Intelligent service resource scheduling method and system based on big data

Abstract

The invention provides an intelligent business resource scheduling method and system based on big data, and relates to the technical field of data processing, and the method comprises the steps: 1, capturing equipment layer sensor data, business layer order data and logistics layer vehicle track data in real time through distributed data collection nodes, and carrying out the real-time acquisition of the data; a dynamic time window division algorithm is adopted to carry out time sequence alignment processing on the heterogeneous data flow, a ternary state vector matrix with a unified timestamp is generated, the row dimension of the matrix represents a collection time sequence, and the column dimension of the matrix corresponds to equipment layer operation parameters, a service layer task queue state and logistics layer transportation node coordinates respectively. According to the invention, through multi-source data fusion, three-dimensional space analysis and an intelligent algorithm, resource gap accurate identification and cross-level scheduling optimization are realized, and the business processing efficiency and the resource utilization rate are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to an intelligent business resource scheduling method and system based on big data. BACKGROUND

[0002] The existing resource scheduling method independently processes device layer (such as sorting device load), business layer (such as order queue) and logistics layer (such as vehicle trajectory) data, resulting in isolated and unsynchronized time sequence of each layer data. For example, real-time load data collected by device layer sensors cannot be dynamically mapped to business layer task allocation strategy, and logistics layer vehicle transportation delay cannot be fed back to sorting task migration decision in time, causing lag and deviation of resource scheduling decision.

[0003] Dependence on fixed threshold or historical experience to perform scheduling (such as departure according to fixed time period per day, task allocation according to historical warehouse efficiency average), cannot cope with real-time changing business scenarios.

[0004] For example, when sudden orders cause task backlog in a warehouse, static rules are difficult to trigger cross-warehouse task migration in time; when temporary congestion occurs in the road network, vehicle scheduling is still performed according to the original route, eventually causing task processing period to be extended and resource utilization to be low. SUMMARY

[0005] The technical problem to be solved by the present application is to provide an intelligent business resource scheduling method and system based on big data, which realizes dynamic and accurate diagnosis of resource gap.

[0006] To solve the above technical problems, the technical solution of the present application is as follows: In a first aspect, an intelligent business resource scheduling method based on big data, the method comprising: Step 1, real-time capture of device layer sensor data, business layer order data and logistics layer vehicle trajectory data by distributed data acquisition nodes, time sequence alignment processing of heterogeneous data streams by dynamic time window division algorithm, generation of a three-element state vector matrix with uniform time stamp, the row dimension of the matrix representing the time sequence of acquisition, and the column dimension corresponding to the device layer operating parameter, the business layer task queue state and the logistics layer transportation node coordinate respectively; Step 2, based on the three-element state vector matrix, extracting device layer real-time load rate, business layer task backlog and logistics layer vehicle arrival time deviation from the state vector in the current time window, mapping the three indicators to three-dimensional Euclidean space to form a dynamic state point by range normalization algorithm; connecting the state points of the three consecutive time windows to build a dynamic triangle, determining the dominant factor type of the current resource gap by calculating the coordinate offset of the triangle centroid and the offset direction angle, and generating a real-time processing capacity correction factor according to the proportional relationship between the centroid offset amplitude and the preset threshold value; Step 3, based on the real-time processing capacity correction factor, combined with the real-time sorting capacity matrix of each warehouse and the transportation road network topology data, the Hungarian algorithm is used to calculate the final migration path of the cross-warehouse sorting task, and according to the space-time relationship between the current position of the vehicle and the target warehouse, a vehicle scheduling priority queue with weight value is dynamically generated.

[0007] Further, by means of the distributed data acquisition node, real-time capture of equipment layer sensor data, business layer order data and logistics layer vehicle trajectory data is realized, a dynamic time window division algorithm is used for time sequence alignment processing of heterogeneous data streams, and a three-element state vector matrix with unified timestamp is generated. The row dimension of the matrix represents the time sequence of the acquisition, and the column dimension corresponds to the equipment layer running parameter, the business layer task queue state and the logistics layer transportation node coordinate respectively, including: The sensor data is collected by the edge computing node of the equipment layer in real time, and the sensor data includes device current, temperature and vibration frequency, and the sensor data is aggregated into an equipment layer running parameter vector, which includes real-time load rate and health status score of the device; The equipment layer running parameter vector is input into the business layer task queue dynamic evaluation module, combined with the task queue length, order emergency level and device real-time load rate of real-time order data, a business layer task queue state vector is generated, which includes task backlog, priority weighting coefficient and device load matching degree index; Based on the priority weighting coefficient in the business layer task queue state vector, the real-time coordinates and road condition data of the vehicle-mounted GPS module and the traffic API interface are called through the logistics layer dynamic path planning engine to calculate the estimated arrival time deviation of each transportation node, and a logistics layer transportation node vector is generated, including vehicle coordinates, space-time weight and task correlation score; Taking the generation time of the logistics layer transportation node vector as the trigger condition, and obtaining the latest data of the equipment layer running parameter vector and the business layer task queue state vector, for the data time sequence deviation of the equipment layer and the logistics layer, a dynamic interpolation algorithm is used to forcibly align the timestamp of the equipment layer running parameter vector to the generation time axis of the logistics layer transportation node vector; The time sequence aligned equipment layer running parameter vector, business layer task queue state vector and logistics layer transportation node vector are spliced according to the unified time axis to generate a three-element state vector matrix.

[0008] Further, the equipment layer running parameter vector is input into the business layer task queue dynamic evaluation module, combined with the task queue length, order emergency level and device real-time load rate of real-time order data, a business layer task queue state vector is generated, which includes task backlog, priority weighting coefficient and device load matching degree index, including: Based on the equipment layer running parameter vector, the maximum loadable task quantity of the equipment per unit time is calculated as the capacity baseline value according to the real-time load rate of the equipment; The task queue length of the business layer real-time order data is subtracted from the capacity baseline value, if the queue length is greater than or equal to the capacity baseline value, a positive task backlog is generated, if the queue length is less than the capacity baseline value, it is marked as negative redundant capacity; According to the difference between the delivery deadline and the current time in the business layer real-time order data, the order emergency level parameter is dynamically generated, and the order emergency level parameter is combined with the real-time load rate of the equipment in proportion to generate the priority weighting coefficient; Through the covariance of the real-time load rate of the equipment and the task queue length, the equipment load matching degree index is calculated, if the covariance value is less than the preset threshold, it is marked as a load imbalance state; The task backlog, priority weighting coefficient and load matching degree index are packaged into a business layer task queue state vector according to a unified dimension.

[0009] Further, based on the three-element state vector matrix, the equipment layer real-time load rate, the business layer task backlog and the vehicle arrival time deviation of the logistics layer are extracted from the state vector in the current time window, and the three indexes are mapped to a three-dimensional Euclidean space to form a dynamic state point through a range normalization algorithm; connecting the state points of the three consecutive time windows to build a dynamic triangle, calculating the coordinate offset and the offset direction angle of the center of gravity of the triangle to determine the dominant factor type of the current resource gap, and generating a real-time processing capacity correction factor according to the proportional relationship between the center of gravity offset amplitude and the preset threshold, including: Based on the row vector of the three-element state vector matrix, the real-time load rate is extracted from the equipment layer running parameter vector, the task backlog is extracted from the business layer task queue state vector, and the vehicle arrival time deviation is extracted from the logistics layer transportation node vector; The real-time load rate, task backlog and arrival time deviation are subjected to range normalization processing, and the normalized equipment layer index is mapped to the X-axis coordinate of the three-dimensional Euclidean space, the business layer index is mapped to the Y-axis coordinate, and the logistics layer index is mapped to the Z-axis coordinate to generate a dynamic state point in the current time window; The dynamic state points of the adjacent three time windows are connected in time sequence to build a dynamic triangle topology, the center of gravity coordinates of the dynamic triangle are calculated, and the current center of gravity coordinates and the previous period center of gravity coordinates are subjected to vector difference operation to obtain the center of gravity offset vector; The projection components of the center of gravity offset vector in the X-axis, Y-axis and Z-axis directions are decomposed, and the dominant factor type is determined according to the proportional relationship of the absolute values of the projection components in each axis direction; According to the ratio of the module length of the barycentric offset vector to the preset threshold value, a correction factor generation rule is dynamically selected, when the ratio is in the interval of 1.2 to 1.5 times the threshold value, a correction factor in linear relationship with the offset module length is generated; when the ratio exceeds 1.5 times the threshold value, an exponential correction factor proportional to the square of the offset amplitude is generated.

