; Calculate the difference between the field value of the spatial neighboring object and the middle value in the window, and generate the neighboring object deviation value. When the abnormal record to be verified is marked as a high value abnormality and the neighboring object deviation value is greater than 0, generate a same-direction change mark. When the abnormal record to be verified is marked as a low value abnormality and the neighboring object deviation value is less than 0, generate a same-direction change mark. The number of objects and the number of same-direction change markers in the spatial neighbor object set are counted, the ratio of the number of same-direction change markers to the number of objects is calculated, and a spatial consistency score is generated. When the number of objects in the spatial neighbor set is 0, the spatial consistency score is recorded as 0; The spatial consistency score is associated with and stored in relation to anomaly type, central field value, field type, collection time stamp, business object stamp, and spatial neighbor object set; The initial anomaly markers are constrained and validated based on spatial consistency scores to distinguish between data noise and real logistics disturbances. Correction values ​​are generated for data noise, and disturbance markers are generated for real logistics disturbances. The correction values, disturbance markers, and spatial consistency scores are then written into the trusted logistics status record. For real-world logistics disturbances, generate disturbance markers, and write correction values, disturbance markers, and spatial consistency scores into the trusted logistics status record, specifically as follows: Read the exception type, central field value, field type, data source flag, collection time flag, and business object flag from the initial exception flag; Read the windowed middle value, corrected outage threshold, and absolute deviation of the median from the field-level outage threshold; Read the spatial consistency score, the set of spatially neighboring objects, and the number of same-direction change markers from the spatial consistency score; When the anomaly type is marked as normal, the value of the center field is used as the correction value, the disturbance mark is recorded as the no-disturbance mark, and the spatial consistency score is recorded as 1; When the anomaly type is high-value anomaly marker or low-value anomaly marker, and the spatial consistency score is less than the spatial consistency judgment threshold of 0.5, the anomaly corresponding to the value of the center field is judged as data noise. To address data noise, the window midpoint is used as a correction value, and data noise identification results are generated. When the anomaly type is a high-value anomaly marker or a low-value anomaly marker, and the spatial consistency score is greater than or equal to the spatial consistency judgment threshold of 0.5, the anomaly corresponding to the value of the center field is determined to be a real logistics disturbance. For real logistics disturbances, the value of the central field is retained as a correction value, and a high-value disturbance label or a low-value disturbance label is generated according to the anomaly type. The correction value, disturbance marker, spatial consistency score, data noise identification result, field type, data source marker, collection time marker, and business object marker are associated and stored to generate a reliable logistics status record.

[0023] In this embodiment, generating a multi-source trusted logistics status frame includes: Read the collection time stamp, business object stamp, correction value, disturbance stamp and spatial consistency score from the trusted logistics status record, perform time merging according to the unified scheduling time window, and generate time-aligned logistics status record; Based on time-aligned logistics status records, the system reads business object tags and spatial location fields, mapping order objects, vehicle objects, warehouse objects, road objects, and delivery area objects to unified logistics spatial nodes, generating spatial node mapping records, where: Generate spatial node mapping records, specifically as follows: Read the business object marker, field type, field value, and collection time marker from the time-aligned logistics status record; Identify order objects, vehicle objects, warehouse objects, road objects, and delivery area objects based on business object tags; Extract the order association region, vehicle location, warehouse node, road segment, and delivery area from the field values ​​to generate a spatial location field; Based on the spatial location field, order objects are mapped to order-associated region nodes, vehicle objects are mapped to vehicle current location nodes, warehouse objects are mapped to warehouse nodes, road objects are mapped to road segment nodes, and delivery area objects are mapped to delivery area nodes; The order-related area nodes, vehicle current location nodes, warehouse nodes, road segment nodes, and delivery area nodes are uniformly coded to generate unified logistics space nodes; The unified logistics space nodes are associated and stored with business object tags, field types, collection time tags, and spatial location fields to generate space node mapping records. Based on the disturbance markers and spatial consistency scores in the spatial node mapping records and trusted logistics status records, state associations between nodes are established, generating logistics node association records, where: Generate logistics node association records, specifically: Read the unified logistics space node, business object tag, field type, collection time tag, and spatial location field from the space node mapping record; Read the disturbance markers, spatial consistency scores, field types, collection time markers, and business object markers from the trusted logistics status records; Read the task execution stage, vehicle carrying status, order fulfillment status, business object marker, and collection time marker from the time-aligned logistics status record; Based on business object tags, field types, and collection time tags, the unified logistics space nodes are matched with corresponding disturbance tags, spatial consistency scores, task execution stages, vehicle carrying status, and order fulfillment status. Establish associations between the order objects, vehicle objects, warehouse objects, road objects, and delivery area objects located in the same unified logistics space node; Establish adjacency relationships between unified logistics space nodes that have adjacent spatial location fields or upstream and downstream delivery relationships; Based on the association of the same node and the association of adjacent nodes, write the associated node number, associated object tag, disturbance tag, spatial consistency score, task execution stage, vehicle carrying status and order fulfillment status; The associated node number, associated object tag, disturbance tag, spatial consistency score, task execution stage, vehicle carrying status, order fulfillment status and collection time tag are associated and stored to generate a logistics node association record; Based on the logistics node association records, read the task execution stage, vehicle carrying status and order fulfillment status, bind the task status, and generate task-bound logistics status records; The time-aligned logistics status records, spatial node mapping records, logistics node association records, and task-bound logistics status records are fused to generate a multi-source trusted logistics status frame, in which: Generate a multi-source trusted logistics status frame, specifically as follows: Read the collection time stamp, field type, field value, correction value, disturbance flag, and spatial consistency score from the time-aligned logistics status record; Read the unified logistics space node, business object tag, spatial location field, and collection time tag from the spatial node mapping record; Read the associated node number, associated object tag, disturbance tag, spatial consistency score, and collection time tag from the logistics node association record; Read the task execution stage, vehicle carrying status, order fulfillment status, business object marker, and collection time marker from the task-bound logistics status record; Based on the collection time stamp and business object stamp, the time-aligned logistics status record, spatial node mapping record, logistics node association record and task-bound logistics status record are matched by field matching. Write the correction value, disturbance marker, spatial consistency score, unified logistics space node, associated node number, task execution stage, vehicle carrying status and order fulfillment status into the same status frame record; The status frame records are sorted according to the collection time stamp to generate multi-source trusted logistics status frames.

