Cross-domain heterogeneous system cooperative control system and method based on semantic abstraction

Abstract

This invention provides a cross-domain heterogeneous system collaborative control system and method based on semantic abstraction, comprising: receiving operational intentions; performing semantic abstraction on physical entities and their relationships to construct an operational intention ontology model, thereby decoupling tasks from devices; acquiring perception data, task data, and runtime context data, and generating state information and constraint information associated with the operational intentions through time alignment, data fusion, and semantic association processing; parsing preset semantic mapping rules based on the operational intentions, state information, and constraint information to generate an execution action sequence; encapsulating the execution action sequence into a directed semantic execution chain and distributing it to multiple cross-domain heterogeneous execution systems; receiving execution feedback data and dynamically updating the operational intention ontology model to form closed-loop control. This system enables execution systems from different domains, using different protocols, and with different control logics to work collaboratively within a unified semantic framework, improving the collaborative capabilities and dynamic adaptability of cross-domain heterogeneous systems.

Description

Technical Field

[0001] This invention relates to the field of cross-domain heterogeneous system collaborative control technology, and in particular to a cross-domain heterogeneous system collaborative control system and method based on semantic abstraction. Background Technology

[0002] In existing technologies, multi-system collaborative control is typically implemented using a layered execution architecture. Upper-layer business platforms, manufacturing execution systems, scheduling platforms, or management systems generate task requests based on production plans, process rules, energy consumption strategies, or manual input, and send these requests to the middle-layer control platform. The middle-layer control platform parses the task requests through interface adaptation, protocol conversion, rule engines, or process orchestration modules, and breaks them down into multiple sets of control instructions or parameters for different execution systems.

[0003] Taking industrial manufacturing scenarios as an example, when the upper-level system issues a task request to increase production line capacity and maintain stable operation, the middle-layer control platform can break down the task and issue it to multiple heterogeneous systems or devices: It issues conveyor speed adjustment instructions to the production line's programmable logic controller (PLC) control system, which controls the conveyor mechanism, motor start / stop, and cycle synchronization; it issues gripping frequency or work path adjustment instructions to the industrial robot control system, which performs handling, assembly, palletizing, or loading / unloading actions; it issues inspection cycle or inspection mode switching instructions to the machine vision inspection system, which performs product appearance inspection, size recognition, and defect judgment; it issues load adjustment instructions to the energy management system, which monitors workshop electricity consumption, compressed air, and equipment energy consumption; it issues temperature and humidity adjustment instructions to the environmental control system, which maintains the temperature, humidity, and cleanliness conditions of the production area; and it issues operating status synchronization instructions to the safety interlock system, which monitors the status of protective doors, emergency stop devices, and entry into dangerous areas, and triggers shutdown protection in abnormal situations.

[0004] After each control command is issued, different heterogeneous systems execute corresponding actions according to their respective control protocols, action templates, local logic programs, or drive interfaces. Examples include accelerating conveyor lines, adjusting robot cycle times, increasing detection frequency in vision systems, switching operating modes in air conditioning systems, or performing peak-shaving control in energy systems. After the actions are completed, each system feeds back equipment status, execution results, production line cycle times, detection data, energy consumption parameters, or alarm information to the intermediate control platform. The intermediate control platform aggregates and processes this feedback information and uploads it to the upper-level business platform for subsequent task adjustments, status display, production optimization, or the next round of control requests. This forms a cyclical execution process consisting of upper-level task generation, intermediate-level command distribution, lower-level heterogeneous system execution, and feedback transmission.

[0005] In existing technologies, due to the lack of a unified cross-level semantic expression and mapping mechanism, computer systems have difficulty automatically deconstructing and transforming macro-level intentions, such as optimizing production processes or performing complex environmental reconnaissance, into structured control instruction sequences that can be accurately identified by multiple heterogeneous execution systems at the underlying level, such as industrial robots, mobile terminals, and IoT sensors.

[0006] High-level intentions often contain fuzzy spatiotemporal constraints and logical dependencies, while low-level execution units can only process deterministic discrete signals. Because high-level task semantics lack a unified ontological expression mechanism, semantic distortion inevitably occurs as task instructions descend from the decision-making layer to the physical layer. For example, high-level instructions are typically fuzzy, qualitative, and goal-oriented, such as "maintain comfort," while low-level heterogeneous devices, such as PLCs, sensors, and robotic arms, can only recognize precise, quantitative, and parameterized instructions, such as "valve opening 45%" or "rotation speed 1200 rpm."

[0007] When faced with sudden interference or task changes, the subsystems cannot adapt and adjust like an organic whole. Due to the lack of global intent information, the underlying devices often can only execute preset mechanical replacement strategies, such as forcibly switching to backup mode when a fault occurs, which can easily cause secondary disasters. This makes it difficult for existing technologies to support continuous, consistent, and highly reliable collaborative execution for complex task objectives, ultimately reducing the effectiveness of collaborative control of cross-domain heterogeneous systems. Summary of the Invention

[0008] This invention provides a cross-domain heterogeneous system collaborative control system and method based on semantic abstraction. By semantically abstracting the operational intent and constructing a unified ontology model and directed semantic execution chain, execution systems from different domains, with different protocols and different control logics can perform collaborative control under a unified semantic framework, thereby improving the collaborative capability and dynamic adaptability of cross-domain heterogeneous systems.

[0009] To achieve the aforementioned objectives, the present invention provides the following technical solution: a cross-domain heterogeneous system collaborative control system based on semantic abstraction, comprising: The semantic abstraction modeling module receives operational intentions, identifies the physical entities affected by these intentions, extracts spatial topological relationships and state coupling relationships between different physical entities, and performs semantic abstraction of the operational intentions based on the physical entities and their relationships to construct an operational intention ontology model. The data acquisition and semantic instantiation module acquires input data and performs semantic instantiation processing on the input data based on a preset knowledge representation structure to generate state and constraint information associated with the operational intentions. The rule parsing and action generation module parses preset semantic mapping rules based on operational intentions, state information, and constraint information to generate a sequence of execution actions. The execution chain generation and distribution module generates directed semantic execution chains based on the execution action sequences and sends these chains to multiple controlled cross-domain heterogeneous execution systems. The feedback reception and model update module receives execution feedback data returned by multiple controlled cross-domain heterogeneous execution systems and updates the operational intention ontology model based on the execution feedback data to enable collaborative control of cross-domain heterogeneous systems.