[0010] Further, the projection components of the barycentric offset vector in the X-axis, Y-axis and Z-axis directions are decomposed, and the dominant factor type is determined according to the proportion relationship of the absolute values of the axial projection components, including: When the component in the X-axis direction is the maximum projection component, it is determined as the equipment resource gap dominant type; When the Y-axis direction component is the maximum projection component, it is determined as the task backlog dominant type; When the Z-axis direction component is the maximum projection component, it is determined as the transportation delay dominant type.

[0011] Further, based on the real-time processing capacity correction factor, the real-time sorting capacity matrix of each warehouse and the transportation road network topology data are combined, and the Hungarian algorithm is used to calculate the final migration path of the cross-warehouse sorting task, and a vehicle scheduling priority queue with a weight value is dynamically generated according to the space-time relationship between the current position of the vehicle and the target warehouse, including: The real-time processing capacity correction factor is used to dynamically adjust the concurrent task upper limit threshold of each warehouse sorting equipment, and the adjusted concurrent task upper limit threshold is obtained; The adjusted concurrent task upper limit threshold is combined with the historical sorting efficiency data of the warehouse to generate an updated sorting capacity matrix, wherein the historical sorting efficiency data includes the task processing success rate and the average sorting time per unit time; According to the updated sorting capacity matrix and the real-time passing time between nodes in the transportation road network topology, a bipartite graph model is constructed with the warehouse as the vertex and the cross-warehouse task migration cost as the edge weight, wherein the edge weight value is generated by the sorting capacity difference between the source warehouse and the target warehouse, the time cost weight of the transportation path, and the logistics layer task correlation score; The Hungarian algorithm is used to solve the final matching path of the bipartite graph model, and the vertex matching combination with the minimum edge weight sum is calculated by iteration, and the final path and migration task quantity of the cross-warehouse task migration are output, and a sorting optimization instruction set containing the target warehouse code, task allocation list and time window constraint is generated; Based on the target warehouse code and the migration task quantity in the sorting optimization instruction set, the geographical position and task associated vehicle information of the corresponding target warehouse are extracted from the logistics layer transportation node vector, and the space-time cost of reaching the target warehouse is calculated combined with the real-time GPS coordinates of the vehicle; According to the matching degree of the space-time cost weight and the remaining loading capacity of the vehicle, a scheduling priority index is generated, and the vehicle scheduling instruction queue is generated in descending order of the scheduling priority index.

[0012] Further, the Hungarian algorithm is used to solve the final matching path of the bipartite graph model, and the vertex matching combination with the minimum total edge weight is calculated by iteration, and the final path and the task migration amount of cross-warehouse task migration are output, and the sorting optimization instruction set containing the target warehouse code, task allocation list and time window constraint is generated, including: Based on the edge weight value of the bipartite graph model, the initial matching relationship between warehouses is established, and each source warehouse vertex is assigned an initial matching weight, wherein the source warehouse vertex is marked as the maximum edge weight value, and the target warehouse vertex is marked as zero; By dynamically adjusting the matching relationship between the source warehouse and the target warehouse, the warehouse pairing combination with the minimum total migration cost is iteratively calculated, and in each iteration, the current edge weight value and the difference between the warehouse processing capacity are compared, and when the remaining processing capacity of the target warehouse is greater than the task amount to be migrated, the matching path is updated; When all source warehouse vertices determine the final matching path, the vertex matching combination with the minimum total edge weight is generated according to the adjusted difference in the marking matrix, and the final path set of cross-warehouse task migration and the corresponding migration task amount are output; Based on the pairing relationship between the source warehouse and the target warehouse in the final path set, combined with the migration task amount and the time window constraint, a sorting optimization instruction set containing the unique code of the target warehouse, the task allocation ratio and the sorting deadline is generated.

[0013] In a second aspect, an intelligent business resource scheduling system based on big data includes: A data alignment module is configured to capture device layer sensor data, business layer order data and logistics layer vehicle trajectory data in real time through a distributed data acquisition node, and perform time sequence alignment processing on heterogeneous data streams using a dynamic time window division algorithm to generate a three-element state vector matrix with a unified timestamp, wherein the row dimension of the matrix represents the time sequence of acquisition, and the column dimension corresponds to the device layer operating parameters, the business layer task queue state and the logistics layer transportation node coordinates, respectively. An index analysis module is configured to extract three indexes, i.e., device layer real-time load rate, business layer task backlog and logistics layer vehicle arrival time deviation, from the state vectors in the current time window based on the three-element state vector matrix, map the three indexes to a dynamic state point in a three-dimensional Euclidean space through a range normalization algorithm, construct a dynamic triangle by connecting the state points of three consecutive time windows, determine the dominant factor type of the current resource gap by calculating the coordinate offset of the triangle centroid and the offset direction angle, and generate a real-time processing capacity correction factor according to the proportional relationship between the centroid offset amplitude and a preset threshold value. The task scheduling module is configured to calculate the final migration path of the cross-warehouse sorting task by using the Hungarian algorithm based on the real-time processing capacity correction factor, the real-time sorting capacity matrix of each warehouse, and the transportation road network topology data, and dynamically generate a vehicle scheduling priority queue with a weight value according to the space-time relationship between the current position of the vehicle and the target warehouse.

[0014] In a third aspect, a computing device includes: one or more processors; a storage device storing one or more programs, when executed by the one or more processors, cause the one or more processors to implement the method.

[0015] In a fourth aspect, a computer-readable storage medium stores a program, which, when executed by a processor, implements the method.

[0016] The above scheme of the present application at least has the following beneficial effects: By collecting three-layer data of equipment, business, and logistics through distributed nodes, and combining time sequence alignment technology, a three-element state vector matrix of unified space-time reference is constructed to eliminate decision bias caused by data heterogeneity. The load rate, task backlog, and transportation deviation are mapped to a three-dimensional Euclidean space, and the dominant factor (equipment overload / task surge / logistics delay) of the resource bottleneck is intuitively identified through dynamic triangular barycentric offset analysis, avoiding the one-sidedness of traditional single-index monitoring. Based on the barycentric offset amplitude, a correction factor is dynamically generated to adjust the warehouse processing capacity exponentially or linearly, so that the system remains flexible during business peaks and avoids resource waste during troughs.

[0017] The Hungarian algorithm is combined to solve the cross-warehouse task migration problem, and the sorting capacity difference, transportation time efficiency, and historical cooperation relationship are comprehensively considered to realize resource allocation with minimum end-to-end cost. By fusing real-time traffic and vehicle loading state, a priority queue with weight is generated to preferentially schedule vehicles with low space-time cost and high matching degree, reducing invalid transportation and waiting time. A complete closed loop is formed from data collection to scheduling execution, and through continuous monitoring and correction factor iteration, the system can automatically adapt to business fluctuations and improve overall operation efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 is a flowchart of an intelligent business resource scheduling method based on big data provided by an embodiment of the present application.