[0024] In this embodiment, generating the logistics situation prediction result includes: An improved FEDformer model is constructed, comprising a perturbation-fidelity dual-track decomposition and reconstruction layer, a logistics heterogeneous causal spectrum interaction layer, and a joint feature-based reliable frequency band calibration layer, wherein: The improved FEDformer model is constructed as follows: The FEDformer model retains the backbone of time series decomposition structure, frequency domain attention structure, encoder structure, decoder structure and prediction output structure, and makes structural modifications to input representation, perturbation preservation, node interaction and prediction reliability calibration around the smart logistics dynamic scheduling scenario. A logistics status input mapping structure is set at the model input end to map the correction value, disturbance mark, spatial consistency score, unified logistics space node, associated node number, task execution stage, vehicle carrying status and order fulfillment status in the multi-source trusted logistics status frame into a logistics status input vector. After the time series decomposition structure, a perturbation-fidelity dual-track decomposition and reconstruction layer is added to divide the logistics state input vector into a normal state input track and a perturbation state input track. The normal state input track reads the correction value and the task execution stage, while the perturbation state input track reads the perturbation mark and the spatial consistency score to generate normal state logistics representation and perturbation state logistics representation. A heterogeneous causal spectrum graph interaction layer for logistics is added to the frequency domain attention structure. Based on the unified logistics space node and associated node numbering, the spectrum interaction relationship between order nodes, vehicle nodes, warehouse nodes, road nodes and delivery area nodes is constructed to generate a causal spectrum interaction representation. Before the prediction decoding structure, a joint feature credible frequency band calibration layer is added. Causal spectrum interaction representation, data source labeling, spatial consistency score and frequency band residual record are read to perform credibility calibration on logistics prediction features under different frequency bands and generate credible frequency band representation. The forecast output structure is improved by expanding the forecast objectives to include changes in order fulfillment, vehicle operation, warehousing operations, road traffic, delivery area risk, and forecast reliability, thus forming the logistics situation forecast results. The perturbation-fidelity dual-track decomposition and reconstruction layer specifically includes: Status correction value reading end: Reads the correction value, field type, collection time stamp, business object stamp and task execution stage from the model input sequence to form the basis for normal state input; Perturbation label splitting zone: Read the perturbation labels and spatial consistency scores in the model input sequence, write the data corresponding to the unperturbed label into the normal input track, and write the data corresponding to the high-value perturbation label and the low-value perturbation label into the perturbed input track; Dual-track state reconstruction channel: Perform trend decomposition on the correction values ​​in the normal state input track, and perform disturbance intensity encoding on the disturbance markers and spatial consistency scores in the disturbance dynamic input track to generate normal state logistics representation and disturbance dynamic logistics representation; Perturbation-preserving output: The normal logistics representation and the perturbation logistics representation are aligned and output according to the collection time marker, business object marker and unified logistics space node, so that the real logistics perturbation is expressed independently in the prediction input; The heterogeneous causal spectrum diagram interaction layer of logistics specifically includes: Logistics node reading end: Reads unified logistics space nodes, associated node numbers, business object tags and field types, and identifies order nodes, vehicle nodes, warehouse nodes, road nodes and delivery area nodes; Causal connection construction area: Based on the associated node number, associated object tag, disturbance tag and collection time tag in the logistics node association record, establish the state transmission relationship between orders, vehicles, warehouses, roads and delivery areas; Spectrum interaction channel: Maps the normal state logistics representation and the disturbed state logistics representation to the frequency domain feature space, and calculates the spectrum interaction intensity between different logistics nodes according to the state transfer relationship; Interactive representation output: Associate the spectral interaction intensity with the corresponding logistics node, collection time marker, and business object marker to generate a causal spectral interaction representation; The joint feature-based reliable frequency band calibration layer specifically includes: Trusted feature reading end: Reads causal spectrum interaction representation, data source marker, spatial consistency score and perturbation marker to form frequency band calibration input; Frequency band residual calibration area: Based on the difference in frequency band characteristics before and after prediction decoding, frequency band residual records are generated and collected according to field type and unified logistics space node; Credible weight adjustment channel: Based on the data source credibility, spatial consistency score and frequency band residual record corresponding to the data source label, a frequency band credibility weight is generated, and the logistics prediction characteristics under different frequency bands are weighted and calibrated. Trusted frequency band output end: The weighted and calibrated logistics prediction features are associated with field types, collection time stamps, business object stamps and unified logistics space nodes to generate a trusted frequency band representation. Read the time-aligned logistics status records, spatial node mapping records, logistics node association records, and task-bound logistics status records from multi-source trusted logistics status frames, and generate the model input sequence according to the prediction time window, where: The model input sequence is generated according to the prediction time window, specifically: Read the time-aligned logistics status record, spatial node mapping record, logistics node association record, and task-bound logistics status record from the multi-source trusted logistics status frame; Based on the collection time marker in the time-aligned logistics status record, the multi-source reliable logistics status frame is divided into windows according to the predicted time window of 30 minutes. The prediction time window is moved by a sliding step of 5 minutes, and the correction value, disturbance mark, spatial consistency score, unified logistics space node, associated node number, task execution stage, vehicle carrying status and order fulfillment status are extracted within the same prediction time window. Based on the acquisition time stamp, the data within the same prediction time window are sorted by time to generate a window state sequence; The window state sequence is associated and stored with the prediction time window number, unified logistics space node, and business object label to generate the model input sequence; The model input sequence is fed into the perturbation-fidelity dual-track decomposition and reconstruction layer. A normal input track is constructed based on the correction value, and a perturbation-dynamic input track is constructed based on the perturbation label and spatial consistency score. This generates normal and perturbation-dynamic logistics representations, where: Generate normal-state and disturbed-state logistics representations, specifically as follows: Read the correction values, disturbance markers, spatial consistency scores, unified logistics spatial nodes, associated node numbers, task execution stages, vehicle carrying status, order fulfillment status, prediction time window numbers, and business object markers from the model input sequence; By extracting the state correction value reading end in the perturbation-fidelity dual-track decomposition and reconstruction layer, the correction value, task execution stage, vehicle carrying status, order fulfillment status, prediction time window number and business object tag are generated to generate a state correction input record. The disturbance markers and spatial consistency scores are read through the disturbance marker diversion area, and then matched with the prediction time window number and business object markers and status correction input records. When the disturbance flag is a disturbance-free flag, the corresponding correction value, task execution stage, vehicle load status, and order fulfillment status are written into the normal state input track. When the disturbance marker is a high-value disturbance marker or a low-value disturbance marker, the corresponding correction value, disturbance marker, spatial consistency score, unified logistics space node and associated node number are written into the disturbance input track; The normal state input track is encoded with time trend through a dual-track state reconstruction channel to generate a normal state logistics representation. The disturbance intensity of the disturbance dynamic input track is encoded by the dual-track state reconstruction channel to generate a disturbance dynamic logistics characterization. The output is aligned according to the prediction time window number, business object tag, and unified logistics space node by perturbation; Normal and disturbed logistics representations are input into the heterogeneous causal spectrum graph interaction layer. A spectrum interaction graph is constructed based on order nodes, vehicle nodes, warehouse nodes, road nodes, and delivery area nodes. A causal spectrum interaction representation is generated according to the logistics node association records, where: A causal spectrum interaction representation is generated based on the logistics node association records, specifically as follows: Read the prediction time window number, business object label, unified logistics space node, associated node number, correction value, disturbance label and spatial consistency score from the normal logistics representation and the disturbed logistics representation; By using the logistics node reading end in the logistics heterogeneous causal spectrum diagram interaction layer, order nodes, vehicle nodes, warehouse nodes, road nodes, and delivery area nodes are identified based on unified logistics space nodes and business object tags; Read the associated node number, associated object tag, disturbance tag, spatial consistency score and collection time tag from the logistics node association record, and write the unified logistics space nodes that have the same node association or adjacent node association into the spectrum interaction graph; By constructing a causal connection region, the order of node state changes is determined based on the collection time stamp. The node that first shows a disturbance mark in time is taken as the causal source node, and the associated node that shows a state change later in time is taken as the causal response node. When the time difference between the causal source node and the causal response node is less than or equal to the causal time delay threshold of 30 minutes, a directed causal connection is established, and the perturbation mark, spatial consistency score and associated node number are written into the directed causal connection. Through the spectrum interaction channel, the normal state logistics characterization and the disturbed state logistics characterization are mapped to node frequency band features, and divided into low frequency trend band (0 to 0.33), mid frequency fluctuation band (greater than 0.33 and less than or equal to 0.66), and high frequency disturbance band (greater than 0.66 and less than or equal to 1) according to the frequency index ratio. For each directed causal connection, read the node frequency band characteristics of the causal source node and the causal response node, calculate the average characteristic energy in the low-frequency trend band, the mid-frequency fluctuation band and the high-frequency disturbance band respectively, and generate the source node frequency band energy and the response node frequency band energy. Calculate the mean absolute value of the feature difference between causal source nodes and causal response nodes within the same frequency band, and generate the frequency band similarity by dividing 1 by 1 and adding the mean absolute value of the feature difference. The time delay attenuation coefficient is generated by dividing 1 by 1 and adding the number of minutes of the acquisition time difference. When the disturbance label is a high-value disturbance label or a low-value disturbance label, the disturbance weight is recorded as 1.2; when the disturbance label is an undisturbed label, the disturbance weight is recorded as 1. The source node frequency band energy, response node frequency band energy, frequency band similarity, time delay attenuation coefficient, disturbance weight, and spatial consistency score are multiplied to generate the frequency band transmission strength. By fusing node frequency band characteristics, frequency band transmission strength, disturbance markers, spatial consistency scores, and directed causal connections, a record of inter-node spectrum interaction is generated. Through the interactive representation output end, the spectrum interaction records between nodes are aligned and output according to the prediction time window number, unified logistics space node, business object label and associated node number to generate a causal spectrum interaction representation. The causal spectral interaction representation is input into the joint feature-based reliable frequency band calibration layer, and a reliable frequency band representation is generated by combining data source labeling, spatial consistency scoring, and frequency band residual records, where: A reliable frequency band characterization is generated by combining data source markers, spatial consistency scores, and frequency band residual records, specifically as follows: Read the node frequency band characteristics, frequency band transmission strength, disturbance markers, spatial consistency scores, prediction time window numbers, unified logistics spatial node and business object markers from the causal spectrum interaction representation; By using the trusted feature reading end in the joint feature trusted frequency band calibration layer, data source labels are matched from the model input sequence based on the prediction time window number, unified logistics space node and business object label; By using the frequency band residual calibration area in the joint feature credible frequency band calibration layer, the absolute value of the difference between the node frequency band feature within the same prediction time window and the frequency band feature before prediction decoding is calculated to generate a frequency band residual record. By adjusting the trust weights in the joint feature trust band calibration layer, a trust band weight is generated based on the data source trustworthiness, spatial consistency score, and band residual record corresponding to the data source label. The node frequency band features are multiplied by the frequency band confidence weights to generate calibration frequency band features; By linking and storing the calibration band features, band confidence weights, band residual records, prediction time window numbers, unified logistics space nodes, and business object tags in the joint feature trusted band calibration layer, a trusted band representation is generated. Based on reliable frequency band representation, predictive decoding is performed to generate logistics situation prediction results, including: The logistics situation prediction results are generated as follows: Read the calibration band features, band confidence weights, band residual records, prediction time window numbers, unified logistics space nodes, and business object tags from the trusted band representation; The prediction start time is determined based on the prediction time window number, and the prediction time window is divided into 30-minute segments according to a prediction step size of 5 minutes to generate prediction time slices. The calibration frequency band features are input into the improved prediction and decoding structure of the FEDformer model, and the node prediction values ​​corresponding to the order status, vehicle status, warehouse status, road status, environmental status and fulfillment status are decoded according to the prediction time slice. Based on unified logistics space node and business object labeling, node prediction values ​​are divided into order node prediction values, vehicle node prediction values, warehouse node prediction values, road node prediction values, and delivery area node prediction values. The average confidence weight of the same unified logistics space node in each prediction time slice is calculated based on the frequency band confidence weight to generate the node prediction confidence. The average frequency band residual of the same unified logistics space node in each prediction time slice is calculated based on the frequency band residual record, and the node prediction confidence is matched and corrected with the average frequency band residual to generate prediction confidence. When the average frequency band residual is greater than the frequency band residual threshold of 0.3, or the node prediction confidence is less than the confidence threshold of 0.6, a disturbance impact marker is generated. A stable prediction label is generated when the average frequency band residual is less than or equal to the frequency band residual threshold of 0.3 and the node prediction confidence is greater than or equal to the confidence threshold of 0.6. The prediction time slice, node prediction value, disturbance impact marker, stability prediction marker, prediction confidence, unified logistics space node and business object marker are associated and stored to generate logistics situation prediction results; The improved FEDformer model was trained using a combination of logistics situation prediction error, disturbance reconstruction error, causal spectrum interaction error, and reliable frequency band calibration error as the joint optimization objective. The network parameters of the disturbance fidelity dual-track decomposition reconstruction layer, the logistics heterogeneous causal spectrum interaction layer, and the joint feature reliable frequency band calibration layer were continuously optimized. When the change in the joint loss value over five consecutive training rounds was less than 0.001, the improved FEDformer model was considered to have completed convergence training. The improved FEDformer model is trained as follows: A training sample set is constructed from historical multi-source reliable logistics status frames, historical logistics node association records, and actual scheduling execution records. The training sample set includes 10,000 sets of logistics scheduling samples. Each set of logistics scheduling samples includes the correction value within the prediction time window of 30 minutes, disturbance marker, spatial consistency score, unified logistics spatial node, associated node number, task execution stage, vehicle carrying status, and order fulfillment status, and includes labels for actual order fulfillment changes, actual vehicle operation changes, actual warehousing operation changes, actual road traffic changes, and actual delivery area risk changes. The training sample set is input into the improved FEDformer model. Normal logistics representation and disturbed logistics representation are generated by the state correction value reading end, disturbance mark diversion area, dual-track state reconstruction channel and disturbance hold output end in the disturbance fidelity dual-track decomposition and reconstruction layer. Causal spectrum interaction representation is generated by the logistics node reading end, causal connection construction area, spectrum interaction channel and interaction representation output end in the logistics heterogeneous causal spectrum graph interaction layer. Confidential frequency band representation is generated by the credible feature reading end, frequency band residual calibration area, credible weight adjustment channel and credible frequency band output end in the joint feature credible frequency band calibration layer. Finally, the logistics situation prediction result is generated by prediction decoding. The logistics situation prediction error is generated by squaring the difference between the node prediction value and the actual change label and taking the average. The disturbance reconstruction error is generated by squaring the difference between the disturbance dynamic logistics characterization and the actual disturbance intensity and taking the average. The causal spectrum interaction error is generated by squaring the difference between the frequency band transmission intensity and the actual propagation intensity and taking the average. The reliable frequency band calibration error is generated by squaring the difference between the frequency band reliable weight and the actual prediction hit mark and taking the average. The logistics situation prediction error is multiplied by 0.40, the disturbance reconstruction error by 0.20, the causal spectrum interaction error by 0.25, and the reliable frequency band calibration error by 0.15, and then summed to generate a joint loss value. Based on the joint loss value, the gradient value of each trainable parameter is calculated. The gradient value is multiplied by the learning rate of 0.001 and then subtracted from the corresponding trainable parameter to update the network parameters. This process is repeated until the change in the joint loss value in 5 consecutive training rounds is less than 0.001, at which point the improved FEDformer model is considered to have completed convergence training.

