Cross-domain job flow collaborative scheduling and experience-driven incremental evolution system

Abstract

The application provides a cross-domain job flow collaborative arrangement and experience-driven incremental evolution system, and belongs to the technical field of equipment management; comprising: a collaborative action flow abstract extraction engine, which abstracts a multi-device collaborative job sequence into a dynamic process framework containing device roles and capability derivation rules; an MCP capability packaging middleware, which packages heterogeneous device capability actions into standardized MCP tools, the return results of which contain standard execution feedback loads for feeding back measured physical quantities; and a self-adaptive instantiation solver, which, in the execution process of the dynamic process framework, recursively calibrates effective job capability coefficients according to the measured physical quantities, and dynamically determines the termination condition of repetitive action processes based on the comparison between the accumulated job quantity and the preset job completion threshold; the application changes from open-loop fixed-time execution to closed-loop target state-driven execution, and solves the technical problems that the solidification process parameters are invalid due to equipment changes and working condition fluctuations, and that collaborative experience is difficult to be generalized and reused across scenes.

Description

Technical Field

[0001] This invention relates to the field of equipment management technology, and in particular to a cross-domain workflow collaborative orchestration and experience-driven incremental evolution system. Background Technology

[0002] In complex operational scenarios such as mining and smart logistics, various types of unmanned equipment from different manufacturers are typically required to complete collaborative tasks according to strict temporal and spatial constraints. Typical high-frequency collaborative operation scenarios include, but are not limited to: collaborative loading of unmanned trucks and excavators in open-pit mines; collaborative picking of automated guided vehicles (AGVs) and robotic arms in warehousing and logistics; and collaborative inspection of drones and unmanned trucks in mine inspections. These scenarios share the common characteristic of involving the sequential collaboration of multiple heterogeneous devices. The actions of each device must be precisely coordinated in time and space with the actions of other devices to complete a single operational task.

[0003] Currently, the following technical solutions are mainly used for the arrangement and execution of the above-mentioned multi-device collaborative operations.

[0004] The first approach is centralized scheduling based on manually pre-set process templates. In this approach, engineers use an industrial control platform or low-code orchestration tools to pre-configure the action sequences, execution order, interlocking conditions, and fixed parameters of each device for each collaborative task. When a task needs to be performed, the scheduling system issues control commands to each device sequentially according to the pre-set template. The main drawback of this approach is that the parameters in the process template (such as loading times, waiting times, and movement speeds) are all fixed values. Once the equipment model changes or the operating conditions change, the original fixed parameters are often no longer applicable, requiring manual adjustment and configuration by technicians. In production environments with frequently changing equipment combinations, this repeated manual reconfiguration is labor-intensive, slow in response, and severely restricts the flexibility and overall efficiency of collaborative operations.

[0005] The second approach involves deploying an independent intelligent control system on each piece of equipment. For example, excavators could be equipped with a material level sensing and autonomous digging decision-making module, while unmanned trucks could be equipped with a path planning and load monitoring module. Each device makes local decisions based on its own sensor inputs. While this approach improves the adaptability of individual machines to local environments, the decisions of each device are independent, lacking a unified global coordination layer to perceive the overall progress of the operation. When the efficiency of one piece of equipment fluctuates due to changes in operating conditions, other devices cannot automatically sense this and synchronously adjust their own action parameters, leading to idle periods or task backlogs, making it difficult to maintain an optimized overall work cycle.

[0006] The third approach involves introducing machine learning or reinforcement learning methods to train an end-to-end collaborative strategy model using a large amount of simulation or historical data. This model then directly outputs control commands for each device. However, this approach requires massive amounts of labeled data or interactive simulation environments, resulting in high training costs, limited generalization ability, and poor interpretability of the model's decision-making process. Consequently, it struggles to meet the stringent requirements of mining, logistics, and other fields for operational safety, auditability, and deterministic execution.

[0007] Furthermore, in the existing solutions described above, the excellent execution experience accumulated from a successful collaborative task (including reasonable parameter configuration and efficient equipment coordination) usually disappears after the task ends and cannot be structured into reusable digital assets. Even if a task achieves a relatively good collaborative effect, its experience cannot be systematically accumulated, retrieved, and reused across scenarios, resulting in the need for a large amount of manual programming or model training to be redone each time similar but not exactly the same task requirements are encountered.

[0008] Therefore, there is an urgent need for a cross-domain workflow collaborative orchestration and experience-driven incremental evolution system to solve the technical problems of low orchestration efficiency and difficulty in generalizing and reusing experience when existing multi-device collaborative workflow templates are reused across devices and working conditions due to the inability to adaptively adjust execution parameters. Summary of the Invention

[0009] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a cross-domain workflow collaborative orchestration and experience-driven incremental evolution system. It aims to abstract multi-device collaborative operation sequences into a dynamic process skeleton containing capability derivation rules, and to perform online recursive calibration of operation capability coefficients using measured physical quantities fed back by standardized protocols during execution. It also aims to dynamically terminate repetitive action processes based on a comparison of cumulative operation volume with the safety threshold of the target equipment's rated capacity. This enables the standardization and cross-scenario reuse of work processes, allowing the reused processes to automatically adapt and adjust when faced with equipment changes and operating condition fluctuations.