[0010] A collaborative control method for cross-domain heterogeneous systems based on semantic abstraction, including the following: S1: Receive the operation intent, identify the physical entity affected by the operation intent, extract the spatial topological associations and state coupling relationships between different physical entities, and semantically abstract the operation intent based on the physical entities and their associations to construct an operation intent ontology model; S2: Obtain input data and perform semantic instantiation processing on the input data based on a preset knowledge representation structure to generate state information and constraint information associated with the operation intent; S3: Based on the operation intent, state information, and constraint information, parse the preset semantic mapping rules to generate an execution action sequence; S4: Generate a directed semantic execution chain based on the execution action sequence and send the directed semantic execution chain to multiple controlled cross-domain heterogeneous execution systems; S5: Receive execution feedback data returned by multiple controlled cross-domain heterogeneous execution systems and update the operation intent ontology model based on the execution feedback data to perform collaborative control of cross-domain heterogeneous systems.

[0011] The above technical solution has at least the following advantages compared with the existing technology: 1. By constructing an operational intent ontology model through semantic abstraction, complex operational intents are transformed into structured semantic representations. Input data is semantically instantiated to generate state and constraint information associated with the operational intent, giving the input data clear semantic meaning. This provides accurate information support for the subsequent generation of execution action sequences, ensuring that the execution actions conform to the operational intent and the actual system state. Then, based on the operational intent, state information, and constraint information, a preset semantic mapping rule is parsed to generate an execution action sequence, realizing the transformation from abstract semantics to concrete executable actions. This provides clear operational guidance for the execution system, enabling it to perform corresponding operations according to the operational intent. Next, a directed semantic execution chain is generated and sent to multiple controlled cross-domain heterogeneous execution systems, clarifying the sequence and dependencies between actions. This ensures that execution systems from different domains and with different structures can work collaboratively and execute actions in the correct order, achieving effective collaborative control of cross-domain heterogeneous systems. Finally, execution feedback data is received and the operational intent ontology model is updated, allowing the model to be dynamically adjusted and optimized based on actual execution conditions. This improves the model's accuracy and adaptability, providing a more reliable basis for subsequent collaborative control and realizing a closed-loop collaborative control process.

[0012] 2. After screening candidate actions, this invention further performs state evolution analysis based on execution parameters and current state data to obtain the state evolution results after action execution. This result is then compared with constraint boundary information to determine the sequence of actions to be executed. Compared to existing technologies that directly output control commands based on static rules, this invention comprehensively considers future state change trends and constraint boundaries before action execution, improving the matching degree between action decisions and actual operating states, making the control process more suitable for dynamically changing physical entity control scenarios.

[0013] 3. This invention constructs a directed semantic execution chain that includes perception source information, state information, intent information, constraint information, execution action sequence information, and association information, and sends it as a unified control carrier to multiple cross-domain heterogeneous execution systems. Compared with the existing technology of sending independent control commands to different devices, this invention can maintain the consistent transmission of control context information, enabling execution systems with different domains, different protocols, and different control logics to operate collaboratively under a unified semantic framework, thereby improving the consistency and scalability of multi-system collaborative control.

[0014] 4. When a system failure occurs, this invention enables the execution end to read the intent, state, and constraint information from the directed semantic execution chain, and accordingly determine the constraint satisfaction of alternative actions. If the conditions are met, the alternative action is executed; otherwise, degradation processing or fault feedback data is output. Compared to the existing technology's fault switching based on a fixed backup strategy, this invention enables the execution end to autonomously process based on task semantics and constraints, improving the continuous operation capability and collaborative stability of cross-domain heterogeneous systems under abnormal scenarios. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0016] Figure 1 This is a schematic diagram of the cross-domain heterogeneous system collaborative control system based on semantic abstraction according to the present invention. Figure 2 This is a flowchart of the rule parsing and action generation process of the present invention; Figure 3 This is a flowchart illustrating the execution of the action processing fault tolerance mechanism of the present invention; Figure 4 This is a 24-hour cross-domain collaborative energy consumption optimization curve of the present invention; Figure 5 This is a comparison diagram of the physical entity semantic abstraction before and after the present invention. Detailed Implementation

[0017] 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, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0018] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms “first,” “second,” and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an,” “a,” or “the,” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising,” “including,” or “including,” and similar terms mean that the element or object preceding the word encompasses the element or object listed following the word and its equivalents, without excluding other elements or objects. The terms “connected,” “linked,” or “connected,” and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect.

[0019] It should be noted that the terms "up", "down", "left", "right", "front", and "back" used in this invention are only used to indicate relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0020] This invention addresses the problem of reduced effectiveness in existing cross-domain heterogeneous system collaborative control by providing a semantically abstract-based collaborative control system and method for cross-domain heterogeneous systems that can enhance the collaborative capability and dynamic adaptability of such systems.

[0021] The cross-domain heterogeneity in this application is not limited to a specific industry scenario, but refers to the simultaneous application of the same control task to multiple execution systems with different sources, types, protocols, or controlled objects. Examples include production line equipment in industrial manufacturing systems, load regulation equipment in energy systems, air conditioning and lighting equipment in building systems, irrigation and environmental control equipment in agricultural systems, and conveying and robotic equipment in logistics systems. Although these systems all belong to execution systems, they differ significantly in object modeling methods, action expression methods, control interfaces, feedback forms, and update mechanisms, thus constituting typical cross-domain heterogeneous execution systems.

[0022] Example 1: In agricultural planting scenarios, such as digital control scenarios for greenhouses, multi-span greenhouses, plant factories, or open-field planting. In this scenario, it is necessary to coordinate multiple controlled cross-domain heterogeneous execution systems simultaneously to complete control tasks such as planting environment, irrigation behavior, water and fertilizer supply, ventilation and dehumidification, and supplemental lighting. Cross-domain heterogeneous execution systems may include, but are not limited to: irrigation control systems for controlling drip irrigation valves, water pumps, and zoned irrigation actuators; fertigation control systems for controlling fertilizer-liquid ratios, fertilizer application rates, and supply rhythms; environmental control systems for controlling fans, wet curtains, skylights, roller blinds, and heating equipment; supplemental lighting control systems for controlling the switching, brightness, or duration of supplemental lighting; and sensing and monitoring systems for collecting data such as soil moisture, temperature and humidity, light intensity, carbon dioxide concentration, and leaf temperature. For the aforementioned massive amounts of multi-source heterogeneous sensing data, big data processing technologies are introduced for data access and preprocessing before semantic abstraction. Specifically, based on distributed big data stream processing frameworks such as Flink or Spark Streaming, high-frequency time-series data streams such as temperature and humidity, light intensity, and carbon dioxide concentration are received and cleaned in real time, providing high-throughput, low-latency standardized data input for subsequent semantic abstraction.

