An experimental intervention action generation and execution control method and system based on artificial intelligence

Abstract

The application provides an experiment intervention action generation and execution control method and system based on artificial intelligence, and relates to the technical field of experiment automation. The method comprises the following steps: obtaining an experiment intervention decision; generating an experiment intervention action based on the experiment intervention decision and real-time experiment state data; determining execution parameters of the experiment intervention action, wherein the execution parameters comprise execution time, execution sequence, execution strength and / or execution range; driving an execution mechanism to complete the experiment intervention action according to the execution parameters; obtaining an experiment state change after execution of the experiment intervention action, and adjusting the experiment intervention action and / or the execution parameters based on the experiment state change. Through the above scheme, dynamic generation, parameterized execution and closed-loop optimization of the experiment intervention action are realized, and the automation degree, execution precision and intervention effectiveness of the experiment process are improved.

Description

Technical Field

[0001] This invention relates to the field of experimental automation and intelligent control technology, and in particular to a method and system for generating and executing experimental intervention actions based on artificial intelligence. Background Technology

[0002] In the automated synthesis of nanomaterials using liquid-phase methods, researchers need to perform key experimental interventions such as temperature regulation, stirring speed control, and precursor feeding during the reaction process. Under current technology, all intervention actions in this experiment are pre-set based on human experience, such as fixed heating rates, stirring speed ranges, and single-time precursor feeding dosage and timing. Once these intervention parameters are set, they cannot be dynamically adjusted according to the real-time experimental conditions. During the experiment, unintended agglomeration of nanoparticles occurs due to local overheating of the system.

[0003] After online imaging equipment detects this morphological anomaly, traditional systems are unable to automatically match and generate corresponding intervention actions based on the real-time state. They rely on manual intervention to determine the cause of the anomaly and manually adjust relevant parameters, resulting in a significant lag in experimental intervention. Furthermore, the manually adjusted parameters lack precise quantitative basis. At the same time, after the system executes the manually adjusted parameters, it cannot provide real-time feedback on the improvement effect of the parameter adjustment on the particle aggregation state, and cannot further optimize subsequent intervention operations based on this effect. Current technology has technical defects such as a lack of dynamic generation capability for experimental intervention actions, insufficient flexibility of intervention parameters relying on human experience, disconnect between execution control and real-time experimental state with weak feedback loop, and difficulty in supporting fine intervention in complex experimental processes.

[0004] Therefore, the existing technology has at least the following drawbacks: First, the intervention actions mainly rely on pre-set parameters or manual judgment, and are difficult to dynamically generate based on the real-time experimental status; second, the execution time, execution order, execution intensity, and execution scope of the intervention actions lack quantitative determination basis, resulting in insufficient execution control precision; third, the changes in the experimental status after intervention lack an effective feedback utilization mechanism, making it difficult to form a closed-loop optimization control for subsequent interventions. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide an experimental intervention action generation and execution control method and system based on artificial intelligence, which addresses the problems in existing experimental automation systems that intervention actions are difficult to dynamically generate based on real-time experimental status, lack quantitative determination basis for intervention action execution parameters, and are difficult to use for subsequent intervention adjustments after execution.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: Firstly, an artificial intelligence-based method for generating and controlling experimental intervention actions, the method comprising: Obtaining experimental intervention decisions; Experimental intervention actions are generated based on experimental intervention decisions and real-time experimental status data. Determine the execution parameters for the experimental intervention; The actuator is driven to perform experimental intervention actions according to the execution parameters; The experimental state changes after the implementation of the experimental intervention are obtained, and the experimental intervention and / or execution parameters are adjusted based on the changes in the experimental state.

[0007] Furthermore, the experimental intervention decision originates from at least one of the upstream artificial intelligence decision-making module, path planning module, stage identification module, risk control module, or manual input.

