Main transport capacity intelligent dynamic scheduling method and system

Abstract

The invention discloses a main transport capacity intelligent dynamic scheduling method and system, and particularly relates to the technical field of intelligent scheduling, and the method comprises the following steps: receiving and normalizing a multi-source scheduling request, constructing a scheduling conflict graph, dividing conflict sub-graphs, and for each scheduling request in each conflict sub-graph, carrying out the scheduling of each conflict sub-graph; the method comprises the following steps: respectively calculating a resource competition index and a historical abnormal index according to an isolated scheduling request which does not belong to any conflict sub-graph, inputting the resource competition index and the historical abnormal index into a machine learning model to obtain priorities, preferentially scheduling requests with higher priorities, controlling other requests to enter a scheduling waiting pool according to a delay threshold value, and finally generating a scheduling execution instruction according to a preset strategy. Driving the main transport capacity resource to complete the task; according to the method, resource conflicts are identified by constructing the scheduling conflict graph, intelligent priority classification of the scheduling requests is realized in combination with the resource competition index and the historical anomaly index, resource path allocation and time coordination are completed by adopting the preset strategy, the scheduling execution instruction is generated, and the scheduling decision-making efficiency and execution reliability are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent scheduling, more particularly, to a main transport capacity intelligent dynamic scheduling method and system. BACKGROUND

[0002] With the rapid development of urban transportation, instant delivery, online transport capacity and emergency support, etc., the main transport capacity resource as the core bearing capacity undertakes a large number of key tasks in daily logistics, service scheduling and resource response. The so-called main transport capacity resource is the deployable entity resource used to complete the core task execution, such as transport vehicles, service operation personnel, emergency delivery equipment, etc. How to realize efficient and reliable main transport capacity scheduling under the background of dynamic growth of task demand and limited resource supply has become a core issue in intelligent scheduling system.

[0003] The existing scheduling system mostly uses rules or semi-static way to allocate tasks, which can basically meet the demand in the case of low concurrency or sufficient resources, but in actual operation, multiple scheduling requests are often concentrated in specific high-frequency periods (such as morning and evening peak) or resource-intensive areas (such as warehouse parks and transportation hubs), and the tasks are highly overlapped in time period and space area, which is easy to cause resource scheduling conflict. The current scheduling system is prone to problems such as resource being occupied multiple times, frequent command conflict and scheduling failure rate increasing. Moreover, the existing scheduling scheme mostly takes the current task field as the basis for scheduling decision, lacks comprehensive judgment ability for the historical behavior of scheduling request (such as whether the past response is successful, whether the delay is frequent), and is difficult to evaluate the stability and risk level of the request source, so that the system cannot manage the risk isolation and delayed response of high-risk requests. Therefore, a main transport capacity intelligent dynamic scheduling method and system are proposed to solve the above problems. SUMMARY

[0004] To achieve the above purpose, the present application provides the following technical scheme:

[0005] A main transport capacity intelligent dynamic scheduling method, comprising the following steps:

[0006] Receiving scheduling request information from different sources, the scheduling request information including target time period, target space area, resource category, service level and historical scheduling performance, and forming a current scheduling request set after standardization;

[0007] Building a scheduling conflict graph based on the overlapping relationship of each request in the scheduling request set in time period and space area, each node in the graph representing a scheduling request, and if the time period and space area of two requests have intersection, a connection edge is established between the corresponding nodes; by analyzing the connectivity of the graph, the node set with mutual association relationship is divided into several conflict subgraphs;

[0008] For each scheduling request in each conflict subgraph, and the isolated scheduling request not belonging to any conflict subgraph, the resource competition index and the historical anomaly index are calculated respectively, the resource competition index is calculated according to the current resource density, the number of requests and the service level in the target area, and the historical anomaly index is calculated according to the historical scheduling performance of the request source in the past execution;

[0009] The resource competition index and the historical anomaly index of each scheduling request are input into the pre-trained machine learning model, and the priority level thereof is output, the scheduling request with the first priority level directly enters the current scheduling period, and the scheduling delay threshold of the request with the remaining priority level is calculated according to the priority level, the resource competition index and the historical anomaly index, the request whose delay threshold is not reached enters the scheduling waiting pool, and participates in the construction of the next round of scheduling conflict graph together with the newly received scheduling request;

[0010] The resource path allocation and time coordination are performed on all requests in the current scheduling period according to the preset strategy, the scheduling execution instruction avoiding resource conflict is generated, and the instruction is sent to the scheduling execution interface to drive the main operating resource to complete the corresponding task allocation and dynamic scheduling process.

[0011] In a preferred embodiment, in the process of forming the current scheduling request set by the scheduling request information, the target time period of the scheduling request information is standardized, a uniform time granularity interval is set as a fixed time interval, and the target time period of each scheduling request is divided into a plurality of continuous periods;

[0012] The target space area is uniformly divided into a two-dimensional grid area, each grid unit is set according to the preset standard, and the coordinates of the target space area of all scheduling requests are mapped to the corresponding grid unit;

[0013] The service level is represented by an integer level, which is divided into m levels, the historical scheduling performance records include the number of successful scheduling responses, the number of failed scheduling responses and the average response time of the scheduling request source in the same target time period and target space area in the past preset time window, and after normalization, the current scheduling request set for constructing the scheduling conflict graph is formed.

[0014] In a preferred embodiment, in the process of constructing the scheduling conflict graph, the connection edge between nodes is established only when the intersection of the target time period is not less than one period and the grid unit mapped by the target space area exists complete overlap, the connection edge is undirected edge, and it is indicated that the two scheduling requests have conflict possibility;

[0015] The node set with direct or indirect connection edge relationship is extracted as a conflict subgraph through connectivity analysis of the graph, all scheduling requests in a conflict subgraph are identified as existing resource scheduling competition relationship in the current period, the conflict subgraphs have no intersection and cover all nodes with established connection edges, and the nodes without established connection edges are isolated scheduling requests and do not belong to any conflict subgraph.

[0016] In a preferred embodiment, the calculation process of the resource competition index comprises the following steps:

[0017] The number of scheduling requests in the current scheduling request set that are completely consistent with the target space region of the scheduling request is counted, and the number is taken as a first reference parameter of resource competition;

[0018] The number of main operating force resources available in the target space region in the current scheduling period is counted, and the number is taken as a second reference parameter;

[0019] The service level of the scheduling request is obtained, and the service level is a third reference parameter;

[0020] The calculation method of the resource competition index is as follows: the first reference parameter is multiplied by a preset first coefficient to obtain a weighted demand pressure; then the weighted demand pressure is divided by the sum of the second reference parameter and a preset second coefficient to obtain a unit resource pressure index; then the difference between the service level and the maximum value of the service level is calculated, and multiplied by a preset third coefficient to obtain a priority weight; finally, the unit resource pressure index is taken as a natural logarithm after being added by one, and then multiplied by the value of the priority weight added by one to obtain the final resource competition index.