[0019] Figure 2 is a schematic diagram of an intelligent business resource scheduling system based on big data provided by an embodiment of the present application. DETAILED DESCRIPTION

[0020] Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it is understood that the present disclosure can be embodied in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure can be thoroughly understood and fully conveyed to those skilled in the art.

[0021] As Figure 1 shown, an embodiment of the present application proposes an intelligent business resource scheduling method based on big data, which comprises the following steps: Step 1, real-time capture equipment layer sensor data, business layer order data and logistics layer vehicle trajectory data through distributed data acquisition nodes, adopt dynamic time window division algorithm for time sequence alignment processing of heterogeneous data stream, generate three-state vector matrix with unified timestamp, the row dimension of the matrix represents the time sequence of acquisition, and the column dimension corresponds to the equipment layer running parameter, the business layer task queue state and the logistics layer transportation node coordinate respectively; Step 2, based on the three-state vector matrix, extract the equipment layer real-time load rate, the business layer task backlog and the logistics layer vehicle arrival time deviation from the state vector in the current time window, map the three indexes to three-dimensional Euclidean space to form a dynamic state point through range normalization algorithm; connect the state points of the continuous three time windows to build a dynamic triangle, calculate the coordinate offset of the triangle barycenter and the offset direction angle, determine the dominant factor type of the current resource gap, and generate a real-time processing capacity correction factor according to the proportional relationship between the barycenter offset amplitude and the preset threshold value; Step 3, based on the real-time processing capacity correction factor, combine the real-time sorting capacity matrix of each warehouse and the transportation road network topology data, adopt Hungarian algorithm to calculate the final migration path of cross-warehouse sorting tasks, and simultaneously generate a vehicle scheduling priority queue with weighted value according to the spatio-temporal relationship between the current position of the vehicle and the target warehouse.

[0022] In the embodiment of the present application, multi-dimensional data (such as equipment state, order queue, vehicle trajectory) of equipment layer, business layer and logistics layer are collected to build a full-link business view, avoiding one-sidedness of single-dimensional decision.

[0023] Dynamic time windowing and interpolation algorithms align heterogeneous data time series, addressing data frequency discrepancies across different systems and ensuring spatiotemporal consistency in data analysis. A ternary state vector matrix is ​​generated, transforming complex business data into a standardized, computable matrix structure. This provides a unified input format for subsequent algorithms (such as three-dimensional spatial analysis and matrix operations), reducing model adaptation costs. Equipment load rates, task backlogs, and transportation deviations are mapped to three-dimensional Euclidean space. Dynamic state points and triangular topology are used to visualize resource state evolution, allowing rapid identification of the dominant factors (equipment / task / transportation) responsible for current resource gaps and accurate attribution of issues.

[0024] A real-time processing capacity correction factor (linear or exponential) is dynamically generated based on the magnitude of the center of gravity offset. Minor deviations use linear adjustments, while severe deviations trigger exponential corrections. This ensures the system responds appropriately to varying degrees of resource pressure, avoiding over-adjustment or under-response. The Hungarian algorithm solves a bipartite graph model, integrating factors such as sorting capacity, transportation costs, and task relevance to calculate the optimal cross-warehouse task migration path, balancing warehouse loads, reducing overall processing costs, and improving sorting efficiency. Priority queues are generated based on real-time vehicle location, spatiotemporal costs (such as road conditions and distance), and loading capacity, enabling dynamic matching of tasks and vehicles, reducing idle travel and waiting time, and improving logistics resource utilization.

[0025] In a preferred embodiment of the present invention, in step 1, distributed data acquisition nodes are used to capture device layer sensor data, business layer order data, and logistics layer vehicle trajectory data in real time. A dynamic time window partitioning algorithm is used to perform time series alignment processing on the heterogeneous data streams to generate a ternary state vector matrix with a unified timestamp. The row dimension of the matrix represents the acquisition time series, and the column dimension corresponds to the device layer operating parameters, the business layer task queue status, and the logistics layer transportation node coordinates. The matrix may include: Step 100: Using edge computing nodes at the device layer to collect sensor data in real time, the sensor data includes device current, temperature, and vibration frequency, and aggregate the sensor data into a device layer operating parameter vector, which includes the device's real-time load rate and health status score. Step 101: Input the device layer operation parameter vector into the business layer task queue dynamic evaluation module, and generate the business layer task queue dynamic evaluation module based on the task queue length, order urgency level and device real-time load rate of the real-time order data. The service layer task queue state vector includes the task backlog, priority weighting coefficient, and device load matching index, specifically including: Step 1010: Calculate the maximum load capacity of the device per unit time as the capacity baseline value based on the device real-time load rate in the device layer operation parameter vector; Step 1011, difference operation is performed on the task queue length of the service layer real-time order data and the capacity baseline value, if the queue length is greater than or equal to the capacity baseline value, a positive task backlog is generated, if the queue length is less than the capacity baseline value, it is marked as a negative redundant capacity; Step 1012, according to the difference between the delivery deadline and the current time in the service layer real-time order data, an order emergency level parameter is dynamically generated, and the order emergency level parameter is combined with the equipment real-time load rate in proportion to generate a priority weighting coefficient; Step 1013, the equipment load matching degree index is calculated through the covariance of the equipment real-time load rate and the task queue length, if the covariance value is less than a preset threshold, it is marked as a load imbalance state; Step 1014, the task backlog, the priority weighting coefficient and the load matching degree index are packaged into a service layer task queue state vector according to a unified dimension; Step 102, based on the priority weighting coefficient in the service layer task queue state vector, the real-time coordinates and traffic data of the vehicle-mounted GPS module and the traffic API interface are called through the logistics layer dynamic path planning engine to calculate the estimated arrival time deviation of each transportation node, and a logistics layer transportation node vector containing vehicle coordinates, space-time weight and task correlation degree score is generated; Step 103, taking the generation time of the logistics layer transportation node vector as a trigger condition, and obtaining the latest data of the equipment layer running parameter vector and the service layer task queue state vector, for the data time sequence deviation of the equipment layer and the logistics layer, a dynamic interpolation algorithm is used to forcibly align the time stamp of the equipment layer running parameter vector to the generation time axis of the logistics layer transportation node vector; Step 104, the equipment layer running parameter vector, the service layer task queue state vector and the logistics layer transportation node vector after time sequence alignment are spliced according to a unified time axis to generate a ternary state vector matrix.

[0026] In the embodiment of the application, the equipment current , temperature and vibration frequency are collected in real time by the equipment layer edge computing node, and these original data directly reflect the basic state of equipment operation. The real-time load rate is calculated by the ratio of equipment current to rated current , wherein is the rated current of the equipment. The value range of , represents the equipment empty load, represents full load, and more than 1 is regarded as overload.

[0027] The weighted comprehensive evaluation method is adopted to calculate the health state score combined with the degree of deviation of temperature and vibration frequency from the threshold value Let temperature threshold be , and vibration frequency threshold be , then , where and are the weights of temperature and vibration frequency, respectively, and . Assign weights according to device types, such as for high-temperature-sensitive devices, and vice versa for mechanically vibration-sensitive devices. Finally ranges from , with higher values indicating better device health status. The device layer operating parameter vector is represented as .

[0028] Step 1010, calculate the maximum loadable task quantity per unit time of the device according to the real-time load rate , where is the rated task processing capacity of the device. For example, if a device is rated to process 100 tasks per hour, and the current load rate is , then tasks can be loaded.