[0010] To achieve the above objectives, this application proposes a cross-domain job flow collaborative orchestration and experience-driven incremental evolution system, comprising: The collaborative action flow abstraction and extraction engine is used to abstract at least one successfully executed multi-device collaborative operation sequence into a dynamic process skeleton containing device roles and capability derivation rules. MCP capability encapsulation middleware is used to encapsulate the capability actions of heterogeneous devices into standardized MCP tools, wherein the return result of the MCP tool includes a standard execution feedback load for feeding back the measured physical quantity of the action. An adaptive instantiated solver, connected to the MCP capability encapsulation middleware and the collaborative action flow abstraction extraction engine, is used to perform online recursive calibration of the effective operational capability coefficients based on the measured physical quantities in the standard execution feedback load during the execution of the dynamic process skeleton; and to dynamically adjust the execution parameters of the remaining actions in the dynamic process skeleton based on the calibrated coefficients.

[0011] As a further solution, a device capability map construction module is also included, which is connected to the adaptive instantiation solver and is used to maintain a capability vector model for each physical device registered through the MCP capability encapsulation middleware; the capability vector model includes at least static capability attributes and dynamic state information.

[0012] As a further solution, the adaptive instantiation solver performs online recursive calibration of the effective workability coefficients as follows: The online recursive calibration is performed using the following formula: λ_i=(1-β)*λ_{i-1}+β*(mi / Q_ideal); Where λ_i is the effective operational capability coefficient after the i-th calibration, m_i is the measured physical quantity obtained from the standard execution feedback load in the i-th calibration, Q_ideal is the preset single ideal operational quantity, and β is the learning rate. The ideal workload Q_ideal for a single task is calculated by the first capability derivation rule generated by the collaborative action flow abstraction extraction engine, based on the static capability attributes and empirical parameters of the corresponding device in the device capability map construction module.

[0013] As a further solution, the dynamic process skeleton includes a repetitive action process with a preset total number of iterations, and the adaptive instantiated solver is also used for: The measured physical quantities after each action is executed are obtained through the MCP capability encapsulation middleware, and the cumulative workload is accumulated. The cumulative workload is continuously compared with the preset workload completion threshold. Once the cumulative workload reaches the workload completion threshold, an instruction to terminate the current repetitive action process is immediately sent to the MCP capability encapsulation middleware. The task completion threshold is set as a preset safe percentage range of the rated capacity of the target equipment affected by the repetitive action process; the preset total number of times is calculated by the second capability derivation rule based on the rated capacity of the target equipment and the initial ideal single task quantity when the dynamic process skeleton is instantiated.

[0014] As a further solution, the adaptive instantiation solver is also used to skip all remaining actions of the repetitive action process in the dynamic process skeleton and directly proceed to the next action after the repetitive action process when the cumulative workload reaches the job completion threshold but the number of times the repetitive action process has been executed has not reached the preset total number of times.

[0015] As a further solution, before the current action is executed, the adaptive instantiated solver is also used for: If the sum of the predicted cumulative workload and the estimated workload of the current action exceeds the rated capacity limit of the target equipment, the execution parameters of the current action will be automatically modified to a partial workload mode, or an overload warning signal will be generated.

[0016] As a further solution, when the collaborative action flow abstraction extraction engine generates the dynamic process skeleton, it identifies the repetitive equipment action sequence in it as a reusable substructure, and replaces the fixed number of action repetitions in the original sequence with a second capability derivation rule; the second capability derivation rule is configured to calculate the required number of action repetitions based on the ratio of the target equipment capacity to the ideal workload in a single operation when instantiated.

[0017] As a further solution, the adaptive instantiation solver is also used to calculate the job cycle balance index using the static capability attributes of multiple devices bound to the current job provided by the device capability map construction module before the dynamic process skeleton starts to execute, and to generate optimization suggestions for the current device allocation based on the job cycle balance index. The work cycle balance index TBI is calculated as follows: TBI = (N_Auxiliary Equipment * CT_Main Equipment) / (N_Main Equipment * CT_Auxiliary Equipment); where CT_Main Equipment is the single work cycle time of the main equipment, CT_Auxiliary Equipment is the single transport cycle time of the auxiliary equipment, N_Main Equipment is the number of main equipment bound in the current work, and N_Auxiliary Equipment is the number of auxiliary equipment bound in the current work.

