Communication dispatching method and system for cross-department closed-loop disposal of urban management events

Abstract

The present application belongs to the technical field of city management and communication dispatching, and specifically discloses a communication dispatching method and system for cross-department closed-loop disposal of city management events. The present application adopts a closed-loop external architecture of a central dispatching system + external disposal terminal, the external terminal has multi-modal perception fusion and edge decision-making capability, which reduces the central computing power pressure and realizes the dual improvement of dispatching response speed and scene adaptability; an interpretable causal network model is introduced, the optimal intervention instruction is generated through counterfactual reasoning, which replaces the traditional fixed rule driving mode, improves the instruction accuracy and reduces invalid execution; a closed-loop adaptive iteration mechanism is constructed, the deviation between the actual state and the target state is dynamically calculated, the intervention instruction is continuously optimized, the whole-process closed-loop regulation and control of perception-decision-making-execution-feedback is realized, and the disadvantages of the prior art, such as lack of iteration optimization, low resource utilization rate and repeated occurrence of similar problems, are solved.

Description

Technical Field

[0001] This invention belongs to the field of urban management and communication dispatch technology, specifically involving a communication dispatch method and system for cross-departmental closed-loop handling of urban management events. Background Technology

[0002] Urban management is a core support for ensuring the orderly operation of cities and improving the quality of public services. With the acceleration of urbanization, urban management events are characterized by fragmentation, dynamism, and high concurrency, covering multiple scenarios such as facility maintenance, environmental governance, and traffic management. This places stringent demands on the real-time performance, accuracy, and resource optimization capabilities of dispatching. As a key hub in urban management, communication dispatching aims to achieve end-to-end coordination from status awareness to instruction generation, execution feedback, and optimization iteration, ensuring that all types of urban management events are handled quickly, efficiently, and accurately.

[0003] In existing urban management communication and dispatch practices, mainstream solutions mostly adopt a centralized architecture and a fixed rule-driven model: a central system collects urban status data, generates intervention instructions based on a preset rule base, and then issues them to the implementing departments through simple communication methods. During execution, there is a lack of dynamic adaptation to local real-time scenarios, and only simple results of completion or incompleteness are provided after execution, without forming an effective effect verification and iterative optimization mechanism. Furthermore, existing dispatch communication protocols lack unified standards; instruction transmission only includes basic execution content, failing to clearly define key information such as resource constraints, target trajectories, and effect verification. Moreover, the security protection of the communication process is insufficient, posing potential risks of instruction tampering and identity forgery.

[0004] However, the above-mentioned existing technologies have certain shortcomings in dealing with complex urban management scenarios: (1) The centralized architecture leads to the separation of decision-making and execution. The external execution unit lacks edge computing and autonomous decision-making capabilities, cannot adjust the execution plan according to local real-time perception information, and has a delayed scheduling response and poor scenario adaptability.

[0005] (2) The generation of intervention instructions relies on a fixed rule base, fails to explore the deep causal relationship between urban status and intervention actions, cannot cope with the dynamic urban management needs, and is prone to problems such as low accuracy of instructions and ineffective execution. Summary of the Invention

[0006] In view of this, in order to solve the problems mentioned in the background technology, a communication scheduling method and system for cross-departmental closed-loop handling of urban management incidents is proposed.

[0007] The objective of this invention can be achieved through the following technical solution: The first aspect of this invention provides a communication scheduling method for cross-departmental closed-loop handling of urban management events, including: S1, dividing the target urban management area into multiple grid units, defining and dynamically updating a set of quantifiable state vectors for each grid unit, and forming a unified digital state field across the entire area.

[0008] S2. For the target event and its corresponding influence grid target state vector, deduce the set of atomic intervention actions through a pre-constructed interpretable causal network model, and combine them to generate an intervention instruction set.

[0009] S3. The intervention instruction set is encoded according to the predetermined state task protocol to obtain a state task protocol packet, and then sent to the corresponding external processing terminal with edge computing capability.

[0010] S4. The external processing terminal autonomously plans and executes the atomic intervention action sequence according to the state task protocol package, and collects the state vector of the target grid before and after execution, calculates the state change gradient and actual resource consumption, and encapsulates and transmits the feedback data packet according to the protocol format.

[0011] S5. The digital state field is incrementally updated according to the feedback data packet, and based on the updated differences, the interpretable causal network model is re-triggered to perform counterfactual reasoning and intervention instruction synthesis, and new state task protocol packets are iteratively issued.

[0012] The second aspect of the present invention provides a communication and scheduling system for cross-departmental closed-loop handling of urban management events, comprising: an urban state field construction module, an automatic intervention instruction synthesis module, a state task protocol encoding module, a protocol execution feedback module, and a closed-loop adaptive iteration module.

[0013] The urban state field construction module is connected to the automatic intervention instruction synthesis module, the automatic intervention instruction synthesis module is connected to the state task protocol encoding module, the state task protocol encoding module is connected to the protocol execution feedback module, and the protocol execution feedback module is connected to the closed-loop adaptive iteration module.

[0014] The city state field construction module divides the target city management area into multiple grid units, defines and dynamically updates a set of quantifiable state vectors for each grid unit, and forms a unified digital state field for the entire region.

[0015] The automatic intervention instruction synthesis module, for the target event and its corresponding influence grid target state vector, deduces the set of atomic intervention actions through a pre-built interpretable causal network model, and combines them to generate an intervention instruction set.

[0016] The status task protocol encoding module encodes the intervention instruction set according to a predetermined status task protocol to obtain a status task protocol packet, and sends it to the corresponding external processing terminal with edge computing capabilities.

[0017] The protocol execution feedback module, external processing terminal, autonomously plans and executes atomic intervention action sequence according to the state task protocol package, and collects the state vector of the target grid before and after execution, calculates the state change gradient and actual resource consumption, and encapsulates feedback data packets according to the protocol format for back transmission.

[0018] The closed-loop adaptive iterative module incrementally updates the digital state field based on the feedback data packet, and based on the updated differences, re-triggers the interpretable causal network model to perform counterfactual reasoning and intervention instruction synthesis, and iteratively issues new state task protocol packets.