[0021] In a preferred embodiment, the calculation process of the historical abnormality index comprises the following steps:

[0022] First, search for all historical records of the current scheduling request source in the historical scheduling performance that are completely consistent with the target time period and the target space region, and then count the number of successful scheduling times in the time period and the region as a fourth reference parameter, the number of failed scheduling times as a fifth reference parameter, and record the average response time as a sixth reference parameter;

[0023] The calculation method of the historical abnormality index is as follows:

[0024] First, multiply the fifth reference parameter by a preset fifth coefficient to obtain a failure weight;

[0025] Divide the failure weight by the sum of the fourth reference parameter and a preset fourth coefficient to obtain an abnormality proportion index;

[0026] Divide the sixth reference parameter by a standard response time threshold to obtain a relative response ratio;

[0027] The relative response ratio is substituted into a hyperbolic tangent function as an input to obtain a response time stability weight;

[0028] The abnormality proportion index is squared, and a product of the squared abnormality proportion index and a product of the response time stability weight and a preset sixth coefficient is added to obtain a historical abnormality index.

[0029] In a preferred embodiment, the normalization before forming the current scheduling request set includes format unification and processing by a data deduplication rule, which considers scheduling requests that are completely consistent in time period, have an overlap rate of target space area greater than ninety percent, and have the same resource category as repeated, and only keeps the request with a high service level.

[0030] In a preferred embodiment, the machine learning model is a fuzzy inference engine, the input is a combination of the resource competition index and the historical abnormality index of each scheduling request, and the output is the priority level of each scheduling request.

[0031] In a preferred embodiment, the calculation of the scheduling delay threshold includes the following steps:

[0032] Step one, set the priority level of the current scheduling request as symbol P, the resource competition index as symbol Q, and the historical abnormality index as symbol R;

[0033] Step two, multiply the resource competition index Q by a preset resource weight coefficient A to obtain a first intermediate value, and raise the first intermediate value to the power of N to obtain a resource pressure factor S1, which is calculated as: S1=(A·Q) N ;

[0034] Step three, multiply the historical abnormality index R by a preset abnormality risk coefficient B to obtain a second intermediate value, divide the second intermediate value by a normalization parameter E to obtain a normalized ratio, and substitute the ratio into a hyperbolic tangent function to obtain an abnormality stability factor S2, which is calculated as:

[0035] Step four, multiply the priority level P by a preset priority control coefficient C to obtain a priority level factor S3, which is calculated as: S3=P·C;

[0036] Step five, add the resource pressure factor S1, the abnormality stability factor S2, and the priority level factor S3 to obtain a comprehensive scheduling parameter S4; substitute the parameter into an exponential growth function controlled by a preset scheduling growth control factor γ, and then add a preset basic scheduling delay value T0 to finally obtain the scheduling delay threshold T, which is calculated as: T=T0+exp(γ·S4).

[0037] In a preferred embodiment, a main transport capacity intelligent dynamic scheduling system specifically comprises:

[0038] A request access module receives scheduling request information from different sources, the scheduling request information including target time period, target space area, resource category, service level and historical scheduling performance, and forms a current scheduling request set after standardization;

[0039] A conflict mapping module constructs a scheduling conflict graph based on the overlapping relationship of each request in the scheduling request set in the time period and the space area, each node in the graph representing a scheduling request, and if the time period and the space area of two requests have an intersection, a connection edge is established between the corresponding nodes; by analyzing the connectivity of the graph, the node set having mutual association is divided into several conflict subgraphs;

[0040] A data evaluation module calculates the resource competition index and the historical abnormal index for each scheduling request in each conflict subgraph and the isolated scheduling request not belonging to any conflict subgraph, respectively, the resource competition index being calculated according to the current resource density, the number of requests and the service level in the target area, and the historical abnormal index being calculated according to the historical scheduling performance of the request source in the past execution;

[0041] A priority classification module inputs the resource competition index and the historical abnormal index of each scheduling request into a pre-trained machine learning model, and outputs the priority level thereof, the scheduling request of the first priority level directly entering the current scheduling period; the scheduling requests of the remaining priority levels calculate their scheduling delay threshold values according to their priority levels, resource competition indexes and historical abnormal indexes, and the requests not reaching the delay threshold values enter a scheduling waiting pool and participate in the next round of scheduling conflict graph construction together with the newly received scheduling requests;

[0042] A scheduling execution module performs resource path allocation and time coordination on all requests in the current scheduling period according to a preset strategy, generates scheduling execution instructions avoiding resource conflicts, and sends the instructions to a scheduling execution interface to drive the main transport resource to complete corresponding task allocation and dynamic scheduling process.

[0043] The technical effects and advantages of the present application are as follows:

[0044] The present application explicitly expresses whether there is an overlapping relationship between each scheduling request in the target time period and the target space area by constructing a scheduling conflict graph after standardizing the scheduling request information, and clearly identifies the request pairs having the possibility of scheduling conflict through the mechanism of establishing connection edges in the scheduling conflict graph. Further, by analyzing the connectivity of the graph structure, the requests having association are divided into several conflict subgraphs, effectively realizing the structured modeling and grouping processing of potential resource conflicts, and improving the perception ability of the scheduling system to the resource competition relationship and the graph structure processing efficiency.

[0045] After identifying the conflicting subgraph, the present invention calculates two characteristic parameters for each scheduling request: the resource contention index and the historical anomaly index. These parameters are then fed into a pre-trained machine learning model as combined inputs to output the priority level of the scheduling request. This allows the scheduling system to comprehensively determine the response priority of a scheduling request based on its current environment and historical behavior, without relying on fixed rules. Requests with the first priority level are directly included in the current scheduling cycle, while remaining requests are managed through a delay mechanism, achieving an intelligent closed-loop system for scheduling requests, from state perception to priority decision-making.

[0046] Within the scheduling cycle, this invention uses a preset strategy to allocate resource paths and coordinate time for all requests selected for execution. By adjusting the time and resource matching relationships between tasks, scheduling execution instructions are generated to ensure that scheduling requests do not conflict at the temporal and spatial levels. The generated scheduling execution instructions are sent to the scheduling execution interface, driving the primary transport resources to complete the task allocation and scheduling process according to the instructions. This system embodies a complete scheduling closed loop from conflict identification, priority judgment, delay control, to execution drive, improving the dynamic scheduling efficiency of primary transport resources and the reliability of task execution. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings;

[0048] Figure 1 This is a schematic diagram of a method for intelligent dynamic scheduling of main transport capacity in the present invention.

[0049] Figure 2 This is a schematic diagram of a main transport capacity intelligent dynamic scheduling system in the present invention. DETAILED DESCRIPTION

[0050] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0051] Reference Figure 1 - Figure 2 The following examples were obtained:

[0052] Example 1: A method for intelligent dynamic scheduling of primary transport capacity, comprising the following steps:

[0053] Dispatch request receiving and normalization: By receiving dispatch request information from different sources, the input basis of the dispatch process is formed. The dispatch request information includes five dimensions of target time period, target space area, resource category, service level and historical dispatch performance, ensuring that the collected information has the key elements required for dispatch, such as time, space, resource type, priority order and execution history. In order to ensure the uniformity and calculation efficiency of subsequent processing, all received dispatch request information needs to be standardized through uniform format processing, and finally form the current dispatch request set. The core significance of this process is to establish a clear structure, consistent field and comparable data dispatch information set, so that the subsequent graph structure construction and index calculation are based on clear, orderly and operable data structure, avoiding the inconsistency of source data interfering with the intelligence of dispatch.