[0025] In this embodiment, generating the scheduling risk status table includes: Read the prediction time window, node prediction value, disturbance impact marker and prediction confidence from the logistics situation prediction results, and generate prediction result analysis record; Based on the prediction results, the logistical operation risk characteristics associated with orders, vehicles, warehousing, roads, and delivery areas are extracted from the analysis records to generate object-level risk characteristic records, including: Generate object-level risk characteristic records, specifically: Read the prediction time slice, node prediction value, disturbance impact marker, stable prediction marker, prediction confidence, unified logistics space node and business object marker from the prediction result parsing record; Identify order objects, vehicle objects, warehouse objects, road objects, and delivery area objects based on business object tags; The node prediction values ​​are matched with the order fulfillment status, vehicle operation status, warehouse operation status, road traffic status, and delivery area status to extract order delay risk characteristics, vehicle capacity risk characteristics, warehouse congestion risk characteristics, road traffic risk characteristics, and regional fulfillment risk characteristics. Each risk characteristic is associated with disturbance impact markers, stability prediction markers, prediction confidence, prediction time slices, and unified logistics space nodes for storage, generating object-level risk characteristic records. Risk propagation relationships are established based on object-level risk feature records, disturbance impact markers, and logistics node association records, generating logistics operation risk association records, including: Generate logistics operation risk association records, specifically: Read the order delay risk characteristics, vehicle capacity risk characteristics, warehouse congestion risk characteristics, road traffic risk characteristics, regional fulfillment risk characteristics, disturbance impact markers, prediction credibility, prediction time slices, unified logistics space nodes, and business object markers from the object-level risk characteristic records; Read the associated node number, associated object tag, disturbance tag, spatial consistency score, and collection time tag from the logistics node association record; Based on unified logistics space nodes, business object tags, and associated object tags, object-level risk characteristic records are matched with logistics node associated records; Filter records with object-level risk characteristics marked with disturbance effects to generate risk source object records; Based on logistics node association records, retrieve business object tags that are associated with the same node or adjacent nodes as the risk source object records, and generate risk response object records; Based on the prediction time slice and the collection time stamp, the time difference between the risk source object and the risk response object is calculated, and the associated objects with a time difference less than or equal to the propagation time delay threshold of 30 minutes are retained as valid propagation objects. The logistics operation risk characteristics in the risk source object record are normalized to generate the risk source intensity. The risk source intensity is multiplied by 0.40, the prediction confidence by 0.25, the spatial consistency score by 0.20, and the disturbance weight by 0.15. The four products are then added together to generate the risk propagation intensity. The disturbance weight is generated based on the disturbance label. When the disturbance label is a high-value disturbance label or a low-value disturbance label, the disturbance weight is recorded as 1. When the disturbance label is a disturbance label without a disturbance label, the disturbance weight is recorded as 0. The risk propagation intensity is associated with and stored in relation to risk source object records, risk response object records, disturbance impact markers, prediction credibility, unified logistics space nodes, and prediction time slices to generate logistics operation risk association records. Based on the correlation records and prediction reliability of logistics operation risks, risk level labels are assigned to the characteristics of logistics operation risks, generating risk label records, including: Generate risk marker records, specifically as follows: Read the risk source object record, risk response object record, risk propagation intensity, disturbance impact marker, prediction credibility, unified logistics spatial node and prediction time slice from the logistics operation risk association record; Read the order delay risk characteristics, vehicle capacity risk characteristics, warehouse congestion risk characteristics, road traffic risk characteristics, and regional fulfillment risk characteristics from the object-level risk characteristic records; Based on unified logistics spatial nodes, business object tags, and prediction time slices, the logistics operation risk association records are matched with object-level risk characteristic records; The risk characteristics of order delay, vehicle capacity, warehouse congestion, road traffic, and regional fulfillment are normalized to generate the risk intensity of the object. The comprehensive risk score is generated by multiplying the target risk intensity by 0.50, the risk propagation intensity by 0.30, and the prediction credibility by 0.20. When the overall risk score is less than 0.3, a low-risk label is generated; A medium-risk label is generated when the comprehensive risk score is greater than or equal to 0.3 and less than 0.7; A high-risk marker is generated when the overall risk score is greater than or equal to 0.7. The comprehensive risk score, low-risk marker, medium-risk marker, or high-risk marker are associated with the risk source object record, risk response object record, unified logistics space node, and prediction time slice and stored to generate a risk marker record. By integrating the fields of object-level risk characteristic records, logistics operation risk association records, and risk tag records, a scheduling risk status table is generated, in which: Generate a scheduling risk status table, specifically as follows: Read the order delay risk characteristics, vehicle capacity risk characteristics, warehouse congestion risk characteristics, road traffic risk characteristics, regional fulfillment risk characteristics, prediction time slices, unified logistics space nodes, and business object tags from the object-level risk characteristic records; Read the risk source object record, risk response object record, risk propagation intensity, disturbance impact marker, and prediction credibility from the logistics operation risk association record; Read the comprehensive risk score, low-risk marker, medium-risk marker, high-risk marker, unified logistics spatial node, and prediction time slice from the risk marking record; Based on unified logistics spatial nodes, business object tags, and prediction time slices, the object-level risk characteristic records, logistics operation risk association records, and risk tag records are matched by fields. The characteristics of logistics operation risks, the intensity of risk propagation, the marker of disturbance impact, the marker of stability prediction, the credibility of prediction, the comprehensive risk score and the marker of risk level are written into the same risk status record. The risk status records are sorted according to the predicted time slice and the unified logistics space node to generate a scheduling risk status table.