[0008] Furthermore, based on experimental intervention decisions and combined with real-time experimental status data, experimental intervention actions matching the experimental status data are automatically generated; the experimental status data includes morphological or performance data of the experimental object, experimental environment parameter data, experimental device operating parameter data, and experimental process status data, including: Real-time acquisition of experimental status data, including morphological or performance data of experimental objects, experimental environment parameter data, experimental device operating parameter data, and experimental process status data; Based on the intervention objectives and constraints included in the experimental intervention decision, the preset triggering conditions are dynamically adjusted to determine the currently applicable set of triggering conditions; The experimental status data acquired in real time is matched and judged against each trigger condition in the set of currently applicable trigger conditions; When the experimental state data meets a certain triggering condition, a corresponding candidate intervention action is generated from the predefined intervention action types according to the intervention action type corresponding to the triggering condition, and the basic attributes of the intervention action are determined. The correspondence between the preset triggering condition and the intervention action type includes: when the sensor parameters, image features or execution deviations in the experimental state data meet the parameter adjustment triggering condition, a parameter adjustment action is generated, and the type of parameter to be adjusted is determined. When the stage change trend and risk boundary in the experimental status data approach the condition switching trigger condition, a condition switching action is generated, and the type of condition to be switched is determined; when the composition concentration and morphological color changes in the experimental status data meet the substance introduction or removal trigger condition, a substance introduction or removal action is generated, and the substance type and operation type are determined; when the process events and human-machine collaboration requirements in the experimental status data meet the process control trigger condition, an experimental process control action is generated, and the control type is determined.

[0009] Furthermore, determining execution parameters for the candidate intervention action includes: Obtain candidate intervention actions and their basic attributes; The execution parameter range for each candidate intervention action is determined based on real-time experimental status data and intervention constraints. Within the range of execution parameters, determine the execution time, execution order, execution intensity, and / or execution range of each candidate intervention action.

[0010] Furthermore, in some implementations, the method also includes estimating the execution effect corresponding to the candidate execution parameters based on historical execution data, rule models, experience models, simulation models, or knowledge bases, and optimizing the execution parameters based on the estimation results.

[0011] Furthermore, the experimental intervention actions are performed according to the execution parameters, including: Determine the execution relationship between multiple experimental intervention actions based on the execution order; Generate execution instructions based on action type and execution parameters; The execution command is sent to the corresponding execution mechanism to complete the experimental intervention.

[0012] Furthermore, during the execution of experimental intervention actions, constraint verification is performed to check whether the execution parameters fall within the allowable execution boundaries. When it is detected that the actual execution deviates from the allowable execution boundaries, parameter correction, execution is suspended, or execution instructions are reissued.

[0013] Furthermore, in some implementations, a manual confirmation request is generated before the experimental intervention is performed, and the intervention is executed or the execution parameters are adjusted based on the manual confirmation result.

[0014] Furthermore, it also includes: Obtain changes in experimental status after the implementation of experimental intervention actions; The changes in the experimental state are compared and analyzed with the intervention target to determine whether the intervention effect has achieved the expected results. When the intervention effect does not meet expectations, adjust the type of experimental intervention, execution parameters, and / or execution order; The adjusted experimental intervention actions and / or execution parameters are fed back to the intervention action generation and execution control process to form a closed-loop control.

[0015] Secondly, an artificial intelligence-based experimental intervention action generation and execution control system includes: The experimental intervention decision acquisition module is used to acquire experimental intervention decisions, which are derived from artificial intelligence decision-making systems or human input, and are used to indicate the need for intervention in the experimental process. The intervention action generation module is used to automatically generate experimental intervention actions that match the experimental status data based on experimental intervention decisions and in conjunction with real-time experimental status data. The experimental status data includes the morphological or performance data of the experimental object, experimental environment parameter data, experimental device operating parameter data, and experimental process status data. The experimental intervention actions are generated from predefined intervention action types according to the matching of the experimental status data type with preset conditions or target trajectories. The predefined intervention action types include parameter adjustment actions, condition switching actions, substance introduction or removal actions, and experimental process control actions. The execution parameter determination module is used to process experimental intervention actions and determine corresponding execution parameters for each experimental intervention action to obtain the determined execution parameters; the execution parameters include execution time, execution order, execution intensity or execution range; The experimental intervention execution module is used to drive the execution mechanism to complete the corresponding experimental intervention actions according to the determined execution parameters; The execution feedback module is used to obtain the changes in experimental status after the execution of experimental intervention actions, and to use the changes in experimental status to adjust and optimize the intervention actions. The changes in experimental status include changes in the parameters representing the experimental object, changes in process and environmental parameters, changes in the status of the device and actuator, changes in experimental stages and trends, and changes related to safety and risks.