[0054] Dispatch conflict graph construction and subgraph division: The normalized dispatch request set is converted into a graph structure for processing, and each dispatch request is mapped to a node in the graph. According to the target time period and target space area of the dispatch request, it is judged whether there is an intersection relationship between any two dispatch requests; if the time period of the two has an intersection, and the space area overlaps, a connection edge is established between the corresponding nodes. Through this process, a complete dispatch conflict graph is constructed. Further, through the connectivity analysis of the graph, the node set with connection edges is divided into conflict subgraphs, and each subgraph represents a group of dispatch requests with mutual scheduling competition relationship. The node that does not produce a connection edge with any other node is processed as an isolated dispatch request. The significance of this step is to make the dispatch conflict explicit and structured through the graph structure, which helps to identify which dispatch requests have direct competition relationship in resources or time. The connected subgraph division of the graph enables the dispatch conflict to be grouped and processed, facilitating the parallel execution of local scheduling strategy, and providing independent judgment units for subsequent priority evaluation and dispatch delay judgment.

[0055] In the present application, the dispatch conflict graph is a kind of undirected graph structure constructed based on the time period overlap and space area overlap relationship between dispatch requests. In the graph: each node represents a normalized dispatch request; if the target time period of two dispatch requests has an intersection and the mapped grid unit of its target space area completely overlaps, a connection edge is established between them. The connection edge reflects the possibility of resource conflict. The finally formed graph structure reflects the potential dispatch conflict relationship network between dispatch requests.

[0056] In graph theory, the "connectivity" of a graph refers to whether there is a path connecting any two nodes in the graph. If there is a direct or indirect connection between any two nodes in a group of nodes, the subgraph is called a "connected subgraph"; if there is no connection between a node and any other node, the node is called an "isolated node". In the scheduling scenario, this means that there is a direct or indirect link between scheduling requests in the conflict subgraph, which needs to be analyzed and processed; and the isolated request can be evaluated independently because it does not conflict with any other request.

[0057] In actual execution, the connectivity analysis of the graph mainly includes the following steps: input the constructed scheduling conflict graph, which contains all the scheduling request nodes and their connection edges. By traversing the node set in the graph, a depth-first search (DFS) or breadth-first search (BFS) algorithm can be used to mark all nodes directly or indirectly connected to a node that has not been visited. Each complete access sequence determines a connected subgraph, and all visited nodes are classified into a conflict subgraph. Continue to search from unvisited nodes until all nodes are processed. In the statistical results, the graph is divided into several disjoint conflict subgraph sets, and the remaining unconnected single nodes are isolated scheduling requests.

[0058] Data calculation and input modeling: For each scheduling request node in the scheduling conflict graph and the isolated scheduling request not included in any subgraph, numerical evaluation is performed, mainly calculating two indicators: resource competition index and historical anomaly index. The resource competition index is calculated based on the current resource density, the number of scheduling requests, and the service level in the target space region, which measures the resource competition strength of the scheduling request under the current resource distribution. The historical anomaly index evaluates the performance of the request source in historical scheduling execution, including success rate, failure rate, response time, etc., which is used to evaluate the execution stability of the scheduling request. Through the above two indexes, the system can quantitatively evaluate the resource demand risk and execution success rate of the scheduling request before scheduling strategy, which provides identifiable and comparable feature input for the subsequent decision model, enhancing the response ability and decision accuracy of the entire system in dynamic complex environment.

[0059] Priority classification and scheduling delay management: The resource competition index and historical anomaly index of each scheduling request in the last step are input as a set of features, and the machine learning model is classified via a pre-trained model. The model outputs the priority level of each scheduling request. The scheduling request classified as the first priority level is considered the most urgent task and is directly executed in the current scheduling period. The remaining requests need to calculate a threshold for controlling the scheduling delay based on their priority level, resource competition index, and historical anomaly index. If the current time has not reached the scheduling delay threshold, the request will temporarily enter the scheduling waiting pool and wait with the newly received scheduling requests to participate in the next round of conflict graph construction. If the current time reaches the scheduling delay threshold, the request enters the scheduling and is considered as one of all requests in the current scheduling period, realizing an intelligent closed loop from scheduling request attribute evaluation to scheduling execution timing control. It avoids congestion or conflict caused by treating all requests equally, while introducing a three-dimensional comprehensive evaluation mechanism of "time + level + risk" to achieve dynamic hierarchical scheduling. This mechanism makes the scheduling execution not only responsive, but also stable and predictive, which is a key component of the intelligent scheduling system.

[0060] Resource path allocation and execution driving: In the current scheduling period, all scheduling requests that enter the execution state will be allocated resource paths and time coordinated according to the preset strategy, generating scheduling execution instructions to guide the specific behavior of resources. The execution instructions will include resource allocation scheme, task execution time window, and target space area, etc. The final instruction is issued through the scheduling execution interface to drive the main operating resources to complete dynamic response according to the task requirements. The conversion from decision model output to entity resource action is realized, ensuring the formation of a closed loop between scheduling logic and real operation. By reasonably allocating resource paths and coordinating task execution time, resource conflicts can be effectively reduced, idle running can be reduced, and response efficiency can be improved, thereby supporting the "intelligent + dynamic" scheduling characteristics of the entire system to truly execute.

[0061] In the process of forming the current scheduling request set, the key elements in the scheduling request information need to be standardized first to ensure the comparability and computational operability of all requests. The standardization of the target time period: the target time period refers to the time range in which each scheduling request expects to complete the scheduling task, which is the time dimension input to represent the scheduling behavior. In order to improve the scheduling accuracy and comparison efficiency of the system, the target time period needs to be standardized to a fixed time granularity interval. The uniform time granularity interval refers to taking the smallest schedulable time unit preset by the system as a period, such as five minutes, ten minutes, etc. Under the fixed time interval, the original target time period of each scheduling request will be divided into several consecutive periods, i.e. the scheduling demand of the scheduling request is refined into a time expression that spans several standard periods. For example, if the target time period of a scheduling request is from 9:00 to 10:00, and the system sets the time granularity to ten minutes per period, then the request will be divided into six consecutive standard periods: 9:00-9:10, 9:10-9:20, …, to 9:50-10:00. This approach facilitates subsequent determination of whether any two requests overlap in the time dimension.

[0062] Uniform grid of target spatial region: The target spatial region refers to the spatial range involved in the scheduling request, which is usually input in the form of geographic coordinates (such as latitude and longitude). In order to achieve spatial conflict analysis and consistent matching of regions, the entire service area needs to be uniformly divided. This division method uses a two-dimensional grid region model, i.e. the target geographic region is divided into several planar grid cells. The size of each grid cell is set by the system according to the preset standard, typically one kilometer by one kilometer. The target spatial region of each scheduling request is mapped to a specific grid cell through coordinates, to establish mapping consistency in the spatial domain. For example, if the target region of a scheduling request is longitude 121.50 and latitude 31.22, the system will map this coordinate to the grid cell number where it is located. This mapping allows spatial conflict determination between different requests to be based on whether the grid numbers are the same, simplifying the geometric calculation complexity.