[0029] Step 1011, let the task queue length of the business layer real-time order data be . If is greater than or equal to , then the positive task backlog is ; if is less than , then the negative redundant capacity is .

[0030] Step 1012, calculate the order urgency level parameter according to the difference between the order delivery deadline and the current time , using a piecewise function: When (the emergency threshold, such as 1 hour), ; When (the normal threshold, such as 4 hours), ; When , , .

[0031] Combine the order urgency level and device load rate to calculate the priority weighting coefficient , i.e. , where is the urgency weight, ranging from , to highlight the priority of urgent orders.

[0032] Step 1013: The device load matching index is calculated by the device real-time load rate and task queue length The covariance of Evaluate load balancing. is the number of historical data samples, and For the The load rate and queue length of samples are ,in, is the average real-time load rate of the device, is the average length of the task queue. ( is a preset threshold, such as 0.1), marked as a load imbalance state.

[0033] Step 1014: Backlog the task , priority weighting coefficient and load matching index Encapsulated as a business layer state vector .

[0034] Step 102: Based on the service layer priority weighting coefficient , the dynamic path planning engine at the logistics layer calls the vehicle GPS coordinates and traffic API traffic data (such as congestion coefficient , value range , 1 means smooth traffic, 0.5 means severe congestion), calculate the estimated arrival time deviation of each transportation node ,Right now ,in is the transport distance, is the historical average speed, is the real-time speed. The transport node vector of the logistics layer is represented as .

[0035] Step 103: Generate time using logistics layer vector As a benchmark, the device layer vector Linear interpolation is used to align timestamps. The two most recent sampling times at the device layer are and , the corresponding vector is , then the aligned vector .

[0036] Step 104: Align the 、 and Splice along the time axis to form a ternary state vector matrix The rows correspond to time series, and the columns correspond to three-layer data indicators.

[0037] By fine collection and calculation of device, business, and logistics three-layer data, a comprehensive business status view is constructed to avoid the limitations of single-dimensional data and provide complete data support for resource scheduling. Key indicators such as device health, task priority, and transportation deviation are quantitatively modeled, and through clear calculation rules and parameter value ranges, complex business logic is converted into calculable and comparable standardized data. Based on real-time data and dynamic threshold adjustment, changes in device load, order urgency, and road conditions can be quickly responded to, ensuring that the data is always synchronized with the actual business status and improving the timeliness of scheduling decisions. Through the calculation of device load matching degree and task backlog, the bottleneck link in the business process (such as device overload and task accumulation) can be accurately located.

[0038] In a preferred embodiment of the present application, step 2, based on the three-state vector matrix, extracts the device layer real-time load rate, the business layer task backlog, and the logistics layer vehicle arrival time deviation from the state vector in the current time window, and maps the three indicators to a three-dimensional Euclidean space to form a dynamic state point through a range normalization algorithm; connecting the state points of the three consecutive time windows to construct a dynamic triangle, calculating the coordinate offset and direction angle of the triangle's centroid, determining the dominant factor type of the current resource gap, and generating a real-time processing capacity correction factor according to the proportional relationship between the centroid offset magnitude and the preset threshold, which can include: Step 200, based on the row vector of the three-state vector matrix, extracts the real-time load rate from the device layer running parameter vector, the task backlog from the business layer task queue state vector, and the vehicle arrival time deviation from the logistics layer transportation node vector; Step 201, range normalization is performed on the real-time load rate, task backlog, and arrival time deviation, and the normalized device layer indicators are mapped to the X-axis coordinate of the three-dimensional Euclidean space, the business layer indicators are mapped to the Y-axis coordinate, and the logistics layer indicators are mapped to the Z-axis coordinate, to generate a dynamic state point for the current time window; Step 202, connect the dynamic state points of the three adjacent time windows in time sequence to construct a dynamic triangle topology, calculate the centroid coordinates of the dynamic triangle, and perform vector difference operation on the current centroid coordinates and the previous period centroid coordinates to obtain the centroid offset vector; Step 203, decompose the projection components of the centroid offset vector in the X-axis, Y-axis, and Z-axis directions, and determine the dominant factor type according to the proportional relationship of the absolute values of the projection components, which specifically includes: When the component in the X-axis direction is the maximum projection component, it is determined as the device resource gap dominant type; When the Y-axis direction component is the maximum projection component, it is determined as the task backlog dominant type; When the Z-axis direction component is the maximum projection component, it is determined as the transportation delay dominant type; Step 204, according to the ratio of the module length of the barycentric offset vector to the preset threshold value, a dynamic selection correction factor generation rule is selected, when the ratio is in the interval of 1.2 to 1.5 times the threshold value, a correction factor in linear relationship with the offset module length is generated; when the ratio exceeds 1.5 times the threshold value, an exponential correction factor in direct proportion to the square of the offset amplitude is generated.

[0039] In the embodiment of the application, three types of core indicators are extracted from the ternary state vector matrix by row (time window): Device layer real-time load rate: directly corresponding to the first element of the device layer running parameter vector in the matrix (such as L calculated in step 100), reflecting the current workload of the device (value range , 1 is full load).

[0040] Business layer task backlog: corresponding to the first element of the business layer state vector (such as B calculated in step 1011), if it is positive, it indicates task backlog, and if it is negative, it indicates capacity redundancy.

[0041] Logistics layer vehicle arrival time deviation: corresponding to the estimated arrival time deviation value in the logistics layer vector (such as D calculated in step 102), positive value indicates delay, and negative value indicates early arrival.

[0042] Step 201, calculate the maximum value (Max) and the minimum value (Min) in the historical data of each indicator respectively, and normalize the value by the formula value = (X- Min) / (Max- Min). , scale the indicator to the interval. For example, if the historical maximum value of the task backlog is 200, the minimum value is -50, and the current value is 100, the normalized value is .

[0043] Three-dimensional space mapping: The normalized device load rate is taken as the axis coordinate (range ), the task backlog is taken as the axis coordinate (which may be negative after normalization, and needs to be adjusted to a reasonable interval, such as by translation to make the minimum value 0), and the vehicle arrival time deviation is taken as the axis coordinate (also normalized to ), forming a point in the three-dimensional space, representing the resource state of the current time window.

[0044] Step 202, take the state points of three consecutive time windows in time sequence, denoted as , , , connect three points to form a triangle, reflecting the state of evolution over time. The triangle barycenter coordinates are the average of the three point coordinates, that is, the barycenter . Subtract the vector of the current triangle barycenter (such as moment) from the previous period barycenter (such as moment) to get the offset vector , the vector direction indicates the trend of the offset, and the module length indicates the offset amplitude; wherein, is the normalized coordinate value of the device layer real-time load rate in the current time window (such as the moment) mapped to the axis of the three-dimensional Euclidean space. This value reflects the current load state of the device (the normalized range is , the larger the value, the higher the load); is the normalized coordinate value of the device layer load rate in the previous period (such as the moment), which is used as a historical reference benchmark; is the normalized coordinate value of the business layer task backlog in the current time window mapped to the axis, which can be positive (task backlog) or negative (capacity redundancy), reflecting the pressure state of the business layer task queue; is the normalized coordinate value of the business layer task backlog in the previous period, which is used to compare the change trend of the current task backlog; is the normalized coordinate value of the logistics layer vehicle arrival time deviation in the current time window mapped to the axis, a positive value indicates vehicle arrival delay, and a negative value indicates early arrival, reflecting the timeliness of logistics transportation; is the normalized coordinate value of the logistics layer time deviation in the previous period, which is used to measure the fluctuation of transportation efficiency.