[0026] In this embodiment, generating the scheduling constraint mask table includes: Read the object-level risk characteristic records, logistics operation risk association records, and risk marker records from the scheduling risk status table to generate the basis for candidate combination generation, including: The criteria for generating candidate combinations are as follows: Read the logistics operation risk characteristics, risk propagation intensity, disturbance impact marker, prediction credibility, comprehensive risk score, risk level marker, unified logistics space node, prediction time slice and business object marker from the scheduling risk status table; Identify order objects, vehicle objects, warehouse objects, road objects, and delivery area objects based on business object tags; The risk level coefficient corresponding to the low-risk label is recorded as 0.2, the risk level coefficient corresponding to the medium-risk label is recorded as 0.6, and the risk level coefficient corresponding to the high-risk label is recorded as 1. When a disturbance impact marker exists, the disturbance coefficient is recorded as 1; when a stability prediction marker exists, the disturbance coefficient is recorded as 0. Multiply the comprehensive risk score by 0.45, the risk propagation intensity by 0.25, the risk level coefficient by 0.20, and the disturbance coefficient by 0.10, and then add the four products together to generate the basic scheduling priority score. Multiply the basic scheduling priority score by the prediction confidence to generate a candidate combination priority score. Filter the risk status records with a candidate combination priority score greater than or equal to 0.4 to generate a record of key scheduling objects. Based on the unified logistics space nodes and logistics operation risk characteristics in the key scheduling object records, the order objects, vehicle objects and path association nodes that participate in the generation of candidate combinations are determined. The order object, vehicle object, path association node, risk level marker, disturbance impact marker, candidate combination priority score and prediction time slice are associated and stored to generate the basis for candidate combination generation. Based on the candidate combination generation criteria and the spatial node mapping records in the multi-source trusted logistics status frames, order objects, vehicle objects, and route objects are matched to generate order-vehicle-route candidate combinations, where: Generate order-vehicle-route candidate combinations, specifically: Read the order objects, vehicle objects, path association nodes, risk level markers, disturbance impact markers, candidate combination priority scores, and prediction time slices from the candidate combination generation criteria; Read the spatial node mapping records in the multi-source trusted logistics status frame to obtain the order-related area node, vehicle current location node, warehouse node, road segment node, and delivery area node; Based on path association nodes, road segments connecting the vehicle's current location node, order association area node, warehouse node, and delivery area node are extracted from road segment nodes to generate path objects; Based on the order association area node, vehicle current location node, warehouse node, road segment node and delivery area node, establish spatial matching relationship between order object, vehicle object and route object; Based on the connection relationship between road segment nodes, determine whether there are continuous road segments between the vehicle's current location node and the order-related area node, and between the order-related area node and the delivery area node, and generate the path connectivity status. When the path connectivity status is connected and there is a road segment node connection relationship between the vehicle's current location node and the order-related area node, an order-vehicle-path candidate is generated. The order object, vehicle object, route object, route connectivity status, risk level marker, disturbance impact marker, candidate combination priority score, and prediction time slice are associated and stored to generate order-vehicle-route candidate combinations. Based on the order-vehicle-route candidate combination and the task-bound logistics status record, extract logistics execution constraints to generate execution constraint records, where: Generate execution constraint records, specifically: Read the order object, vehicle object, route object, route connectivity status, risk level marker, disturbance impact marker, candidate combination priority score, and prediction time slice from the order-vehicle-route candidate combination; Read the task execution stage, vehicle carrying status, order fulfillment status, business object marker, and collection time marker from the task-bound logistics status record; Based on the tags of order objects, vehicle objects, and business objects, the candidate combinations of order-vehicle-route are matched with the logistics status records bound to the task; When the order fulfillment status is unassigned, an order assignability constraint is generated. When the order fulfillment status is in the execution state, a semi-locking constraint is generated for the order; When the order fulfillment status is "completed", an order cannot be assigned a constraint. When the vehicle is in an idle or loadable state, a vehicle loadability constraint is generated. When the vehicle is fully loaded, a constraint is generated that the vehicle cannot be loaded. When the task execution phase is the waiting-to-depart phase or the loading preparation phase, the generated task can have adjustable constraints. When the task execution phase is the transportation execution phase, a task semi-locking constraint is generated; When the task execution phase is the acceptance completion phase, the generated task cannot have its constraints adjusted. When the path connectivity status is connected, executeable constraints are generated for the path. When the path connectivity status is not connected, an unexecutable constraint is generated. The order assignable constraints, order semi-locked constraints, order non-assignable constraints, vehicle load-bearing constraints, vehicle non-load-bearing constraints, task adjustable constraints, task semi-locked constraints, task non-adjustable constraints, path executable constraints, and path non-executable constraints are associated and stored with risk level markers, disturbance impact markers, candidate combination priority scores, and prediction time slices to generate execution constraint records. Based on the execution constraint record, the candidate combinations of order-vehicle-route are marked as executable, semi-locked, and non-executable, generating a candidate combination execution status record, where: Generate a record of the execution status of candidate combinations, specifically as follows: Read the order assignable constraints, order semi-locked constraints, order non-assignable constraints, vehicle load-bearing constraints, vehicle non-load-bearing constraints, task adjustable constraints, task semi-locked constraints, task non-adjustable constraints, path executable constraints, path non-executable constraints, risk level markers, disturbance impact markers, candidate combination priority scores, and prediction time slices from the execution constraint record; Read the order object, vehicle object, path object, and path connectivity status from the order-vehicle-path candidate combination; Based on order objects, vehicle objects, route objects, and prediction time slices, execution constraint records are matched with order-vehicle-route candidate combinations; When the execution constraint record contains order assignability constraint, vehicle capacity constraint, task adjustability constraint and path executable constraint at the same time, the corresponding order-vehicle-path candidate combination will be marked as executable. When there is a semi-locked order constraint or a semi-locked task constraint in the execution constraint record, and there are no constraints on unassignable orders, uncarryable vehicles, unadjustable tasks, or unexecuted paths, the corresponding order-vehicle-path candidate combination will be marked as semi-locked. When any of the following constraints exist in the execution constraint record: order cannot be assigned, vehicle cannot be carried, task cannot be adjusted, or path cannot be executed, the corresponding order-vehicle-path candidate combination will be marked as unexecuted. The executable, semi-locked, or non-executable states are associated with order objects, vehicle objects, path objects, risk level markers, disturbance impact markers, candidate combination priority scores, and prediction time slices for storage, generating candidate combination execution status records; The execution status records of candidate combinations are written into the order-vehicle-route candidate combination, generating a scheduling constraint mask table, where: Generate a scheduling constraint mask table, specifically as follows: Read the order object, vehicle object, route object, route connectivity status, risk level marker, disturbance impact marker, candidate combination priority score, and prediction time slice from the order-vehicle-route candidate combination; Read the executable, semi-locked, or non-executable states from the candidate combination execution status record; Based on the order object, vehicle object, route object, and predicted time slice, the execution status record of the candidate combination is written into the corresponding order-vehicle-route candidate combination; When the execution status of a candidate combination is recorded as executable, an executable mask flag is written. When the candidate combination execution status record is in a semi-locked state, write a semi-locked mask flag; When the execution status of a candidate combination is recorded as non-executable, an non-executable mask flag is written. The order object, vehicle object, route object, executable mask marker, semi-locked mask marker, non-executable mask marker, risk level marker, disturbance impact marker, candidate combination priority score, and prediction time slice are associated and stored to generate a scheduling constraint mask table.