[0063] Hierarchical setting of service level: Service level is an important attribute for scheduling requests to express their urgency and priority. Service level is represented by an integer value and divided into m levels, where m is a positive integer representing the total number of levels. For example, if m = 5, then the levels increase from level 1 (lowest) to level 5 (highest), where level 5 may represent emergency scheduling or high-priority guarantee services. This integer classification not only facilitates direct reference in subsequent priority judgment models, but also serves as one of the factors for calculating the resource competition index (see claim 4), affecting the priority decision of scheduling execution.

[0064] Structured record of historical dispatch performance: Historical dispatch performance refers to the performance data record formed by the dispatch request source in the past dispatch process, usually including three dimensions of successful dispatch response times, failed dispatch times and average response time. To enhance the timeliness and context matching of data, these data must be limited within a preset time window, and only the historical records of the target time period and the target space area consistent with the current dispatch request are counted.

[0065] For example, if the target time period of the current dispatch request is the morning peak (such as 8:00-9:00 am) and the target area is a specific business district, the system will only retrieve the dispatch data submitted by the source in the same period and area within the past 30 days, and count how many times it has successfully responded (such as successful order acceptance), failed (such as cancellation, no response), and the average response time in seconds. This structured record is not only used for historical anomaly index calculation (see claim 5), but also serves as the basis for evaluating the reliability of the dispatch request.

[0066] After the above target time period standardization, target space area grid mapping, service level unified coding and historical dispatch performance standard statistical processing, all dispatch requests are uniformly organized into a current dispatch request set. This request set is the only input set for building a dispatch conflict graph, and is used to build nodes, judge conflict edges, divide conflict subgraphs and calculate dispatch priority levels in subsequent processing steps. The structured and standardized characteristics ensure that the dispatch process has high consistency, scalability and implementability in terms of logic, data and algorithm.

[0067] In the process of building a dispatch conflict graph, a node in the graph is first generated according to each dispatch request in the current dispatch request set. The current dispatch request set is formed by receiving dispatch request information from different sources and standardizing the target time period, target space area, resource category, service level and historical dispatch performance of the dispatch request.

[0068] Two conditions need to be met to establish a connection edge between nodes in the scheduling conflict graph. First, there must be an intersection between the target time periods of the two scheduling requests, and the length of the intersection must be no less than one period. This period is the minimum time granularity unit set by the scheduling system, for example, ten minutes per period. If the time periods of the two requests intersect only below a period, it is considered not to have a time conflict. Second, the target spatial regions of the two scheduling requests must completely fall into the same grid cell after grid mapping. The grid cell is a standard two-dimensional regional unit that the system divides the service space into, and the size is set by the preset parameters, for example, one kilometer by one kilometer. Only when both scheduling requests meet the conditions of time intersection no less than one period and spatial region completely overlapping, the system establishes a connection edge between the two corresponding nodes. The connection edge is an undirected edge, that is, it does not distinguish between first or direction, but only indicates that there is a possibility of conflict between the two requests. The so-called conflict possibility means that the two scheduling requests have a competition relationship for scheduling resources in the same time and space dimensions, so they must be uniformly evaluated and prioritized.

[0069] After establishing the connection edge, the system performs graph connectivity analysis on the complete graph structure. Graph connectivity analysis is a graph structure processing technique used to identify node sets that have direct or indirect connection paths in connection edge relationships. If two nodes are reachable through one or more connection edges, they belong to the same connected component. In this method, each connected component is identified as a conflict subgraph.

[0070] All scheduling requests in a conflict subgraph are considered by the system to have a resource scheduling competition relationship within the current scheduling period. This competition relationship may be caused by time overlap, spatial overlap, or both, so it needs to be uniformly included in the same priority judgment and delay scheduling policy range for evaluation and processing. Conflict subgraphs do not overlap with each other, that is, there is no situation where the same scheduling request is included in multiple subgraphs at the same time. Each conflict subgraph exists independently in the graph structure, and the node set it contains does not appear in other conflict subgraphs. For those nodes in the current scheduling request set that do not meet the connection edge construction condition with any other request, the system determines them as isolated scheduling requests. These isolated scheduling requests do not belong to any conflict subgraph, but they can still enter the calculation process of the resource competition index and the historical anomaly index, and be determined by the pre-trained machine learning model to belong to the priority level, and participate in the scheduling execution strategy together with other scheduling requests, maintaining the consistency of the scheduling logic.

[0071] For example, if the target time period of the scheduling request A is from 9:00 to 10:00, the target space region corresponds to the grid unit X after mapping, the time period of the scheduling request B is from 9:20 to 9:30, and the space coordinate also corresponds to the grid unit X after mapping, and the period is set to 10 minutes, since the time intersection of A and B is 10 minutes, which meets the time overlap requirement, and the space positions are completely the same, a connection edge will be established between the two, and finally belong to one conflict subgraph. Through this method, the scheduling conflict graph and its conflict subgraph not only effectively express the conflict relationship of the scheduling resources in the time and space dimensions, but also provide a clear and operable structural basis for the subsequent priority evaluation and resource allocation of the scheduling request. Moreover, all scheduling requests in the scheduling conflict graph belong to only one conflict subgraph, or are identified as isolated scheduling requests. The division in the scheduling conflict graph is based on the connectivity analysis of the graph, and there is no case where the same scheduling request belongs to multiple conflict subgraphs.

[0072] The calculation process of the resource competition index includes the following steps:

[0073] The number of scheduling requests that are completely consistent with the target space region in the current scheduling request set is counted, and this number is taken as the first reference parameter Y1 of resource competition;

[0074] The number of main operating force resources available in the target space region within the current scheduling period is counted, and this number is taken as the second reference parameter Y2;

[0075] The service level of the scheduling request is obtained, which is the third reference parameter Y3;

[0076] The calculation method of the resource competition index is:

[0077] The first reference parameter Y1 is multiplied by a preset first coefficient y1 to obtain a weighted demand pressure JP:

[0078] JP=Y1·y1; then the weighted demand pressure JP is divided by the sum of the second reference parameter Y2 and a preset second coefficient y1 to obtain a unit resource pressure index U: Then, the difference between the service level Y3 and the maximum service level Ymax is calculated, and multiplied by a preset third coefficient y3 to obtain a priority weight W: W=y3·(Ymax-Y3);

[0079] Finally, the unit resource pressure index is added by one, and the natural logarithm is taken, and then multiplied by the value of the priority weight added by one to obtain the final resource competition index: Q=ln(U+1)·(W+1). The first coefficient, the second coefficient, the third coefficient, and the maximum service level can be adjusted according to the actual use scene, and the purpose is to adjust the weight sensitivity and avoid nonlinear interference of extreme values on the calculation result.