[0045] Step 203, analyze the absolute value of the projection component of the offset vector on the three coordinate axes : If is the largest, it means that the change of device load rate is the main cause of resource gap (such as device overload or idling), and it is determined that the device resource gap is dominant.

[0046] If is the largest, it means that the task backlog fluctuation is the core problem (such as order surge or processing efficiency insufficient), and it is determined that the task backlog is dominant.

[0047] If is the largest, it means that the vehicle arrival time deviation is significant (such as road conditions deterioration or unreasonable path planning), and it is determined that the transportation delay is dominant.

[0048] Step 204, calculate the module length of the barycenter offset vector Ratio to preset threshold .

[0049] Dynamic selection rule: When 1.2≤ , it is considered that the resource gap is in a medium pressure state, and a linear correction is adopted: correction factor (such as , , make increase linearly with ).

[0050] When , it is determined that the resource gap is serious, and an exponential correction is adopted: correction factor (such as ), through the square amplification effect, the processing capacity is forced to adjust, and the pressure is quickly relieved.

[0051] By mapping the three types of indexes of equipment, business and logistics to three-dimensional space, the abstract resource state is converted into a measurable geometric model, and the coordination state and change trend of each link are intuitively displayed, which is convenient for quickly positioning the bottleneck. Based on the axial component analysis of the barycenter offset, the core source (equipment, task or transportation) of the resource gap can be accurately judged, and the problem can be solved directionally. According to the offset amplitude, the linear or exponential correction rule is dynamically switched, so that the system remains stable under light pressure and quickly responds under heavy pressure. By constructing a triangular topology through three consecutive time windows, the state evolution trend is analyzed through the time dimension, rather than isolated single point data, which enhances the ability to capture the dynamic changes of resource state. Normalization processing ensures that indexes of different dimensions can be directly compared, provides uniform scale input parameters for subsequent correction factor generation and scheduling algorithm, and reduces the model calculation complexity.

[0052] In a preferred embodiment of the present application, step 3, based on the real-time processing capacity correction factor, combines the real-time sorting capacity matrix of each warehouse and the transportation road network topology data, and adopts the Hungarian algorithm to calculate the final migration path of the cross-warehouse sorting task, and dynamically generates a vehicle scheduling priority queue with a weighted value according to the spatio-temporal relationship between the current position of the vehicle and the target warehouse, which can include: Step 300, based on the real-time processing capacity correction factor, dynamically adjusting the concurrent task upper limit threshold of each warehouse sorting equipment, obtaining the adjusted concurrent task upper limit threshold; Step 301, combining the adjusted concurrent task upper limit threshold with the historical sorting efficiency data of the warehouse, generating an updated sorting capacity matrix, wherein the historical sorting efficiency data includes the task processing success rate and the average sorting time per unit time; Step 302, according to the updated sorting capacity matrix and the real-time travel time between nodes in the transportation network topology, a bipartite graph model is constructed with the warehouse as the vertex and the cross-warehouse task migration cost as the edge weight, wherein the edge weight value is generated by the sorting capacity difference between the source warehouse and the target warehouse, the time cost weight of the transportation path and the logistics layer task correlation score; Step 303, the Hungarian algorithm is used to solve the final matching path of the bipartite graph model, and the vertex matching combination with the minimum total edge weight is calculated by iteration, and the final path and the migration task quantity of the cross-warehouse task migration are output, and the sorting optimization instruction set containing the target warehouse code, the task allocation list and the time window constraint is generated, which specifically includes: Step 3030, based on the edge weight value of the bipartite graph model, the initial matching relationship between warehouses is established, and each source warehouse vertex is assigned an initial matching weight, wherein the source warehouse vertex is marked as the maximum edge weight value, and the target warehouse vertex is marked as zero; Step 3031, by dynamically adjusting the matching relationship between the source warehouse and the target warehouse, the warehouse pairing combination with the minimum total migration cost is calculated iteratively, and in each iteration, the current edge weight value and the warehouse processing capacity difference are compared, and when the remaining processing capacity of the target warehouse is greater than the migration task quantity, the matching path is updated; Step 3032, when all source warehouse vertices determine the final matching path, the vertex matching combination with the minimum total edge weight is generated according to the adjusted difference value in the marking matrix, and the final path set of the cross-warehouse task migration and the corresponding migration task quantity are output; Step 3033, based on the source warehouse and target warehouse pairing relationship in the final path set, combined with the migration task quantity and the time window constraint, the sorting optimization instruction set containing the target warehouse unique code, the task allocation proportion and the sorting deadline is generated; Step 304, based on the target warehouse code and the migration task quantity in the sorting optimization instruction set, the geographical position and the task associated vehicle information of the corresponding target warehouse are extracted from the logistics layer transportation node vector, and the space-time cost of reaching the target warehouse is calculated combined with the real-time GPS coordinates of the vehicle; Step 305, according to the matching degree of the space-time cost weight and the remaining loading capacity of the vehicle, a scheduling priority index is generated, and the vehicle scheduling instruction queue is generated in descending order of the scheduling priority index.

[0053] In the embodiments of the present application, the concurrent task upper limit of the sorting equipment of each warehouse is dynamically adjusted according to the real-time processing capacity correction factor (such as linear or exponential correction value) generated in step 2. For example, if the original concurrent task upper limit of a warehouse is 100, and the correction factor is 1.2 (indicating that the processing capacity needs to be improved), the adjusted upper limit threshold is 100 x 1.2 = 120 tasks. The correction factor is generated according to the severity of the resource gap. A small linear correction (such as 1.1-1.3 times) is used for a mild gap, and a large exponential correction (such as 1.5 times or more) is used for a severe gap, so as to flexibly adapt to the current business pressure.

[0054] Step 301, combine the adjusted concurrent task upper limit threshold with the historical sorting efficiency data of the warehouse. The historical data includes two core indicators: Task processing success rate per unit time: statistics of the proportion of the number of tasks successfully completed by the warehouse in the past period to the total number of received tasks, for example, a warehouse processed 1000 tasks in the past week, successfully completed 950 tasks, the success rate is 95%.

[0055] Average sorting time: calculate the average time required to complete a single task, for example, a warehouse processed 100 tasks for a total of 200 minutes, the average sorting time is 2 minutes / task.

[0056] Integrate these data with the adjusted concurrent upper limit to form a matrix. For example, assuming there are 3 warehouses, the matrix is: .

[0057] Step 302, divide all warehouses into two disjoint sets: source warehouse set (warehouses whose task quantity exceeds their processing capacity and need to migrate tasks outside, such as warehouses A, B, and C) and target warehouse set (warehouses with idle processing capacity and can receive migrated tasks, such as warehouses D, E, and F). Take the warehouses in these two sets as the vertices of the bipartite graph, for example, the source warehouse A corresponds to a vertex on the left side of the graph, and the target warehouse D corresponds to a vertex on the right side of the graph. Each vertex is labeled with the warehouse number, current processing capacity status, and other key information.

[0058] For each pair of source warehouse and target warehouse (such as source warehouse A and target warehouse D), their processing capacity data is obtained respectively. The processing capacity data contains two aspects: one is the adjusted concurrent task upper limit threshold obtained in step 301, and the other is the historical task processing success rate. Subtract the concurrent task upper limit of the target warehouse D from the concurrent task upper limit of the source warehouse A, and take the absolute value if the result is negative. For example, if the upper limit of A warehouse is 80 tasks and the upper limit of D warehouse is 120 tasks, the difference is 40; if the upper limit of A warehouse is 120 and the upper limit of D warehouse is 80, the difference is 40. Subtract the historical task processing success rate of the target warehouse D from the historical task processing success rate of the source warehouse A, and take the absolute value. Assuming that the success rate of A warehouse is 90% and the success rate of D warehouse is 95%, the difference is 5%. The two differences are combined according to certain weights (such as the concurrent upper limit weight accounting for 70% and the success rate weight accounting for 30%), and a value in the interval of 0-100 is obtained as the sorting capacity difference cost. For example, in the above example, the cost value is 35 after weight calculation (the value is only an example).