[0016] Thirdly, a computing device, comprising: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0017] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0018] The above-described solution of the present invention has at least the following beneficial effects: This method employs a closed-loop control approach covering the entire process: acquiring intervention decisions, generating matching intervention actions, determining optimized execution parameters, driving the actuator, and acquiring state changes to optimize actions. It combines multi-dimensional real-time experimental state data acquisition, dynamic adjustment of preset trigger conditions, and precise state-action mapping for intervention action generation. Relying on a knowledge base, it enables the pre-simulation of intervention action execution effects and optimization of execution parameters. Simultaneously, it constructs a full-dimensional experimental state change feedback mechanism and adaptive adjustment logic for intervention actions, incorporating real-time safety monitoring and human-machine collaborative confirmation techniques. This effectively overcomes the technical problems of existing experimental intervention technologies, such as the lack of dynamic generation capabilities for intervention actions, insufficient flexibility in setting intervention parameters based on human experience, disconnect between execution control and real-time experimental state with weak feedback loops, and difficulty in supporting fine-grained interventions in complex experiments. This achieves intelligent matching and generation of experimental intervention actions and real-time experimental states, and precise and safe determination of execution parameters, realizing intelligent closed-loop control of the entire experimental intervention process. It improves the accuracy, consistency, and refinement of experimental execution, enhances the automation and intelligence level of the experimental system, and allows experimental intervention strategies to be continuously optimized based on execution feedback, better adapting to the refined intervention needs of various complex experiments.

[0019] Therefore, the present invention has at least the following beneficial effects: First, it can dynamically generate experimental intervention actions that match the current experimental state based on real-time experimental status data, thereby improving the timeliness and pertinence of intervention action generation. Secondly, it can parameterize the execution time, execution sequence, execution intensity and / or execution range of experimental intervention actions, thereby improving the accuracy and standardization of experimental intervention execution; Third, it can adjust subsequent intervention actions and / or execution parameters based on changes in the experimental state after the implementation of the intervention actions, forming a closed-loop execution control of the intervention and improving the effectiveness and stability of the intervention. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating an artificial intelligence-based experimental intervention action generation and execution control method provided by an embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of an artificial intelligence-based experimental intervention action generation and execution control system provided by an embodiment of the present invention. Detailed Implementation

[0022] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0023] like Figure 1 As shown, an embodiment of the present invention proposes a method for generating and controlling experimental intervention actions based on artificial intelligence, the method comprising the following steps: Step 1: Obtain experimental intervention decisions. These decisions, derived from an AI decision-making system or human input, indicate the need for intervention in the experimental process. Step 2: Based on experimental intervention decisions and combined with real-time experimental status data, automatically generate experimental intervention actions that match the experimental status data. The experimental status data includes the morphological or performance data of the experimental object, experimental environment parameter data, experimental device operating parameter data, and experimental process status data. The experimental intervention actions are generated from predefined intervention action types according to the matching of the experimental status data type with preset conditions or target trajectories. The predefined intervention action types include parameter adjustment actions, condition switching actions, substance introduction or removal actions, and experimental process control actions. Step 3: Process the experimental intervention actions and determine the corresponding execution parameters for each experimental intervention action to obtain the determined execution parameters; the execution parameters include execution time, execution order, execution intensity or execution range; Step 4: Based on the determined execution parameters, drive the actuator to complete the corresponding experimental intervention action; Step 5: Obtain the changes in experimental state after the implementation of the experimental intervention, and use the changes in experimental state to adjust and optimize the intervention. The changes in experimental state include changes in the parameters representing the experimental object, changes in process and environmental parameters, changes in the state of the device and actuator, changes in experimental stage and trend, and changes related to safety and risk.

[0024] In this embodiment of the invention, experimental intervention decisions originating from an artificial intelligence decision-making system or manual input are acquired. This is combined with multi-dimensional real-time experimental state data from the experimental object, environment, device, and process levels to automatically generate predefined types of experimental intervention actions, such as parameter adjustments and condition switching. Execution parameters, such as execution time and sequence, are determined for the intervention actions, driving the execution mechanism to complete the actions. The technical means of acquiring multi-dimensional experimental state changes after execution and using them for intervention action adjustment and optimization effectively overcomes the technical problems of existing experimental intervention technologies, such as the inability to dynamically generate intervention actions with the real-time experimental state, the disconnect between intervention execution and experimental state, and the lack of subsequent optimization and adjustment capabilities for intervention actions, making it difficult to achieve precise experimental intervention. This achieves precise matching and automated generation of experimental intervention actions with the real-time experimental state, ensuring standardized and precise execution of experimental intervention actions. It constructs a complete link for experimental intervention from decision-making, generation, execution to feedback optimization, effectively improving the automation and accuracy of experimental intervention. Simultaneously, it provides effective data support for the continuous optimization of experimental intervention strategies, adapting to various automated experimental intervention needs.

