Multi-scene accompanying diagnosis service-oriented group intelligence resource allocation and scheduling method and system

Abstract

The present application relates to the technical field of resource allocation scheduling, in particular to a group intelligence resource allocation scheduling method and system for multi-scene accompanying diagnosis service, which can convert a medical service process into a process node with physical area mapping and configure a benchmark probability quality function; receive a state signal to obtain a time stamp, calculate a deviation value based on actual time consumption and the function, and generate a dynamic impedance factor with time attenuation characteristics by collecting deviations in the same area; use the factor to perform translation transformation and variance inflation on the function of the unexecuted node to generate a modified function; calculate the matching degree weight by discrete convolution of the modified function and an order time window distribution function, solve the optimal matching matrix and convert it into a scheduling instruction for issuance. It can be understood that the present application dynamically quantifies environmental congestion resistance to accurately predict task time consumption and reduce the risk of order misplacement in cross-area flow, thereby realizing global optimization of group intelligence resource allocation scheduling for multi-scene accompanying diagnosis service.

Description

Technical Field

[0001] This invention relates to the field of resource allocation and scheduling technology, specifically to a group-based intelligent resource allocation and scheduling method and system for multi-scenario medical escort services. Background Technology

[0002] Accompanying patients during medical visits is a professional assistance service that provides companionship, assistance with queuing for medication, and assistance with lab tests. Large medical institutions encompass dispersed physical areas for consultations and examinations, creating a complex, multi-scenario medical environment. Given the massive flow of patients, effectively coordinating a large number of accompanying service personnel and achieving efficient resource allocation and scheduling is crucial for improving management efficiency.

[0003] Currently, the industry generally views mobile companions as distributed processing nodes, relying on centralized platforms to conduct business dispatch under a swarm intelligence model. Conventional resource allocation and scheduling usually depend on the current location of personnel and the fixed historical average time of each department to drive the flow. The system tends to directly assign new orders to the service personnel who are closest in space or who can finish their current tasks the fastest according to static estimates.

[0004] However, this scheduling logic fails to fully consider the dynamic changes in the medical setting. The time taken for staff to move between different departments is highly susceptible to disruptions caused by sudden surges in patient traffic. Fixed, static time estimates cannot quantify real-time environmental congestion, leading to significant deviations in the prediction of task completion times. When staff move frequently between different departments, prediction errors in certain stages are carried over and accumulate, ultimately causing a misalignment between subsequent connection times and the patient's expected start time, resulting in dispatch conflicts and service delays. Summary of the Invention

[0005] The purpose of this invention is to provide a swarm intelligence resource allocation and scheduling method and system for multi-scenario escort services, in order to solve the problem that in the complex multi-scenario escort services of hospitals, the failure to effectively quantify real-time environmental congestion fluctuations leads to distorted prediction of task consumption time, which in turn causes order conflict and order continuation misalignment in the resource allocation and scheduling under the underlying swarm intelligence architecture.

[0006] According to a first aspect of the present invention, a swarm intelligence resource allocation and scheduling method for multi-scenario escort services is provided, comprising:

[0007] The patient's medical service process is transformed into process nodes with physical region mapping relationships, and a baseline probability quality function is configured for the process nodes; Receive the status signal of the process node sent by the mobile terminal, extract the actual time consumption of the process node from the status signal, and calculate the status deviation value based on the actual time consumption and the benchmark probability quality function; Based on the location identification tag carried by the state deviation value, the state deviation values ​​of the same physical area are grouped into the same state deviation value set, and the dynamic impedance factor with time decay characteristics is calculated based on the state deviation value set. The dynamic impedance factor is used to perform translation and variance expansion on the baseline probability mass function corresponding to the process node that has not yet been executed, to generate a modified probability mass function. Discrete convolution calculation is performed based on the modified probability quality function and the target start time window distribution function of the order to be assigned to obtain the matching degree weight; The optimal matching matrix is ​​calculated based on the matching degree weight, and the order to be assigned corresponding to the optimal matching matrix is ​​converted into a scheduling instruction and sent to the mobile terminal.

[0008] Preferably, configuring a baseline probability mass function for the process node includes: Extract the environmental feature vector of the process node; Calculate the similarity between the environmental feature vector and each known node in the template library; use the historical time consumption probability distribution data corresponding to the known node with the highest similarity value in the template library as the target template probability mass function; Obtain the actual number of samples collected locally and calculate the confidence weight; sum the target template probability quality function and the actual local time consumption probability distribution according to the confidence weight, and output the baseline probability quality function.

[0009] Preferably, the step of calculating the similarity between the environmental feature vector and each known node in the template library, and using the historical time consumption probability distribution data corresponding to the known node with the highest similarity value in the template library as the target template probability quality function, includes: Calculate the similarity value between the environmental feature vector and the environmental feature vector corresponding to each known node in the template library; If all similarity values ​​are lower than the preset similarity threshold, then the global average probability mass function preset in the template library is called as the target template probability mass function. The global average probability quality function is a discrete probability distribution obtained by performing an indiscriminate arithmetic average of the historical time probability distribution data of all known nodes in the template library.

[0010] Preferably, after receiving the status signal of the process node sent by the mobile terminal, the method further includes: Read the current completion timestamp of the process node from the status signal; Query the historical execution log and extract the historical completion timestamp of the immediately preceding process node; Call the spatial matrix to obtain the minimum walking time between the starting physical region of the previous process node mapping and the ending physical region of the process node mapping; The time difference is calculated by subtracting the historical completion timestamp from the current completion timestamp. If the time difference is less than the minimum walking time, the status signal is determined to be invalid.

[0011] Preferably, based on the location identification tags carried by the state deviation values, state deviation values ​​of the same physical area are grouped into the same state deviation value set, and a dynamic impedance factor with time decay characteristics is calculated based on the state deviation value set, including: Set a time sliding window, traverse all state deviation values ​​received within the current time sliding window, extract the state deviation values ​​mapped to the same physical region, and form a set of state deviation values. The median absolute deviation algorithm is used to identify outlier data points, and these outlier data points are removed from the state deviation value set, leaving the remaining normal data points to form a valid sample subset. The mean of the effective sample subset is calculated, and the mean of the effective sample subset is weighted by a preset state update smoothing coefficient. The impedance factor of the previous cycle, which decays according to the time decay constant, is then added to generate the dynamic impedance factor.

[0012] Preferably, the baseline probability mass function corresponding to the process node that has not yet been executed is transformed by translation and variance inflation using the dynamic impedance factor to generate a modified probability mass function, including: Extract the dynamic impedance factor of the physical region mapped to the process node that has not yet been executed in the current cycle, which is the latest calculated dynamic impedance factor. The mathematical expectation of the baseline probability mass function is added to the dynamic impedance factor to obtain a new mathematical expectation, thereby completing the translation transformation; The original variance of the baseline probability mass function is calculated, and the original variance of the baseline probability mass function is expanded by multiplying the preset uncertainty penalty coefficient with the absolute value of the dynamic impedance factor to obtain a new variance, thereby completing the variance inflation. Discrete distribution data is reconstructed by combining the normal distribution probability density function with the new mathematical expectation and the new variance. The reconstructed discrete distribution data is then normalized to output the corrected probability mass function.

[0013] Preferably, the step of performing discrete convolution calculation based on the modified probability quality function and the target launch time window distribution function of the orders to be assigned to obtain the matching degree weights includes: For each unexecuted process node in the execution state task, the modified probability mass function is subjected to discrete autocorrelation accumulation processing along the execution sequence to generate the final completion time distribution function; Extract the physical region of the current task endpoint and the physical region of the order to be assigned after all tasks have been completed, and call the spatial matrix to obtain the spatial transfer time between the physical region of the current task endpoint and the physical region of the order to be assigned. The spatial transfer time is shifted in the future direction by the final completion time distribution function to generate the predicted arrival time distribution. Based on the predicted arrival time distribution and the target start time window distribution function of the orders to be assigned, the discrete convolution is performed to calculate and output the matching weight.

[0014] Preferably, the step of performing discrete autocorrelation accumulation processing on the modified probability quality function corresponding to all unexecuted process nodes in the execution state task along the execution sequence to generate the final completion time distribution function includes: When it is determined that there is no task in the execution state that is currently being executed, the probability of the final completion time distribution function at the time point representing the current moment is assigned to 1, and the probability at other time points is assigned to 0.