[0023] like Figure 1 As shown, embodiments of the present invention provide a cross-domain heterogeneous system collaborative control system based on semantic abstraction, including: The semantic abstraction modeling module receives operational intentions, identifies the physical entities affected by these intentions, and extracts spatial topological relationships and state coupling relationships between different physical entities. Based on the physical entities and their relationships, it performs semantic abstraction on the operational intentions to construct an operational intention ontology model. An operational intention refers to a high-level control objective proposed by a user, scheduling system, business rule system, or upper-level decision-making system. This operational intention is not directly equivalent to the action of a specific underlying device but rather represents the task semantics oriented towards physical entities. For example, "replenish soil moisture." Semantic abstraction extracts key concepts, relationships, and attributes from specific operational intention descriptions, removes irrelevant details, and transforms them into a more general and abstract semantic representation. In this embodiment, physical entities may include: individual crops or crop groups, root zone soil, a specific environmental area within the greenhouse, an irrigation zone, a ventilation zone, a supplemental lighting zone, and a fertilizer supply unit. Therefore, the collaborative control in this invention does not involve controlling a single device in isolation but rather coordinates multiple agricultural execution systems around the state changes and control requirements of the aforementioned physical entities.

[0024] Furthermore, the operational intent ontology model includes intent nodes, state nodes, constraint nodes, and action nodes. Intent nodes are used to represent task semantics that do not include device control parameters. State nodes are used to represent the object state, environment state, or context state of a physical entity. Constraint nodes are used to represent execution constraint information related to the physical entity, which includes at least one of time constraints, resource constraints, security constraints, or performance constraints. Action nodes are used to represent action description information associated with the execution object corresponding to the physical entity. Intent nodes, state nodes, constraint nodes, and action nodes establish semantic relationships through directed association edges, which represent causal relationships, dependency relationships, or constraint relationships between nodes.

[0025] For example, the planting management platform issues an operational intent: "Suppress the risk of gray mold during the flowering and fruit setting period of tomatoes and avoid adverse effects on fruit setting stability." The semantic abstraction modeling module first performs semantic abstraction on this operational intent, forming intent node 1 for suppressing disease risk and intent node 2 for maintaining fruit setting stability; then it identifies the physical entities related to this operational intent, including: tomato crop population, canopy air environment inside the greenhouse, ventilation regulation area, and water and fertilizer supply area.

[0026] In this embodiment, intent nodes can represent suppressing disease risk, replenishing root zone moisture, and maintaining a suitable microclimate; state nodes can represent high humidity inside the greenhouse, increased risk of leaf dew formation, insufficient root zone moisture, and the crop being in the flowering and fruiting stage; constraint nodes can represent prohibiting sudden drops in greenhouse temperature, ensuring that the amount of irrigation per irrigation does not exceed a set threshold, and avoiding the risk of flower and fruit drop; action nodes can represent slightly opening the top ventilation, low-speed fan disturbance, zoned drip irrigation, and low-concentration water and fertilizer supply; semantic relationships are established between nodes through directed association edges. For example, high humidity inside the greenhouse can be associated with an increased risk of gray mold, which can be associated with suppressing disease risk; the flowering and fruiting stage can be associated with avoiding drastic temperature fluctuations; and suppressing disease risk can be associated with slightly opening the top ventilation / low-speed fan disturbance. Unstructured, colloquial natural language instructions are transformed into a structured, clean set of key information, providing a high-quality data source for subsequent graph construction; a high-level abstract expression of the task is achieved, allowing the same task intent to be adapted to multiple different execution schemes.

[0027] Specifically, the construction and training of the operational intent ontology model is a systematic project from knowledge definition to intelligent extraction and then to safety fine-tuning. First, based on the knowledge of agricultural experts, a standardized rule skeleton containing four types of nodes (intent, state, constraint, and action) and directed related edges is defined. Second, a high-quality supervised dataset containing entities and relationships is constructed by manually annotating a large amount of agricultural operation text. Then, a pre-trained large language model and graph neural network are used for joint training, giving the system the ability to automatically parse and generate structured topological graphs from arbitrarily complex natural language. Finally, a reinforcement learning mechanism is introduced for logical fine-tuning, and agricultural constraints are added as a penalty to the reward function, so that the model can automatically identify and cut off the associated paths that violate the safety bottom line, such as the sudden drop in temperature during flowering, at the graph construction stage, thereby completing logical self-consistency and risk filtering at the source of semantic abstraction.

[0028] The data acquisition and semantic instantiation module is used to acquire input data and perform semantic instantiation processing on the input data based on a preset knowledge representation structure to generate state information and constraint information associated with the operation intention.

[0029] Furthermore, the input data includes sensing data, task data, and operational context data. Sensing data is collected by agricultural sensors and monitoring equipment, such as soil moisture, soil conductivity, greenhouse temperature, greenhouse relative humidity, light intensity, carbon dioxide concentration, leaf temperature, crop canopy humidity, water pump operating status, and valve opening / closing status. Task data is input by users or the scheduling system, such as the daily irrigation plan, disease early warning tasks, water-saving control strategies, high-temperature period control instructions, and growth stage management tasks. Operational context data reflects the current planting background, such as the current crop variety, current growth stage including seedling stage, flowering stage, fruit setting stage, and fruit expansion stage, historical irrigation records, historical disease occurrence records, equipment maintenance status, current weather forecast information, electricity price information, and water supply limit.

[0030] The input data is semantically instantiated based on a pre-defined knowledge representation structure, including: The predefined knowledge representation structure refers to a predefined framework for transforming unstructured / semi-structured input data into structured graph nodes. This framework specifies standard description paradigms for the types, attributes, and relationships of each type of data. Logically, this structure defines the field elements that must be included to generate a standard node, including: node type identifiers to distinguish between state nodes and constraint nodes; entity attribute sets such as numerical values, units, and timestamps; semantic tags such as condensation risk and sensitive periods; and association pointer interfaces for connecting other nodes.