[0059] Using transportation network topology data, determine the transportation route between the source warehouse and the target warehouse. For example, there are three optional routes R1, R2, R3 from source warehouse B to target warehouse E.

[0060] Through real-time traffic data (such as calling traffic API to obtain), the expected travel time of each route is calculated respectively. Assuming that R1 is expected to take 30 minutes, R2 is expected to take 40 minutes (due to partial road congestion), and R3 is expected to take 35 minutes, then select the shortest time R1 route, and its travel time 30 minutes as the basic time cost. According to the historical traffic data fluctuation, set a time cost adjustment coefficient (such as early peak coefficient 1.2, flat peak coefficient 1, and late peak coefficient 1.3). If it is currently in the early peak, the final time cost is 30x1.2=36 minutes, and it is converted into a cost value in the interval of 0-100 (such as proportional conversion, assuming that the corresponding cost value is 60).

[0061] Calculate the task association score of the logistics layer: Extract the association information of the task and the target warehouse from the historical data of the logistics layer. For example, count the number of tasks sent from source warehouse C to target warehouse F in the past month, and if there are 100 times of cooperation, a basic association score (such as 60 points) is given. Considering the geographical location adaptability, calculate the straight-line distance (through latitude and longitude calculation) between the source warehouse and the target warehouse, and if the distance is close (such as less than 10 kilometers), add 10 points; if the distance is far (such as more than 50 kilometers), subtract 10 points. Consider other factors such as historical cooperation on-time delivery rate between warehouses, and fine-tune the score. Finally, a task association score in the interval of 0-100 is obtained (such as the final score is 75).

[0062] Generate a comprehensive cost value: The sorting ability difference cost, the transportation path time cost, and the task correlation degree score calculated above are weighted and summed according to preset weights (for example, sorting ability accounts for 40%, time cost accounts for 40%, and correlation degree accounts for 20%). For example, the sorting ability cost is 35, the time cost is 60, and the correlation degree cost is 75, and the calculation can obtain: 35*0.4+60*0.4+75*0.2=14+24+15=53. The 53 is the comprehensive cost value of the edge between the source warehouse and the target warehouse. The above calculation process is repeated for each pair of source warehouse and target warehouse to assign a comprehensive cost value to all edges in the bipartite graph, and the construction of the bipartite graph model is completed.

[0063] In step 3030, the migration costs (calculated in step 302) of the source warehouses (such as A, B, and C) and the target warehouses (such as D, E, and F) are arranged into a matrix. For example: ; For each source warehouse, the maximum value in its corresponding row is taken as the initial weight. For example: The maximum cost of source warehouse A is 72 (corresponding to target warehouse E), so the initial weight of A is set to 72. Similarly, the weight of B is set to 65 (corresponding to D), and the weight of C is set to 63 (corresponding to E). The weights of all target warehouses (D, E, and F) are initialized to 0.

[0064] In step 3031, starting from the source warehouse with the highest weight (such as B with 65), try to assign its tasks to the target warehouse with the maximum cost (D). Check if the remaining processing capacity of target warehouse D is greater than the amount of tasks to be migrated from B. Assuming that the adjusted concurrent upper limit of D is 120, and the allocated task amount is 50, then the remaining capacity is 70. If B needs to migrate 60 tasks, it meets the condition, and records the matching (B→D, task amount 60). For the next source warehouse (such as A with 72), try to match E. If the remaining capacity of E is insufficient, calculate the second largest cost edge (such as A→D with 53), and check the remaining capacity of D (70-60=10). If 10 is less than the task amount of A (such as 30), then it cannot be matched. By adjusting the weight of the source warehouse (such as reducing the weight of A to 53), find other feasible paths (such as A→F with 41). If the remaining capacity of F is sufficient, update the matching (A→F, task amount 30).

[0065] Repeat the above process for each source warehouse. In each iteration, prefer the smallest cost edge under the current weight, and check the remaining capacity of the target warehouse. If the total cost decreases after adjustment (such as from the initial 72+65+63=200 to 53+65+28=146), accept the adjustment; otherwise, back up.

[0066] Step 3032, record the adjustment difference in each iteration (such as source warehouse weight reduction value, target warehouse weight increase value). According to the label matrix, select the vertex matching combination that minimizes the total cost. For example, the final matching is: A→F (task amount 30, cost 41) B→D (task amount 60, cost 65) C→E (task amount 40, cost 63) The total migration cost is 41+65+63=169 (global optimal solution).

[0067] Step 3033, if multiple source warehouses migrate tasks to the same target warehouse (such as D receiving tasks from B and another source warehouse), allocate according to the proportion of each task amount. For example: B migrates 60 tasks to D, and another source warehouse migrates 40 tasks to D, then B's allocation ratio is 60%. According to the order urgency level (such as the priority weighting coefficient in step 1012) and warehouse processing capacity, calculate the deadline. For example: Emergency orders (coefficient=1) need to be completed within 2 hours, and ordinary orders (coefficient=0.5) can be relaxed to 4 hours. Combined with the historical average sorting time of the warehouse (such as 2 minutes / task in step 301), the final deadline = current time + (task amount × average sorting time × emergency coefficient).

[0068] Step 304, get the target warehouse code (such as D) from the sorting optimization instruction set, and its geographic location is latitude and longitude (116.4, 39.9). Extract associated vehicle information from logistics layer data: Vehicle V1: current location (116.3, 39.8), remaining loading capacity 80 pieces; Vehicle V2: current location (116.5, 40.0), remaining capacity 50 pieces.

[0069] Space cost calculation: Calculate the straight-line distance through the map API: V1 to D distance ≈ 11.2 km; V2 to D distance ≈ 14.5 km.

[0070] Time cost calculation: Combined with real-time traffic: current period congestion coefficient of East City 1.3 (relatively congested), West City coefficient 1.0 (smooth). Assuming the average speed of the vehicle is 30 km / h, considering the road conditions: V1 required time = 11.2 ÷ 30 × 1.3 ≈ 0.485 hours ≈ 29 minutes; V2 required time = 14.5 ÷ 30 × 1.0 ≈ 0.48 hours ≈ 29 minutes.

[0071] Comprehensive space-time cost: Set the space cost weight 0.4, the time cost weight 0.6, and after normalization: V1 cost=0.4x +0.6x ≈0.91; V2 cost=0.4x +0.6x =1.00。

[0072] Step 305, the task quantity demand is 60, and the matching degree of the vehicle residual capacity and the task quantity is calculated: V1 matching degree=80÷60≈1.33 (complete matching); V2 matching degree=50÷60≈0.83 (two times of transportation are required).

[0073] Priority index=space-time cost reciprocal x matching degree, for example: V1 index= x1.33≈1.46.

[0074] V2 index= x0.83≈0.83.

[0075] In descending order of priority index: vehicle V1 (index 1.46) → priority scheduling; vehicle V2 (index 0.83) → alternative (if V1 is unavailable).