[0019] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The present invention adopts a closed-loop external architecture of central scheduling system + external processing terminal. The external terminal has multimodal perception fusion and edge decision-making capabilities. It can autonomously plan execution schemes in combination with local real-time scenarios, reduce the pressure on central computing power, and achieve a dual improvement in scheduling response speed and scenario adaptability. It solves the problems of slow response and poor adaptability of traditional centralized architecture.

[0020] 2. This invention introduces an interpretable causal network model, which mines the causal dependency between state variables and intervention actions based on historical urban operation data. It generates optimal intervention instructions through counterfactual reasoning, replacing the traditional fixed rule-driven mode, improving the accuracy of instructions, reducing ineffective execution, and solving the problems of weak instruction targeting and low processing efficiency in existing technologies.

[0021] 3. This invention designs a standardized state task protocol, which clearly defines the core information such as the target state trajectory, resource constraints, and verification observation points. Combined with a secure communication mechanism and dual verification of integrity and legality, it ensures the uniformity, security, and verifiability of command transmission, avoiding execution ambiguity and tampering risks caused by the inconsistency and insufficient security protection of traditional protocols.

[0022] 4. This invention constructs a closed-loop adaptive iterative mechanism, which updates the digital state field based on execution feedback, dynamically calculates the deviation between the actual state and the target state, drives the continuous optimization of intervention instructions, and realizes closed-loop control of the entire process of perception-decision-execution-feedback. This solves the shortcomings of existing technologies, such as lack of iterative optimization, low resource utilization, and recurrence of similar problems, and improves the level of precision in urban management and the efficiency of resource allocation. Attached Figure Description

[0023] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating the implementation steps of the method of the present invention.

[0025] Figure 2 This is a schematic diagram of the system module connections of the present invention. Detailed Implementation

[0026] The technical solutions of 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.

[0027] Example 1 Please see Figure 1 As shown, the present invention provides a communication scheduling method for cross-departmental closed-loop handling of urban management events. The specific steps are as follows: S1, the target urban management area is divided into multiple grid units, and a set of quantifiable state vectors is defined and dynamically updated for each grid unit to form a unified digital state field across the entire region.

[0028] It should be noted that the state vector includes facility integrity calculated based on IoT device data, environmental cleanliness calculated based on the fusion of visual and sensor data, and public space accessibility calculated based on pedestrian and vehicle flow data.

[0029] As an exemplary embodiment of the present invention, the specific implementation process of S1 includes: dividing the target urban management area into multiple grid units according to a preset spatial granularity, and assigning a unique identifier to each grid unit.

[0030] In one specific example, the preset spatial granularity can be 100m×100m for the core urban area, 300m×300m for the general urban area, and 500m×500m for the suburbs.

[0031] Raw facility operation data from IoT devices deployed in each grid cell is acquired, normalized, and the facility integrity is calculated.

[0032] In one specific example, the raw facility operation data includes, but is not limited to, runtime, number of fault alarms, and parameter deviation thresholds.

[0033] Specifically, the original facility operation data is mapped to the [0,1] interval using the min-max normalization algorithm, and the facility integrity value is obtained by weighted summation of each monitoring parameter in the normalized original facility operation data according to preset weights.

[0034] The minimum and maximum values ​​of the data used in the min-max normalization algorithm are determined based on statistical data of the operation of similar facilities over the past three years.

[0035] Simultaneously collect image data from visual sensors and pollutant concentration data from environmental sensors, perform spatiotemporal alignment and feature fusion on the image data and pollutant concentration data, and generate environmental cleanliness data.

[0036] Specifically, spatiotemporal alignment is performed based on GPS positioning information and NTP time synchronization protocol. Visual feature vectors of the image data are extracted using CNN, and environmental feature vectors of the pollutant concentration data are extracted using MLP. The visual feature vectors and environmental feature vectors are weighted and summed to obtain an environmental cleanliness score, such as a score range of 0-100.

[0037] It can access traffic data output from pedestrian density detection equipment and traffic flow monitoring equipment in real time, and calculate the smoothness of public space based on the traffic efficiency model.

[0038] Specifically, the traffic data includes, but is not limited to, the pedestrian flow per unit time obtained by pedestrian density monitoring equipment and the vehicle flow per unit time and average traffic speed obtained by traffic flow monitoring equipment.

[0039] In a specific example, the specific calculation formula for the traffic efficiency model is as follows: The range of public space accessibility is [0,1]. The regional carrying capacity is specifically determined based on urban planning standards.

[0040] The state vector is defined by three dimensions: facility integrity, environmental cleanliness, and accessibility of public spaces. The state vector of each grid unit is refreshed according to a preset time granularity.

[0041] In one specific example, the preset time granularity can be: 5 minutes / time in the core urban area, 15 minutes / time in the general urban area, and 30 minutes / time in the suburbs.

[0042] The refreshed state vector is written into a unified data structure to form a globally unified digital state field, which is then stored in a distributed state database for subsequent use.

[0043] Specifically, the unified data structure is in key-value pair form and includes the following required fields: (1) grid unique identifier, which is a string type such as CD-01-0023, the first two digits are the city code, the middle two digits are the area code, and the last four digits are the grid number; (2) data refresh timestamp, which is a timestamp type, accurate to the second; (3) facility integrity, which is a floating point type, ranging from [0,1]; (4) environmental cleanliness, which is a floating point type, ranging from [0,100]; (5) public space accessibility, which is a floating point type, ranging from [0,1].

[0044] The digital state field adopts a master-slave architecture with sharded storage, and stores the state in a distributed manner in the database, supporting fast retrieval and retrieval by GridID and timestamp.

[0045] This invention adopts a closed-loop external architecture consisting of a central scheduling system and an external processing terminal. The external terminal has multimodal perception fusion and edge decision-making capabilities, and can autonomously plan execution schemes in combination with local real-time scenarios, reducing the pressure on central computing power and achieving a dual improvement in scheduling response speed and scenario adaptability. This solves the problems of slow response and poor adaptability in traditional centralized architectures.

[0046] S2. For the target event and its corresponding influence grid target state vector, deduce the set of atomic intervention actions through a pre-constructed interpretable causal network model, and combine them to generate an intervention instruction set.