[0080] As an important indicator for evaluating the tension of the resource where the scheduling request is located in the current space region and time period, the resource competition index first takes the number of other scheduling requests whose target space region is completely consistent with the current scheduling request as the first reference parameter for evaluating the demand intensity. This parameter reflects how many competing scheduling requests exist in the same geographical region under the same time granularity. The more the number is, the more intensive the scheduling demand in the region is in the time period, and the greater the possibility of scheduling conflict is, so the scheduling priority needs to be increased accordingly. Secondly, the number of main operating force resources available in the space region in the current period is counted as the second reference parameter. This parameter reflects the strength of the scheduling supply end in the current period. The less the number of resources is, the greater the pressure is when facing the same number of scheduling requests, so it is introduced as a balancing item for demand pressure in the calculation. By dividing the number of demands by the number of resources, the pressure value that a unit of resource needs to bear can be obtained. The greater the value is, the more nervous the resource is, and the higher the scheduling difficulty is. In addition, the service level of the scheduling request itself also needs to be considered. The service level is set in integer form, and the greater the value is, the higher the priority of the request is. In the calculation, the service level is introduced in order to reflect the role of the service priority strategy in resource scheduling. Specifically, by introducing the difference between the current service level and the maximum level allowed by the system and assigning a weight factor, more restrictions can be given to low-level requests, and higher competition weight can be given to high-level requests, so as to realize the differentiated scheduling strategy of priority.

[0081] In order to enhance the smoothness and discriminability of the index in the growth of the value, one is added to the unit resource pressure index and the natural logarithm is taken, so that the index does not amplify excessively when facing extreme high values, while maintaining the non-linear characteristics of growth. Then, the result is multiplied by the value after adding one to the priority weight, realizing the dynamic integration of supply and demand pressure and service priority. The regional intensity of scheduling requests and resource supply situation are integrated, which can truly restore the resource use competition situation in the scheduling scenario. By introducing the grade difference control mechanism, the response priority of high-level requests in competition is enhanced, which is conducive to guaranteeing the priority execution of key scheduling tasks. The logarithmic function is used to control the growth rate of the index in the high pressure situation, which ensures the stability and comparability of the calculation results of the index. In summary, the calculation method of the resource competition index takes into account the dynamics of the scheduling environment, the distinguishability between requests and the controllability of the model as a whole, which is one of the indispensable evaluation bases in the intelligent dynamic scheduling process of main operating force.

[0082] The calculation process of the historical abnormality index includes the following steps:

[0083] First, search all historical records in the historical scheduling performance that are completely consistent with the target time period and the target space area of the current scheduling request source, then respectively count the number of successful scheduling times in the time period and the area as the fourth reference parameter F1, and the number of failed scheduling times as the fifth reference parameter F2, and record the average response time as the sixth reference parameter F3;

[0084] The calculation method of the historical abnormality index is as follows:

[0085] First, multiply the fifth reference parameter F2 by a preset fifth coefficient y5 to obtain a failure weight SF:

[0086] SF=F2·y5; divide the failure weight SF by the sum of the fourth reference parameter F1 and a preset fourth coefficient y4 to obtain an abnormality proportion indicator YC; Divide the sixth reference parameter F3 by a standard response time threshold Tstd to obtain a relative response ratio XD; Substitute the relative response ratio XD into a hyperbolic tangent function to obtain a response time stability weight WD; WD=tanh(XD); square the abnormality proportion indicator YC, and add the product of the response time stability weight WD and a preset sixth coefficient y6 to obtain the historical abnormality index R: R=Y 2 C+WD·y6; The fourth coefficient, the fifth coefficient, the sixth coefficient, and the standard response time threshold are used to adjust the comprehensive proportion of abnormal frequency and response delay, and can be set according to specific use scenarios.

[0087] The purpose of establishing the historical abnormality index is to comprehensively quantify the stability, success rate, and response efficiency of the source of the scheduling request in the historical execution process, and then to assist in judging the risk level of the current scheduling request. It consists of two parts: the first part: historical frequency evaluation of scheduling failure: the system first counts the number of successful scheduling times and the number of failed scheduling times of the current scheduling request in the target time period and the target space area. By introducing a failure weight coefficient, the failure times are converted into a weighted measure. Divide the failure weight by the sum of the success times and the smoothing factor to obtain an abnormality proportion indicator. This indicator reflects the proportion of failures relative to successes, that is, the frequency of abnormalities. Through square processing, high failure frequency is amplified in the index, and its risk weight is increased.

[0088] The second part: stability measurement of response time, to avoid the influence of delayed response of successful scheduling requests on overall efficiency, the system records the average response time of the source in the area in the time period, and compares it with the standard response time. Through the hyperbolic tangent function, the response ratio is converted into a continuous stability weight value, so that the longer the response time, the closer the output value to the upper limit of the positive value, so as to reflect the risk of "slow response" requests in the final index.

[0089] The final historical abnormality index is formed by adding the results of both, taking into account not only the risk of failure, but also the potential for response delay, and has the following characteristics: identifying the source of abnormal scheduling behavior: identifying unreliable or inefficient executors through abnormality proportion; preventing high-delay tasks from occupying resources: warning about requests that are submitted frequently but respond slowly; enhancing scheduling safety and success rate prediction ability: playing a key input role in priority level classification and scheduling delay judgment.

[0090] Before forming the current scheduling request set, to ensure the structural consistency and semantic accuracy of scheduling request information in the system, the scheduling requests from different sources need to be standardized. The standardization process mainly includes two aspects: format unification and data deduplication processing.

[0091] Format unification refers to the process of converting scheduling request information from different sources and different data formats into a standard structured format before entering the scheduling processing link, to meet the requirements of subsequent calculation and graph construction. Specifically, each scheduling request should include the following five types of information fields:

[0092] Target time period: represents the start and end time range of the request expected to be scheduled for execution, which needs to be expressed in a standard format using the system-defined time granularity; target space area: the service range or resource usage area corresponding to the request, which must be normalized through a unified two-dimensional grid mapping method; resource category: represents the type of main operating force resources required by the scheduling request, such as transportation vehicles, service personnel, or facility types; service level: represents the importance or urgency of the scheduling request in integer form; historical scheduling performance: the success rate, number of failures, and average response time record of the scheduling request source in the same time period and space area in history. Through format unification, all scheduling request data will be converted into standard input items with consistent structure and complete fields, laying a foundation for subsequent scheduling conflict graph construction and index calculation.

[0093] Another key link in the standardization process is data deduplication processing. Due to the existence of scheduling request redundancy, multiple submissions or overlapping sources in the actual scheduling system, if not effectively deduplicated, may lead to scheduling conflict graph inflation, resource misallocation and other consequences. Therefore, the system sets the following deduplication rules: if two or more scheduling requests meet the following three consistency conditions, they are considered duplicate requests, and only the request with the higher service level is retained for subsequent scheduling processing. The deduplication judgment dimensions are as follows: target time period completely consistent: the start and end points of the scheduling time of two scheduling requests are completely the same, and the standard period range after division is equal; target space area overlap rate greater than ninety percent: the space coordinate area after two-dimensional grid mapping overlaps more than ninety percent of the smaller area, indicating that the service area of the two requests is basically the same; resource category is the same: indicates that the types of resources required by two requests are within the completely consistent category definition, and have the same resource calling intention. Among multiple scheduling requests that meet the above three conditions, only the request record with the highest service level is retained, because the high-level request has higher priority, and the deduplication retention principle follows the scheduling priority protection strategy.