[0076] The warehouse processing capacity is adjusted in real time through the correction factor, so that the system can quickly respond to resource gaps such as equipment load and task backlog, and avoid scheduling lag caused by fixed threshold. Combined with the Hungarian algorithm to solve the bipartite graph model, the sorting capacity, transportation cost and task correlation are comprehensively considered to realize the global optimal solution of the cross-warehouse task migration path, and reduce the overall operating cost. The sorting instruction set containing task proportion and time constraint is generated to ensure that the task allocation meets the efficiency demand and meets the order delivery time limit. Based on the space-time cost and loading capacity, the priority queue of vehicles is dynamically generated to reduce the empty running rate and waiting time, improve the utilization rate of logistics resources, and reduce the transportation energy consumption. The whole process relies on real-time data (such as equipment state and road condition information) and historical analysis (such as warehouse efficiency), so that the scheduling strategy is more accurate, and the subjectivity and error of manual intervention are reduced.

[0077] As shown in Figure 2 , the embodiment of the application further provides an intelligent business resource scheduling system based on big data, comprising: A data alignment module is configured to capture device layer sensor data, business layer order data and logistics layer vehicle trajectory data in real time through a distributed data collection node, and perform time sequence alignment processing on heterogeneous data streams by using a dynamic time window division algorithm to generate a three-element state vector matrix with unified timestamps, wherein the row dimension of the matrix represents a time sequence of collection, and the column dimension corresponds to device layer operation parameters, business layer task queue status and logistics layer transportation node coordinates, respectively; An index analysis module is configured to extract three indexes, i.e., device layer real-time load rate, business layer task backlog and logistics layer vehicle arrival time deviation, from the state vectors in the current time window based on the three-element state vector matrix, map the three indexes to a three-dimensional Euclidean space by using a range normalization algorithm to form a dynamic state point, connect the state points of three consecutive time windows to construct a dynamic triangle, determine the dominant factor type of the current resource gap by calculating the coordinate offset and offset direction angle of the triangle barycenter, and generate a real-time processing capacity correction factor according to the proportional relationship between the barycenter offset amplitude and a preset threshold value. A task scheduling module is configured to calculate the final migration path of cross-warehouse sorting tasks by using a Hungarian algorithm based on the real-time processing capacity correction factor, the real-time sorting capacity matrix of each warehouse and transportation road network topology data, and dynamically generate a vehicle scheduling priority queue with a weight value according to the spatiotemporal relationship between the current position of the vehicle and the target warehouse.

[0078] It should be noted that the system corresponds to the above method, and all implementation manners in the above method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0079] Embodiments of the application also provide a computing device, comprising a processor and a memory storing a computer program, wherein the computer program is executed by the processor to perform the above method. All implementation manners in the above method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0080] Embodiments of the application also provide a computer readable storage medium storing instructions, wherein the instructions are executed on a computer to make the computer perform the above method. All implementation manners in the above method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0081] The above is the preferred embodiment of the application. It should be noted that, for those skilled in the art, without departing from the principles of the application, several improvements and refinements can be made, which should also be considered as the protection scope of the application.

Claims

1. A method for intelligent business resource scheduling based on big data, characterized in that: The method comprises: Step 1: Use distributed data collection nodes to capture device-layer sensor data, business-layer order data, and logistics-layer vehicle trajectory data in real time. Use a dynamic time window partitioning algorithm to perform time-series alignment on heterogeneous data streams to generate a ternary state vector matrix with a unified timestamp. The row dimension of the matrix represents the acquisition time series, and the column dimensions correspond to the device-layer operating parameters, the business-layer task queue status, and the logistics-layer transportation node coordinates. Step 2: Based on the ternary state vector matrix, three indicators are extracted from the state vector in the current time window: the real-time load rate of the equipment layer, the task backlog of the business layer, and the vehicle arrival time deviation of the logistics layer. These three indicators are mapped to a three-dimensional Euclidean space to form dynamic state points using the range normalization algorithm. The state points of three consecutive time windows are connected to form a dynamic triangle. By calculating the coordinate offset and offset direction angle of the triangle's center of gravity, the dominant factor type of the current resource gap is determined. A real-time processing capacity correction factor is generated based on the proportional relationship between the center of gravity offset amplitude and the preset threshold. In step 3, based on the real-time processing capacity correction factor, combined with the real-time sorting capacity matrix of each warehouse and the transportation network topology data, the Hungarian algorithm is used to calculate the final migration path of the cross-warehouse sorting task. At the same time, a vehicle scheduling priority queue with a weighted value is dynamically generated according to the spatiotemporal relationship between the vehicle's current location and the target warehouse.

2. The intelligent business resource scheduling method based on big data according to claim 1 is characterized in that: Distributed data acquisition nodes are used to capture device-layer sensor data, business-layer order data, and logistics-layer vehicle trajectory data in real time. A dynamic time window partitioning algorithm is used to perform time-series alignment on heterogeneous data streams to generate a ternary state vector matrix with a unified timestamp. The row dimension of the matrix represents the acquisition time series, and the column dimension corresponds to the device-layer operating parameters, the business-layer task queue status, and the logistics-layer transportation node coordinates, including: The edge computing nodes at the device layer collect sensor data in real time, including device current, temperature, and vibration frequency, and aggregate the sensor data into a device-layer operating parameter vector, which includes the device's real-time load rate and health status score; The device layer operating parameter vector is input into the business layer task queue dynamic evaluation module. The business layer task queue state vector is generated by combining the task queue length, order urgency level and real-time load rate of the real-time order data. The state vector includes the task backlog, priority weighting coefficient and device load matching index. Based on the priority weighted coefficient in the task queue state vector at the business layer, the dynamic path planning engine at the logistics layer calls the real-time coordinates and road condition data from the vehicle GPS module and the traffic API interface to calculate the estimated arrival time deviation of each transportation node, and generate a logistics layer transportation node vector containing vehicle coordinates, spatiotemporal weights and task relevance scores; The generation time of the transport node vector at the logistics layer is used as the trigger condition. The latest data of the equipment layer operating parameter vector and the business layer task queue status vector are obtained. In view of the data timing deviation between the equipment layer and the logistics layer, a dynamic interpolation algorithm is used to forcibly align the timestamp of the equipment layer operating parameter vector to the generation time axis of the transport node vector at the logistics layer. The time-aligned device layer operating parameter vector, business layer task queue state vector, and logistics layer transportation node vector are spliced ​​according to a unified time axis to generate a ternary state vector matrix.

3. The intelligent business resource scheduling method based on big data according to claim 2 is characterized in that: The device layer operating parameter vector is input into the business layer task queue dynamic evaluation module. Combined with the task queue length, order urgency level and real-time load rate of the real-time order data, the business layer task queue state vector is generated. The state vector contains the task backlog, priority weighting coefficient and device load matching index, including: Based on the real-time load rate of the device in the device layer operating parameter vector, the maximum amount of tasks that the device can carry per unit time is calculated as the capacity baseline value; The difference between the task queue length of the real-time order data at the business layer and the capacity baseline value is calculated. If the queue length is greater than or equal to the capacity baseline value, a positive task backlog is generated. If the queue length is less than the capacity baseline value, it is marked as negative redundant capacity. Dynamically generate order urgency level parameters based on the difference between the delivery deadline and the current time in the real-time order data at the business layer. Combine the order urgency level parameters with the real-time load rate of the equipment to generate a priority weighting coefficient. The device load matching index is calculated based on the covariance between the device's real-time load rate and the task queue length. If the covariance value is less than the preset threshold, it is marked as a load imbalance state. The task backlog, priority weighting coefficient and load matching index are encapsulated into the business layer task queue state vector according to a unified dimension.