[0047] It should be noted that the interpretable causal network model takes the current digital state field and the target state vector as input, and uses the causal graph structure and conditional probability distribution between state variables as internal representation. It performs counterfactual reasoning to derive one or more atomic intervention actions that can drive the state field to evolve towards the target state with the highest probability, and combines them to generate an intervention instruction set containing the execution subject identifier, target grid identifier, target state value, atomic intervention action sequence and resource constraints.

[0048] The atomic intervention action refers to the smallest unit operation that can be executed independently in urban management, including but not limited to single management actions such as facility maintenance, environmental cleanup, and traffic management.

[0049] As an exemplary embodiment of the present invention, the specific construction process of the explainable causal network model includes: collecting historical city operation datasets, the datasets containing historical monitoring values ​​of state vectors of each grid unit, records of executed atomic intervention actions, and state vector deviation values ​​before and after the execution of intervention actions.

[0050] Specifically, the historical monitoring values ​​of the state vector include historical monitoring values ​​of facility integrity, environmental cleanliness, and public space accessibility for each grid unit; the atomic intervention action record includes action type, executing entity, execution time, and resource consumption.

[0051] A constraint-based PC algorithm is used to extract the causal dependency between state variables and the original intervention action, generating a causal graph structure in the form of a directed acyclic graph.

[0052] Specifically, the process of extracting causal dependencies using the constraint-based PC algorithm, i.e. the causal discovery algorithm, includes: (1) screening out statistically correlated variable pairs through mutual information tests, such as garbage cleaning actions and environmental cleanliness; (2) eliminating spurious variable pairs based on conditional independence tests, such as chi-square tests, for example, although rain and environmental cleanliness are correlated, they have no direct causal relationship; (3) determining the causal direction of the variable pairs, such as determining through the direction discrimination rules in causal inference that it is garbage cleaning → cleanliness improvement, rather than the reverse, and finally generating a directed acyclic graph.

[0053] A conditional probability distribution parameter is assigned to each directed edge in the causal graph structure using a Bayesian network, thereby generating a conditional probability table and completing the training of the interpretable causal network model.

[0054] Specifically, Bayesian networks are preferred to assign conditional probability distribution parameters. The specific process includes: (1) Data preprocessing: filling missing values ​​in the historical city operation dataset, such as using the KNN algorithm, k=5, and outlier removal, i.e., the 3σ principle; (2) Parameter learning: using the maximum likelihood estimation method to learn the conditional probability table of the Bayesian network, i.e., CPT, for example: P(environmental cleanliness improvement and garbage cleaning)=0.9, P(environmental cleanliness improvement and garbage cleaning not implemented)=0.05; (3) Model validation: using cross-validation methods such as 5-fold cross-validation, when the model prediction accuracy is ≥85%, it is considered that the training of the interpretable causal network model is completed.

[0055] As an exemplary embodiment of the present invention, the specific generation process of the intervention instruction set includes: loading a trained interpretable causal network model, and using the real-time state vector of the target grid in the current digital state field and the preset target state vector as model input.

[0056] In the causal graph of an interpretable causal network model, feasible causal paths from the current state node to the target state node are selected.

[0057] It should be noted that the screening criteria include: (1) Path direction: from the current state node to the target state node, with no reverse or loop paths; (2) Path length: contains a maximum of 3 intermediate nodes, such as the current cleanliness score of 60 points → garbage cleaning → cleanliness score of 80 points → road washing → cleanliness score of 90 points, with a length of 2, which meets the requirements, so as to avoid the path being too long and causing low reasoning efficiency; (3) Causal strength: the causal strength, i.e., the conditional probability, of each directed edge in the path is ≥50%, and weak causal association paths, such as paths with causal strength <30%, are eliminated.

[0058] For each feasible causal path, perform counterfactual reasoning to calculate the joint probability that all dimensions of the target state variable reach the expected value under different atomic intervention actions.

[0059] Specifically, the probability calculation of the counterfactual reasoning adopts the posterior probability accumulation method, which includes: (1) For a single causal path, assuming that a certain atomic intervention action is applied, the posterior probability of each intermediate node is calculated based on the conditional probability table of the model; (2) The joint probability of the target state node reaching the expected value is accumulated. For example, the probability that garbage cleaning makes the cleanliness score change from 60 to 80 is 70%, and the probability that road washing makes the score change from 80 to 90 is 80%. Then the joint probability = 70% × 80% = 56%; (3) If there are multiple parallel causal paths, such as simultaneously involving the improvement of environmental cleanliness and public space accessibility, the product of the probability of achieving the target of each path is calculated as the total driving probability of the combined action.

[0060] Select one or more atomic intervention actions with the highest joint probability. If multiple joint intervention actions exist, retain all candidate atomic intervention actions with a joint probability exceeding a set limit, such as 70%.

[0061] The execution dependencies of atomic intervention actions are determined based on the action dependency matrix, and the execution dependencies are combined with the resource consumption model to perform topological sorting and resource feasibility verification on the selected atomic intervention actions.

[0062] It should be noted that the action dependency matrix is ​​a structured data carrier that records the execution relationship between various atomic intervention actions in the urban management scenario. It supports dynamic updates and is used to clarify the constraint rules of different atomic intervention actions in terms of execution order and compatibility. Its essence is a two-dimensional matrix, where each row and column is a unique identifier of the atomic intervention action. The matrix elements are used to mark the dependency type between the corresponding row and column actions. Specifically, it includes three core relationships: (1) Sequential dependency: marked as A→B, that is, action A must be executed before action B, such as garbage cleaning must be completed before road washing; (2) Mutually exclusive dependency: marked as A⊥B, that is, action A and action B cannot be executed simultaneously or sequentially, such as traffic closure and traffic diversion have logical conflicts; (3) No dependency: marked as A∥B, that is, action A and action B can be executed in parallel or in any order, such as street light maintenance and green belt cleaning have no related constraints.

[0063] Specifically, the topological sorting can be performed using the Kahn algorithm to determine the order of action execution. The specific process includes: (1) constructing an adjacency list of a directed acyclic graph with action dependencies as input; (2) calculating the in-degree of each action and outputting the action sequence step by step according to the principle of prioritizing actions with an in-degree of 0; (3) output format: [action1ID→action2ID→action3ID], such as [garbage cleaning→road washing→facility inspection]. If there are multiple actions without dependencies, they are sorted from low to high resource consumption to optimize resource utilization efficiency.