[0025] In a preferred embodiment of the present invention, step 1 above may include: The system receives intervention requests from an AI decision-making system or human input. The AI ​​decision-making system may include one or more of the following upstream decision-making modules: path planning, stage identification, risk control, or others. When an upstream decision-making module determines that intervention is necessary based on the current experimental state, experimental objective, or abnormal situation, it sends an intervention decision request to the experimental intervention action generation system. Specifically, during the experimental system's operation, the experimental path dynamic planning module continuously and synchronously receives real-time experimental state data output by the experimental stage identification module and preset experimental objective parameters, while simultaneously monitoring the overall progress of the experiment. As a core component of the AI ​​decision-making system, this module continuously performs the determination of experimental intervention requests. During the determination process, the module calculates the deviation between the real-time experimental state parameters and the experimental objective parameters. The deviation is the difference between the real-time experimental state parameter value and the experimental objective parameter value. When the absolute value of the deviation value exceeds the system's preset intervention threshold, or when the module identifies an abnormal evolution trend in the experiment and reaches a critical node for stage switching, and the experiment needs to maintain a specific target state, the experimental path dynamic planning module will immediately send a standardized experimental intervention decision request to the experimental intervention action generation system, thus triggering and sending the intervention request from the artificial intelligence decision-making system. At the same time, the experimental intervention action generation system maintains a real-time receiving status for the human operation terminal. Experimenters can actively input experimental intervention requests to the experimental intervention action generation system through the human operation terminal based on their observations and judgments during the experiment. After receiving the human-input intervention request, the system completes the reception of the human intervention request. The experimental intervention action generation system achieves comprehensive reception of both artificial intelligence decision-making system and human input intervention requests.

[0026] Experimental intervention decisions include at least the current experimental stage information, intervention objectives, and intervention constraints. Intervention objectives include promoting reaction convergence, suppressing abnormal trends, maintaining the target state, or triggering a stage switch. Intervention constraints include safety thresholds, equipment capability ranges, actuator limitations, or time windows. Specifically, after receiving the intervention request, during the experimental intervention decision generation process, real-time data from the experimental stage identification module is extracted across all dimensions to clearly define and label the current experimental stage information. This information accurately reflects the specific progress of the experiment, including but not limited to the reaction initiation stage, reaction in progress stage, product formation stage, and post-processing stage. The corresponding intervention objective is determined based on the triggering reason of the intervention request. If intervention is triggered due to a deviation exceeding a threshold between the real-time experimental state and the target parameters, the intervention objective is set to suppress abnormal trends. If intervention is triggered to advance the experimental process, the intervention objective is set to promote reaction convergence. If intervention is triggered to maintain a stable state, the intervention objective is set to maintain the target state. If intervention is triggered to reach a node in the experiment, the intervention objective is set to trigger a stage switch. For manually input intervention requests, the actual data is directly processed according to the provided data. The instructions input by the test personnel determine the intervention target, ensuring that the intervention target accurately matches the actual experimental needs. Next, the intervention constraints are determined and quantified. For safety thresholds, calculations are performed based on the inherent characteristic parameters of the experimental materials and the safety design standards of the experimental apparatus. Safety thresholds can be pre-set or dynamically updated based on the characteristics of the experimental materials, apparatus design standards, risk level requirements, or historical operating data. Other safety thresholds such as gas concentration and pH value are also quantified using this logic. For the equipment capability range, the factory rated parameters and actual operating calibration parameters of each actuator are extracted to determine the minimum and maximum operable parameters of each actuator, thereby defining the upper and lower boundaries of the equipment capability. For actuator limitations, the operating permissions of faulty or uncalibrated actuators are eliminated based on the real-time operating status of each actuator and the equipment linkage rules, clarifying the range of actuators that can perform normal operations. For time windows, calculations are performed based on the total preset duration of the current experimental stage and the actual running time. Time windows can be determined based on the progress of the current experimental stage, the scheduled operation period, the available equipment period, or process constraints. Determine the effective execution time range of the intervention operation; integrate the current experimental stage information after calibration, the determined intervention target, the calculated and verified safety threshold, equipment capacity range, execution mechanism limitations, time window and other intervention constraints to obtain a complete experimental intervention decision containing all core elements.