[0018] As a further solution, the standard execution feedback payload and the return result of the MCP tool are the same data packet, the content of which is based on the JSONS chema definition of the MCP tool and includes at least one of the measured physical quantity fields, including the actual payload, operation duration or energy consumption of this action.

[0019] As a further solution, it also includes a natural language orchestration and scheduling interface and a semantic retrieval engine; the semantic retrieval engine is used to retrieve the most matching combined capability component containing the dynamic process skeleton from the capability market based on the natural language instructions input by the user; the natural language orchestration and scheduling interface is used to fill the parsed device entity into the device role, complete the component instantiation, and trigger the adaptive instantiation solver to work.

[0020] Compared with related technologies, the cross-domain job flow collaborative orchestration and experience-driven incremental evolution system provided by this invention has the following advantages: 1. This invention abstracts multi-device collaborative operation sequences into a dynamic process skeleton that includes device roles and capability derivation rules. During instantiation, the parameters are evaluated based on the static capability attributes of the actual bound devices, enabling the same process skeleton to transparently adapt to combinations of devices of different brands and models, thus achieving true generalization and reuse of collaborative operation experience.

[0021] 2. This invention solves the problems of underload and overload caused by material characteristic fluctuations and equipment performance changes by presetting a work completion threshold based on the rated capacity of the target equipment in the process skeleton, and using the measured physical quantities returned by the MCP tool to compare the cumulative work volume in real time and dynamically determine the termination during the execution process. This transforms the control logic of the collaborative process from open-loop fixed-number execution to closed-loop target state driven execution.

[0022] 3. By introducing an online recursive calibration mechanism for effective operational capability coefficients, this invention enables the system to continuously perceive the degree of deviation of the current working condition from the ideal benchmark during execution, and adjust the prediction and planning of subsequent actions in real time accordingly, thereby achieving autonomous adaptability to dynamic working environments.

[0023] 4. The entire closed-loop adaptive mechanism of this invention is built on the MCP standardized protocol layer. It uses the enhanced standard execution feedback payload in the tool return results to achieve data closure without introducing additional bypass data channels or heavyweight intelligent inference modules. It has good engineering feasibility and cross-vendor compatibility. Attached Figure Description

[0024] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0025] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0026] Figure 1 This invention provides a schematic diagram of the steps of a cross-domain job flow collaborative orchestration and experience-driven incremental evolution system. The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0028] Please see Figure 1 This embodiment provides a cross-domain job flow collaborative orchestration and experience-driven incremental evolution system. The system mainly includes: a collaborative action flow abstraction and extraction engine, an MCP capability encapsulation middleware, a device capability graph construction module, an adaptive instantiation solver, a semantic retrieval engine, a natural language orchestration and scheduling interface, and a capability marketplace.

[0029] The collaborative action flow abstraction and extraction engine is used to abstract at least one successfully executed multi-device collaborative operation sequence into a dynamic process skeleton containing device roles and capability derivation rules. MCP capability encapsulation middleware is used to encapsulate the capability actions of heterogeneous devices into standardized MCP tools, wherein the return result of the MCP tool includes a standard execution feedback load for feeding back the measured physical quantity of the action. An adaptive instantiated solver, connected to the MCP capability encapsulation middleware and the collaborative action flow abstraction extraction engine, is used to perform online recursive calibration of the effective operational capability coefficients based on the measured physical quantities in the standard execution feedback load during the execution of the dynamic process skeleton; and to dynamically adjust the execution parameters of the remaining actions in the dynamic process skeleton based on the calibrated coefficients.

[0030] The collaborative action flow abstraction engine is used to abstract at least one successfully executed multi-device collaborative task sequence into a dynamic process skeleton containing device roles and capability inference rules. Each device role in the dynamic process skeleton represents a type of device performing a specific function (such as "loading device" or "transportation device"), rather than being bound to a specific device instance. Capability inference rules are used to calculate specific execution parameters based on the actual bound device attributes during instantiation, thereby replacing fixed empirical values.

[0031] MCP Capability Encapsulation Middleware is used to encapsulate the capabilities and actions of heterogeneous devices into standardized MCP tools.

[0032] Specifically, the "slewing," "digging," and "unloading" actions of excavators, the "parking" and "departure" actions of unmanned trucks, the "takeoff," "waypoint flight," and "hovering and shooting" actions of drones, the "moving to shelf" and "lifting" actions of AGVs, and the "grabbing" and "placing" actions of robotic arms are all encapsulated into independent MCP tools. Each MCP tool includes a name, description, input parameter JSON schema, and output JSON schema. The standard execution feedback payload defined in the output JSON schema is used to return the measured physical quantities of the action after the tool has completed execution.