4. The intelligent business resource scheduling method based on big data according to claim 3 is characterized in that: Based on the ternary state vector matrix, three indicators are extracted from the state vector in the current time window: the real-time load rate of the equipment layer, the task backlog of the business layer, and the vehicle arrival time deviation of the logistics layer. These three indicators are mapped to the three-dimensional Euclidean space to form dynamic state points through the range normalization algorithm. Connect the state points of three consecutive time windows to construct a dynamic triangle. By calculating the coordinate offset and offset direction of the triangle's center of gravity, the dominant factor type of the current resource gap is determined. Based on the proportional relationship between the center of gravity offset and the preset threshold, a real-time processing capacity correction factor is generated, including: Based on the row vectors of the ternary state vector matrix, the real-time load rate is extracted from the equipment layer operating parameter vector, the task backlog is extracted from the business layer task queue state vector, and the vehicle arrival time deviation is extracted from the logistics layer transportation node vector. Perform range normalization on the real-time load rate, task backlog, and arrival time deviation. Then, map the normalized device-layer indicators to the X-axis coordinates of a three-dimensional Euclidean space, the business-layer indicators to the Y-axis coordinates, and the logistics-layer indicators to the Z-axis coordinates to generate the dynamic state points of the current time window. Connect the dynamic state points of three adjacent time windows in sequence according to the time series to construct a dynamic triangle topology structure, calculate the center of gravity coordinates of the dynamic triangle, and perform vector difference operation on the current center of gravity coordinates and the center of gravity coordinates of the previous period to obtain the center of gravity offset vector; Decompose the projection components of the center of gravity offset vector in the X-axis, Y-axis, and Z-axis directions, and determine the dominant factor type based on the absolute value ratio of the projection components in each axis. According to the ratio of the modulus of the center of gravity offset vector to the preset threshold, the correction factor generation rule is dynamically selected. When the ratio is in the range of 1.2 to 1.5 times the threshold, a correction factor is generated that is linearly related to the offset modulus; when the ratio exceeds 1.5 times the threshold, an exponential correction factor is generated that is proportional to the square of the offset amplitude.

5. The intelligent business resource scheduling method based on big data according to claim 4 is characterized in that: Decompose the projection components of the center of gravity offset vector in the X-axis, Y-axis, and Z-axis directions, and determine the dominant factor type based on the absolute value ratio of the projection components in each axis, including: When the component in the X-axis direction is the maximum projection component, it is determined to be the equipment resource gap dominant type; When the Y-axis direction component is the maximum projection component, it is determined to be the task backlog dominant type; When the Z-axis direction component is the maximum projection component, it is determined to be a transportation delay dominant type.

6. The intelligent business resource scheduling method based on big data according to claim 5 is characterized in that: Based on the real-time processing capacity correction factor, combined with the real-time sorting capacity matrix of each warehouse and the transportation network topology data, the Hungarian algorithm is used to calculate the final migration path of the cross-warehouse sorting task. At the same time, based on the spatiotemporal relationship between the current location of the vehicle and the target warehouse, a vehicle scheduling priority queue with weighted values ​​is dynamically generated, including: Dynamically adjust the upper limit threshold of concurrent tasks of each warehouse sorting device based on the real-time processing capacity correction factor to obtain the adjusted upper limit threshold of concurrent tasks; Combine the adjusted concurrent task upper limit threshold with the warehouse's historical sorting efficiency data to generate an updated sorting capacity matrix, where the historical sorting efficiency data includes the task processing success rate per unit time and the average sorting time. Based on the updated sorting capacity matrix and the real-time travel time between nodes in the transportation network topology, a bipartite graph model is constructed with warehouses as vertices and cross-warehouse task migration costs as edge weights. The edge weights are generated by the sorting capacity difference between the source and target warehouses, the time cost weight of the transportation path, and the logistics-level task relevance score. The Hungarian algorithm is used to solve the final matching path of the bipartite graph model. By iteratively calculating the vertex matching combination with the minimum sum of edge weights, the final path and migration task volume of cross-warehouse task migration are output, and a sorting optimization instruction set containing the target warehouse code, task assignment list, and time window constraints is generated. Based on the target warehouse code and migration task volume in the sorting optimization instruction set, the geographical location of the corresponding target warehouse and the task-associated vehicle information are extracted from the transportation node vector of the logistics layer, and the time and space cost of reaching the target warehouse is calculated in combination with the real-time GPS coordinates of the vehicle; The dispatch priority index is generated according to the matching degree between the time-space cost weight and the remaining loading capacity of the vehicle, and the vehicles are arranged in descending order according to the dispatch priority index to generate a vehicle dispatch instruction queue.

7. The intelligent business resource scheduling method based on big data according to claim 6 is characterized in that: The Hungarian algorithm is used to solve the final matching path of the bipartite graph model. By iteratively calculating the vertex matching combination with the minimum sum of edge weights, the final path and migration task volume of cross-warehouse task migration are output, and a sorting optimization instruction set containing the target warehouse code, task assignment list, and time window constraints is generated, including: Based on the edge weights of the bipartite graph model, an initial matching relationship between warehouses is established. Each source warehouse vertex is assigned an initial matching weight, where the source warehouse vertex is marked with the maximum edge weight and the target warehouse vertex is marked with zero. By dynamically adjusting the matching relationship between the source warehouse and the target warehouse, the warehouse pairing combination with the minimum total migration cost is iteratively calculated. In each iteration, the current edge weight is compared with the difference between the warehouse processing capacity. When the remaining processing capacity of the target warehouse is greater than the amount of tasks to be migrated, the matching path is updated. When all source warehouse vertices have determined the final matching paths, a vertex matching combination with the minimum edge weight sum is generated based on the adjusted differences in the labeling matrix, and the final path set for cross-warehouse task migration and the corresponding migration task amount are output; Based on the pairing relationship between the source warehouse and the target warehouse in the final path set, combined with the migration task volume and time window constraints, a sorting optimization instruction set containing the target warehouse unique code, task allocation ratio and sorting deadline is generated.

8. An intelligent business resource scheduling system based on big data, the system implementing the method according to any one of claims 1 to 7, characterized in that: include: The data alignment module is used to capture device-layer sensor data, business-layer order data, and logistics-layer vehicle trajectory data in real time through distributed data acquisition nodes. It uses a dynamic time window partitioning algorithm to perform time-series alignment on heterogeneous data streams, generating a ternary state vector matrix with a unified timestamp. The row dimension of the matrix represents the acquisition time series, and the column dimensions correspond to the device-layer operating parameters, the business-layer task queue status, and the logistics-layer transportation node coordinates. The indicator analysis module is used to extract three indicators from the state vector in the current time window based on the ternary state vector matrix: the real-time load rate of the equipment layer, the task backlog of the business layer, and the vehicle arrival time deviation of the logistics layer. The three indicators are mapped to the three-dimensional Euclidean space to form dynamic state points through the range normalization algorithm; Connect the state points of three consecutive time windows to construct a dynamic triangle. By calculating the coordinate offset and offset direction of the triangle's center of gravity, the dominant factor type of the current resource gap is determined. Based on the proportional relationship between the center of gravity offset and the preset threshold, a real-time processing capacity correction factor is generated. The task scheduling module is used to calculate the final migration path of cross-warehouse sorting tasks based on the real-time processing capacity correction factor, combined with the real-time sorting capacity matrix of each warehouse and the transportation network topology data, using the Hungarian algorithm. At the same time, it dynamically generates a vehicle scheduling priority queue with a weighted value based on the spatiotemporal relationship between the vehicle's current location and the target warehouse.

9. A computing device, characterized in that include: one or more processors; A storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a program, which, when executed by a processor, implements the method according to any one of claims 1 to 7.

Images