[0015] Preferably, the step of solving the optimal matching matrix based on the matching degree weight, and converting the order to be assigned corresponding to the optimal matching matrix into a scheduling instruction and sending it to the mobile terminal, includes: Construct a weighted bipartite graph with the set of mobile terminals and the set of orders to be assigned as the two vertices; When the matching degree weight is greater than the preset safety margin threshold, a connection edge is established between the corresponding mobile terminal vertex and the vertex to be assigned in the weighted bipartite graph, and the matching degree weight is assigned as the edge weight. The optimal matching matrix is ​​obtained by using the maximum weight matching algorithm to optimize the weighted bipartite graph. Extract the associated mobile terminal identifier and the order identifier to be assigned from the optimal matching matrix, retrieve the business details data, encapsulate it into a standardized scheduling instruction, and send it to the corresponding mobile terminal.

[0016] According to a second aspect of the present invention, a swarm intelligence resource allocation and scheduling system for multi-scenario escort services is provided, for executing the swarm intelligence resource allocation and scheduling method for multi-scenario escort services described in any one of the above claims, comprising: The node configuration module is used to transform the medical service process into process nodes with physical region mapping relationships, and to configure the benchmark probability quality function for the process nodes. The deviation calculation module is used to receive the status signal of the process node sent by the mobile terminal, obtain the timestamp of the completion of the process node and the previous process node, extract the actual time of the process node, and calculate the status deviation value based on the actual time and the benchmark probability mass function. The impedance factor generation module is used to extract the state deviation values ​​mapped to the same physical region, collect the state deviation values ​​of the same physical region to form a state deviation value set, and generate a dynamic impedance factor with time decay characteristics. The probability correction module is used to perform translation and variance expansion on the baseline probability mass function corresponding to the process node that has not yet been executed using the dynamic impedance factor, so as to generate a corrected probability mass function. The matching degree calculation module is used to perform discrete convolution calculation based on the modified probability quality function and the target start time window distribution function of the order to be assigned, and obtain the matching degree weight; The scheduling instruction issuing module is used to solve the optimal matching matrix based on the matching degree weight, and convert the order to be assigned corresponding to the optimal matching matrix into a scheduling instruction to be issued to the mobile terminal.

[0017] The technical solution provided by this invention may include the following beneficial effects: 1. This invention transforms medical processes into process nodes mapped to physical regions, and generates a baseline probability quality function by combining environmental feature templates and local sample size with confidence weighting. By quantifying the inherent time differences of various departments in multi-scenario escort services, it effectively integrates prior experience with local real data, establishing a more accurate expected time benchmark.

[0018] 2. This invention generates a dynamic impedance factor with time decay by calculating the deviation between the actual time consumption and the expected state, and then performs a translation transformation and variance expansion on the probability mass function accordingly. This objectively quantifies the on-site crowd gathering situation, enables real-time capture of environmental fluctuations in multiple scenarios, makes the corrected time consumption distribution more consistent with actual physical conditions, and improves the dynamic adaptability of scheduling.

[0019] 3. This invention obtains matching degree weights by performing discrete autocorrelation accumulation on the node distribution, combining spatial transfer time with order time windows through discrete convolution calculation, and constructing a weighted bipartite graph to solve for the optimal match. This allows for a more accurate measurement of the time fit between personnel accepting orders, reducing the risk of business delays in cross-regional transfers across multiple scenarios, and ultimately achieving global optimization of scheduling. Attached Figure Description

[0020] Figure 1 This is an architecture diagram of the group intelligent resource allocation and scheduling system for multi-scenario escort services of the present invention; Figure 2This is a flowchart of the group intelligent resource allocation and scheduling method for multi-scenario escort medical services according to the present invention; Figure 3 This is a schematic diagram of the logic for generating the baseline probability mass function of the present invention; Figure 4 This is a schematic diagram of the state deviation value verification and calculation process of the present invention; Figure 5 This is a schematic diagram of the dynamic impedance factor evolution data flow of the present invention; Figure 6 This is a schematic diagram of the expected time consumption probability distribution correction operation of the present invention; Figure 7 This is a schematic diagram of the discrete convolution calculation of order matching degree weights in this invention; Figure 8 This is a schematic diagram of the bipartite graph optimization solution scheduling decision topology of the present invention; Figure 9 A comparison curve of the probability mass function correction of the present invention; Figure 10 This is a time-distributed discrete convolution matching curve diagram of the present invention.

[0021] Among them, 101 is the node configuration module; 102 is the deviation calculation module; 103 is the impedance factor generation module; 104 is the probability correction module; 105 is the matching degree calculation module; and 106 is the scheduling instruction issuance module. Detailed Implementation

[0022] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see the appendix Figure 1 This invention provides a swarm intelligence resource allocation and scheduling system for multi-scenario medical companion services. This system is deployed on a server and interacts with multiple mobile terminals via a wireless communication network. The server is equipped with a processing unit and a storage unit. The system includes: a node configuration module 101, a deviation calculation module 102, an impedance factor generation module 103, a probability correction module 104, a matching degree calculation module 105, and a scheduling instruction issuance module 106. Each of these modules is stored as computer program instructions in the storage unit and is read and executed by the processing unit.

[0024] The node configuration module 101 transforms the medical service process into process nodes with physical region mapping relationships, extracts the environmental feature vector of the process node, calculates the similarity between the environmental feature vector and the known nodes in the template library to extract the target template probability quality function, obtains the number of samples actually collected locally to calculate the confidence weight, and outputs the baseline probability quality function by weighting the target template probability quality function and the local actual time consumption probability distribution according to the confidence weight.

[0025] The deviation calculation module 102 obtains the timestamps of the completion of the current process node and the previous process node, calls the spatial matrix to obtain the minimum walking time between the corresponding physical areas, determines that the state signal is valid when the time difference is greater than or equal to the minimum walking time, and extracts the actual time consumption minus the mathematical expectation of the baseline probability mass function to obtain the state deviation value.

[0026] The impedance factor generation module 103 sets a time sliding window, collects the state deviation values ​​of the same physical region to form a state deviation value set, applies the median absolute deviation algorithm to remove outlier data points to obtain a valid sample subset, combines the preset state update smoothing coefficient to weight the mean of the valid sample subset, and superimposes the impedance factor of the previous period decaying according to the time decay constant to generate a dynamic impedance factor.

[0027] The probability correction module 104 performs a translation transformation by adding the mathematical expectation of the baseline probability mass function to the dynamic impedance factor, expands the variance of the baseline probability mass function by multiplying the uncertainty penalty coefficient and the absolute value of the dynamic impedance factor, and reconstructs the discrete distribution data based on the new mathematical expectation and the new variance, and outputs the corrected probability mass function.

[0028] The matching degree calculation module 105 performs discrete autocorrelation accumulation processing on the corrected probability quality function corresponding to the unfinished process nodes in the execution status task to generate the final completion time distribution function, obtains the spatial transfer time from the physical region of the current task endpoint to the physical region of the order to be assigned, performs discrete convolution calculation based on the final completion time distribution function, the target start time window distribution function and the spatial transfer time, and outputs the matching degree weight.

[0029] The scheduling instruction issuing module 106 constructs a weighted bipartite graph with the set of mobile terminals and the set of orders to be assigned as the two vertices. When the matching degree weight is greater than the preset safety margin threshold, a connection edge is established and a matching degree weight is assigned. The maximum weight matching algorithm is called to optimize the weighted bipartite graph to obtain the optimal matching matrix. The orders to be assigned corresponding to the optimal matching matrix are converted into scheduling instructions and issued to the corresponding mobile terminals through the wireless communication network.

[0030] See attached document Figure 2 This invention provides a group intelligent resource allocation and scheduling method for multi-scenario escort services, comprising the following steps: S1 transforms the patient's medical service process into process nodes with physical region mapping relationships and configures a baseline probability quality function for the process nodes. S2, receive the status signal of the process node sent by the mobile terminal, extract the actual time of the process node, and calculate the status deviation value based on the actual time and the benchmark probability quality function; S3, extract the state deviation value mapped to the same physical region, and generate a dynamic impedance factor with time decay characteristics; S4. The baseline probability mass function corresponding to the process node that has not yet been executed is transformed by translation and variance expansion using the dynamic impedance factor to generate the corrected probability mass function. S5, based on the modified probability quality function and the target start time window distribution function of the order to be assigned, perform discrete convolution calculation to obtain the matching degree weight; S6: Solve for the optimal matching matrix based on the matching degree weight, and send scheduling instructions to the mobile terminal.