[0031] The system performs correlation and organization processing on perceived data, task data, and operational context data. This correlation and organization processing includes time alignment, data fusion, and semantic correlation processing. Time alignment is specifically achieved by using a time window function to forcibly truncate and align perceived data, task data, and operational context data to the same timestamp; for example, unifying all data to 14:30:00 on April 14, 2026. Data fusion is specifically achieved by activating a rule engine for cross-validation, combining soil moisture of 35% and a no-rainfall forecast to determine the need for immediate irrigation. Semantic correlation is specifically achieved by combining input data, such as relative humidity inside the greenhouse exceeding a set threshold, nighttime temperature difference decreasing, leaf surface temperature approaching the dew point, and crops being in the flowering and fruiting stage. This eliminates the spatiotemporal asynchrony errors of multi-source data, enabling a comprehensive and accurate understanding of the current greenhouse situation by integrating multi-dimensional information such as weather, soil, and tasks.

[0032] Based on the preset knowledge representation structure, the associated data is formatted and encapsulated according to the above description paradigm. Based on the preset knowledge representation structure, the associated data is mapped to state nodes and constraint nodes. Based on the preset knowledge representation structure, the above data can be instantiated as: state node A with increased risk of dew formation, state node B with high risk of gray mold, and state node C with the crop in a sensitive growth stage. At the same time, based on the context and task data, it can be instantiated as: constraint node A prohibiting strong ventilation actions that cause a sharp drop in greenhouse temperature, and constraint node B avoiding affecting pollination and fruit setting stability.

[0033] Establish associations between state nodes, constraint nodes, and intent nodes to form a node association structure in the operation intent ontology model.

[0034] like Figure 2 The flowchart shown illustrates the rule parsing and action generation process. The rule parsing and action generation module is used to parse preset semantic mapping rules based on operation intent, state information, and constraint information to generate a sequence of execution actions.

[0035] Furthermore, based on the operational intent, state information, and constraint information, the preset semantic mapping rules are parsed to generate a sequence of execution actions, including: The pre-defined semantic mapping rules are solidified at the system level into a basic association graph, namely a pattern graph, which includes states, intentions, and actions. The nodes of this graph contain abstract state classes, intention classes, and action classes, and the edges contain semantic relationships such as trigger and applicable.

[0036] The intent nodes, state nodes, and constraint nodes are assembled into semantic parsing input; the semantic parsing input containing specific intent nodes, state nodes, and constraint nodes is used as a node set, and the connection relationships in the operation intent ontology model are extracted as an edge set to construct a query subgraph of the current scene in memory.

[0037] This method employs a subgraph isomorphic matching algorithm and a pattern matching inference mechanism. Based on semantic parsing input, it matches according to preset semantic mapping rules to generate a candidate action set. Specifically, it traverses the basic association graph, searching for subgraphs that are topologically isomorphic to the query subgraph (i.e., have consistent connection logic) and whose corresponding node feature vector similarity is higher than a preset threshold (e.g., cosine similarity greater than 0.85). The action-class leaf nodes from all successfully matched subgraphs are extracted, deduplicated, and the candidate action set is generated. Examples include: candidate action 1 for high-power exhaust dehumidification, candidate action 2 for top micro-ventilation, candidate action 3 for low-speed fan disturbance, candidate action 4 for initiating spray dust suppression, and candidate action 5 for spraying pesticides. The candidate action set is then filtered to determine the sequence of actions to be executed. This method achieves explicit enumeration and decoupling of the control strategy search space, breaking away from the traditional single PID control logic based on state feedback. Through graph matching algorithms, it enables parallel strategy activation across devices and systems, providing a complete initial feasible region for global optimization.

[0038] Furthermore, the candidate action set is filtered to determine the sequence of actions to be executed, including: Based on the association between candidate actions and constraint nodes, the candidate actions in the candidate action set are retrieved. The specific association filtering logic includes the following steps: In the global knowledge graph, starting from each candidate action node, a breadth-first search of the directed graph is performed to check whether there is a directed path pointing to the current constraint node within a preset number of hops, such as 3 hops or less. If the path is unreachable, the constraint association information is determined to be conflict-free, and the candidate action is retained. If the path is reachable, the following hierarchical conflict determination process is executed: First, extract the logical predicates of all connections on the path, such as triggering and causing. At the same time, a first-order logical conflict determination rule is built in: When the predicate sequence on the path and the constraint node's restriction attributes, such as prohibition and not exceeding, are logically contradictory, for example, if the path logic is an action... If enabling the action causes a temperature increase, and the constraint attribute prohibits temperature increases, then a causal path that triggers a violation is identified. If the logical contradiction is confirmed, it is directly classified as a positive conflict, and the candidate action is removed. In the second step, if the predicate sequence on the path does not constitute a direct logical contradiction with the constraint attribute of the constraint node, the influence weight and distance decay parameter of each associated edge on the path are read. The influence weight and distance decay parameter are obtained from historical execution feedback data, simulation samples, or a preset knowledge base. Based on the influence weight of each associated edge, the transmission level of the edge, and the risk boundary corresponding to the constraint node, the influence score of the candidate action relative to the current constraint node is calculated. When the influence score exceeds the conflict warning threshold corresponding to the constraint node, the candidate action is determined not to meet the current constraint conditions and is removed from the candidate action set.

[0039] For example: Candidate action 1, although beneficial for rapid dehumidification, may cause a sudden drop in greenhouse temperature, which is incompatible with the constraint of avoiding drastic temperature fluctuations; Candidate action 5 may be incompatible with the constraint of avoiding affecting pollination during the flowering and fruit setting period; Candidate action 2 is basically compatible with the constraints; Candidate action 3 is basically compatible with the constraints; Thus, the constraint association information corresponding to each candidate action is determined, and the candidate action set is filtered according to the constraint association information to retain the candidate actions that are more suitable for the current scenario; The constraint association information is defined as the algebraic or logical conflict measure between the control primitive and the boundary of the system constraint manifold. In the graph structure, it is represented by the causal path weight that triggers the violation between the action node and the constraint node.