[0027] In this embodiment, generating candidate scheduling feasible domains includes: Read the order-vehicle-route candidate combinations, candidate combination execution status records, and risk marker records from the scheduling constraint mask table, and generate candidate combination filtering records, where: The following steps are taken to generate candidate combination screening records: Read the order object, vehicle object, path object, executable mask marker, semi-locked mask marker, non-executable mask marker, risk level marker, disturbance impact marker, candidate combination priority score, and prediction time slice from the scheduling constraint mask table; Read the executable, semi-locked, and non-executable states from the candidate combination execution status record; Read the comprehensive risk score, low-risk label, medium-risk label, and high-risk label from the risk labeling record; Based on order objects, vehicle objects, route objects, and predicted time slices, the scheduling constraint mask table, candidate combination execution status records, and risk label records are matched. Write the matched order object, vehicle object, route object, candidate combination execution status, mask mark, comprehensive risk score, risk level mark, disturbance impact mark, and candidate combination priority score into the same filtering record; The candidate combination screening records are sorted from highest to lowest priority score to generate candidate combination screening records; Candidate combination screening records are classified according to executable status, semi-locked status and non-executable status. Order-vehicle-route candidate combinations corresponding to non-executable status are removed, and a set of executable candidate combinations is generated. Read the task-bound logistics status record and logistics operation risk association record corresponding to the semi-locked state, determine the vehicle holding range, order holding range, and route adjustment range for the semi-locked candidate combination, and generate an adjustment range limitation record, wherein: Generate an adjustment range limitation record, specifically as follows: Read the order objects, vehicle objects, route objects, prediction time slices, and candidate combination priority scores marked as semi-locked in the candidate combination filtering record; Based on the order object, vehicle object, and predicted time slice, read the task execution stage, vehicle carrying status, order fulfillment status, business object marker, and collection time marker from the task-bound logistics status record; Based on order objects, vehicle objects, route objects, and prediction time slices, read risk source object records, risk response object records, risk propagation intensity, disturbance impact markers, prediction credibility, and unified logistics space nodes from the logistics operation risk association records; When the task execution phase is the transportation execution phase, and the vehicle carrying status is either carrying capacity or fully loaded, the vehicle object is marked as a vehicle object to be kept, and a vehicle keeping range is generated. When the order fulfillment status is in the execution state and the order object belongs to the risk response object record, the order object is marked as a hold order object, and an order hold range is generated; When a path object contains road nodes or delivery area nodes corresponding to a risk source object record, the road segments in the path object that are connected to the risk source object record are marked as adjustable road segments. When a path object does not contain a road node or delivery area node corresponding to a risk source object record, the path object is marked as a path object to be kept. Based on the adjustable road segment, adjacent road segment nodes are read and included in the path adjustment range if the distance between the starting point and the ending point of the adjustable road segment is less than or equal to 5km. When the risk propagation intensity is greater than or equal to 0.7, the range of route adjustment will be expanded to adjacent road segment nodes that are both less than or equal to 10km from the start and end points of the adjustable road segment; The vehicle holding range, order holding range, route adjustment range, risk propagation intensity, disturbance impact marker, prediction confidence, risk level marker, and prediction time slice are associated and stored to generate an adjustment range limitation record; The executable candidate combination set and adjustment range limitation record are associated, and the schedulable object tag, restricted adjustment tag, and risk level tag are written to generate a candidate scheduling feasibility domain, wherein: The candidate scheduling feasible domain is generated as follows: Read the order objects, vehicle objects, route objects, candidate combination priority scores, and prediction time slices from the executable candidate combination set; Read the vehicle holding range, order holding range, route adjustment range, risk propagation intensity, disturbance impact marker, and risk level marker from the adjustment range limitation record; Based on order objects, vehicle objects, route objects, and predicted time slices, the set of executable candidate combinations is matched with the adjustment range limit records; When an order object, vehicle object, and route object are within the vehicle holding range, order holding range, and route adjustment range, write a schedulable object flag; Write a restricted adjustment flag when the order object, vehicle object, or route object is within the adjustment range boundary; The schedulable object tag, restricted adjustment tag, risk level tag, candidate combination priority score, predicted time slice, order object, vehicle object and route object are associated and stored to generate candidate scheduling feasible domains.