[0031] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0032] See attached document Figure 1 , Figure 2 and Figure 3 In this embodiment, the node configuration module 101 executes step S1, which transforms the patient's medical service process into process nodes with physical region mapping relationships, and configures a baseline probability quality function for the process nodes.

[0033] In a preferred embodiment, step S1 includes the following sub-steps: S11, the node configuration module 101 transforms the patient's medical service process into process nodes with physical area mapping relationships. The medical service process typically includes sequential steps such as registration, waiting, consultation, payment, and medication pickup. Specifically, the node configuration module 101, through an interface with the hospital information system, reads the medical service catalog document to extract the standard steps for the patient's medical treatment within the hospital. For example, it extracts standard steps such as registration, internal medicine consultation, blood tests, payment, and medication pickup.

[0034] As a preferred approach, the node configuration module 101 constructs a directed acyclic graph based on the sequential constraints set by the medical logic and the time dependencies of business execution. For example, registration is required before an in-person consultation with an internal medicine doctor, and an in-person consultation with an internal medicine doctor is required before blood tests can be performed.

[0035] The node configuration module 101 sets each vertex in the directed acyclic graph as a process node. The node configuration module 101 acquires the coordinate data of the hospital's physical environment and constructs a spatial mapping function. The node configuration module 101 uses the spatial mapping function to bind each process node to an actual physical area within the hospital, completing the association and transformation between logical nodes and physical space. For example, the process node of outpatient blood drawing is mapped to the physical coordinate boundary of Area B on the third floor of the medical technology building.

[0036] S12, after completing the binding of the underlying graph structure and physical space, the node configuration module 101 extracts the environmental feature vector of the process node, calculates the similarity between the environmental feature vector and each known node in the template library to extract the target template probability mass function.

[0037] Specifically, the node configuration module 101 extracts the department attribute code, the number of daily appointment slots, and the floor depth parameter of the process node to be processed, and combines them to form the environmental feature vector of the process node.

[0038] The template library consists of structured data tables pre-installed in local storage. It records historical anonymized business data extracted from other mature, large-scale medical institutions. Internally, it stores environmental feature vectors for multiple known nodes and the historical time consumption probability distribution data corresponding to each known node. For example, the environmental feature vector corresponding to a certain known node might be a numerical vector representing the cardiology department, a daily quota of 600 appointments, and a floor depth of 4 floors. Its corresponding historical time consumption probability distribution data might be a discrete array representing a 20% probability of a 10-minute consumption time, a 50% probability of a 15-minute consumption time, and a 30% probability of a 20-minute consumption time.

[0039] To address the lack of local operational data in the early stages of business operations, the node configuration module 101 calculates the similarity between the extracted environmental feature vector and the corresponding environmental feature vectors of each known node in the template library. For example, if the node to be processed is a newly established respiratory medicine node, and the extracted environmental feature vector represents a daily quota of 500 sources and a floor depth of 3 floors, the node configuration module 101 performs a cosine similarity calculation between this feature vector and the vectors of each known node in the template library. The result is a similarity of 0.85 with the known cardiovascular medicine node and 0.42 with the known dermatology node, thus completing the numerical comparison. Regarding the specific process of calculating vector similarity, those skilled in the art can use the cosine similarity algorithm to calculate the ratio of the vector inner product to the product of the magnitudes. The calculation of the similarity of multidimensional feature vectors is a well-known technique in the field and will not be elaborated upon here.

[0040] The node configuration module 101 compares all similarity values, selects the known node with the highest similarity value, and extracts the historical time consumption probability distribution data corresponding to this known node as the target template probability quality function. The target template probability quality function provides the cold start data foundation in the absence of local data support.

[0041] To avoid computational errors caused by failing to find a known node in the template library that meets the similarity requirements, as a preferred approach, when all similarity values ​​are below a preset similarity threshold (e.g., 0.3), the node configuration module 101 directly calls the global average probability quality function preset in the template library as the target template probability quality function. This global average probability quality function is a discrete probability distribution obtained by performing an indiscriminate arithmetic mean on the historical time probability distribution data of all known nodes in the template library, and is used as the basis data when there are no similar features to match.

[0042] S13, In order to enable the data features of the node to gradually conform to the actual physical environment as local business is carried out, the node configuration module 101 obtains the number of samples actually collected locally, calculates the confidence weight accordingly, and sums the target template probability quality function with the local actual time consumption probability distribution according to the confidence weight, and outputs the benchmark probability quality function.

[0043] The node configuration module 101 queries local data records to count the number of check-in tasks actually occurring within the current physical area for this process node, and uses this number of check-in tasks as the actual sample size collected locally. .

[0044] The node configuration module 101 generates confidence weights using an exponential calculation relationship. The specific calculation formula is as follows: ; In the formula, This is a preset learning rate constant. This learning rate constant is used to control the convergence rate of the local data confidence level as the number of samples increases. Its value range is set between 0.01 and 0.1. Those skilled in the art can adaptively calibrate it according to the actual daily average patient flow of the hospital where the business is located.

[0045] Based on the above derivation, the confidence weight is derived. Then, the node configuration module 101 extracts the local actual time consumption probability distribution. The local actual time consumption probability distribution is a probability quality function generated by discretizing and statistically analyzing historical time consumption records of completing the process node task locally. Specifically, the node configuration module 101 divides the time axis into multiple discrete time intervals of equal length, for example, every 5 minutes. It counts the frequency of historical time consumption records falling within each time interval, and divides the frequency of each interval by the total number of historical records to obtain the probability of occurrence for the corresponding time interval. This constitutes the local actual time consumption probability distribution, whose physical meaning lies in reflecting the time consumption fluctuation pattern in the current real-world environment. It is a time variable.

[0046] Node configuration module 101 retrieves the target template probability mass function Combined with the calculated confidence weights Perform a weighted summation using the following formula: ; Node configuration module 101 outputs the baseline probability mass function Through the above method, the weighted summation mechanism ensures that when the number of local samples is extremely small, the baseline probability quality function tends to call upon the prior experience provided by the template library; as the number of local samples gradually accumulates, the calculation basis will gradually shift to the local real data.

[0047] The baseline probability quality function is the underlying discrete probability data used by a swarm intelligence resource allocation and scheduling system for multi-scenario medical escort services to characterize the expected time consumption of process nodes when no abnormal fluctuations occur in the real-time physical environment.

[0048] See attached document Figure 1 , Figure 2 and Figure 4 In this embodiment, the deviation calculation module 102 performs step S2, receives the status signal of the process node sent by the mobile terminal, extracts the actual time consumed by the process node, and calculates the status deviation value based on the actual time consumed and the benchmark probability quality function.

[0049] In a preferred embodiment, step S2 includes the following sub-steps: S21, in order to establish the spatiotemporal correlation between business actions and the physical environment, the deviation calculation module 102 obtains the timestamps of the completion of the current process node and the previous process node, and calls the spatial matrix to obtain the minimum walking time between the corresponding physical areas.

[0050] In practice, after the accompanying service personnel click to confirm the completion of the business action on a mobile terminal (such as a smartphone, tablet, or customized handheld PDA device), the mobile terminal sends a status signal to the server through the wireless communication network.

[0051] The deviation calculation module 102 receives the status signal, which encapsulates the identification code of the current process node and the current completion timestamp when the business completion action occurred. At the same time, the deviation calculation module 102 queries the historical execution log on the server side and extracts the historical completion timestamp of the next preceding process node in the same service task.

[0052] To establish time constraints in physical space, the deviation calculation module 102 calls a spatial matrix pre-stored in the storage unit. This spatial matrix is ​​a two-dimensional array, whose row and column indices correspond to the physical area numbers within the hospital.

[0053] As a preferred approach, the numerical elements within the spatial matrix are calculated by pre-measuring the actual physical distance traveled between the geometric center points of each physical region and dividing it by a standard adult walking speed (e.g., 1.2 m / s). This calculated lower limit time serves as the minimum walking time between two corresponding physical regions. The deviation calculation module 102 performs an address search within the spatial matrix based on the starting physical region mapped from the previous process node and the ending physical region mapped from the current process node, thereby obtaining the minimum walking time between the corresponding physical regions.

[0054] S22, after extracting the above-mentioned spatiotemporal parameters, the deviation calculation module 102 determines that the status signal is valid when the time difference is greater than or equal to the minimum walking time, and determines that the status signal is invalid when the time difference is less than the minimum walking time.