[0040] Furthermore, after filtering the candidate action set based on constraint association information, the process also includes: The execution parameters corresponding to the selected candidate actions are extracted. State evolution analysis is performed based on the execution parameters and the state data corresponding to the current state node to obtain the state evolution results. The state evolution results are compared with the constraint boundary information corresponding to the constraint nodes, and the sequence of execution actions is determined based on the comparison results. State evolution analysis is defined as a forward temporal digital twin extrapolation based on the controlled object's mechanism model, such as the partial differential equations of greenhouse microclimate thermodynamics and fluid dynamics. It predicts the system's state trajectory on future discrete time scales by integrating and solving in the virtual state space. The state evolution results are defined as a multidimensional temporal state matrix or state curve cluster output by the forward temporal digital twin extrapolation. In terms of specific data structure, it includes a timestamp sequence, a set of predicted state variable values, and a rate of change characteristic value. The timestamp sequence specifically represents: a set of future discrete time nodes set in the extrapolation, such as 5 minutes, 10 minutes, and 20 minutes; and a set of predicted state variable values... The specific manifestations of the combination are: at each time stamp, the predicted numerical vector of key indicators of the controlled physical entity, such as absolute humidity inside the greenhouse, greenhouse temperature, relative humidity of leaves, carbon dioxide concentration, etc.; the specific manifestation of the rate of change characteristic value is: the derivative or difference of the predicted values ​​of state variables between adjacent time stamps, such as cooling rate and dehumidification rate, used to characterize the severity of environmental changes; comparing the state evolution results with the constraint boundary information, which is essentially substituting the set of predicted values ​​of state variables and the rate of change characteristic value in the above state evolution results into the safety interval inequalities pre-set by the constraint nodes, such as the predicted cooling rate being less than the maximum temperature drop threshold and the cooling rate being less than 1.5℃ / hour, to solve whether the system of inequalities holds, thereby determining the final safe execution action sequence; the execution action sequence is defined as a discrete optimal control command vector sequence with strict temporal logic, continuous time index and determined control increment generated after constraint feasibility verification and multi-objective optimization.

[0041] For example, for the candidate actions 2 and 3 retained after screening, the corresponding execution parameters are extracted, such as: ventilation opening angle, fan speed, duration, and action execution sequence. Then, a state evolution analysis is performed in conjunction with the current state data, such as analyzing the trends of humidity change, temperature change, leaf condensation risk change, and crop pollination environment stability change within the next 5, 10, and 20 minutes. It is assumed that the analysis shows that top micro-ventilation can reduce humidity by 5% and greenhouse temperature by 0.5℃ within 10 minutes, and low-speed fan disturbance can reduce the risk of local condensation, but has limited improvement on overall humidity. The combined action can reduce humidity by 8% and greenhouse temperature by 1.2℃, which is still within the allowable range. The above state evolution results are then compared with the constraint boundary information, such as: the greenhouse temperature drop must not exceed 2℃, the relative humidity fluctuation during the flowering and fruit setting period is within the allowable range, and continuous low-temperature stimulation is prohibited. Based on the comparison results, the final action sequence is determined to be to first execute top micro-ventilation, and then execute low-speed fan disturbance.

[0042] The execution chain generation and distribution module is used to generate directed semantic execution chains based on the sequence of execution actions by using knowledge graph serialization encapsulation technology, such as converting to JSON-LD or RDF standard data stream format, combined with the depth-first traversal algorithm of graph structure, and sending the directed semantic execution chains to multiple controlled cross-domain heterogeneous execution systems.

[0043] Furthermore, the directed semantic execution chain includes perceptual source information, state information, intent information, constraint information, execution action sequence information, and association information; the association information includes at least one of causal association information, dependency association information, or constraint association information; the directed semantic execution chain is used to represent the information transmission relationship between perceptual source information, state information, intent information, constraint information, execution action sequence information, and association information related to physical entities.

[0044] Taking the disease risk suppression task as an example, the generated directed semantic execution chain can include: perception source information, state information, intent information, constraint information, execution action sequence information, and association information. Perception source information includes humidity sensors inside the greenhouse, leaf temperature sensors, and crop stage identification modules. State information includes the formation of a high risk of gray mold, an increase in dew formation risk, and the crop being in the flowering and fruiting stage. Intent information includes: suppressing disease risk. Constraint information includes: avoiding a sudden drop in greenhouse temperature and avoiding affecting pollination and fruit setting. Execution action sequence information includes: slightly opening the top for ventilation and disturbing the low-speed fan. Association information includes: state-triggered intent, constraint-limiting actions, and action response status. Subsequently, this execution chain is sent to the skylight execution unit in the environmental control system, the low-speed fan unit in the fan control system, the disease monitoring subsystem, and the agricultural management platform. By synchronizing the complete semantic chain, rather than a single instruction, to the monitoring system and the human-machine platform, global cognitive synchronization is achieved.

[0045] The feedback receiving and model updating module is used to receive execution feedback data returned by multiple controlled cross-domain heterogeneous execution systems, and update the operation intent ontology model based on the execution feedback data to carry out collaborative control of cross-domain heterogeneous systems.

[0046] Furthermore, the operational intent ontology model is updated based on execution feedback data, including: The execution feedback data is mapped to the updated status data. For example, if the received execution feedback data includes: the skylight execution failed, the low-speed fan is running normally, the humidity has decreased by 2%, the risk of leaf condensation still exists, and the greenhouse temperature remains stable, the status node will be updated to: ventilation capacity is limited, the risk of disease has not been eliminated, and the current dehumidification method has limited effect.

[0047] Based on the updated state data, the semantic relationships between nodes in the operation intent ontology model are adjusted. For example, the executable association of the top micro-ventilation action node is reduced, the candidate weights of local low-disturbance ventilation and supplementary auxiliary dehumidification are enhanced, and the set of optional actions under the current constraints is adjusted. Based on the adjusted operation intent ontology model, subsequent semantic parsing inputs are generated, and the next round of action generation and collaborative control is entered.