[0094] Specific implementation example: taking a certain transportation scheduling system as an example, assume that the following two scheduling requests are received at the same time: Request A: time period is 10:00 to 11:00, space area is area X, resource category is "light truck", service level is 3; Request B: time period is 10:00 to 11:00, space area is area X', which has an overlap rate of 92% with area X, resource category is also "light truck", service level is 4. According to the deduplication rule analysis: the time period is completely consistent; the space area overlap rate is greater than ninety percent; the resource category is completely the same. Therefore, request A and request B constitute a duplicate relationship, only the request B with higher service level is retained as an effective record participating in the construction of the standardized scheduling request set, and request A will be excluded.

[0095] In the present application, in order to reasonably sort and respond to each scheduling request in the current scheduling request set, a scheduling priority level determination method based on a machine learning model is introduced. The machine learning model adopts the structure form of a fuzzy inference engine, and its core function is to receive the input feature combination of the scheduling request and output its corresponding scheduling priority level, thereby providing classification basis for subsequent scheduling decision. In the present application, the fuzzy inference engine realizes the automatic classification of the priority response degree of the scheduling request in the complex scheduling environment by constructing the fuzzy membership function of the input variable and multiple conditional rules. The input of the fuzzy inference engine is the feature combination of the resource competition index and the historical abnormal index calculated after normalization of each scheduling request. The resource competition index is used to represent the scheduling resource competition intensity faced by the request in the current target time period and target space region, and the historical abnormal index is used to reflect the execution stability and risk level of the request source in the historical similar scene. The calculation method of the above two indexes is disclosed in detail in the contents disclosed in the preceding claim 4 and claim 5.

[0096] The fuzzy inference engine is pre-configured with multiple rule sets, for example: "if the resource competition index is high and the historical abnormal index is low, then the priority level is high", and similar reasoning rules. Each rule is composed of the fuzzification expression of the input variable and the classification mapping of the output variable. The input variable is first mapped to the corresponding fuzzy set through the fuzzy membership function, such as "high", "medium", "low", and then the rule matching and weight superposition processing are performed, and finally a clear scheduling priority level is output through the aggregation operation and de-fuzzification operation.

[0097] The scheduling priority level is a key control parameter in the system for determining whether the scheduling request can be immediately responded. The level division can be multi-level, such as the first level is the highest priority, the second level is the second highest priority, and so on, for example, it can be divided into six levels. The system sets the scheduling request with the first priority level to directly enter the current scheduling period for resource allocation and path scheduling; the requests with the remaining levels enter the scheduling waiting pool according to the delay threshold mechanism, waiting for the next scheduling period.

[0098] Example: Taking a scheduling request as an example, the resource competition index is calculated as 8.2, and the historical abnormal index is 1.1. The input combination is sent to the fuzzy inference engine, first the indices are fuzzified according to the pre-defined "medium-high" membership function; then the system finds that the combination matches the rule: "if the resource competition index is high and the historical abnormal index is medium, then the priority level is two", then after weight aggregation calculation and de-fuzzification processing, the final output of the priority level of the request is two. By introducing the fuzzy inference engine, the system can integrate the information of environmental complexity and historical behavior performance in two dimensions when processing multiple source scheduling requests, complete the intelligent evaluation of the priority level of the scheduling request, and take into account flexibility, interpretability and scalability.

[0099] The scheduling delay threshold is an important parameter in the scheduling system for controlling the scheduling request to enter the scheduling period opportunity. The smaller the value, the earlier the scheduling request enters the scheduling process; the larger the value, the longer the waiting time is needed to relieve the scheduling conflict or avoid resource risk. The calculation of the scheduling delay threshold includes the following steps:

[0100] Step one, set the priority level of the current scheduling request as symbol P, the resource competition index as symbol Q, and the historical abnormal index as symbol R;

[0101] Step two, multiply the resource competition index Q by the preset resource weight coefficient A to obtain a first intermediate value, and raise the first intermediate value to the power index N to obtain the resource pressure factor S1, which is calculated as: S1=(A·Q) N ; Multiply the resource competition index Q by the resource weight coefficient A to adjust the influence intensity of the competition degree on the system; then raise the result to the power index N to realize nonlinear weighting and reflect the sharp rise of the scheduling load when the resource is excessively tense. The resource pressure factor S1 can effectively identify the scheduling congestion trend under high-density requests and amplify its contribution to the delay threshold.

[0102] Step three, multiply the historical abnormal index R by the preset abnormal risk coefficient B to obtain a second intermediate value, divide the second intermediate value by the normalization parameter E to obtain a normalized ratio, and then substitute the ratio into the hyperbolic tangent function to obtain the abnormal stability factor S2, which is calculated as: Multiply the historical abnormal index R by the preset abnormal risk coefficient B to adjust the actual interference intensity of the scheduling decision; then divide by the normalization parameter E to ensure comparability between different values; then perform nonlinear mapping on the normalized ratio through the hyperbolic tangent function to finally generate the abnormal stability factor S2. This process can effectively suppress the extreme fluctuations of abnormal data while retaining its risk warning function for scheduling safety.

[0103] Step four, multiply the priority level P by the preset priority control coefficient C to obtain the priority level factor S3, which is calculated as: S3=P·C; multiply the priority level P by the preset priority control coefficient C to form the priority level factor S3, which is used to clearly define the response weight possessed by the scheduling request itself. This step directly reflects the institutional guarantee of the system for service priority level, i.e., high-priority requests should be scheduled earlier, and low-priority requests can be appropriately delayed.

[0104] Step five, add the resource pressure factor S1, the abnormal stability factor S2 and the priority level factor S3 to obtain a comprehensive scheduling parameter S4; input the parameter into an exponential growth function controlled by a preset scheduling growth control factor γ, and then add a preset basic scheduling delay value T0 to finally obtain a scheduling delay threshold T, and the calculation expression is: T = T0 + exp(γ·S4). The comprehensive scheduling parameter S4 is input into the scheduling delay threshold function, representing the comprehensive pressure value required for the current request scheduling to avoid, and then the S4 is amplified in an exponential form using the exponential growth function, so that the higher the comprehensive pressure is, the faster the scheduling time is increased, thereby playing a role in actively inhibiting scheduling conflicts. Finally, the result is superimposed on the system preset basic scheduling delay value T0, and the final scheduling delay threshold T is output, which is used to determine whether the scheduling request can be immediately executed or needs to be waited for a future period.