[0028] In this embodiment, the output of intelligent logistics scheduling results includes: Read the schedulable object markers, restricted adjustment markers, risk level markers, and order-vehicle-route candidate combinations from the candidate scheduling feasible domain to generate the scheduling execution basis, where: The basis for generating the scheduling execution is as follows: Read the schedulable object flags, restricted adjustment flags, risk level flags, candidate combination priority scores, predicted time slices, order objects, vehicle objects, and route objects from the candidate scheduling feasible domain; Identify candidate combinations of order-vehicle-route based on order objects, vehicle objects, and route objects; Filter candidate combinations of orders, vehicles, and routes that are marked as schedulable objects, and generate executable scheduling combination records; For order-vehicle-route candidate combinations marked with restricted adjustment, retain the limited range of the corresponding vehicle object, order object or route object, and generate a restricted scheduling combination record; Based on risk level marking and candidate combination priority scoring, the executable scheduling combination records and restricted scheduling combination records are sorted. The sorted order objects, vehicle objects, route objects, risk level markers, restricted adjustment markers, candidate combination priority scores, and predicted time slices are associated and stored to generate the basis for scheduling execution. Based on the scheduling execution criteria, matching order objects, vehicle objects, and route objects generates vehicle allocation results; Order delivery priorities are ranked based on vehicle allocation results and risk level markings to generate a delivery sequence; Based on delivery sequence, route objects, and restricted adjustment tags, a route execution plan and a departure plan are generated, where: Based on delivery sequence, route objects, and restricted adjustment tags, a route execution plan and departure plan are generated, specifically as follows: Read the order object, vehicle object, delivery priority, and predicted time slice from the delivery sequence; Read the path object, restricted adjustment flag, risk level flag, and candidate combination priority score from the scheduling execution basis; Based on order objects, vehicle objects, and predicted time slices, the delivery sequence is matched with route objects; When the path object does not have a restricted adjustment flag, the delivery node, road segment node and delivery area node are written in order of delivery priority to generate the path execution plan; When a path object has a restricted adjustment flag, retain the vehicle object, order object or path object corresponding to the restricted adjustment flag, and write the alternative road segment node within the allowed adjustment range to generate a restricted path execution plan. Determine the departure order based on vehicle object, delivery priority, route execution plan, and predicted time slice; When the risk level is marked as high risk, the corresponding vehicle object will be written into the priority departure queue. When the risk level is marked as low risk or medium risk, the candidate combination is written into the regular departure queue according to the priority score from high to low. The path execution plan, restricted path execution plan, priority departure queue, regular departure queue, order object, vehicle object, path object and prediction time slice are associated and stored to generate path execution plan and departure plan; Based on vehicle allocation results, delivery sequence, route execution plan, departure plan, and restricted adjustment flags, a rescheduling record is generated, and the intelligent logistics scheduling result is output.

[0029] refer to Figure 3 A smart logistics scheduling system based on big data analytics includes the following modules: The data acquisition and preprocessing module is used to collect multi-source logistics operation data and perform standardized processing to generate raw logistics status data. The Hampel filtering module is used to execute the Hampel filtering algorithm with spatial proximity consistency constraints based on the original logistics status data to generate reliable logistics status records. The state frame construction module is used to perform spatiotemporal alignment, spatial node mapping, and task state binding on trusted logistics state records to generate multi-source trusted logistics state frames. The model prediction module is used to build an improved FEDformer model and generate logistics situation prediction results based on multi-source reliable logistics status frames. The risk status generation module is used to extract the risk characteristics of the stream operation and generate a scheduling risk status table; The constraint mask generation module is used to generate a scheduling constraint mask table by combining the scheduling risk status table and logistics execution constraints; The feasible region generation module is used to filter candidate combinations based on the scheduling constraint mask table and generate candidate scheduling feasible regions. The scheduling output module is used to generate vehicle allocation results, delivery sequence, route execution plan, departure plan and rescheduling record based on candidate scheduling feasible domains, and output intelligent logistics scheduling results.

[0030] Example 1: In a continuous logistics scheduling cycle, the system receives multi-source logistics operation data from the order platform, vehicle terminals, warehousing system, road interface, and fulfillment system. This includes 1320 order operation records, 5280 vehicle operation records, 2760 warehousing operation records, 3840 road passage records, 1680 environmental impact records, and 5360 historical fulfillment records, totaling 20240 records. The data contains issues such as vehicle positioning drift, warehousing scanning delays, road status lags, and missing fulfillment status records. Duplicate records account for 2.3%, records with abnormal formats account for 1.5%, missing fields account for 3.7%, and there are 108 records of vehicle location jumps.

[0031] After the data enters the processing flow, the system standardizes field names, units of measurement, and time formats, generating 19,810 standard logistics operation data entries. After duplicate record removal, missing field marking, and abnormal format removal, 19,082 cleaning logistics operation data entries are formed. The system writes field types, field values, data source markers, collection time markers, and business object markers into the cleaning logistics operation data to generate raw logistics status data. Traditional methods only perform abnormal replacement based on a single field sequence, failing to distinguish between location drift and actual logistics disturbances, and easily treating road congestion and warehouse backlog as noise.

[0032] The system generates field-level window sequences based on raw logistics status data and applies Hampel filtering to vehicle speed, road traffic speed, warehouse outbound time, and order backlog. In the field-level window, the speed sequence for a vehicle object is 31 km / h, 33 km / h, 89 km / h, 32 km / h, and 34 km / h, with a median of 33 km / h, a median absolute deviation of 2, and a correction anomaly threshold of 8.4. 89 km / h is marked as a high-value anomaly. The traffic speed sequence for a road object is 42 km / h, 39 km / h, 21 km / h, 18 km / h, and 19 km / h. Neighboring roads synchronously decrease speed within the same window. Ordinary Hampel filtering replaces 18 km / h with 39 km / h, but the method of this invention continues to perform spatial proximity consistency judgment.

[0033] In spatial proximity consistency determination, the system retrieves spatially nearby objects using a proximity distance threshold of 3 km and a time matching threshold of 15 min. For vehicle speed jump records, there are 6 nearby objects and 1 object changing in the same direction, resulting in a spatial consistency score of 0.17, which is less than the spatial consistency determination threshold of 0.5. Therefore, it is judged as data noise, and the correction value is changed from 89 km / h to 33 km / h. For road traffic decrease records, there are 8 nearby objects and 7 objects changing in the same direction, resulting in a spatial consistency score of 0.88, which is greater than 0.5. Therefore, it is judged as real logistics disturbance, retaining a speed of 18 km / h and generating a low-value disturbance marker. A total of 152 initial anomaly markers were obtained, including 94 data noise markers and 58 real logistics disturbance markers. Traditional Hampel filtering incorrectly filters out 47 real disturbances, while this invention incorrectly filters out 9.