[0064] The resource feasibility verification specifically needs to meet the following hard indicators, and the thresholds are synchronized in real time by the central scheduling system: (1) Manpower threshold: The total number of staff required for the action is less than or equal to the current available manpower pool capacity of the executing entity. For example, if there are 5 available cleaners, the action requires 2 people, which meets the requirements; (2) Equipment threshold: The total number of required equipment is less than or equal to the upper limit of regional equipment inventory. For example, if there are 3 sweepers in the region, the action requires 1 vehicle, which meets the requirements; (3) Time threshold: The total duration of the action sequence is less than or equal to the task completion time limit. For example, if the task time limit is 1 hour, the total duration of the action is 40 minutes, which meets the requirements; (4) Cost threshold: The total direct cost is less than or equal to the upper limit of cost in the resource constraints. For example, if the upper limit of cost is 500 yuan, the total cost of the action is 300 yuan, which meets the requirements; If any one of the indicators is not met, the resource is deemed infeasible and the action combination is removed.

[0065] The validated atomic intervention action sequence is combined with the executor identifier, target grid identifier, target state value, and resource constraints in a structured manner to generate a standardized intervention instruction set.

[0066] As an exemplary embodiment of the present invention, the resource constraints include an upper limit for human resource allocation, an upper limit for the number of devices called, and a task completion time limit. In the process of generating the intervention instruction set, a multi-objective optimization constraint solution is implemented. The specific process includes: pre-associating each candidate atomic intervention action with its standard resource consumption template.

[0067] Specifically, the standard resource consumption template is structured data, including action ID, personnel configuration (i.e., the number and skill level of the required personnel), equipment configuration (i.e., the type and quantity of special equipment and energy consumption per unit time), estimated time (including the baseline value ± the floating range), and resource consumption priority (e.g., 1-5 levels, with level 1 being the lowest consumption).

[0068] The template is updated quarterly based on historical execution data. When the atomic action process is optimized, such as by adding equipment or when resource costs change, such as by adjusting labor costs, an immediate update is triggered. After the update, it is synchronized to the central scheduling system and all external terminals.

[0069] Based on the selected set of atomic intervention actions, an integer programming model is constructed with the dual objectives of minimizing total resource consumption and maximizing state evolution probability.

[0070] Specifically, the integer programming model includes decision variables and an objective function: (1) Decision variables: ,in This is the sequence number of the atomic intervention action. This indicates that the action to intervene in this atom has been selected. (2) Objective function: ,in For atomic intervention actions The total resource consumption, including the weighted sum of manpower, equipment, and time; ,in For atomic intervention actions The state evolution probability contribution value is output by the causal model.

[0071] Introduce hard constraints to ensure that the total manpower demand does not exceed the current available manpower pool capacity of the executing entity, the total number of equipment calls does not exceed the regional equipment inventory limit, and the total time of the action sequence does not exceed the task deadline.

[0072] Specifically, the constraints include: (1) Human constraints: ,in For atomic intervention actions Manpower required (2) Equipment constraints: (1) Maximum available manpower capacity, i.e., available manpower pool capacity; ,in For atomic intervention actions Required number of equipment. (3) Time constraints: (This refers to the upper limit of equipment inventory;) ,in For atomic intervention actions Estimated time: This refers to the task completion deadline, or the task end time.

[0073] It should be noted that the available manpower pool capacity specifically comes from the real-time manpower dispatch system of the executing entity, including the number of on-duty personnel and personnel that can be urgently deployed; the equipment inventory limit specifically comes from the real-time inventory data of the regional equipment management platform, including the status of equipment in use, idle, and awaiting maintenance; and the task completion time limit is specifically determined according to the level of urban management events, such as emergency events. Important events routine events .

[0074] Solve the integer programming model to obtain the combination of atomic intervention actions that satisfies all hard constraints and is Pareto optimal.

[0075] Specifically, a branch and bound algorithm combining depth-first search and pruning optimization is used to solve the integer programming model. The termination conditions are: finding a Pareto optimal solution set, running time ≤ 10 seconds, or the difference between the optimal and suboptimal solutions ≤ 5%. Among these, the algorithm prioritizes... The highest combination, Choose when they are the same The smallest combination.

[0076] The optimal combination is used as the final atomic intervention action sequence that passes the resource feasibility check, and is then used to generate a standardized intervention instruction set.

[0077] This invention introduces an interpretable causal network model, which mines the causal dependency between state variables and intervention actions based on historical urban operation data. It generates optimal intervention instructions through counterfactual reasoning, replacing the traditional fixed rule-driven mode. This improves the accuracy of instructions, reduces ineffective execution, and solves the problems of weak instruction targeting and low processing efficiency in existing technologies.

[0078] S3. The intervention instruction set is encoded according to the predetermined state task protocol to obtain a state task protocol packet, and then sent to the corresponding external processing terminal with edge computing capability.

[0079] It should be noted that the protocol explicitly specifies the target location, the expected state change curve, the allowed upper limit of resource costs, and the conditions for verifying the effectiveness of actions.

[0080] The protocol explicitly defines the target state trajectory, the execution logic and resource constraints of atomic actions, and key state observation points for verifying the effectiveness of actions.

[0081] As an exemplary embodiment of the present invention, the state task protocol encoding in step S3 includes: defining a data structure for the state task protocol, wherein the data structure includes a protocol header field, a target grid identifier field, a target state trajectory field, an atomic intervention action sequence field, a resource cost upper limit field, and a verification observation point field.

[0082] Write the unique identifier of the target grid in the intervention instruction set into the target grid identifier field.

[0083] The target state vector is expanded into a discrete sequence of state points along the time dimension and written into the target state trajectory field.

[0084] Specifically, the step of expanding the target state vector into a discrete state point sequence along the time dimension involves discretizing the target state vector into N state points according to the principle of equal distribution of the total duration of atomic actions, where N = number of actions × 2.