[0105] In the current scheduling period, the system needs to perform resource path allocation and time coordination operations on all scheduling requests that have been determined to be schedulable, according to the uniformly set scheduling control principle, that is, the preset strategy, to finally form a schedulable execution instruction. The preset strategy refers to a multi-dimensional scheduling processing rule set that is set in advance by the system to ensure the rationality of resource allocation, the smoothness of task connection and the conflict avoidance in the scheduling execution stage. This strategy system mainly includes the following three key components:

[0106] Resource priority matching rule: used for preferentially selecting the most suitable resource in the case of multiple resource options; time coordination sorting rule: used for time sequence optimization when scheduling requests exist in the target time period; conflict avoidance rule: used for logical restriction that the same main operating force resource is not allocated to multiple conflict tasks at the same time. The goal of the combination of the above three rules is to maximize scheduling efficiency, reduce resource conflict rate, and ensure that the task instruction can be directly executed, while meeting the basic requirements of scheduling requests. Genetic algorithm and the like in the prior art can be used to achieve this, and therefore, no more detailed description is given here.

[0107] When performing resource path allocation, the system first classifies the requests according to the resource category field of each scheduling request to ensure that requests of the same type use the same type of main operating force resource. For example, if a scheduling request requires a “medium-sized transport vehicle”, the system only selects the target resource from the available vehicles in this resource category. Then the system preferentially selects the resource that is currently idle in the target space area and is not pre-allocated by other requests in the future period. When there are multiple options for resources, they are sorted according to their historical response efficiency (which can be read from historical scheduling performance) and current state (such as power, location, etc.), and the resource unit with fast response and high accessibility is preferentially selected. This matching behavior is controlled by the resource priority matching rule.

[0108] For all the selected requests in the current scheduling period, the system needs to determine whether there is an intersection between the target time periods of multiple requests, especially when they fall into the same resource jurisdiction area. In this case, the system prioritizes the requests according to the priority level field of the scheduling request and allocates the early executable time period to the higher priority level. The remaining scheduling requests are appropriately delayed in time to avoid overlapping between task time periods. At the same time, if there is a gap between requests, the system can also perform time window compression operation, i.e. make the tasks more closely connected, to improve the efficiency of resource use. This part is realized by the time coordination sorting rule.

[0109] After resource matching and time coordination are completed, the system performs conflict detection: determines whether the same resource is referenced by two or more scheduling requests in the same period. If so, according to the priority level, resource competition index and other parameters of the request, the conflict is eliminated or the low-priority request is returned to the scheduling waiting pool. This mechanism is controlled by the conflict avoidance rule. After the above resource and time arrangement is completed, the system converts the tasks required to be executed by each scheduling request into scheduling execution instructions. The scheduling execution instruction is a structured information set for controlling the action of the main transport resource, usually including: task number; target resource number; execution time period (start and end); execution path or position coordinates; special constraint conditions (such as task interval limit, load upper limit, etc.). The system packages the above scheduling execution instructions according to the protocol supported by the scheduling execution interface, and issues them to the main transport resource control terminal to drive it to automatically enter the execution state, thereby completing the landing of the scheduling instruction.

[0110] For example, the current scheduling period contains three scheduling requests A, B and C, among which requests A and B both require the use of "heavy cargo truck", the target space area is the same industrial area, the target time period is 14:00-15:00, the priority level is five and three respectively, and the historical abnormal index is equivalent. The system will prioritize request A and assign it resource number V1, and arrange it to start execution at 14:00. B is adjusted to execute after 15:00 due to its lower priority level. Request C is independently assigned to resource number V2 due to different resource categories, and no conflict is generated. Finally, three scheduling execution instructions are generated and sent to the corresponding resource terminal for execution.

[0111] Embodiment 2: A main transport intelligent dynamic scheduling system, specifically comprising:

[0112] A request access module receives scheduling request information from different sources, which includes target time period, target space area, resource category, service level and historical scheduling performance, and forms a current scheduling request set after standardization;

[0113] The conflict graphing module constructs a scheduling conflict graph based on the overlapping relationship of each request in the scheduling request set in the time period and the space region. Each node in the graph represents a scheduling request. If the time period and the space region of two requests both have an intersection, a connection edge is established between the corresponding nodes. Through connectivity analysis of the graph, the node set having mutual correlation is divided into several conflict subgraphs.

[0114] The data evaluation module calculates the resource competition index and the historical abnormality index for each scheduling request in each conflict subgraph and the isolated scheduling request not belonging to any conflict subgraph. The resource competition index is calculated according to the current resource density, the number of requests and the service level in the target region. The historical abnormality index is calculated according to the historical scheduling performance of the request source in the past execution.

[0115] The priority classification module inputs the resource competition index and the historical abnormality index of each scheduling request into a pre-trained machine learning model to output the priority level thereof. The scheduling request of the first priority level directly enters the current scheduling period. The scheduling request of the remaining priority levels calculates the scheduling delay threshold thereof according to the priority level, the resource competition index and the historical abnormality index. The request not reaching the delay threshold enters a scheduling waiting pool to participate in the next round of scheduling conflict graph construction together with the newly received scheduling request.

[0116] The scheduling execution module performs resource path allocation and time coordination on all the requests in the current scheduling period according to a preset strategy, generates scheduling execution instructions avoiding resource conflicts, and sends the instructions to a scheduling execution interface to drive the main operating force resources to complete corresponding task allocation and dynamic scheduling process.

[0117] The above formulas are dimensionless values calculated. The formulas are obtained by collecting a large amount of data to simulate a formula of the most recent real situation. The preset parameters in the formulas are set by a person skilled in the art according to the actual situation.

[0118] It should be understood that the sequence of the above processes does not mean the execution order in the various embodiments of the present application. The execution order of the processes should be determined according to their functions and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0119] Those skilled in the art can appreciate that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0120] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, device and unit can refer to the corresponding processes in the foregoing method embodiments, and will not be described here.

[0121] The above merely describes specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for intelligent dynamic scheduling of main transport capacity, characterized in that: The following steps are involved: Receive scheduling request information from different sources, including target time period, target spatial area, resource category, service level and historical scheduling performance, and form the current scheduling request set after normalization; A scheduling conflict graph is constructed based on the overlapping relationships between the time periods and spatial regions of each request in the scheduling request set. Each node in the graph represents a scheduling request. If the time periods and spatial regions of two requests intersect, a connecting edge is established between the corresponding nodes. By performing connectivity analysis on the graph, the set of nodes with interrelated relationships is divided into several conflict subgraphs. For each scheduling request within each conflict subgraph, as well as isolated scheduling requests that do not belong to any conflict subgraph, we calculate the resource contention index and historical anomaly index. The resource contention index is calculated based on the current resource density, number of requests, and service level in the target area. The historical anomaly index is calculated based on the historical scheduling performance of the request source in past executions. The resource contention index and historical anomaly index of each scheduling request are input into a pre-trained machine learning model to output its priority level. Scheduling requests with priority level 1 enter the current scheduling cycle directly. Requests with other priority levels have their scheduling delay thresholds calculated based on their priority level, resource contention index, and historical anomaly index. Requests that do not meet the delay threshold enter the scheduling waiting pool and participate in the next round of scheduling conflict graph construction together with newly received scheduling requests. All requests within the current scheduling cycle are allocated resource paths and time coordinated according to preset strategies, and scheduling execution instructions are generated to avoid resource conflicts. The instructions are sent to the scheduling execution interface to drive the main transport resources to complete the corresponding task allocation and dynamic scheduling process.