[0048] The cross-domain heterogeneous system collaborative control method based on semantic abstraction provided in this application includes: S1: Receive the operation intent, identify the physical entity affected by the operation intent, and extract the spatial topological associations and state coupling relationships between different physical entities. Based on the physical entities and their associations, perform semantic abstraction on the operation intent and construct an operation intent ontology model. This achieves complete decoupling between tasks and devices. In traditional systems, operation intents are often directly bound to specific devices, resulting in poor scalability. However, this invention, through semantic abstraction, allows the operation intent to describe only what to do, without specifying what to use to do it. The same task intent can be adapted to multiple execution schemes, significantly improving the versatility and flexibility of cross-domain heterogeneous system collaborative control. At the same time, this step makes the constraints originally implicit in the code logic explicit as independent nodes, facilitating the dynamic adjustment and reuse of constraints, and laying a structured knowledge foundation for subsequent intelligent decision-making.

[0049] S2: Acquire input data and perform semantic instantiation processing on the input data based on a preset knowledge representation structure to generate state information and constraint information associated with the operation intention; This achieves deep integration of multi-source heterogeneous data. In traditional systems, data from different protocols and formats often form data silos, which are difficult to process uniformly. However, this invention uses a unified knowledge representation structure to map data from various sensors, management systems, and environmental monitoring equipment to specific node instances in the ontology model, breaking down data barriers and achieving unified management of cross-domain data. More importantly, this step enables the static ontology model to obtain real-time context and has dynamic environmental perception capabilities, providing accurate and comprehensive information support for subsequent intelligent decision-making; Through semantic enhancement, the original data is given clear semantic labels and contextual relationships, greatly improving the depth and accuracy of understanding the input data.

[0050] S3: Based on the operation intent, state information, and constraint information, it parses the preset semantic mapping rules to generate an execution action sequence. It can simultaneously consider multiple constraints such as time, resources, security, and performance. The generated solution is not only syntactically correct, but also meets all constraints at the semantic level. Through state evolution analysis, it predicts the execution effect, ensuring that the generated action sequence is truly feasible in the actual environment, and significantly improving the reliability and accuracy of decision-making.

[0051] S4: Generate a directed semantic execution chain based on the sequence of execution actions, and send the directed semantic execution chain to multiple controlled cross-domain heterogeneous execution systems. Through the protocol adaptation layer, the system can convert the unified semantic execution chain into the protocol format supported by each execution system, realizing unified management and control of heterogeneous devices using different communication protocols, such as ZigBee, Modbus, CAN, ROS, etc., significantly reducing the complexity of cross-domain heterogeneous execution system integration. At the same time, the directed semantic execution chain contains correlation information such as causal correlation, dependency correlation, and constraint correlation to support the collaborative cooperation between multiple controlled cross-domain heterogeneous execution systems, realizing true distributed collaborative execution.

[0052] S5: Receives execution feedback data from multiple controlled cross-domain heterogeneous execution systems and updates the operation intent ontology model based on the execution feedback data to perform cross-domain heterogeneous system collaborative control. It can continuously optimize execution strategies and automatically generate subsequent semantic parsing inputs based on the updated operation intent ontology model, realizing seamless connection between tasks and automated collaboration throughout the entire process, providing strong support for continuous operation in complex scenarios.

[0053] Example 2: Based on Example 1, as follows Figure 3The flowchart shown illustrates the execution of the action processing fault tolerance mechanism. When the agricultural execution system detects a fault event during action execution, it can perform fault tolerance based on the context information in the directed semantic execution chain. After sending the directed semantic execution chain to multiple controlled cross-domain heterogeneous execution systems, it also includes: Information about relevant nodes in a directed semantic execution chain is parsed by multiple controlled cross-domain heterogeneous execution systems.

[0054] Based on the parsing results, corresponding actions are executed and execution feedback data is output. The parsing results include the parsed target physical entity identifier, specific control command parameters such as opening angle and rotation speed, and the timing dependency conditions for action execution.

[0055] When a controlled cross-domain heterogeneous execution system detects a fault event during the execution of a corresponding action, it reads the intent information, state information, and constraint information from the directed semantic execution chain.

[0056] The constraint compliance of alternative actions is determined based on intent information, state information, and constraint information.

[0057] If the alternative action meets the constraints, execute the alternative action.

[0058] If the alternative action does not meet the constraints, a degradation action is performed or fault feedback data is output. The fault feedback data includes the specific location of the faulty entity, such as mechanical jamming or communication timeout, a real-time environmental state snapshot at the time of the fault, and a record of the actual actions performed after degradation.

[0059] For example, during the above execution process, if the top micro-ventilation actuator malfunctions and the skylight cannot be opened, the corresponding environmental control system reads the intent information, state information, and constraint information in the directed semantic execution chain. Then, it evaluates the alternative action of starting a high-power fan for forced ventilation. Although this alternative action can quickly reduce humidity, it may cause the greenhouse temperature to drop rapidly and affect the environment of the flower pot. Therefore, it does not meet the current constraint conditions, so the alternative action is not executed. Instead, a degradation process is performed, such as retaining only the low-speed fan disturbance, shortening the execution cycle and increasing the state feedback frequency, and outputting fault feedback data to the upper-level collaborative control system.

[0060] Example 3: Based on Examples 1 and 2, Example 3 is an application of the cross-domain heterogeneous system collaborative control system based on semantic abstraction in the energy consumption optimization of a park's comprehensive energy scenario. To verify the effectiveness of the cross-domain heterogeneous system collaborative control system based on semantic abstraction in the comprehensive energy dispatch scenario, an energy consumption optimization example is constructed using a cross-domain heterogeneous physical system consisting of office buildings, data centers, water supply systems, and public lighting systems in an industrial park as the object. The controlled execution systems in the park include HVAC systems, energy storage and power distribution systems, server cluster dispatch systems, water supply pumping station systems, and intelligent lighting systems. The original control protocols, communication interfaces, and dispatch logic of each system are independent of each other, making it difficult to form a unified collaborative control.

[0061] like Figure 4 The 24-hour cross-domain collaborative energy consumption optimization curve shown is used to verify the effectiveness of the cross-domain collaborative control strategy proposed in this application at the energy optimization level. The system energy consumption data within a 24-hour period is selected for comparative analysis. The gray curve represents the energy consumption change under the traditional distributed scheduling mode. Affected by factors such as independent operation of equipment, spatiotemporal conflicts of resources, and peak-valley load imbalance, the overall energy consumption is relatively high and fluctuates drastically. The green curve represents the energy consumption performance after adopting the cross-domain semantic collaborative scheduling of this application. Through computing power-power joint scheduling, dynamic resource negotiation and spatiotemporal conflict resolution mechanism, the smooth distribution of load and energy efficiency optimization are achieved, the overall energy consumption is significantly reduced, and the peak-valley difference is significantly reduced.