[0033] The equipment capability map construction module connects to the adaptive instantiated solver to maintain a capability vector model for each physical device registered through the MCP capability encapsulation middleware. The capability vector model contains at least static capability attributes and dynamic state information. Static capability attributes refer to the physical capability boundary parameters of the equipment as specified at the factory or calibrated, such as the bucket capacity and maximum digging radius of an excavator, the rated load and cargo box volume of an unmanned truck, and the lifting height of an AGV. Dynamic state information refers to variable state data such as the equipment's real-time position, current speed, fuel level, or battery level.

[0034] The adaptive instantiation solver is the core decision-making module of this system, connected to the MCP capability encapsulation middleware, the collaborative action flow abstraction and extraction engine, and the equipment capability map construction module, respectively. During the execution of the dynamic process skeleton, it performs three functions: First, before execution begins, it uses the static capability attributes of the currently bound equipment in the equipment capability map to calculate the preset total number of repetitive action processes using the second capability derivation rule, and calculates the ideal workload per cycle using the first capability derivation rule; Second, during execution, it performs online recursive calibration of the effective work capability coefficients based on the measured physical quantities in the standard execution feedback load; Third, based on the calibrated coefficients and cumulative workload, it dynamically determines whether to terminate the current repetitive action process and dynamically adjusts the execution parameters of the remaining actions.

[0035] The semantic retrieval engine and the natural language orchestration and scheduling interface constitute the human-computer interaction layer. The semantic retrieval engine retrieves the most matching combined capability components from the capability marketplace based on user-input natural language commands. The natural language orchestration and scheduling interface populates the parsed device entities into device roles, completes component instantiation, and triggers the adaptive instantiation solver. The capability marketplace stores, retrieves, and versions the combined capability components.

[0036] The present invention will be described in detail below with reference to specific embodiments: Example 1 This embodiment takes the most typical open-pit mine scenario of unmanned trucks and excavators cooperating in loading as an example to explain in detail the implementation of the system of the present invention.

[0037] (I) MCP tool packaging and device capability map construction In the initial stage of system deployment, the MCP capability encapsulation middleware is used to encapsulate the capabilities and actions of various equipment involved in the operation site into standardized MCP tools. Taking excavators and unmanned trucks as examples: The MCP tools corresponding to excavators include "Excavator.Slew", "Excavator.Digging", and "Excavator.Unloading". The input parameter schema for the "Excavator.Unloading" tool is defined as {"target_truck_id":"string","bucket_speed":"float"}, representing the target truck identifier and unloading speed, respectively. Its output schema includes execution status fields and standard execution feedback payloads, defining the actual_payload field (the actual unloading payload in tons) and the cycle_time field (the actual cycle time in seconds). Optionally, it may also include an energy field (the energy consumption of this action in kWh).

[0038] The MCP tools corresponding to unmanned trucks include "Truck.Parking", "Truck.Departure", and "Truck.Cargo Box Status Monitoring". Among them, the input parameter schema for the "Truck.Parking" tool is defined as {"loading_point_coord":"object","approach_speed":"float"}.

[0039] Meanwhile, the equipment capability map construction module establishes a capability vector model for each physical device registered through the MCP capability encapsulation middleware. For example, the capability vector model established for excavator E02 includes: static capability attributes BucketCap=4.5m³ (bucket capacity) and CT_exc_norm=35s (rated single cycle time); dynamic status information includes its real-time position coordinates and current fuel level. The capability vector model established for unmanned truck T10 includes: static capability attributes Payload=60 tons (rated load capacity) and CT_truck_norm=900s (rated transport cycle time).

[0040] (II) Generation of Dynamic Process Skeleton and Capability Derivation Rules After a successful collaborative loading operation is recorded, the collaborative action flow abstraction extraction engine analyzes the action sequence of the operation. The sequence includes the following key actions: excavator E02 turns to the loading point, unmanned truck T10 parks in the loading position, excavator performs a 3-bucket loading action, and T10 drives away.

[0041] The collaborative action flow abstraction and extraction engine first transforms the action sequence into a directed acyclic graph model and identifies reusable substructures from the graph. In this embodiment, the core reusable substructure is the "loading loop," which is the process by which an excavator repeatedly performs unloading actions on a truck until loading is complete.

[0042] For this reusable substructure, the collaborative action flow abstraction extraction engine parameterizes and extracts the repetitive equipment action sequences, extracting the equipment roles "loading equipment" (originally corresponding to excavators) and "transportation equipment" (originally corresponding to unmanned trucks), as well as dynamic variables such as loading point coordinates and speed curves.

[0043] The key processing step involves replacing the fixed action repetition count "3 buckets" in the original sequence with a second capability derivation rule. This second capability derivation rule is configured to calculate the required number of action repetitions based on the ratio of the target equipment capacity to the ideal workload per cycle during instantiation. Specifically, the expression for the second capability derivation rule is N = ceil(Target_Payload / Q_ideal), where Target_Payload is the rated load of the transport equipment, and Q_ideal is the ideal workload per cycle. Here, Q_ideal is a parameter slot during the process skeleton generation stage, and its specific value will be calculated by the first capability derivation rule based on the actual bound equipment attributes during the execution stage.