[0055] In practice, the deviation calculation module 102 subtracts the historical completion timestamp of the previous process node from the current completion timestamp to calculate the time difference between nodes. This time difference represents the absolute physical time that the accompanying person takes to move from the previous business completion point to the current business completion point. The deviation calculation module 102 compares the calculated time difference with the obtained minimum walking time.

[0056] If the time difference is less than the minimum walking time, it indicates that the physical movement time does not match the actual physical displacement time, and is considered abnormal data. To avoid erroneous data contaminating the subsequent calculation environment, the deviation calculation module 102 determines that the status signal is invalid and intercepts and discards it. At the same time, to prevent the algorithm from getting stuck due to data loss, the deviation calculation module 102 sends a re-check-in command to the corresponding mobile terminal, requiring the terminal to resubmit the status signal after obtaining high-precision positioning again.

[0057] Conversely, when the time difference is greater than or equal to the minimum walking time, the deviation calculation module 102 determines that the state signal conforms to physical common sense, marks it as valid data, and allows it to enter the subsequent calculation stage. This verification mechanism helps to reduce the interference of abnormal data on scheduling.

[0058] S23, for the state signal that is determined to be valid, the deviation calculation module 102 extracts the actual time consumption minus the mathematical expectation of the benchmark probability mass function to obtain the state deviation value.

[0059] Specifically, the deviation calculation module 102 analyzes the valid status signals and extracts the actual time consumed by the process node in the current physical region. The actual time taken is usually automatically generated by subtracting the arrival timestamp of the mobile terminal from the current completion timestamp.

[0060] To measure the deviation between the current time consumption and historical experience, the deviation calculation module 102 retrieves the baseline probability mass function configured for the process node in step S1. The deviation calculation module 102 performs discrete expectation operation on the benchmark probability mass function to calculate the mathematical expectation of the benchmark probability mass function. The specific calculation formula is as follows: ; In the formula, This represents the total number of discrete time intervals (for example, dividing 1 hour into 12 intervals of 5 minutes each). The value can be 12); For the first The middle representative value of a time interval (e.g., for a time interval of 0 to 5 minutes, where the time representative value is...) Take 2.5 minutes); This represents the baseline probability of completing the task within this time interval. Mathematical expectation. In a physical sense, it represents the expected average time taken for this process node based on historical data and template settings.

[0061] After calculating the expected value, the deviation calculation module 102 performs the difference calculation, and the specific calculation formula is as follows: ; Deviation calculation module 102 extracts actual time consumption Subtract the expected value of the baseline probability mass function The state deviation value is obtained. Therefore, the above operation, by subtracting the inherent time expectations of different departments, extracts the time deviation reflecting the degree of on-site congestion. This operation reduces the interference caused by the inherent time differences between different departments and different services, and isolates the abnormal delay time caused by on-site queuing, congestion and other immediate environmental factors.

[0062] State deviation value It is a continuous-time variable with a positive and negative sign. When the state deviation value is positive, it indicates that the actual time taken exceeds the expectation, reflecting possible congestion or business blockage in the current physical area. When the state deviation value is negative, it indicates that the business flow is smooth and the actual efficiency is higher than the historical baseline expectation. Obtaining the state deviation value completes the preliminary quantification of real-time micro-environment fluctuations.

[0063] See attached document Figure 1 , Figure 2 and Figure 5 In this embodiment, the impedance factor generation module 103 executes step S3, which, based on the location identifier tag carried by the state deviation value, gathers the state deviation values ​​of the same physical area into the same state deviation value set, and calculates the dynamic impedance factor with time decay characteristics based on the state deviation value set.

[0064] In a preferred embodiment, step S3 includes the following sub-steps: S31, the impedance factor generation module 103 sets a time sliding window and collects the state deviation values ​​of the same physical region to form a state deviation value set.

[0065] Specifically, the impedance factor generation module 103 uses the current calculation trigger time as a reference and extends it in the historical direction by a preset time length (for example, setting the time length to 15 minutes) to define a closed time sliding window. As the time sequence progresses, the time sliding window slides forward continuously.

[0066] The impedance factor generation module 103 iterates through all received state deviation values ​​within the current time sliding window. By extracting the location identifier labels carried by the state deviation values, it filters and aggregates those state deviation values ​​that map to the same hospital physical area. Through this spatial aggregation operation, the impedance factor generation module 103 summarizes multiple state deviation values ​​mapped to the same physical area into a discrete one-dimensional array, thereby obtaining a set of state deviation values.

[0067] The state deviation value set is a local data group reflecting the degree to which the actual business time consumption of a specific physical area deviates from the expected value in a recent period. To avoid calculation interruption due to the absence of any state deviation values ​​within the time sliding window, as a preferred method, when the number of state deviation values ​​within the time sliding window is zero, the impedance factor generation module 103 will determine whether there are historical impedance records. If the system is in the cold start phase of its first operation, the impedance factor generation module 103 directly assigns the dynamic impedance factor of the current period to the initial value of zero; if historical impedance records already exist, the impedance factor generation module 103 only calculates the natural dissipation value of the impedance factor of the previous period based on the time transition as the dynamic impedance factor of the current period and outputs it directly, thereby skipping subsequent internal calculation steps.

[0068] S32, after successfully obtaining a non-empty set of state deviation values, the impedance factor generation module 103 uses the median absolute deviation algorithm to remove outlier data points and obtain a valid sample subset. During the operation of the escort service, some abnormally large state deviation values ​​may be caused by single accidental factors such as sudden physical discomfort of the escorted person. Such data cannot truly reflect the overall business congestion situation of the corresponding physical area.

[0069] To eliminate the interference of such random outliers, the impedance factor generation module 103 extracts all values ​​from the state deviation value set and calculates the median of the state deviation value set. The impedance factor generation module 103 calculates the absolute difference between each state deviation value in the state deviation value set and the median of the state deviation value set, and then calculates the median of all absolute differences to obtain the median absolute deviation parameter. For the specific calculation method of the median absolute deviation parameter, those skilled in the art can use conventional robust statistical processing methods; the basic numerical calculation logic is well-known in the field and will not be elaborated here.

[0070] After calculating the median absolute deviation parameter, the impedance factor generation module 103 sets a deviation judgment threshold coefficient (for example, setting the deviation judgment threshold coefficient to 3). The impedance factor generation module 103 determines data points whose median value of the set of state deviation values ​​exceeds the product of the deviation judgment threshold coefficient and the median absolute deviation parameter as outlier data points.

[0071] The impedance factor generation module 103 removes these outlier data points from the state deviation value set, retaining the remaining normal data points to form a valid sample subset. This valid sample subset represents a high-quality numerical sequence that objectively reflects the general environmental state of the physical area after filtering out individual extreme interferences. If removing outlier data points results in an empty valid sample subset, it indicates that the overall data fluctuation within the current window is abnormal. In this case, the impedance factor generation module 103 directly extracts the median of the state deviation value set before the removal operation and assigns this median directly to the mean of the valid sample subset in subsequent formulas. This ensures that the parameters for subsequent calculation formulas are complete.

[0072] S33, the impedance factor generation module 103 combines the preset state update smoothing coefficient to weight the mean of the effective sample subset, and superimposes the impedance factor of the previous cycle that decays according to the time decay constant to generate a dynamic impedance factor.

[0073] The impedance factor generation module 103 performs an arithmetic mean operation on all values ​​within the effective sample subset to obtain the mean of the effective sample subset. The mean of the effective sample subset reflects the average delay status of a specific physical region within the current time sliding window.

[0074] To maintain a smooth transition of state parameters over time and avoid drastic jumps in data indicators caused by instantaneous fluctuations in pedestrian flow, the impedance factor generation module 103 retrieves the impedance factor generated for a specific physical region in the previous calculation cycle and records it as the impedance factor of the previous cycle. If the current calculation cycle is the first cycle of a system cold start, and there is no historical impedance data yet, the impedance factor generation module 103 will use the impedance factor from the previous cycle. Initialize to 0. After all calculation parameters are available, the impedance factor generation module 103 performs the calculation using the following formula: ; In the formula, The dynamic impedance factor generated for the current period; The state update smoothing coefficient is a preset value used to adjust the weight of the mean of the current effective sample subset in the overall calculation result. Its value ranges from 0 to 1, and can be 0.4 to balance the sensitivity of response to new states and the smooth inheritance of historical states. The time decay constant is used to control the rate at which the influence of historical data decays over time; for example, it can be set to 0.05. This represents the actual time interval between the current cycle and the previous calculation cycle.