[0034] The system performs temporal merging and spatial node mapping on trusted logistics status records, forming 72 unified logistics spatial nodes, 428 records associated with the same node, and 552 records associated with adjacent nodes. The status frame contains correction values, disturbance markers, spatial consistency scores, unified logistics spatial nodes, associated node numbers, task execution stages, vehicle carrying status, and order fulfillment status, generating multi-source trusted logistics status frames. A certain status frame contains 86 vehicle objects, 1320 order objects, 16 warehouse objects, 42 road objects, and 12 delivery area objects, including 21 high-value disturbance markers and 37 low-value disturbance markers.

[0035] During the training phase, the method of this invention uses the same training sample set as the traditional FEDformer method. There are 10,000 training samples, 2,000 validation samples, and 2,000 test samples. Each training sample set includes correction values ​​within a 30-minute prediction time window, disturbance markers, spatial consistency scores, unified logistics space nodes, associated node numbers, task execution stage, vehicle carrying status, and order fulfillment status, as well as labels for changes in order fulfillment, vehicle operation, warehousing operations, road traffic, and delivery area risk. In one training sample set, there are 72 input nodes, 196 associated edges, 19 disturbance markers, an average spatial consistency score of 0.71, an actual order fulfillment change label of 0.22, an actual vehicle operation change label of 0.28, an actual warehousing operation change label of 0.36, an actual road traffic change label of 0.64, and an actual delivery area risk change label of 0.52.

[0036] After training samples are input into the improved FEDformer model, the perturbation-fidelity dual-track decomposition and reconstruction layer writes the unperturbed records into the normal state input track and the high-value perturbation markers and low-value perturbation markers into the perturbed state input track, generating normal state logistics representations and perturbed state logistics representations. The logistics heterogeneous causal spectrum graph interaction layer establishes causal connections based on order nodes, vehicle nodes, warehouse nodes, road nodes, and delivery area nodes. The low-frequency trend band transmission intensity is 0.43, the mid-frequency fluctuation band transmission intensity is 0.58, and the high-frequency perturbation band transmission intensity is 0.81. The joint feature reliable band calibration layer generates a band reliability weight of 0.86, a band residual of 0.19, and a prediction reliability of 0.84. Traditional FEDformer directly inputs the corrected sequence, failing to retain true perturbation markers, and the average error of logistics situation prediction for the test set is 0.119; the method of this invention achieves an average error of 0.069 for logistics situation prediction in the test set.

[0037] In a simulated scheduling application, the order backlog in a certain delivery area increased from 46 to 128 items, road traffic speed decreased from 40 km / h to 18 km / h, and warehouse outbound time increased from 12 minutes to 26 minutes. Traditional rule-based scheduling rearranged all vehicles, involving 43 vehicles and generating 168 candidate combinations, of which 48 were unexecutable, representing an unexecutable rate of 28.6%, with an average solution generation time of 8.7 minutes. The traditional Hampel filtering plus FEDformer method generated 214 candidate combinations, with 39 unexecutable combinations, representing an unexecutable rate of 18.2%. The method of this invention generated 260 order-vehicle-route candidate combinations, of which 171 were executable, 65 were semi-locked, and 24 were unexecutable, representing an unexecutable rate of 9.2%. Furthermore, it only generated alternative routes within a 10km radius before and after the risky road segment, ensuring that the executing vehicles and orders were not disrupted.

[0038] Comparative experimental results show that the order timeout rate is 10.9% for the traditional rule-based scheduling method, 7.4% for the traditional Hampel filtering plus FEDformer method, and 4.0% for the method of this invention. The invalid detour mileage for vehicles is 342km for the traditional rule-based scheduling method, 226km for the traditional Hampel filtering plus FEDformer method, and 118km for the method of this invention. The average waiting time is 32min for the traditional rule-based scheduling method, 21min for the traditional Hampel filtering plus FEDformer method, and 11min for the method of this invention. The rescheduled vehicle ratio is 52.1% for the traditional rule-based scheduling method, 35.9% for the traditional Hampel filtering plus FEDformer method, and 20.6% for the method of this invention. In the test sample, the method of this invention improves the accuracy of real logistics disturbance identification to 92.8%, reduces the false filtering rate of real disturbances to 6.4%, and increases the on-time departure rate of vehicles to 95.1%, proving that this invention can improve the accuracy of state identification, prediction reliability, and scheduling execution continuity in scenarios with inconsistent multi-source data and dynamic disturbances.

[0039] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A smart logistics scheduling method based on big data analysis, characterized in that, include: Collect multi-source logistics operation data and perform preprocessing to generate raw logistics status data; Based on the original logistics status data, the Hampel filtering algorithm is executed, spatial proximity consistency constraints are introduced, window scale selection, credibility threshold correction and spatial proximity consistency discrimination are performed to generate a reliable logistics status record. Spatiotemporal alignment, spatial node mapping, and task status binding are performed on trusted logistics status records to generate multi-source trusted logistics status frames. An improved FEDformer model is constructed. Multi-source reliable logistics state frames are input into the improved FEDformer model. After processing by the perturbation-fidelity dual-track decomposition and reconstruction layer, the logistics heterogeneous causal spectrum interaction layer, and the joint feature reliable frequency band calibration layer, logistics situation prediction results are generated. Based on the logistics situation prediction results, extract logistics operation risk characteristics and generate a scheduling risk status table; Based on the scheduling risk status table, candidate combinations of order-vehicle-route are generated, and the execution status is marked in combination with logistics execution constraints to generate a scheduling constraint mask table. Based on the scheduling constraint mask table, candidate combination screening and adjustment range limitation are performed to generate candidate scheduling feasible domains; Based on the candidate scheduling feasible domain, vehicle allocation results, delivery sequence, route execution plan, departure plan and rescheduling record are generated, and the intelligent logistics scheduling result is output.

2. The intelligent logistics scheduling method based on big data analysis according to claim 1, characterized in that, The generation of raw logistics status data includes: The system reads order operation records, vehicle operation records, warehousing operation records, road traffic records, environmental impact records, and historical performance records from the logistics business platform to generate multi-source logistics operation data. Standardized logistics operation data is generated by unifying field names, units of measurement, and time formats for multi-source logistics operation data. Perform duplicate record removal, missing field marking, and abnormal format removal on standard logistics operation data to generate clean logistics operation data; Write data source markers, collection time markers, and business object markers into the cleaning logistics operation data to generate raw logistics status data.

3. The intelligent logistics scheduling method based on big data analysis according to claim 1, characterized in that, The generation of a reliable logistics status record includes: Read the field types, field values, data source markers, collection time markers, and business object markers from the raw logistics status data, configure the Hampel filter window scale according to the field types, and generate a field-level window sequence. The absolute deviation of the window median and the middle median is calculated based on the field-level window sequence. The data source credibility is generated by combining the data source label and the data source credibility is used to correct the anomaly judgment threshold and generate the field-level anomaly threshold. Anomaly detection is performed on field values ​​based on field-level anomaly thresholds to generate initial anomaly markers; For the business object corresponding to the initial anomaly label, retrieve the spatial neighboring objects, calculate the same-direction change status of the spatial neighboring objects, and generate a spatial consistency score; The initial anomaly markers are constrained and verified based on the spatial consistency score, which distinguishes between data noise and real logistics disturbances. Correction values ​​are generated for data noise, and disturbance markers are generated for real logistics disturbances. The correction values, disturbance markers, and spatial consistency scores are written into the trusted logistics status record.