[0044] Meanwhile, when generating the dynamic process skeleton, a task completion threshold is preset for the loading cycle. This task completion threshold is set as a preset safe percentage range of the target equipment's rated capacity; in this embodiment, it is set to 95%, meaning that loading is considered complete when the cumulative loading reaches 95% of the transport equipment's rated load capacity.

[0045] The aforementioned process framework, along with equipment roles, first capability derivation rules, second capability derivation rules, and job completion thresholds, are packaged together to generate a combined capability component description file, which is then stored in the capability marketplace. The initial version is marked as "Standard Loading Job v1.0.0".

[0046] (III) Component instantiation and parameter evaluation before execution When the on-site dispatcher inputs the command "Use the E02 excavator at shovel position 3 to perform the standard loading procedure for T10 and T11 trucks" through the natural language orchestration and dispatch interface, the system executes the following processing steps.

[0047] First, the natural language orchestration and scheduling interface sends the instruction to the semantic retrieval engine. The semantic retrieval engine performs intent parsing and semantic vector encoding on the instruction, retrieves the most matching combined capability component "Standard Loading Operation v1.0.0" in the capability market, and extracts dynamic parameters: the actual coordinate value of the loading point coordinates corresponding to "Shovel Position No. 3", the equipment role "Loading Equipment" is bound to the equipment instance "E02", and the equipment role "Transportation Equipment" is bound to the equipment instance list [T10,T11].

[0048] After the component is instantiated, the adaptive instantiation solver is triggered.

[0049] The adaptive instantiation solver first performs pre-execution parameter evaluation before the dynamic process skeleton begins execution. The specific steps are as follows: The first step is to query the static capability attributes of the currently bound equipment from the equipment capability map construction module. Querying C(E02) yields BucketCap=4.5m³ and CT_exc_norm=35s; querying C(T10) and C(T11) yields Payload=60 tons and CT_truck_norm=900s.

[0050] The second step is to calculate the ideal workload Q_ideal for a single operation using the first capability derivation rule. This rule utilizes the static capability attributes and empirical parameters of the corresponding equipment in the equipment capability map construction module. In this embodiment, the calculation formula is Q_ideal = BucketCap × FillFactor_norm × MaterialDensity_norm, where FillFactor_norm is the empirical value of the full bucket coefficient (taken as 0.92), and MaterialDensity_norm is the empirical value of the material density (taken as 2.5 tons / m³). Substituting these values, we get Q_ideal = 4.5 × 0.92 × 2.5 = 10.35 tons.

[0051] The third step is to calculate the preset total number of repetitive actions using the second capacity derivation rule. This second capacity derivation rule is calculated based on the target equipment's rated capacity and the initial ideal workload per cycle during the instantiation of the dynamic process skeleton. In this embodiment, N = ceil(60 / 10.35) = ceil(5.80) = 6 buckets. This preset total number of cycles serves as the initial planned value for the loading cycle.

[0052] The fourth step is to calculate the work cycle balance index (TBI) using the static capability attributes of multiple devices bound to the current job, provided by the equipment capability map construction module. With two trucks (N_auxiliary equipment=2) and one excavator (N_main equipment=1) currently bound, TBI = (2×35) / (1×900)≈0.078. This TBI value is significantly less than 0.8. Based on this, the adaptive instantiation solver generates optimization suggestions for the current equipment allocation and prompts the user through the natural language orchestration and scheduling interface: "The current two trucks will result in a large amount of excavator idle time; it is recommended to add more trucks to meet the cycle balance requirement." After user confirmation, four trucks, T12 to T15, are added, and the transport equipment list is updated to [T10, T11, T12, T13, T14, T15].

[0053] (iv) Online recursive calibration and dynamic termination during execution After completing the parameter evaluation before execution and obtaining the user's confirmation of the equipment allocation, the adaptive instantiated solver begins execution driven by the dynamic process skeleton.

[0054] The loading cycle is a repetitive process with a preset total number of iterations (6). For the first transport device T10, the adaptive instantiated solver sequentially calls tools such as "excavator.slew" and "truck.park" through the MCP capability encapsulation middleware to complete the loading preparation, and then enters the loading cycle.

[0055] First load: Invoke the "Excavator.Unload" tool, targeting a T10 truck. After the tool completes its execution, the measured physical quantities are returned via standard execution feedback: actual_payload = 8.2 tons, cycle_time = 37.2 seconds.