[0075] The continuous numerical variable derived from the above formula is the dynamic impedance factor. Based on the aforementioned rules, a natural base is introduced. The time decay term is due to the objective law that congestion in the physical environment spontaneously dissipates over time. As time progresses, without new abnormal deviations, the historically accumulated impedance effect should decay exponentially until it reaches zero. The dynamic impedance factor is a quantitative parameter used to characterize the service flow resistance caused by on-site environmental factors such as crowding and business backlog in a specific physical area. The dynamic impedance factor takes into account both the current real-time fluctuations and historical objective trends, achieving dynamic quantification of the actual congestion situation in the physical space.

[0076] See attached document Figure 1 , Figure 2 and Figure 6 In this embodiment, the probability correction module 104 executes step S4, which uses the dynamic impedance factor to perform translation transformation and variance expansion on the baseline probability mass function corresponding to the process node that has not yet been executed, and generates the corrected probability mass function.

[0077] In a preferred embodiment, step S4 includes the following sub-steps: S41, the probability correction module 104 adds the mathematical expectation of the baseline probability mass function to the dynamic impedance factor to complete the translation transformation.

[0078] When a service task is in progress, the allocation and scheduling need to predict the expected time consumption of the process nodes that have not yet been executed. The probability correction module 104 extracts the latest calculated dynamic impedance factor of the physical region mapped to the process nodes that have not yet been executed in the current cycle. Simultaneously, the probability correction module 104 retrieves the baseline probability mass function initially configured for the process nodes that have not yet been executed in step S1. The expected value of the baseline probability mass function is calculated according to the aforementioned expected value formula. .

[0079] Since the expected value of the baseline probability mass function only represents the static average time based on historical data and fails to include real-time conditions such as sudden congestion or abnormal smooth flow in a specific physical area, the probability correction module 104 performs an addition operation to ensure that the time prediction for subsequent process nodes that have not yet been executed can truly reflect the real-time changes in the current physical environment and achieve dynamic adaptive adjustment of the expected time. This adds the dynamic impedance factor, which characterizes the real-time environmental deviation, as a time compensation amount to the expected value of the baseline probability mass function. The specific calculation formula is as follows: ; In the formula, This represents the new mathematical expectation generated after the translation transformation. Based on the positive and negative properties of the dynamic impedance factor, when the physical area is congested, the dynamic impedance factor is positive, and its addition results in a larger new mathematical expectation, indicating an increase in expected time consumption. Conversely, when the physical area is unobstructed, the dynamic impedance factor is negative, and its addition results in a smaller new mathematical expectation, indicating a decrease in expected time consumption. This numerical addition operation is represented on the probability distribution chart as shifting the original probability distribution curve to the right or left along the time axis, thus completing the translation transformation.

[0080] Under abnormally smooth conditions, if the dynamic impedance factor is a negative number with a large absolute value, it may cause the calculated new mathematical expectation to be less than zero, which violates the basic logic of non-negative physical time consumption. To avoid such algorithmic calculation errors, as a preferred approach, the probability correction module 104 sets a minimum time threshold (e.g., 0.5 minutes). If the new mathematical expectation obtained from the addition operation is less than the minimum time threshold, the probability correction module 104 forcibly assigns the new mathematical expectation to the minimum time threshold.

[0081] S42, the probability correction module 104 expands the variance of the baseline probability mass function by multiplying the uncertainty penalty coefficient by the absolute value of the dynamic impedance factor. Abnormal fluctuations in the physical environment not only change the average processing time of services but also exacerbate the uncontrollability of service completion times. Both severe congestion and abnormally smooth traffic will interfere with the original stability of processing time. To quantify this volatility, the probability correction module 104 calculates the original variance of the baseline probability mass function. The specific calculation formula is as follows: ; In the formula, This represents the total number of discrete time intervals. For the first The median representative value of each time interval.

[0082] The probability correction module 104 retrieves the pre-set uncertainty penalty coefficient. The uncertainty penalty coefficient measures the amplification effect of physical environment impedance on the time dispersion, and its value is set between 0.1 and 0.5 based on the sensitivity of the business scenario to uncertainty. The probability correction module 104 extracts the absolute value of the dynamic impedance factor. Then, perform the variance inflation calculation using the following formula: ; In the formula, This represents the new variance after inflation. By introducing the absolute value of the dynamic impedance factor, regardless of whether the environment is positively or negatively impeded, the new variance will be greater than or equal to the original variance of the baseline probability mass function. This operation manifests in the probability distribution as making the originally concentrated distribution curve more dispersed, thus reflecting the time uncertainty risk brought about by fluctuations in the real-world environment.

[0083] S43, the probability correction module 104 reconstructs the discrete distribution data based on the new mathematical expectation and variance, and outputs the corrected probability mass function. The underlying scheduling logic relies on discretized time-slice data; the probability correction module 104 needs to remap the shifted and dilated continuous statistical parameters back to the discrete data structure. The probability correction module 104 uses the normal distribution probability density function to calculate the unnormalized probability value for each discrete time interval. The calculation formula is as follows: ; Considering that if the new variance is close to zero, the denominator of the above formula will be zero, causing an operational error, as a compensation mechanism, when the new variance is lower than the set lower limit minimum value (e.g., 0.0001), the probability correction module 104 directly assigns the probability of the corresponding discrete time interval containing the new mathematical expectation to 1, and assigns the probability of the other discrete time intervals to 0, so as to avoid division by zero error.

[0084] After calculating the unnormalized probability values ​​for all discrete time intervals, in order to meet the mathematical requirement that the sum of the probability distributions must be constant at 1, the probability correction module 104 normalizes the unnormalized probability values ​​for all discrete time intervals. The calculation formula is as follows: ; Based on the normalization formula described above, the probability correction module 104 generates a new set of discrete probability arrays and outputs this set of discrete probability arrays as the corrected probability mass function. The corrected probability mass function is a multi-dimensional set of discrete values. It replaces the initial baseline probability mass function and is used in subsequent calculations as the underlying data support for evaluating the actual expected time consumption of each unexecuted process node.

[0085] See attached document Figure 1 , Figure 2 and Figure 7 In this embodiment, the matching degree calculation module 105 performs step S5, which performs discrete convolution calculation based on the corrected probability quality function and the target start time window distribution function of the order to be assigned, and obtains the matching degree weight.

[0086] In a preferred embodiment, step S5 includes the following sub-steps: S51, the matching degree calculation module 105 performs discrete autocorrelation accumulation processing on the corrected probability mass function corresponding to the unfinished process nodes in the execution status task to generate the final completion time distribution function.

[0087] In real-world business scenarios, escort service personnel are often in the process of performing a task, i.e., a task in execution state. An execution state task typically contains multiple sequentially flowing process nodes that have not yet been executed. The matching degree calculation module 105 retrieves the modified probability quality functions corresponding to each of the unexecuted process nodes. To obtain the global probability distribution of the remaining total time for the escort service personnel to complete the execution state task, the matching degree calculation module 105 performs discrete autocorrelation accumulation processing on all the modified probability quality functions mentioned above.

[0088] Specifically, since the total time consumed by multiple serial process nodes is, in terms of probabilistic statistics, equivalent to the sum of the random variables of the time consumed by each independent step, the matching degree calculation module 105 iteratively calculates the process nodes that have not yet been executed, following the set execution sequence of the task execution state, using the discrete convolution formula. The discrete convolution merging formula for any two adjacent probability distributions is as follows: ; In the formula, The total time spent after merging is the variable. For discrete-time variables involved in the summation operation; The previously accumulated probability distribution at time points The probability value; The corrected probability mass function for the process nodes currently participating in the merger at time point The probability value; This is the temporary cumulative probability distribution generated after a single convolutional merge.

[0089] The matching degree calculation module 105 substitutes the corrected probability mass functions of the first and second unexecuted nodes in the execution sequence into the above formula to obtain an initial temporary cumulative probability distribution. Subsequently, the matching degree calculation module 105 uses this temporary cumulative probability distribution as the new... The same convolution calculation is then performed with the corrected probability quality function of the third node in the sequence. Through this successive traversal and multi-round merging operation on all incomplete nodes, the matching degree calculation module 105 converges the discrete local node time consumption distribution into a unified probability array, that is, generates the final completion time distribution function.