2. The method for intelligent dynamic scheduling of main transport capacity according to claim 1, characterized in that: In the process of forming the current scheduling request set from the scheduling request information, the target time period of the scheduling request information is standardized, and a uniform time granularity interval is set as a fixed time interval of one period. The target time period of each scheduling request is divided into several consecutive periods; The target space area is uniformly divided into two-dimensional grid areas. Each grid cell is set according to the preset standard. The target space area coordinates of all scheduling requests are mapped to the corresponding grid cell. The service level is expressed as an integer grade and is divided into m levels. The historical scheduling performance records the number of successful scheduling responses, the number of failed scheduling responses, and the average response time of the scheduling request source in the same target time period and target spatial area within the past preset time window. After normalization, the current scheduling request set is formed to construct the scheduling conflict graph.

3. The method for intelligent dynamic scheduling of main transport capacity according to claim 2, characterized in that: When constructing the scheduling conflict graph, the connection edges between nodes are established only when the intersection of the target time periods is no less than one cycle and the grid cells mapped to the target spatial regions completely overlap. The connection edges are undirected. Through the connectivity analysis of the graph, the set of nodes with direct or indirect connection edges is extracted as a conflict subgraph. All scheduling requests in a conflict subgraph are considered to have resource scheduling competition in the current cycle. The conflict subgraphs have no intersection and cover all nodes with established connection edges. Nodes without established connection edges are isolated scheduling requests and do not belong to any conflict subgraph.

4. The method for intelligent dynamic scheduling of main transport capacity according to claim 3, characterized in that: The calculation process of resource competition index includes the following steps: Count the number of scheduling requests in the current scheduling request set whose target spatial areas are completely consistent with the scheduling request, and use this number as the first reference parameter for resource competition; Count the number of available primary transport resources in the target space within the current scheduling cycle, and use this number as the second reference parameter; Obtaining a service level of the scheduling request, where the service level is a third reference parameter; The calculation method of the resource competition index is as follows: multiply the first reference parameter by the preset first coefficient to obtain the weighted demand pressure; then divide the weighted demand pressure by the sum of the second reference parameter and the preset second coefficient to obtain the unit resource pressure index; then calculate the difference between the service level and the maximum service level, and multiply it by the preset third coefficient to obtain the priority weight; finally, add one to the unit resource pressure index and take the natural logarithm, and then multiply it by the value of the priority weight plus one to obtain the final resource competition index.

5. The method for intelligent dynamic scheduling of main transport capacity according to claim 4, characterized in that: The calculation process of the historical anomaly index includes the following steps: First, retrieve all historical records of the current scheduling request source that are completely consistent with its target time period and target spatial area in the historical scheduling performance. Then, count the number of successful scheduling within the time period and area as the fourth reference parameter, the number of failed scheduling as the fifth reference parameter, and record the average response time as the sixth reference parameter. The calculation method of the historical anomaly index is as follows: First, the fifth reference parameter is multiplied by the preset fifth coefficient as the failure weight; Dividing the failure weight by the sum of the fourth reference parameter and the preset fourth coefficient to obtain an abnormal proportion index; dividing the sixth reference parameter by the standard response time threshold to obtain a relative response ratio; Substituting the relative response ratio as input into the hyperbolic tangent function, we can obtain the response time stability weight; The anomaly ratio index is squared and then added to the product of the response time stability weight multiplied by the preset sixth coefficient to obtain the historical anomaly index.

6. A method for intelligent dynamic dispatching of main transport capacity according to claim 5, characterized in that: Normalization before forming the current scheduling request set includes format unification and processing through data deduplication rules. The deduplication rules are based on the fact that scheduling requests with completely consistent time periods, target spatial area overlap rates greater than 90%, and the same resource categories are considered duplicates, and only the request with the higher service level is retained.

7. The method for intelligent dynamic dispatching of main transport capacity according to claim 6, characterized in that: The machine learning model is a fuzzy reasoner. The input is a combination of the resource competition index and the historical anomaly index of each scheduling request, and the output is the priority level of each scheduling request.

8. The method for intelligent dynamic dispatching of main transport capacity according to claim 7, characterized in that: The calculation of the scheduling delay threshold includes the following steps: Step 1: Set the priority of the current scheduling request to symbol P, the resource contention index to symbol Q, and the historical anomaly index to symbol R; Step 2: Multiply the resource competition index Q by the preset resource weight coefficient a to obtain a first intermediate value, and then raise the first intermediate value to the power index N to obtain the resource pressure factor S1, which is calculated as: S1 = (A·Q) N ; Step 3: Multiply the historical anomaly index R by the preset anomaly risk coefficient B to obtain the second intermediate value, and divide the second intermediate value by the normalization parameter E to obtain the normalized ratio. Substitute the ratio into the hyperbolic tangent function to obtain the anomaly stability factor S2, which is calculated as follows: Step 4: Multiply the priority level P by the preset priority control coefficient C to obtain the priority level factor S3, which is calculated as: S3 = P·C; Step 5: Add the resource pressure factor S1, the abnormal stability factor S2, and the priority factor S3 to obtain the comprehensive scheduling parameter S4; substitute this parameter as input into the exponential growth function controlled by the preset scheduling growth control factor γ, and then add the preset basic scheduling delay value T0 to finally obtain the scheduling delay threshold T, which is calculated as: T = T0 + exp(γ·S4).

9. A main transport capacity intelligent dynamic scheduling system, based on a main transport capacity intelligent dynamic scheduling method according to any one of claims 1 to 8, characterized in that: Specifically include: The request access module receives scheduling request information from different sources. The scheduling request information includes the target time period, target spatial area, resource category, service level and historical scheduling performance, and forms the current scheduling request set after normalization; The conflict graph construction module constructs a scheduling conflict graph based on the overlapping relationships between the time periods and spatial regions of each request in the scheduling request set. Each node in the graph represents a scheduling request. If the time periods and spatial regions of two requests intersect, a connecting edge is established between the corresponding nodes. By performing connectivity analysis on the graph, the node sets with interrelated relationships are divided into several conflict subgraphs. The data evaluation module calculates the resource contention index and historical anomaly index for each scheduling request within each conflict subgraph, as well as for isolated scheduling requests that do not belong to any conflict subgraph. The resource contention index is calculated based on the current resource density, number of requests, and service level in the target area. The historical anomaly index is calculated based on the historical scheduling performance of the request source in past executions. The priority classification module inputs the resource contention index and historical anomaly index of each scheduling request into a pre-trained machine learning model and outputs its priority level. Scheduling requests with priority level 1 enter the current scheduling cycle directly. Requests with other priority levels have their scheduling delay thresholds calculated based on their priority level, resource contention index, and historical anomaly index. Requests that do not meet the delay threshold enter the scheduling waiting pool and participate in the next round of scheduling conflict graph construction together with newly received scheduling requests. The scheduling execution module allocates resource paths and coordinates time for all requests within the current scheduling cycle according to preset strategies, generates scheduling execution instructions to avoid resource conflicts, and sends the instructions to the scheduling execution interface to drive the main transport resources to complete the corresponding task allocation and dynamic scheduling process.

Images