[0056] Upon receiving this feedback, the adaptive instantiation solver performs online recursive calibration of the effective operational capability coefficients. The following formula is used: λ_1=(1-β)×λ_0+β×(m_1 / Q_ideal) Where λ_0 is initially set to 1.0, m_1 = 8.2, Q_ideal = 10.35, and the learning rate β is preset to 0.4 based on material variability. Substituting these values ​​into the calculation: λ_1=(1-0.4)×1.0+0.4×(8.2 / 10.35)=0.6+0.317=0.917 At this point, the cumulative workload M_total = 8.2 tons. The threshold for completing the task is 95% of the target equipment's rated capacity of 60 tons, which is 57 tons. Since M_total = 8.2 has not reached 57 tons, execution continues.

[0057] Second load: Feedback actual_payload = 9.1 tons. Update λ_2 = (1-0.4)×0.917 + 0.4×(9.1 / 10.35) = 0.55 + 0.352 = 0.902. Cumulative workload M_total = 8.2 + 9.1 = 17.3 tons, not reaching the threshold, continue.

[0058] Third load: Feedback actual_payload = 9.8 tons. Update λ_3 = (1-0.4)×0.902 + 0.4×(9.8 / 10.35) = 0.541 + 0.379 = 0.920. Cumulative workload M_total = 27.1 tons, not reaching the threshold, continue.

[0059] 4th bucket: Due to entering a denser area of ​​the stockpile, actual_payload = 12.5 tons. Update λ_4 = (1-0.4)×0.920 + 0.4×(12.5 / 10.35) = 0.552 + 0.483 = 1.035. Cumulative workload M_total = 39.6 tons, below the threshold, continue.

[0060] 5th bucket: Feedback actual_payload = 13.1 tons. Update λ_5 = (1-0.4)×1.035 + 0.4×(13.1 / 10.35) = 0.621 + 0.506 = 1.127. Cumulative workload M_total = 52.7 tons, below the threshold.

[0061] At this point, the adaptive instantiated solver also dynamically adjusts the execution parameters of the remaining actions based on the calibrated effective working capacity coefficient λ_5 = 1.127. For example, the estimated value of the actual load per bucket is updated according to the calibrated coefficient as λ_5 × Q_ideal ≈ 1.127 × 10.35 ≈ 11.66 tons, and the remaining number of buckets required to complete the loading is re-predicted accordingly, so as to dynamically adjust the loading cycle and the timing of subsequent truck parking.

[0062] Before the sixth bucket was executed, the adaptive instantiated solver made an overload prediction: the current cumulative workload was 52.7 tons, and the estimated workload for the sixth bucket was approximately 11.66 tons, for a total estimated value of 52.7 + 11.66 = 64.36 tons, exceeding the target equipment's rated capacity limit of 61.2 tons (60 tons × 102%). Based on the overload prediction mechanism, the adaptive instantiated solver automatically modified the execution parameters for the sixth bucket to a partial workload mode, unloading only approximately 4.3 tons of material.

[0063] Bucket 6: Unloading was performed according to the partial workload mode, with an actual unloading of 4.3 tons. The cumulative workload M_total = 52.7 + 4.3 = 57.0 tons, reaching the work completion threshold of 57 tons.

[0064] At this point, the adaptive instantiated solver detects that the cumulative workload has reached the job completion threshold and immediately issues a command to the MCP capability encapsulation middleware to terminate the current repetitive action process. In this embodiment, the number of executions is 6, which is exactly equal to the preset total number of executions of 6.

[0065] In another possible execution scenario, if the cumulative workload reaches the 57-ton completion threshold after the 5th bucket, the adaptive instantiated solver skips all remaining action nodes of the repetitive action process when the cumulative workload reaches the completion threshold but the number of executions of the repetitive action process has not reached the preset total number. It then directly proceeds to the next action in the dynamic process skeleton following the repetitive action process, i.e., issuing a "drive away" command to the truck. This demonstrates the core advantage of the dynamic termination mechanism of this invention compared to the fixed-number scheme.

[0066] After loading T10 is completed, the adaptive instantiated solver continues to execute the same loading process for trucks T11, T12, etc., in the order of the transport equipment list. Throughout the operation, if the single cycle time is detected to deviate continuously from the rated value, the adaptive instantiated solver can also update the input parameters of the operation cycle balance index accordingly, and assess in real time whether further adjustments to the equipment allocation are needed.

[0067] (v) Experience accumulation and component version iteration Once all loading tasks are completed, the execution log for this operation is fully recorded, including the actual load of each bucket, cycle time, and the trajectory of changes in the effective working capacity coefficient. This data can be stored as an execution profile of the component.

[0068] If process developers analyze the execution profile and find that the initial learning rate β=0.4 is insufficient for response speed in scenarios with drastic changes in material properties, the learning rate can be adjusted to 0.5, and the full bucket coefficient experience value in the first capability derivation rule can be updated. After modification, a new version, "Standard Loading Operation v1.1.0," will be released through the capability marketplace. The new version will be available after sandbox verification, and mining companies subscribing to this component can choose to automatically update to the new version, allowing collaborative operation experience to continuously evolve through ongoing use.