[0090] The final completion time distribution function characterizes the probability distribution of the absolute time points at which the escort service personnel finally complete their tasks after experiencing various real-time impedance interferences. To avoid calculation errors caused by the escort service personnel being idle and having no tasks in progress, as a preferred approach, when it is determined that the escort service personnel have no tasks in progress, the matching degree calculation module 105 forcibly assigns a probability of 1 to the final completion time distribution function at the time point representing the current moment, and assigns a probability of 0 to the probabilities at other time points, thereby ensuring the completeness of subsequent convolution operation parameters.

[0091] S52, the matching degree calculation module 105 obtains the spatial transfer time from the physical region of the current task endpoint to the physical region of the starting point of the order to be assigned. The matching degree calculation module 105 parses the configuration parameters of the task in execution status and determines the physical region of the current task endpoint where the accompanying service personnel are located after all tasks in execution status have been completed.

[0092] Simultaneously, the matching degree calculation module 105 parses and allocates the real-time incoming orders to be assigned, extracting the physical region of the starting point of the order to be assigned, which is required for the business triggering of the order. The matching degree calculation module 105 calls the spatial matrix constructed in the previous steps, takes the physical region of the current task endpoint as the retrieval starting point, and the physical region of the starting point of the order to be assigned as the retrieval endpoint, and extracts the corresponding minimum walking time from the spatial matrix.

[0093] If the spatial matrix lacks distance configuration data between corresponding physical regions, making it impossible to retrieve the minimum walking time, the matching degree calculation module 105 assigns a preset global maximum safe transfer time (e.g., set to 15 minutes) as a substitute value to prevent scheduling interruption. Under normal circumstances, the matching degree calculation module 105 directly uses the extracted minimum walking time as the spatial transfer time from the current task's endpoint physical region to the starting point physical region of the pending order. The spatial transfer time reflects the absolute physical time cost necessary for the accompanying service personnel to cross physical locations between two tasks.

[0094] S53, the matching degree calculation module 105 performs discrete convolution calculation based on the final completion time distribution function, the target start time window distribution function, and the spatial transfer time, and outputs the matching degree weights. The matching degree calculation module 105 obtains the target start time window distribution function pre-constructed for the orders to be assigned.

[0095] Specifically, when an order to be assigned is generated by the system, a preset start time is attached. This preset start time refers to the explicitly specified start time of the service agreement when the order is dispatched (e.g., 9:00 AM on the patient's appointment). Based on this preset start time, the matching degree calculation module 105 constructs a target start time window distribution function. For the specific construction method of the target start time window distribution function, those skilled in the art can use conventional probability density distribution generation methods, which are well-known techniques in the field.

[0096] Here's a brief explanation: The matching degree calculation module 105 uses a preset start-up baseline time as the mathematical expectation, and combines it with the time tolerance threshold configured in the business configuration (e.g., a set 15-minute grace period before and after) as the standard deviation. Based on the Gaussian distribution probability formula, it generates a set of discrete probability density sequences, which is the target start-up time window distribution function. This target start-up time window distribution function has the highest probability value corresponding to the preset start-up baseline time, and gradually decreases in a Gaussian distribution pattern as time deviates from this baseline point. The target start-up time window distribution function reflects the allowable range of time deviation between the business order and the actual service start time.

[0097] To assess the timeliness of personnel accepting orders, the matching degree calculation module 105 shifts the final completion time distribution function along the time axis towards the future by the time length corresponding to the spatial transfer time, generating a predicted arrival time distribution. The predicted arrival time distribution represents the expected arrival probability of the accompanying service personnel reaching the location of the new order. Subsequently, the matching degree calculation module 105 performs discrete convolution calculations based on the predicted arrival time distribution and the target start time window distribution function. The specific calculation formula is as follows: ; In the formula, To calculate the generated matching weights; This represents the total number of discrete time points divided on the time axis. The index of the specific time point for the traversal calculation; The predicted arrival time distribution after spatial translation operation is located at time points. The probability value; The target time window distribution function for orders to be assigned is initiated at time point. The probability values ​​are determined by the following calculation process. Mathematically, this process is equivalent to performing discrete convolution and cross-correlation operations on two discrete distribution sequences at a zero time offset. By calculating the product of their probabilities at the same time point and performing global integration and accumulation, the overlapping area of ​​the two probability distribution curves is obtained.

[0098] The value derived from the discrete convolution formula described above is the output matching weight. The matching weight is a quantitative evaluation parameter; its physical meaning lies in measuring the probability that a personal accompaniment service provider, after completing their current task and moving to the new order location, can respond to the new patient's service request on time. A higher matching weight indicates a closer alignment between the personal accompaniment service provider's time trajectory and the preset start time of the new order. This reduces the risk of wasted time and order delays during task continuation, thus providing a data foundation for optimal order matching in allocation and scheduling.

[0099] See attached document Figure 1 , Figure 2 and Figure 8 In this embodiment, the scheduling instruction issuing module 106 executes step S6, which calculates the optimal matching matrix based on the matching degree weight and issues a scheduling instruction to the mobile terminal.

[0100] In a preferred embodiment, step S6 includes the following sub-steps: S61, the scheduling instruction issuance module 106 constructs a weighted bipartite graph with the mobile terminal set and the unassigned order set as its two vertices. In a concurrent scenario with multiple service personnel and multiple orders, the scheduling instruction issuance module 106 gathers the personnel identifiers corresponding to the currently online mobile terminals that are qualified to accept orders, forming the mobile terminal set. Simultaneously, the scheduling instruction issuance module 106 extracts the order identifiers waiting to be assigned within the current batch of the allocation schedule, forming the unassigned order set. Based on these two sets, the scheduling instruction issuance module 106 constructs the weighted bipartite graph.

[0101] A weighted bipartite graph, as a graph theory model, divides its vertices into two disjoint, independent sets: a set of mobile terminals and a set of orders to be assigned. Each connection in the graph exists only between vertices belonging to different sets, and each connection is accompanied by a quantified numerical weight to characterize the degree of association between the corresponding nodes.

[0102] S62, when the matching degree weight is greater than the preset safety margin threshold, the scheduling instruction issuing module 106 establishes a connection edge and assigns a matching degree weight, and calls the maximum weight matching algorithm to optimize the weighted bipartite graph to obtain the optimal matching matrix. The scheduling instruction issuing module 106 obtains the preset safety margin threshold. The preset safety margin threshold is the minimum matching degree limit set to ensure the quality of basic services, and its value range can be set between 0.3 and 0.5.

[0103] The scheduling instruction issuing module 106 traverses the potential combination relationships between the mobile terminal set and the order set to be assigned. For any pairing of a mobile terminal with any order to be assigned, the scheduling instruction issuing module 106 extracts the corresponding matching degree weight calculated in the preceding order. If the corresponding matching degree weight is greater than a preset safety margin threshold, the scheduling instruction issuing module 106 establishes a connection edge between the corresponding mobile terminal vertex and the order vertex to be assigned in the weighted bipartite graph, and assigns the matching degree weight to the connection edge as the edge weight. If the corresponding matching degree weight is not greater than the preset safety margin threshold, no connection edge is established, thereby pruning unqualified matching branches.

[0104] Under the aforementioned filtering mechanism, some orders to be assigned may become isolated vertices because all their corresponding matching weights are too low to establish any connecting edges. To avoid such isolated orders being overlooked or causing algorithm errors in subsequent solutions, as a preferred approach, the scheduling instruction issuing module 106 performs a graph connectivity check before calling the maximum weight matching algorithm. If a vertex of an order to be assigned without any connecting edges is detected, the scheduling instruction issuing module 106 removes it from the weighted bipartite graph and pushes the order to be assigned into the scheduling queue of the next cycle to wait for re-matching, thereby ensuring the continuous progress of the assignment process.

[0105] After constructing and weighting the weighted bipartite graph, the scheduling instruction issuing module 106 calls the maximum weight matching algorithm to optimize the solution of the weighted bipartite graph. The calculation process is based on the following objective function: ; In the formula, A collection of mobile terminals; The set of orders to be assigned; The index of an element in the collection of mobile terminals; The index of the element in the set of orders to be assigned; The matching weight assigned to the established connection edges; For the purpose of finding the optimal decision variable, its value can only be 0 or 1. A value of 1 represents that an assignment relationship has been established, and a value of 0 represents that an assignment relationship has not been established. In addition, it must satisfy the constraint that each vertex in the weighted bipartite graph is associated with at most one valid connection edge.

[0106] For the specific mathematical iteration process of the maximum weight matching algorithm, those skilled in the art can use conventional graph theory matching optimization methods, such as the Kuhn-Munkres algorithm. Its underlying loop calculation logic is a well-known technology in this field and will not be elaborated here.