[0085] The principle of evenly distributing the total duration of the atomic actions specifically includes: (1) If the total duration of the atomic intervention action sequence is T, such as 40 minutes, it is discrete into N state points, where N = number of actions × 2. For example, if there are 2 actions, it is discrete into 4 state points; (2) Time granularity = T / (N-1). For example, if there are 4 state points in 40 minutes, the time granularity is 13.33 minutes, and the state point timestamps are 0 min, 13.33 min, 26.66 min, and 40 min; (3) The target values ​​of the discrete state points are set linearly increasing or decreasing. For example, if the environmental cleanliness is from 60 points to 90 points, the target values ​​of the 4 state points are 60, 70, 80, and 90.

[0086] The atomic intervention action sequence is encoded into action instruction codes according to the execution order and written into the atomic intervention action sequence field.

[0087] Specifically, the action instruction code adopts a 5-bit fixed-length code, and the encoding rules include: (1) the first 2 bits are the action type code, such as 01=garbage cleaning, 02=road washing, 03=facility maintenance, 04=traffic diversion; (2) the last 3 bits are the execution order code, such as 001-999, assigned according to the topology sorting result; (3) the encoding mapping table is pre-placed in the central dispatch system and external disposal terminal, and supports dynamic updates.

[0088] The maximum allowable human, material, and time consumption values ​​for this task are determined based on resource constraints and written into the resource cost ceiling field.

[0089] Specifically, resource constraints are quantified into specific values ​​in the format of maximum manpower - maximum material resources - maximum time, such as 2 people - 1 sweeper / 1 flushing truck - 40 minutes, and written into the resource cost upper limit field.

[0090] Extract the key state variables and their sampling time windows that need to be verified before and after each action from the atomic intervention action sequence, form a list of verification observation points and write them into the verification observation point field.

[0091] Specifically, for each atomic action, key state variables sampled within 1-5 minutes before execution and key state variables sampled within 3-10 minutes after execution are extracted to form a list of verification observation points, which are then written into the verification observation point field.

[0092] The sampling time window is adapted according to the action type: (1) For short-term actions, such as garbage cleaning, with a duration of ≤30 minutes, the sampling window before execution is within 1 minute before the action starts, and the sampling window after execution is within 3 minutes after the action ends; (2) For long-term actions, such as facility maintenance, with a duration of >30 minutes, the sampling window before execution is within 5 minutes before the action starts, and the sampling window after execution is within 10 minutes after the action ends.

[0093] The complete protocol data structure is serialized and encrypted to generate a status task protocol packet, which is then sent to the designated external processing terminal through a secure communication channel.

[0094] Specifically, the Protobuf algorithm is used to serialize the complete protocol data structure, and then the serialized data is encrypted using the AES-256-GCM encryption algorithm to generate a state task protocol packet. The state task protocol packet is then sent to the designated external processing terminal through a secure communication channel with TLS1.3+PKI two-way authentication. The terminal receives the packet by matching the unique terminal ID.

[0095] As an exemplary embodiment of the present invention, after the data is sent to a designated external processing terminal through a secure communication channel, dual verification of integrity and legality is performed during the protocol packet reception and parsing phase. Specifically, this includes: pre-setting a terminal identity certificate and corresponding private key issued by a trusted certificate authority on the external processing terminal side.

[0096] After generating the status task protocol packet, the central scheduling system uses its own private key to digitally sign the protocol packet and uses the public key of the external processing terminal to encrypt the protocol packet.

[0097] Specifically, after the central scheduling system generates the status task protocol packet, it first uses the SHA-256 algorithm to calculate the hash value of the protocol packet, then uses its own private key to digitally sign the hash value using the RSA-2048 algorithm, and finally uses the public key of the target external processing terminal to encrypt the protocol packet and digital signature.

[0098] After receiving the encrypted protocol packet, the external processing terminal decrypts it using its own private key and then verifies the validity of the digital signature using the public key of the central scheduling system.

[0099] Specifically, after receiving the encrypted protocol packet, the external processing terminal first decrypts it using its own private key to obtain the original protocol packet and digital signature; then it recalculates the hash value of the original protocol packet using the SHA-256 algorithm and compares it with the hash value attached to the digital signature. If they do not match, the integrity verification is deemed to have failed.

[0100] If the signature verification fails or the protocol packet hash value is inconsistent with the content, execution will be refused and a security exception will be reported.

[0101] If the verification passes, the version number, expiration stamp, and execution entity identifier in the protocol header field are further checked to see if they match the local configuration. The protocol packet is only loaded into the execution queue when all legality checks pass.

[0102] Specifically, if any of the following occurs: integrity verification failure, invalid signature, version number incompatibility, expired expiration stamp, or mismatched execution entity identifier, the instruction will be rejected, and the security anomaly will be reported encrypted through the TLS 1.3 secure communication channel. The protocol packet will only be loaded into the terminal execution queue if all verifications pass.

[0103] The security anomaly specifically includes the anomaly type, protocol packet ID, terminal ID, and anomaly timestamp.

[0104] S4. The external processing terminal autonomously plans and executes the atomic intervention action sequence according to the state task protocol package, and collects the state vector of the target grid before and after execution, calculates the state change gradient and actual resource consumption, and encapsulates and transmits the feedback data packet according to the protocol format.

[0105] As an exemplary embodiment of the present invention, the specific implementation process of S4 includes: the external processing terminal has the ability to fuse four types of multimodal perceptions, namely vision, environment, traffic flow, and equipment status, as well as edge decision-making capabilities. Based on the constraints of the status task protocol package and local real-time perception information, it autonomously plans the execution scheme of atomic intervention actions, including personnel division of labor and equipment scheduling, and executes them in sequence.

[0106] The edge decision-making capability mainly focuses on three types of autonomous decision-making scenarios: (1) fine-tuning of execution order, such as adjusting the execution order of actions without dependencies according to the real-time resource occupancy; (2) adaptation of execution parameters, such as automatically increasing the sweeper operation time by 10%-20% when the initial value of environmental cleanliness is lower than expected; (3) decision on handling invalid actions, that is, judging whether to retry based on the environmental context, such as retrying first if the action is invalid due to temporary crowd density; and requesting alternative instructions first if the action is invalid due to equipment failure.

[0107] The decision-making is mainly based on the edge decision-making rule base pre-installed on the external processing terminal. The rule base is generated by training based on historical execution data and supports dynamic updates.