[0069] Example 2 This embodiment illustrates the application of the present invention in a collaborative picking scenario of AGVs and robotic arms in warehousing and logistics.

[0070] In this scenario, the MCP capability encapsulation middleware encapsulates the AGV's "move to shelf" and "lift" actions, and the robotic arm's "grab" and "place" actions, into standardized MCP tools. The equipment capability map construction module establishes a capability vector model for each AGV, including static capability attributes such as rated lifting weight and operating speed, and establishes a capability vector model for each robotic arm, including static capability attributes such as rated gripping weight and working radius.

[0071] The collaborative motion flow abstraction engine extracts a dynamic process skeleton from a successful manual teaching or simulated picking process, including the equipment roles of "AGV" (Automated Guided Vehicle) and "grabbing device" (robotic arm), and replaces the fixed number of grabs with a second capability derivation rule. This second capability derivation rule dynamically calculates the required number of grabs based on the ratio of the target bin capacity to the amount of a single grab operation.

[0072] During instantiation, the adaptive instantiation solver first calculates the ideal grasping capacity per operation based on empirical parameters such as the robotic arm's rated grasping weight and product density using a first capability derivation rule. Then, it calculates the preset total number of grasps using a second capability derivation rule. During execution, the effective operational capacity coefficient is recursively calibrated online based on the actual grasping weight reported by the MCP tool after each grasp. Furthermore, the picking completion condition is dynamically determined based on a comparison between the cumulative grasping volume and the target bin capacity. When the cumulative grasping volume reaches the preset operation completion threshold, the picking cycle is immediately terminated, even if the preset total number of grasps has not been reached, and the process proceeds to the next action.

[0073] Warehouse operators only need to use natural language commands, such as "use the A1 robotic arm and the two AGVs G3 and G5 to pick the Y item on the X shelf to port 3", and the system can automatically complete the entire process of component retrieval, equipment binding, parameter evaluation and closed-loop execution.

[0074] Example 3 This embodiment illustrates the application of the present invention in the scenario of collaborative inspection of unmanned trucks and drones in open-pit mines.

[0075] In this scenario, the MCP capability encapsulation middleware encapsulates the drone's "takeoff," "waypoint flight," and "hovering shooting" actions, as well as the unmanned truck's "follow mode" and "speed setting" actions, into standardized MCP tools. The collaborative action flow abstraction extraction engine abstracts a successful "truck following drone slope inspection" operation sequence into a dynamic process skeleton, where the equipment roles are "inspection equipment" (drone) and "following equipment" (truck), and the dynamic parameters include shooting point coordinates, following distance, etc.

[0076] In this dynamic workflow framework, the repetitive action involves the drone hovering and taking pictures sequentially at multiple preset waypoints. The second capability derivation rule is configured to calculate the number of executable waypoints based on the drone's endurance and the estimated time taken for each waypoint during instantiation. The task completion threshold is set when all preset waypoints have been photographed. Measured physical quantities may include the deviation between the actual shooting position and the preset position for each waypoint.

[0077] The adaptive instantiation solver calculates a reasonable segmentation scheme for the inspection route during instantiation based on static capability attributes such as the drone's endurance and the unmanned truck's maximum following distance in the equipment capability map. During execution, it uses the actual location and battery data fed back by the MCP tool to perform online recursive calibration of the effective operational capability coefficient (such as the ratio of actual cruising speed to rated speed), and dynamically adjusts the estimated arrival time of subsequent waypoints and the truck's following speed based on the calibrated coefficients. When the cumulative number of completed waypoints reaches the preset number of waypoints, the inspection cycle is immediately terminated, triggering a return command.

[0078] Users only need to give the command "Use the D01 drone to inspect the east slope and have the T02 truck follow and record," and the system will automatically match components and dynamically bind the equipment to complete the entire closed loop of vehicle-machine collaborative inspection.

[0079] The above are only some embodiments of this application and do not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A cross-domain workflow collaborative orchestration and experience-driven incremental evolution system, characterized in that, include: The collaborative action flow abstraction and extraction engine is used to abstract at least one successfully executed multi-device collaborative operation sequence into a dynamic process skeleton containing device roles and capability derivation rules. MCP capability encapsulation middleware is used to encapsulate the capability actions of heterogeneous devices into standardized MCP tools, wherein the return result of the MCP tool includes a standard execution feedback load for feeding back the measured physical quantity of the action. An adaptive instantiated solver, connected to the MCP capability encapsulation middleware and the collaborative action flow abstraction extraction engine, is used to perform online recursive calibration of the effective operational capability coefficients based on the measured physical quantities in the standard execution feedback load during the execution of the dynamic process skeleton; and to dynamically adjust the execution parameters of the remaining actions in the dynamic process skeleton based on the calibrated coefficients.