[0062] like Figure 5 The diagram shows a comparison of physical entity semantic abstraction before and after. To address the issues of inconsistent physical device descriptions and high interaction complexity in cross-domain heterogeneous systems, this application proposes a physical entity abstraction method based on semantic ontology. The diagram compares the changes in the number of measurement points for six typical heterogeneous devices—temperature sensors, electricity meters, air conditioners, water pumps, valves, and lighting—before and after semantic abstraction. The blue bars represent the number of original heterogeneous measurement points, which are the original data collection points uploaded by temperature sensors, electricity meters, air conditioning units, water pumps, valves, and lighting equipment. The orange bars represent the number of standardized measurement points, which are those mapped to standardized language after semantic abstraction. The standard state fields, constraint fields, or action fields in the semantic entity do not represent changes in the number of physical device entities. Compared with the blue bar chart, the number of measurement points for all devices has decreased significantly and tends to be consistent. The original heterogeneous measurement points are scattered, have heterogeneous formats, and have high redundancy, making it difficult to directly achieve cross-domain collaboration. After semantic modeling and normalized abstraction, various devices are uniformly encapsulated into standardized semantic entities with consistent interaction interfaces. The number of measurement points is greatly reduced, and the complexity of interaction between systems is significantly reduced. This application provides a unified data foundation and standardized expression capability for the collaborative control of cross-domain heterogeneous systems, which is the core prerequisite for realizing end-to-end collaboration.

[0063] The semantic abstract modeling module receives high-level operational intentions from the management platform to reduce the overall energy consumption of the park 24 hours a day and ensure operational comfort. It then constructs an operational intention ontology model based on standardized semantic entities corresponding to the physical entities in the park. The standardized semantic entities are obtained through semantic abstraction from the original measurement points, control points, and state points of the physical entities. They are used to uniformly represent the state, constraints, and executable actions of different devices in the operational intention ontology model. Intention nodes are defined as reducing overall energy consumption, peak shaving and valley filling, and ensuring office comfort. State nodes are defined as real-time temperature, personnel density, server load, electricity price period, water pressure status, and illuminance level. Constraint nodes are defined as maintaining room temperature between 22℃ and 26℃, computer room temperature not exceeding 28℃, water supply pressure not lower than the set threshold, illuminance meeting office standards, and maximum grid demand limits. Action nodes are defined as air conditioning temperature adjustment, energy storage charging and discharging, task migration, pump station frequency conversion speed regulation, and lighting brightness adjustment. All nodes establish unified semantic relationships through causal edges, dependency edges, and constraint edges.

[0064] The data acquisition and semantic instantiation module periodically collects the following input data: indoor temperature and humidity, return air temperature, and unit power uploaded by the HVAC system; time-of-use electricity price, energy storage status charge, and total load power uploaded by the power system; server CPU utilization and task queue length uploaded by the data center; water supply pressure and instantaneous flow rate uploaded by the water system; and area occupancy status and ambient brightness uploaded by the lighting system. Then, based on a preset knowledge representation structure, the above heterogeneous data is uniformly instantiated into state nodes and constraint nodes. For example, the current peak electricity price period is instantiated as a cost constraint; an unoccupied meeting room is instantiated as a low occupancy state; rising server room load is instantiated as a high heat load state; and sufficient energy storage power is instantiated as an adjustable resource state.

[0065] The rule parsing and action generation module performs semantic rule matching based on the operation intent ontology model to generate a set of candidate actions: when it is during peak electricity price periods and the energy storage capacity is sufficient, the candidate action is to discharge the energy storage; when the personnel density in the office area decreases and the room temperature is below the upper limit, the candidate action is to raise the air conditioning set temperature by 1°C; when the server is running under low load, the candidate action is to migrate some tasks and shut down idle servers; when the area is deserted at night, the candidate action is to turn off some lighting; when water demand decreases, the candidate action is to reduce the water pump frequency. Subsequently, the module combines constraint nodes to filter and eliminate actions that may lead to decreased comfort, equipment exceeding limits, or insufficient water supply, forming the final sequence of execution actions.

[0066] Before issuing actions, the state evolution prediction and action verification module extracts the execution parameters of each action and performs state evolution analysis in conjunction with the current state. Examples include: the room temperature change trend over the next 30 minutes after the air conditioner set temperature is increased; the peak load reduction after energy storage discharge; the hotspot temperature change in the computer room after task migration; the water supply pressure change after the pump station frequency is reduced; and the area illuminance change after lighting is reduced. If the prediction results meet the constraint boundaries, the corresponding action is retained; if the prediction results exceed the constraint conditions, the action magnitude is reduced or replaced with a suboptimal action.

[0067] The execution chain generation and distribution module encapsulates the final action sequence into a directed semantic execution chain. This chain includes: perception source information from real-time data collected by each system; status information such as current load, temperature, electricity price, and occupancy rate; intent information to reduce energy consumption and ensure comfort; constraint information including temperature range, illuminance standards, and lower limits of water supply pressure; action information including air conditioning temperature adjustment, energy storage discharge, task migration, pump station speed regulation, and lighting adjustment; and node association information including the triggering basis and execution order of the actions. This chain is then distributed to the HVAC controller, energy storage EMS system, data center scheduling platform, water pump PLC controller, and lighting gateway for execution.

[0068] The feedback receiving and collaborative update module receives execution feedback data in real time, including: actual power reduction value, regional temperature change, server task completion rate, water pressure stability, and user lighting satisfaction. If an execution system reports an anomaly, such as water pressure falling below the lower limit after the water pump frequency is reduced, the system automatically adjusts the semantic association relationship, increases the weight of water supply constraints, regenerates the subsequent action chain, and stops further frequency reduction actions.

[0069] The following points need to be explained: (1) The accompanying drawings of the embodiments of the present invention only involve the structures involved in the embodiments of the present invention. Other structures can refer to the general design.

[0070] (2) For clarity, the thickness of layers or regions is enlarged or reduced in the drawings used to describe embodiments of the invention, i.e., these drawings are not drawn to scale. It is understood that when an element such as a layer, film, region or substrate is referred to as being “above” or “below” another element, the element may be “directly” located “above” or “below” the other element or there may be intermediate elements.