[0108] Before and after each atomic intervention action, the state vectors of facility integrity, environmental cleanliness, and public space accessibility of the target grid are collected, while the actual consumption of human, equipment, and time resources are recorded.

[0109] Before executing each atomic intervention action, the corresponding key state variable perception module is activated based on the verification observation point field defined in the state task protocol package.

[0110] After the atomic intervention action is completed, the state vector within the time window specified by the verification observation point is collected.

[0111] Specifically, the specified time window is: short-term action ≤ 3 minutes, long-term action ≤ 10 minutes.

[0112] Calculate the changes in state variables before and after execution, and compare them with the minimum effective change threshold preset in the protocol.

[0113] In a specific example, the minimum effective change threshold is set according to the state vector dimension and is strongly correlated with the action type. The specific preset values ​​are as follows, which can be dynamically adjusted through the central scheduling system: (1) Facility integrity: When the action type is facility maintenance, the minimum effective change threshold is ≥0.1, such as from 0.3→0.4; (2) Environmental cleanliness: When the action type is garbage cleaning, the minimum effective change threshold is ≥8 points, and when the action type is road washing, the minimum effective change threshold is ≥15 points; (3) Public space accessibility: When the action type is traffic management, the minimum effective change threshold is ≥0.15, such as from 0.2→0.35.

[0114] The minimum effective change threshold is determined based on 70% of the average effective change of similar actions in history, ensuring that the threshold is neither too lenient nor too strict.

[0115] If the change amount does not reach the minimum effective change threshold, the atomic intervention action is determined to be invalid, and the invalid action identifier and the current environmental context information are recorded.

[0116] Specifically, the invalid action identifier includes action type, execution order, and invalid reason; the environmental context information includes, but is not limited to, weather conditions such as sunny / rainy / windy / wind force level, real-time pedestrian and vehicle flow density within the grid, equipment operating status such as whether the sweeper is malfunctioning / whether the tools are working properly, external interference events such as whether there is temporary construction / traffic control, execution time such as whether it is a peak period / nighttime or other special time period, and status data collection anomaly identifiers such as whether the sensor is malfunctioning / whether the data is valid.

[0117] When an action is invalid, the external processing terminal will first trigger a local retry mechanism. If the retry fails, it will request an alternative action command from the central scheduling system.

[0118] Specifically, when an invalid execution occurs, the terminal will first trigger a local retry mechanism, which will retry a maximum of 2 times with an interval of ≥5 minutes and a cumulative resource consumption of ≤120% of the resource limit. If the retry fails, the terminal will request an alternative action instruction from the central scheduling system.

[0119] The feedback packet ID, target grid identifier, action execution log, pre-execution state vector, post-execution state vector, state change gradient, actual resource consumption, and action validity status are encapsulated into a feedback data packet in Protobuf format and transmitted back to the central scheduling system through a secure communication channel.

[0120] Specifically, the feedback packet ID is a unique identifier associated with the corresponding protocol packet ID; the action execution log includes the action ID, execution time, and execution duration; the pre-execution state vector and the post-execution state vector include facility integrity, environmental cleanliness, and public space accessibility; the state change gradient includes the deviation values ​​of the three dimensions of facility integrity, environmental cleanliness, and public space accessibility before and after execution; the actual resource consumption includes manpower, equipment, and time; and the action validity status includes a sub-action marker for valid / invalid + an invalid reason code.

[0121] This invention designs a standardized state task protocol that clearly defines core information such as target state trajectory, resource constraints, and verification observation points. Combined with a secure communication mechanism and dual verification of integrity and legality, it ensures the uniformity, security, and verifiability of command transmission, avoiding execution ambiguity and tampering risks caused by the inconsistency and insufficient security protection of traditional protocols.

[0122] S5. The digital state field is incrementally updated according to the feedback data packet, and based on the updated differences, the interpretable causal network model is re-triggered to perform counterfactual reasoning and intervention instruction synthesis, and new state task protocol packets are iteratively issued.

[0123] As an exemplary embodiment of the present invention, the specific implementation process of S5 includes: receiving the feedback data packet returned by the external processing terminal, and parsing the pre-execution state vector, post-execution state vector, actual resource consumption and action execution log contained therein.

[0124] By matching the target grid identifier in the feedback data packet with the corresponding grid entry in the distributed state database, the old state data of the grid is overwritten only by the post-execution state vector, the metadata is preserved and the update log is recorded, and the incremental update of the state field is completed.

[0125] Calculate the absolute difference between the updated state vector and the target state vector in three dimensions: facility integrity, environmental cleanliness, and public space accessibility.

[0126] If the absolute difference in any dimension exceeds the tolerance threshold, the updated state vector will be used as the current state input for the next round of inference.

[0127] Specifically, the preset tolerance thresholds are set according to the following dimensions: the tolerance threshold for facility integrity is 0.05, the tolerance threshold for environmental cleanliness is 5 points, and the tolerance threshold for public space accessibility is 0.05, and can be dynamically adjusted according to the event level.

[0128] The preset tolerance threshold is specifically based on urban management industry standards such as the standard for the integrity rate of municipal facilities and the standard for environmental sanitation quality, and the average error of historical implementation data, such as taking 70% of the error of historical compliance cases. This ensures that the threshold is neither too lenient, thus avoiding failure to meet actual management requirements, nor too strict, thus avoiding excessive iteration and waste of resources.

[0129] It should be noted that if the absolute difference of all three dimensions is less than its corresponding tolerance threshold, the iteration process is terminated, the target grid state is marked as compliant, and a stable observation period is started; if the absolute difference of any dimension exceeds the tolerance threshold, the updated state vector is used as the current state input for the new round of inference.

[0130] Re-invoking the interpretable causal network model to perform counterfactual reasoning generates a new set of atomic intervention actions.

[0131] Specifically, the interpretable causal network model is re-invoked, the model parameters are adjusted according to the rule of increasing the weight of the unmet dimension by 20%, and counterfactual reasoning is performed to generate a new set of atomic intervention actions.

[0132] An updated set of intervention instructions is generated based on the new set of atomic intervention actions, and S3-S5 are executed repeatedly until the differences in all dimensions are less than the tolerance threshold, at which point the iteration process is terminated.