4. The intelligent logistics scheduling method based on big data analysis according to claim 1, characterized in that, The generation of multi-source trusted logistics status frames includes: Read the collection time stamp, business object stamp, correction value, disturbance stamp and spatial consistency score from the trusted logistics status record, perform time merging according to the unified scheduling time window, and generate time-aligned logistics status record; Based on time-aligned logistics status records, business object tags and spatial location fields are read, and order objects, vehicle objects, warehouse objects, road objects, and delivery area objects are mapped to unified logistics spatial nodes to generate spatial node mapping records. Based on the disturbance markers and spatial consistency scores in the spatial node mapping records and trusted logistics status records, establish the status association between nodes and generate logistics node association records; Based on the logistics node association records, read the task execution stage, vehicle carrying status and order fulfillment status, bind the task status, and generate task-bound logistics status records; The time-aligned logistics status record, spatial node mapping record, logistics node association record, and task-bound logistics status record are fused together to generate a multi-source trusted logistics status frame.

5. The intelligent logistics scheduling method based on big data analysis according to claim 1, characterized in that, The generated logistics situation prediction results include: An improved FEDformer model is constructed, which includes a perturbation-fidelity dual-track decomposition and reconstruction layer, a logistics heterogeneous causal spectrum graph interaction layer, and a joint feature reliable frequency band calibration layer. Read the time-aligned logistics status record, spatial node mapping record, logistics node association record and task-bound logistics status record from the multi-source trusted logistics status frame, and generate the model input sequence according to the prediction time window; The model input sequence is input into the perturbation-fidelity dual-track decomposition and reconstruction layer. The normal state input track is constructed based on the correction value, and the perturbation dynamic input track is constructed based on the perturbation label and spatial consistency score, generating normal state logistics representation and perturbation dynamic logistics representation. Normal logistics representation and disturbed logistics representation are input into the interaction layer of the heterogeneous causal spectrum graph of logistics. A spectrum interaction graph is constructed based on order nodes, vehicle nodes, warehouse nodes, road nodes and delivery area nodes. Causal spectrum interaction representation is generated according to the logistics node association records. The causal spectrum interaction representation is input into the joint feature trusted band calibration layer, and trusted band representation is generated by combining data source labeling, spatial consistency score and band residual record. Predictive decoding is performed based on reliable frequency band representation to generate logistics situation prediction results; The improved FEDformer model was trained by using a combination of logistics situation prediction error, disturbance reconstruction error, causal spectrum interaction error, and reliable frequency band calibration error as the joint optimization objective. The network parameters of the disturbance fidelity dual-track decomposition reconstruction layer, the logistics heterogeneous causal spectrum interaction layer, and the joint feature reliable frequency band calibration layer were continuously optimized. When the change of the joint loss value in five consecutive training rounds was less than 0.001, the improved FEDformer model was considered to have completed convergence training.

6. The intelligent logistics scheduling method based on big data analysis according to claim 1, characterized in that, The generation of the scheduling risk status table includes: Read the prediction time window, node prediction value, disturbance impact marker and prediction confidence from the logistics situation prediction results, and generate prediction result analysis record; Based on the prediction results, the analysis records are used to extract logistics operation risk characteristics associated with orders, vehicles, warehouses, roads, and delivery areas, generating object-level risk characteristic records; Risk propagation associations are established based on object-level risk feature records, disturbance impact markers, and logistics node association records to generate logistics operation risk association records. Based on the risk association records and prediction reliability of logistics operation, the risk characteristics of logistics operation are labeled with risk levels, and risk label records are generated. By integrating the fields of object-level risk characteristic records, logistics operation risk association records, and risk label records, a scheduling risk status table is generated.

7. The intelligent logistics scheduling method based on big data analysis according to claim 1, characterized in that, The generation of the scheduling constraint mask table includes: Read the object-level risk characteristic records, logistics operation risk association records, and risk marker records from the scheduling risk status table to generate the basis for candidate combination generation; Based on the candidate combination generation criteria and the spatial node mapping records in the multi-source trusted logistics status frames, order objects, vehicle objects and route objects are matched to generate order-vehicle-route candidate combinations. Based on the order-vehicle-route candidate combination and the task-bound logistics status record, extract logistics execution constraints and generate execution constraint records; Based on the execution constraint record, the candidate combinations of order-vehicle-route are marked as executable, semi-locked and non-executable, and an execution status record of the candidate combinations is generated. Write the execution status record of the candidate combination into the order-vehicle-route candidate combination to generate a scheduling constraint mask table.

8. The intelligent logistics scheduling method based on big data analysis according to claim 1, characterized in that, The generated candidate scheduling feasible domain includes: Read the order-vehicle-route candidate combinations, candidate combination execution status records, and risk marker records from the scheduling constraint mask table, and generate candidate combination screening records; Candidate combination screening records are classified according to executable status, semi-locked status and non-executable status. Order-vehicle-route candidate combinations corresponding to non-executable status are removed, and a set of executable candidate combinations is generated. Read the task-bound logistics status record and logistics operation risk association record corresponding to the semi-locked state, determine the vehicle holding range, order holding range and route adjustment range of the semi-locked candidate combination, and generate the adjustment range limitation record; The executable candidate combination set and the adjustment range limit record are associated, and the schedulable object flag, restricted adjustment flag and risk level flag are written to generate the candidate scheduling feasible domain.

9. The intelligent logistics scheduling method based on big data analysis according to claim 1, characterized in that, The output of intelligent logistics scheduling results includes: Read the schedulable object markers, restricted adjustment markers, risk level markers, and order-vehicle-route candidate combinations from the candidate scheduling feasible domain to generate the basis for scheduling execution; Based on the scheduling execution criteria, matching order objects, vehicle objects, and route objects generates vehicle allocation results; Order delivery priorities are ranked based on vehicle allocation results and risk level markings to generate a delivery sequence; Generate route execution plans and departure plans based on delivery sequence, route objects, and restricted adjustment tags; Based on vehicle allocation results, delivery sequence, route execution plan, departure plan, and restricted adjustment flags, a rescheduling record is generated, and the intelligent logistics scheduling result is output.

10. A smart logistics scheduling system based on big data analytics, executing the smart logistics scheduling method based on big data analytics as described in any one of claims 1 to 9, characterized in that, Includes the following modules: The data acquisition and preprocessing module is used to collect multi-source logistics operation data and perform standardized processing to generate raw logistics status data. The Hampel filtering module is used to execute the Hampel filtering algorithm with spatial proximity consistency constraints based on the original logistics status data to generate reliable logistics status records. The state frame construction module is used to perform spatiotemporal alignment, spatial node mapping, and task state binding on trusted logistics state records to generate multi-source trusted logistics state frames. The model prediction module is used to build an improved FEDformer model and generate logistics situation prediction results based on multi-source reliable logistics status frames. The risk status generation module is used to extract the risk characteristics of the stream operation and generate a scheduling risk status table; The constraint mask generation module is used to generate a scheduling constraint mask table by combining the scheduling risk status table and logistics execution constraints; The feasible region generation module is used to filter candidate combinations based on the scheduling constraint mask table and generate candidate scheduling feasible regions. The scheduling output module is used to generate vehicle allocation results, delivery sequence, route execution plan, departure plan and rescheduling record based on candidate scheduling feasible domains, and output intelligent logistics scheduling results.

Images