[0107] The decision variables that maximize the objective function after calculation. The set of elements constitutes the optimal matching matrix. The optimal matching matrix is ​​a two-dimensional Boolean matrix structure, in which the matrix elements record the final assignment relationship. Mathematically, the optimal matching matrix establishes a one-to-one correspondence between mobile terminals and orders to be assigned. At the physical business level, it reflects a group intelligence resource allocation scheme that maximizes the sum of matching weights of the entire escort service group in multi-scenario cross-regional flow.

[0108] S63, the scheduling instruction issuing module 106 converts the orders to be assigned corresponding to the optimal matching matrix into scheduling instructions and issues them to the corresponding mobile terminals via the wireless communication network. The scheduling instruction issuing module 106 parses the matrix elements with a value of 1 in the optimal matching matrix and extracts the associated mobile terminal identifier and the order identifier to be assigned. The scheduling instruction issuing module 106 retrieves the business details data of the corresponding order to be assigned, specifically including parameters such as patient identity information, the physical region of the order's starting point, and the preset start reference time corresponding to the target start time window distribution function. The scheduling instruction issuing module 106 encapsulates the business details data into standardized scheduling instructions according to the communication protocol format.

[0109] The dispatch instruction issuing module 106, based on the extracted mobile terminal identifier, sends the dispatch instruction to the corresponding mobile terminal via the wireless communication network. Upon receiving the dispatch instruction, the mobile terminal displays and reminds the user of the service on its interface, instructing the accompanying staff to execute a new order service after completing the current task, thus completing the dispatch instruction issuance.

[0110] To further aid in understanding the present invention, this embodiment provides a detailed example of the actual operating logic of the group intelligent resource allocation and scheduling method and system for multi-scenario escort services, combined with specific application scenarios.

[0111] In the process of swarm intelligence resource allocation and scheduling for multi-scenario escort services, the escort service involves cross-regional movement between the internal medicine consultation area and the outpatient blood drawing and testing area. The server-side is deployed with a swarm intelligence resource allocation and scheduling system for multi-scenario escort services, which interacts with multiple mobile terminals via a wireless communication network. At this time, five escort service personnel are performing status tasks in the outpatient blood drawing and testing area, each carrying mobile terminal A, B, C, D, and E respectively.

[0112] The node configuration module 101 transforms the medical service process into process nodes with physical region mapping relationships, extracting the environmental feature vector of the process node corresponding to the physical region of outpatient blood drawing and testing. The node configuration module 101 calculates the similarity between the environmental feature vector and known nodes in the template library to extract the target template probability quality function, obtains the actual number of locally collected samples to calculate the confidence weight, and then weights and sums the target template probability quality function with the actual local time consumption probability distribution according to the confidence weight, outputting the baseline probability quality function. For this outpatient blood drawing and testing physical region, the expected value of this baseline probability quality function is 10 minutes, and the original variance of this baseline probability quality function is 2.0.

[0113] After completing the current process node, mobile terminals A, B, C, and D send status signals to the server via the wireless communication network. The deviation calculation module 102 obtains the timestamps of the completion of the current process node and the previous process node, calls the spatial matrix to obtain the minimum walking time between the corresponding physical areas, and determines that the status signal is valid when the time difference is greater than or equal to the minimum walking time.

[0114] The deviation calculation module 102 parses the valid status signals and extracts the actual time consumed by each mobile terminal. The actual time consumed by mobile terminals A, B, and C are 15 minutes, 16 minutes, and 14 minutes, respectively; the actual time consumed by mobile terminal D is 28 minutes. The deviation calculation module 102 extracts the actual time consumed and subtracts the expected value of the baseline probability mass function (10 minutes) to obtain the corresponding status deviation values ​​of 5 minutes, 6 minutes, 4 minutes, and 18 minutes, respectively.

[0115] The impedance factor generation module 103 sets a time sliding window to collect state deviation values ​​from the same outpatient blood sampling and testing physical area to form a state deviation value set. The impedance factor generation module 103 uses the median absolute deviation algorithm to remove the outlier data point at 18 minutes, obtaining a valid sample subset containing values ​​of 5 minutes, 6 minutes, and 4 minutes.

[0116] The impedance factor generation module 103 calculates the mean of the effective sample subset to be 5 minutes. Combining a state update smoothing coefficient of 0.4, a time decay constant of 0.05, and the impedance factor of the previous cycle with an initial value of 0, the impedance factor generation module 103 weights the mean of the effective sample subset with the preset state update smoothing coefficient, and adds the impedance factor of the previous cycle decaying according to the time decay constant to calculate and generate a dynamic impedance factor of 2.0 minutes.

[0117] Mobile terminal E corresponds to an incomplete process node in the task execution state. The probability correction module 104 extracts the newly generated dynamic impedance factor (2.0 minutes), adds the expected value of the baseline probability mass function (10 minutes) to the dynamic impedance factor to complete the translation transformation, and obtains a new expected value of 12 minutes.

[0118] The probability correction module 104 retrieves an uncertainty penalty coefficient of 0.2 and expands the variance of the baseline probability mass function by multiplying the uncertainty penalty coefficient by the absolute value of the dynamic impedance factor, resulting in a new variance of 2.4. The probability correction module 104 then reconstructs the discrete distribution data based on the new expected value and the new variance, and outputs the corrected probability mass function.

[0119] See attached document Figure 9 , Figure 9 The solid line in the figure, without data markers, corresponds to the baseline probability mass function mentioned above. Its probability distribution is relatively concentrated, with the probability peak located at the 10-minute time point. The dashed line in the figure, combined with the asterisk data markers, corresponds to the modified probability mass function mentioned above. Due to the positive resistance of the dynamic impedance factor, the overall distribution curve of the modified probability mass function shifts in the direction of increased time consumption, and the probability peak shifts to the 12-minute time point, reflecting the extension of expected time consumption. At the same time, due to the intervention of the uncertainty penalty coefficient, the variance is increased, and the curve shape of the modified probability mass function is wider than that of the solid line, with an increased distribution span, thus characterizing the increased time uncertainty caused by physical environment congestion.

[0120] The matching degree calculation module 105 performs discrete autocorrelation accumulation processing on the corrected probability quality function corresponding to the incomplete process nodes in the task execution status of the mobile terminal E, generating the final completion time distribution function. The matching degree calculation module 105 obtains the spatial transfer time from the physical region of the current task endpoint to the physical region of the internal medicine consultation starting point of the order to be assigned, and the extracted spatial transfer time value is 3 minutes.

[0121] The matching degree calculation module 105 generates a predicted arrival time distribution by shifting the spatial transfer time of the final completion time distribution function, and shifts the probability peak of this predicted arrival time distribution to the 15-minute time point. Simultaneously, the matching degree calculation module 105 obtains the target start time window distribution function for the orders to be assigned, with the preset start time of these orders being the 16-minute time point. Based on the predicted arrival time distribution and the target start time window distribution function, the matching degree calculation module 105 performs discrete convolution calculations and outputs the matching degree weights.

[0122] See attached document Figure 10 , Figure 10The solid line in the figure, without data markers, corresponds to the predicted arrival time distribution mentioned above, with its probability peak at the 15-minute time point. The dotted line combined with the circled data markers in the figure corresponds to the target initiation time window distribution function mentioned above, with its probability peak at the 16-minute time point. The region where the two distribution curves overlap represents the probability that the mobile terminal E arrives at the new physical region and is within the tolerance range of the pending order. Within the scope of discrete mathematics processing, discrete convolution calculation discretizes and sums the product of the probabilities of these two curves at the corresponding time points. The closer the shapes of the two curves are and the wider the overlap area, the larger the value of the matching degree weight in the calculated output.

[0123] The scheduling instruction issuing module 106 constructs a weighted bipartite graph with the set of mobile terminals and the set of orders to be assigned as its two vertices. When the matching weight between the mobile terminal E and the order to be assigned is greater than the preset safety margin threshold of 0.4, the scheduling instruction issuing module 106 establishes a connection edge and assigns a matching weight.

[0124] The scheduling instruction issuing module 106 calls the maximum weight matching algorithm to optimize the weighted bipartite graph and obtain the optimal matching matrix. The matrix elements of the optimal matching matrix establish a one-to-one correspondence between the mobile terminal E and the order to be assigned. The scheduling instruction issuing module 106 converts the order to be assigned corresponding to the optimal matching matrix into a scheduling instruction and issues it to the mobile terminal E through the wireless communication network, instructing the accompanying service personnel corresponding to the mobile terminal E to execute the order to be assigned, thus completing the swarm intelligence resource allocation and scheduling.