[0133] Specifically, after the iteration terminates, the following closing operations are performed: (1) Result archiving: The initial state vector of the target grid, the state vector after each round of execution, the final target state vector, the resource consumption of each round, and the number of iterations are archived to the historical database for subsequent model optimization; (2) State marking: In the digital state field, the state of the grid is marked as target, and a stable observation period is set, such as 30 minutes in the core urban area and 60 minutes in the general urban area. During the observation period, state data is collected every 10 minutes. If the state regresses, i.e. the difference is ≥ the tolerance threshold, the iteration is automatically restarted; (3) Resource release: The executing entity is notified to release the human and equipment resources occupied by the task and include them in the regional available resource pool.

[0134] This invention constructs a closed-loop adaptive iterative mechanism, which updates the digital state field based on execution feedback, dynamically calculates the deviation between the actual state and the target state, drives the continuous optimization of intervention instructions, and realizes closed-loop control of the entire process of perception-decision-execution-feedback. It solves the shortcomings of existing technologies, such as lack of iterative optimization, low resource utilization, and recurrence of similar problems, and improves the level of precision in urban management and the efficiency of resource allocation.

[0135] Example 2 Please see Figure 2As shown, this invention provides a communication and scheduling system for cross-departmental closed-loop handling of urban management events. The specific modules are distributed as follows: urban state field construction module, automatic intervention instruction synthesis module, state task protocol encoding module, protocol execution feedback module, and closed-loop adaptive iteration module.

[0136] The urban state field construction module is connected to the automatic intervention instruction synthesis module, the automatic intervention instruction synthesis module is connected to the state task protocol encoding module, the state task protocol encoding module is connected to the protocol execution feedback module, and the protocol execution feedback module is connected to the closed-loop adaptive iteration module.

[0137] The city state field construction module divides the target city management area into multiple grid units, defines and dynamically updates a set of quantifiable state vectors for each grid unit, and forms a unified digital state field for the entire region.

[0138] The automatic intervention instruction synthesis module, for the target event and its corresponding influence grid target state vector, deduces the set of atomic intervention actions through a pre-built interpretable causal network model, and combines them to generate an intervention instruction set.

[0139] The status task protocol encoding module encodes the intervention instruction set according to a predetermined status task protocol to obtain a status task protocol packet, and sends it to the corresponding external processing terminal with edge computing capabilities.

[0140] The protocol execution feedback module, external processing terminal, autonomously plans and executes atomic intervention action sequence according to the state task protocol package, and collects the state vector of the target grid before and after execution, calculates the state change gradient and actual resource consumption, and encapsulates feedback data packets according to the protocol format for back transmission.

[0141] The closed-loop adaptive iterative module incrementally updates the digital state field based on the feedback data packet, and based on the updated differences, re-triggers the interpretable causal network model to perform counterfactual reasoning and intervention instruction synthesis, and iteratively issues new state task protocol packets.

[0142] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0143] Those skilled in the art will recognize that the algorithmic steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0144] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0145] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0146] Finally, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A communication and dispatching method for cross-departmental closed-loop handling of urban management incidents, characterized in that: include: S1. Divide the target urban management area into multiple grid units, define and dynamically update a set of quantifiable state vectors for each grid unit, and form a unified digital state field for the entire area. S2. For the target event and its corresponding influence grid target state vector, deduce the set of atomic intervention actions through a pre-constructed interpretable causal network model, and combine them to generate an intervention instruction set; S3. The intervention instruction set is encoded according to the predetermined state task protocol to obtain a state task protocol packet, and then sent to the corresponding external processing terminal with edge computing capability. S4. The external processing terminal autonomously plans and executes the atomic intervention action sequence according to the state task protocol package, and collects the state vector of the target grid before and after execution, calculates the state change gradient and actual resource consumption, and encapsulates and transmits the feedback data packet according to the protocol format. S5. The digital state field is incrementally updated according to the feedback data packet, and based on the updated differences, the interpretable causal network model is re-triggered to perform counterfactual reasoning and intervention instruction synthesis, and new state task protocol packets are iteratively issued. The specific implementation process of S1 includes: The target urban management area is divided into multiple grid units according to a preset spatial granularity, and a unique identifier is assigned to each grid unit; Obtain the raw facility operation data of IoT devices in each grid cell, normalize it, and calculate the facility integrity. Simultaneously collect image data and pollutant concentration data within each grid cell, perform spatiotemporal alignment and feature fusion on the image data and pollutant concentration data, and generate environmental cleanliness; Real-time access to traffic data within each grid cell; calculation of public space accessibility based on traffic efficiency model. The state vector is defined by three dimensions: facility integrity, environmental cleanliness, and public space accessibility. The state vector of each grid unit is refreshed according to a preset time granularity. The refreshed state vector is written into a unified data structure to form a unified digital state field across the entire domain, and stored in a distributed state database for subsequent use. The specific construction process of the interpretable causal network model includes: Collect historical city operation datasets, which include historical monitoring values ​​of state vectors for each grid cell, records of executed atomic intervention actions, and state vector deviation values ​​before and after the intervention actions. Extract the causal dependencies between state variables and the original intervention actions, and generate a causal graph structure in the form of a directed acyclic graph; A conditional probability distribution parameter is assigned to each directed edge in the causal graph structure using a Bayesian network, thereby generating a conditional probability table and completing the training of the interpretable causal network model. The specific generation process of the intervention instruction set includes: The real-time state vector of the target grid in the current digital state field and the preset target state vector are used together as input to the interpretable causal network model; In the causal graph of an interpretable causal network model, select feasible causal paths from the current state node to the target state node; For each feasible causal path, perform counterfactual reasoning to calculate the joint probability that all dimensions of the target state variable reach the expected value under different atomic intervention actions. Select one or more atomic intervention actions with the highest joint probability. If multiple joint probabilities exist, retain all candidate atomic intervention actions whose joint probability exceeds a set limit. The execution dependencies of atomic intervention actions are determined based on the action dependency matrix, and the execution dependencies are combined with the resource consumption model to perform topological sorting and resource feasibility verification on the selected atomic intervention actions. The validated atomic intervention action sequence is combined with the executor identifier, target grid identifier, target state value, and resource constraints in a structured manner to generate a standardized intervention instruction set.