[0071] (3) Where there is no conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other to obtain new embodiments.

[0072] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A cross-domain heterogeneous system collaborative control system based on semantic abstraction, characterized in that, include: The semantic abstract modeling module is used to receive operation intentions, identify the physical entities affected by the operation intentions, extract the spatial topological associations and state coupling relationships between different physical entities, perform semantic abstraction on the operation intentions based on the physical entities and their association relationships, and construct an operation intention ontology model. The data acquisition and semantic instantiation module is used to acquire input data and perform semantic instantiation processing on the input data based on a preset knowledge representation structure to generate state information and constraint information associated with the operation intention. The rule parsing and action generation module is used to parse the preset semantic mapping rules based on the operation intention, state information and constraint information to generate an execution action sequence; The execution chain generation and distribution module is used to generate a directed semantic execution chain based on the sequence of execution actions, and send the directed semantic execution chain to multiple controlled cross-domain heterogeneous execution systems; The feedback receiving and model updating module is used to receive execution feedback data returned by multiple controlled cross-domain heterogeneous execution systems, and update the operation intent ontology model based on the execution feedback data to carry out collaborative control of cross-domain heterogeneous systems.

2. The cross-domain heterogeneous system collaborative control system based on semantic abstraction according to claim 1, characterized in that, The operational intent ontology model includes intent nodes, state nodes, constraint nodes, and action nodes; The intent node is used to represent the semantics of a task that does not include device control parameters; The state node is used to represent the object state, environment state, or context state of a physical entity. The constraint node is used to represent execution constraint information related to the physical entity, and the execution constraint information includes at least one of time constraints, resource constraints, security constraints, or performance constraints; The action node is used to represent the action description information associated with the execution object corresponding to the physical entity; The intention nodes, state nodes, constraint nodes, and action nodes establish semantic relationships through directed association edges, which represent causal relationships, dependency relationships, or constraint relationships between nodes.

3. The cross-domain heterogeneous system collaborative control system based on semantic abstraction according to claim 2, characterized in that, The input data includes perception data, task data, and runtime context data; The semantic instantiation processing of input data based on a preset knowledge representation structure includes: Perform correlation and organization processing on sensing data, task data, and runtime context data; Based on a pre-defined knowledge representation structure, the associated and organized data is mapped into state nodes and constraint nodes; Establish associations between state nodes, constraint nodes, and intent nodes to form a node association structure in the operation intent ontology model.

4. The cross-domain heterogeneous system collaborative control system based on semantic abstraction according to claim 3, characterized in that, The step of parsing the preset semantic mapping rules based on the operation intention, state information, and constraint information to generate an execution action sequence includes: The intent node, state node, and constraint node are assembled into semantic parsing input; Based on semantic parsing input, a set of candidate actions is generated by matching according to preset semantic mapping rules. The candidate action set is filtered to determine the sequence of actions to be executed.

5. The cross-domain heterogeneous system collaborative control system based on semantic abstraction according to claim 4, characterized in that, The process of filtering the candidate action set to determine the sequence of actions to be executed includes: Based on the association between candidate actions and constraint nodes, perform association retrieval on candidate actions in the candidate action set; Determine the constraint association information corresponding to each candidate action; The candidate action set is filtered based on constraint association information.

6. The cross-domain heterogeneous system collaborative control system based on semantic abstraction according to claim 5, characterized in that, After filtering the candidate action set based on constraint association information, the method further includes: Extract the execution parameters corresponding to the filtered candidate actions; State evolution analysis is performed based on execution parameters and state data corresponding to the current state node to obtain state evolution results; The state evolution results are compared with the constraint boundary information corresponding to the constraint nodes; The sequence of actions to be executed is determined based on the comparison results.

7. The cross-domain heterogeneous system collaborative control system based on semantic abstraction according to claim 6, characterized in that, The directed semantic execution chain includes perception source information, state information, intent information, constraint information, execution action sequence information, and association information; The association information includes at least one of causal association information, dependency association information, or constraint association information; The directed semantic execution chain is used to represent the information transmission relationship between the perceived source information, state information, intention information, constraint information, execution action sequence information, and associated information related to physical entities.

8. The cross-domain heterogeneous system collaborative control system based on semantic abstraction according to claim 7, characterized in that, After sending the directed semantic execution chain to multiple controlled cross-domain heterogeneous execution systems, the process further includes: Information about relevant nodes in a directed semantic execution chain is resolved by multiple controlled cross-domain heterogeneous execution systems; Based on the parsing results, perform the corresponding actions and output execution feedback data; When a controlled cross-domain heterogeneous execution system detects a fault event during the execution of a corresponding action, it reads the intent information, state information, and constraint information from the directed semantic execution chain. The constraint compliance of alternative actions is determined based on intent information, state information, and constraint information. If the alternative action meets the constraints, then the alternative action is executed. If the alternative action does not meet the constraints, perform a degradation action or output fault feedback data.

9. The cross-domain heterogeneous system collaborative control system based on semantic abstraction according to claim 8, characterized in that, The update of the operation intent ontology model based on execution feedback data includes: Map the execution feedback data to the updated status data; Adjust the semantic relationships between nodes in the ontology model of operational intent based on updated state data; The adjusted operational intent ontology model is used to generate subsequent semantic parsing input.

10. A collaborative control method for cross-domain heterogeneous systems based on semantic abstraction, characterized in that: The method, applied to the cross-domain heterogeneous system collaborative control system based on semantic abstraction as described in any one of claims 1-9, comprises: S1: Receive the operation intention, identify the physical entity that the operation intention affects, extract the spatial topological association and state coupling relationship between different physical entities, perform semantic abstraction on the operation intention based on the physical entity and the association relationship, and construct the operation intention ontology model. S2: Obtain input data and perform semantic instantiation processing on the input data based on a preset knowledge representation structure to generate state information and constraint information associated with the operation intention; S3: Based on the operation intention, state information and constraint information, the preset semantic mapping rules are parsed to generate an execution action sequence; S4: Generate a directed semantic execution chain based on the sequence of execution actions, and send the directed semantic execution chain to multiple controlled cross-domain heterogeneous execution systems; S5: Receives execution feedback data from multiple controlled cross-domain heterogeneous execution systems and updates the operation intent ontology model based on the execution feedback data to perform collaborative control of cross-domain heterogeneous systems.

Images