[0125] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A group-based intelligent resource allocation and scheduling method for multi-scenario medical escort services, characterized in that, include: The patient's medical service process is transformed into process nodes with physical region mapping relationships, and a baseline probability quality function is configured for the process nodes; Receive the status signal of the process node sent by the mobile terminal, extract the actual time consumption of the process node from the status signal, and calculate the status deviation value based on the actual time consumption and the benchmark probability quality function; Based on the location identification tag carried by the state deviation value, the state deviation values ​​of the same physical area are grouped into the same state deviation value set, and the dynamic impedance factor with time decay characteristics is calculated based on the state deviation value set. The dynamic impedance factor is used to perform translation and variance expansion on the baseline probability mass function corresponding to the process node that has not yet been executed, to generate a modified probability mass function. Discrete convolution calculation is performed based on the modified probability quality function and the target start time window distribution function of the order to be assigned to obtain the matching degree weight; The optimal matching matrix is ​​calculated based on the matching degree weight, and the order to be assigned corresponding to the optimal matching matrix is ​​converted into a scheduling instruction and sent to the mobile terminal.

2. The group intelligent resource allocation and scheduling method for multi-scenario escort services according to claim 1, characterized in that, Configure a baseline probability quality function for the process node, including: Extract the environmental feature vector of the process node; Calculate the similarity between the environmental feature vector and each known node in the template library; use the historical time consumption probability distribution data corresponding to the known node with the highest similarity value in the template library as the target template probability mass function; Obtain the actual number of samples collected locally and calculate the confidence weight; sum the target template probability quality function and the actual local time consumption probability distribution according to the confidence weight, and output the baseline probability quality function.

3. The group intelligent resource allocation and scheduling method for multi-scenario escort services according to claim 2, characterized in that, The calculation of the similarity between the environmental feature vector and each known node in the template library, using the historical time consumption probability distribution data corresponding to the known node with the highest similarity value in the template library as the target template probability quality function, includes: Calculate the similarity value between the environmental feature vector and the environmental feature vector corresponding to each known node in the template library; If all similarity values ​​are lower than the preset similarity threshold, then the global average probability mass function preset in the template library is called as the target template probability mass function. The global average probability quality function is a discrete probability distribution obtained by performing an indiscriminate arithmetic average of the historical time probability distribution data of all known nodes in the template library.

4. The group intelligent resource allocation and scheduling method for multi-scenario escort services according to claim 1, characterized in that, After receiving the status signal of the process node sent by the mobile terminal, the process further includes: Read the current completion timestamp of the process node from the status signal; Query the historical execution log and extract the historical completion timestamp of the immediately preceding process node; Call the spatial matrix to obtain the minimum walking time between the starting physical region of the previous process node mapping and the ending physical region of the process node mapping; The time difference is calculated by subtracting the historical completion timestamp from the current completion timestamp. If the time difference is less than the minimum walking time, the status signal is determined to be invalid.

5. The group intelligent resource allocation and scheduling method for multi-scenario escort services according to claim 1, characterized in that, Based on the location identifiers carried by the state deviation values, state deviation values ​​from the same physical region are grouped into a single state deviation value set. A dynamic impedance factor with time-decrease characteristics is then calculated from this set, including: Set a time sliding window, traverse all state deviation values ​​received within the current time sliding window, extract the state deviation values ​​mapped to the same physical region, and form a set of state deviation values. The median absolute deviation algorithm is used to identify outlier data points, and these outlier data points are removed from the state deviation value set, leaving the remaining normal data points to form a valid sample subset. The mean of the effective sample subset is calculated, and the mean of the effective sample subset is weighted by a preset state update smoothing coefficient. The impedance factor of the previous cycle, which decays according to the time decay constant, is then added to generate the dynamic impedance factor.

6. The group intelligent resource allocation and scheduling method for multi-scenario escort services according to claim 1, characterized in that, The dynamic impedance factor is used to perform a translation transformation and variance inflation on the baseline probability mass function corresponding to the process node that has not yet been executed, generating a modified probability mass function, including: Extract the dynamic impedance factor of the physical region mapped to the process node that has not yet been executed in the current cycle, which is the latest calculated dynamic impedance factor. The mathematical expectation of the baseline probability mass function is added to the dynamic impedance factor to obtain a new mathematical expectation, thereby completing the translation transformation; The original variance of the baseline probability mass function is calculated, and the original variance of the baseline probability mass function is expanded by multiplying the preset uncertainty penalty coefficient with the absolute value of the dynamic impedance factor to obtain a new variance, thereby completing the variance inflation. Discrete distribution data is reconstructed by combining the normal distribution probability density function with the new mathematical expectation and the new variance. The reconstructed discrete distribution data is then normalized to output the corrected probability mass function.

7. The group intelligent resource allocation and scheduling method for multi-scenario escort services according to claim 1, characterized in that, The step of performing discrete convolution calculation based on the modified probability quality function and the target start time window distribution function of the orders to be assigned to obtain the matching degree weights includes: For each unexecuted process node in the execution state task, the modified probability mass function is subjected to discrete autocorrelation accumulation processing along the execution sequence to generate the final completion time distribution function; Extract the physical region of the current task endpoint and the physical region of the order to be assigned after all tasks have been completed, and call the spatial matrix to obtain the spatial transfer time between the physical region of the current task endpoint and the physical region of the order to be assigned. The spatial transfer time is shifted in the future direction by the final completion time distribution function to generate the predicted arrival time distribution. Based on the predicted arrival time distribution and the target start time window distribution function of the orders to be assigned, the discrete convolution is performed to calculate and output the matching weight.

8. The group intelligent resource allocation and scheduling method for multi-scenario escort services according to claim 7, characterized in that, The step of performing discrete autocorrelation accumulation processing on the modified probability quality function corresponding to each of the unexecuted process nodes in the task in the execution state along the execution sequence to generate the final completion time distribution function includes: When it is determined that there is no task in the execution state that is currently being executed, the probability of the final completion time distribution function at the time point representing the current moment is assigned to 1, and the probability at other time points is assigned to 0.

9. The group intelligent resource allocation and scheduling method for multi-scenario escort services according to claim 1, characterized in that, The step of solving the optimal matching matrix based on the matching degree weight, and converting the orders to be assigned corresponding to the optimal matching matrix into scheduling instructions and sending them to the mobile terminal includes: Construct a weighted bipartite graph with the set of mobile terminals and the set of orders to be assigned as the two vertices; When the matching degree weight is greater than the preset safety margin threshold, a connection edge is established between the corresponding mobile terminal vertex and the vertex to be assigned in the weighted bipartite graph, and the matching degree weight is assigned as the edge weight. The optimal matching matrix is ​​obtained by using the maximum weight matching algorithm to optimize the weighted bipartite graph. Extract the associated mobile terminal identifier and the order identifier to be assigned from the optimal matching matrix, retrieve the business details data, encapsulate it into a standardized scheduling instruction, and send it to the corresponding mobile terminal.

10. A swarm intelligence resource allocation and scheduling system for multi-scenario escort services, used to execute the swarm intelligence resource allocation and scheduling method for multi-scenario escort services according to any one of claims 1 to 9, characterized in that, include: The node configuration module is used to transform the medical service process into process nodes with physical region mapping relationships, and to configure the benchmark probability quality function for the process nodes. The deviation calculation module is used to receive the status signal of the process node sent by the mobile terminal, obtain the timestamp of the completion of the process node and the previous process node, extract the actual time of the process node, and calculate the status deviation value based on the actual time and the benchmark probability mass function. The impedance factor generation module is used to extract the state deviation values ​​mapped to the same physical region, collect the state deviation values ​​of the same physical region to form a state deviation value set, and generate a dynamic impedance factor with time decay characteristics. The probability correction module is used to perform translation and variance expansion on the baseline probability mass function corresponding to the process node that has not yet been executed using the dynamic impedance factor, so as to generate a corrected probability mass function. The matching degree calculation module is used to perform discrete convolution calculation based on the modified probability quality function and the target start time window distribution function of the order to be assigned, and obtain the matching degree weight; The scheduling instruction issuing module is used to solve the optimal matching matrix based on the matching degree weight, and convert the order to be assigned corresponding to the optimal matching matrix into a scheduling instruction to be issued to the mobile terminal.

Images