2. The communication scheduling method for cross-departmental closed-loop handling of urban management incidents according to claim 1, characterized in that: The resource constraints include upper limits on human resource allocation, equipment deployment, and task completion time. Furthermore, multi-objective optimization constraints are addressed during the intervention instruction set generation process, which specifically includes: Pre-associate each candidate atomic intervention action with its standard resource consumption template; Based on the selected set of atomic intervention actions, an integer programming model is constructed with the dual objectives of minimizing total resource consumption and maximizing state evolution probability. Introduce hard constraints to ensure that the total manpower demand does not exceed the current available manpower pool capacity of the executing entity, the total number of equipment calls does not exceed the regional equipment inventory limit, and the total time of the action sequence does not exceed the task deadline. Solve the integer programming model to obtain the combination of atomic intervention actions that satisfies all hard constraints and is Pareto optimal; The optimal combination is used as the final atomic intervention action sequence that passes the resource feasibility check, and is then used to generate a standardized intervention instruction set.

3. The communication scheduling method for cross-departmental closed-loop handling of urban management incidents according to claim 1, characterized in that: The state task protocol encoding in step S3 includes: Define the data structure of the state task protocol, which includes a protocol header field, a target grid identifier field, a target state trajectory field, an atomic intervention action sequence field, a resource cost upper limit field, and a verification observation point field; Write the unique identifier of the target grid in the intervention instruction set into the target grid identifier field; The target state vector is expanded into a discrete state point sequence along the time dimension and written into the target state trajectory field; The atomic intervention action sequence is encoded into action instruction codes according to the execution order and written into the atomic intervention action sequence field; Determine the maximum allowable human, material, and time consumption values ​​for this task based on resource constraints, and write them into the resource cost ceiling field; Extract the key state variables and their sampling time windows that need to be verified before and after each action from the atomic intervention action sequence, form a list of verification observation points and write them into the verification observation point field; The complete protocol data structure is serialized and encrypted to generate a status task protocol packet, which is then sent to the designated external processing terminal through a secure communication channel.

4. The communication scheduling method for cross-departmental closed-loop handling of urban management incidents according to claim 3, characterized in that: After being transmitted to the designated external processing terminal through a secure communication channel, the protocol packet reception and parsing phase performs dual verification of integrity and legitimacy, which specifically includes: Pre-install a terminal identity certificate and corresponding private key issued by a trusted certificate authority on the external processing terminal side; After generating the status task protocol packet, the central dispatch system uses its own private key to digitally sign the protocol packet and uses the public key of the external processing terminal to encrypt the protocol packet. After receiving the encrypted protocol packet, the external processing terminal decrypts it using its own private key, and then uses the public key of the central scheduling system to verify the validity of the digital signature. If the signature verification fails or the protocol packet hash value is inconsistent with the content, execution will be refused and a security exception will be reported. If the verification passes, the version number, expiration stamp, and execution entity identifier in the protocol header field are further checked to see if they match the local configuration. The protocol packet is only loaded into the execution queue when all legality checks pass.

5. The communication scheduling method for cross-departmental closed-loop handling of urban management incidents according to claim 1, characterized in that: The specific implementation process of S4 includes: Before executing each atomic intervention action, the corresponding key state variable perception module is activated based on the verification observation point field defined in the state task protocol package. After the atomic intervention action is completed, the state vector within the specified time window of the verification observation point is collected; Calculate the changes in state variables before and after execution, and compare them with the minimum effective change threshold preset in the protocol; If the change amount does not reach the minimum effective change threshold, the atomic intervention action is determined to be invalid, and the invalid action identifier and the current environmental context information are recorded. When an action is invalid, the external processing terminal will first trigger a local retry mechanism. If the retry fails, it will request an alternative action command from the central scheduling system. The feedback packet ID, target grid identifier, action execution log, pre-execution state vector, post-execution state vector, state change gradient, actual resource consumption, and action validity status are encapsulated into a feedback data packet in Protobuf format and transmitted back to the central scheduling system through a secure communication channel.

6. The communication scheduling method for cross-departmental closed-loop handling of urban management incidents according to claim 5, characterized in that: The specific implementation process of S5 includes: By matching the target grid identifier in the feedback data packet with the corresponding grid entry in the distributed state database, the old state data of the grid is overwritten only by the post-execution state vector, the metadata is preserved and the update log is recorded, and the incremental update of the state field is completed. Calculate the absolute difference between the updated state vector and the target state vector in three dimensions: facility integrity, environmental cleanliness, and public space accessibility; If the absolute difference in any dimension exceeds the tolerance threshold, the updated state vector will be used as the current state input for the next round of inference. Re-invoke the interpretable causal network model to perform counterfactual reasoning, generating a new set of atomic intervention actions; An updated set of intervention instructions is generated based on the new set of atomic intervention actions, and S3-S5 are executed repeatedly until the differences in all dimensions are less than the tolerance threshold, at which point the iteration process is terminated.

7. A system for executing the communication scheduling method for cross-departmental closed-loop handling of urban management incidents as described in claims 1-6, characterized in that: include: The urban state field construction module divides the target urban management area into multiple grid units, defines and dynamically updates a set of quantifiable state vectors for each grid unit, and forms a unified digital state field for the entire region. The automatic intervention instruction synthesis module, for the target event and its corresponding influence grid target state vector, deduces the set of atomic intervention actions through a pre-built interpretable causal network model, and combines them to generate an intervention instruction set; The status task protocol encoding module encodes the intervention instruction set according to the predetermined status task protocol to obtain a status task protocol packet, and sends it to the corresponding external processing terminal with edge computing capability. The protocol execution feedback module, external processing terminal, autonomously plans and executes atomic intervention action sequence according to the state task protocol package, and collects the state vector of the target grid before and after execution, calculates the state change gradient and actual resource consumption, and encapsulates feedback data packets according to the protocol format and sends them back. The closed-loop adaptive iterative module incrementally updates the digital state field based on the feedback data packet, and re-triggers the interpretable causal network model to perform counterfactual reasoning and intervention instruction synthesis based on the updated differences, and iteratively issues new state task protocol packets.

Images