2. The cross-domain workflow collaborative orchestration and experience-driven incremental evolution system according to claim 1, characterized in that, It also includes a device capability map construction module, which is connected to the adaptive instantiation solver and is used to maintain a capability vector model for each physical device registered through the MCP capability encapsulation middleware; the capability vector model includes at least static capability attributes and dynamic state information.

3. The cross-domain workflow collaborative orchestration and experience-driven incremental evolution system according to claim 2, characterized in that, The adaptive instantiation solver performs online recursive calibration of the effective job capability coefficients in the following manner: The online recursive calibration is performed using the following formula: λ_i=(1-β)*λ_{i-1}+β*(mi / Q_ideal); Where λ_i is the effective operational capability coefficient after the i-th calibration, m_i is the measured physical quantity obtained from the standard execution feedback load in the i-th calibration, Q_ideal is the preset single ideal operational quantity, and β is the learning rate. The ideal workload Q_ideal for a single task is calculated by the first capability derivation rule generated by the collaborative action flow abstraction extraction engine, based on the static capability attributes and empirical parameters of the corresponding device in the device capability map construction module.

4. The cross-domain workflow collaborative orchestration and experience-driven incremental evolution system according to claim 3, characterized in that, The dynamic process skeleton includes a repetitive action process with a preset total number of iterations, and the adaptive instantiated solver is further used for: The measured physical quantities after each action is executed are obtained through the MCP capability encapsulation middleware, and the cumulative workload is accumulated. The cumulative workload is continuously compared with the preset workload completion threshold. Once the cumulative workload reaches the workload completion threshold, an instruction to terminate the current repetitive action process is immediately sent to the MCP capability encapsulation middleware. The task completion threshold is set as a preset safe percentage range of the rated capacity of the target equipment affected by the repetitive action process; the preset total number of times is calculated by the second capability derivation rule based on the rated capacity of the target equipment and the initial ideal single task quantity when the dynamic process skeleton is instantiated.

5. The cross-domain workflow collaborative orchestration and experience-driven incremental evolution system according to claim 4, characterized in that, The adaptive instantiation solver is further configured to, when the cumulative workload reaches the workload completion threshold but the number of times the repetitive action process has been executed has not reached the preset total number of times, skip all remaining actions of the repetitive action process in the dynamic process skeleton and directly proceed to the next action after the repetitive action process.

6. The cross-domain workflow collaborative orchestration and experience-driven incremental evolution system according to claim 4, characterized in that, Before the current action is executed, the adaptive instantiated solver is also used for: If the sum of the predicted cumulative workload and the estimated workload of the current action exceeds the rated capacity limit of the target equipment, the execution parameters of the current action will be automatically modified to a partial workload mode, or an overload warning signal will be generated.

7. The cross-domain workflow collaborative orchestration and experience-driven incremental evolution system according to claim 3, characterized in that, When the collaborative action flow abstraction extraction engine generates the dynamic process skeleton, it identifies the repetitive equipment action sequence in it as a reusable substructure and replaces the fixed number of action repetitions in the original sequence with a second capability derivation rule. The second capability derivation rule is configured to calculate the required number of action repetitions based on the ratio of the target equipment capacity to the ideal workload in a single operation during instantiation.

8. The cross-domain workflow collaborative orchestration and experience-driven incremental evolution system according to claim 3, characterized in that, The adaptive instantiation solver is also used to calculate the job cycle balance index using the static capability attributes of multiple devices bound to the current job provided by the device capability map construction module before the dynamic process skeleton starts to execute, and to generate optimization suggestions for the current device allocation based on the job cycle balance index. The work cycle balance index TBI is calculated as follows: TBI = (N_Auxiliary Equipment * CT_Main Equipment) / (N_Main Equipment * CT_Auxiliary Equipment); where CT_Main Equipment is the single work cycle time of the main equipment, CT_Auxiliary Equipment is the single transport cycle time of the auxiliary equipment, N_Main Equipment is the number of main equipment bound in the current work, and N_Auxiliary Equipment is the number of auxiliary equipment bound in the current work.

9. The cross-domain workflow collaborative orchestration and experience-driven incremental evolution system according to claim 1, characterized in that, The standard execution feedback payload and the return result of the MCP tool are the same data packet. Its content is based on the JSONS chema definition of the MCP tool and includes at least one of the measured physical quantity fields, which includes the actual payload, operation duration or energy consumption of this action.

10. The cross-domain workflow collaborative orchestration and experience-driven incremental evolution system according to claim 1, characterized in that, It also includes a natural language orchestration and scheduling interface and a semantic retrieval engine; the semantic retrieval engine is used to retrieve the most matching combined capability component containing the dynamic process skeleton from the capability market based on the natural language instructions input by the user; the natural language orchestration and scheduling interface is used to fill the parsed device entity into the device role, complete the component instantiation, and trigger the adaptive instantiation solver to work.

Images