[0027] In this embodiment of the invention, a dual-source intervention request receiving mechanism combining an artificial intelligence decision-making system and human input is established. This ensures that the initiation of experimental intervention requests is fully aligned with the real-time operational status of the experiment, avoiding the lag and blindness in the initiation of intervention requests. Simultaneously, through quantitative calculation and actual parameter verification, the core components of experimental intervention decisions are clarified, providing a scientific quantitative basis for the formulation of intervention decisions, rather than relying solely on human experience. The quantitative calculation and boundary delineation of intervention constraints pre-set safe and feasible operational boundaries for the generation and execution of experimental intervention actions. This avoids the risk of intervention operations exceeding equipment operating capabilities and breaching experimental safety standards from the decision-making source, improving the scientific rigor, accuracy, and standardization of experimental intervention decisions, and laying a solid foundation for the precise generation of intervention actions.

[0028] In a preferred embodiment of the present invention, step 2 above may include: Real-time acquisition of experimental status data, including morphological or performance data of experimental objects, experimental environmental parameters, experimental apparatus operating parameters, and experimental process status data, specifically involves: to acquire multi-dimensional experimental status data in real time, multi-sensor acquisition devices and module data retrieval are used to collect morphological or performance data of experimental objects, experimental environmental parameters, experimental apparatus operating parameters, and experimental process status data in full-dimensional real-time, and the collected data are uniformly transmitted to the data processing module for preprocessing. The morphological or performance data of experimental objects is continuously collected by online imaging equipment, microscopic detection devices, and performance characterization sensors, covering morphological data such as particle size, morphology, and degree of aggregation, and performance data such as reaction rate, crystallinity, and component content; experimental environmental parameters... Data is collected by sensors such as temperature, humidity, pressure, gas concentration, pH value, and conductivity at preset time intervals, which can be flexibly set according to the type of experiment. The operating parameter data of the experimental device is retrieved in real time from the control systems of various actuators such as stirring, feeding, temperature control, and valves, covering operating parameter values, equipment operating status, and fault information. The experimental process status data is output in real time by the experimental stage identification module, covering the current experimental stage, stage duration, process execution progress, and whether key nodes have been reached. The data processing module performs noise reduction and normalization preprocessing on all collected and retrieved data. Noise reduction removes abnormal fluctuation data, and normalization converts parameter values ​​of different dimensions to the same numerical range to ensure the validity and consistency of the experimental status data, thus completing the real-time acquisition of experimental status data.

[0029] Based on the intervention objectives and constraints included in the experimental intervention decision, the preset trigger conditions are dynamically adjusted to determine the currently applicable set of trigger conditions. Specifically, this includes: dynamically adjusting the preset trigger conditions according to the intervention objectives and constraints to determine the currently applicable set of trigger conditions. The system first retrieves a pre-stored basic trigger condition library, which contains basic trigger conditions, judgment indicators, and basic thresholds corresponding to four types of intervention actions: parameter adjustment, condition switching, substance introduction or removal, and experimental process control. Combined with the intervention objectives in the experimental intervention decision, the judgment sensitivity of the trigger conditions is adjusted. For some pre-objectives, to suppress abnormal trends, the basic thresholds of trigger conditions related to abnormal states are lowered by multiplying the basic threshold by a preset sensitivity coefficient. This sensitivity coefficient can be preset or dynamically adjusted according to the intervention objectives and experimental scenarios to improve the sensitivity of anomaly identification. For some pre-objectives, to promote reaction convergence, the basic thresholds of trigger conditions related to reaction progress are raised by multiplying the basic threshold by a preset advancement coefficient, which is set to 1.2 to adapt to the reaction advancement requirements.

[0030] By combining intervention constraints with boundary limits for triggering conditions, and based on a safety threshold, the threshold at which the risk boundary approaches the relevant triggering condition is set as the safety threshold multiplied by a safety warning coefficient. The safety warning coefficient can be set to a warning ratio less than 1 to ensure that a warning is triggered in advance before the safety threshold is reached. Based on the equipment's capability range, all triggering conditions that exceed the minimum and maximum operating parameters of the actuator are eliminated. Based on a time window, an effective execution period is set for each triggering condition, and triggering conditions that exceed the time period are temporarily invalidated. After all adjustments are completed, the system integrates the triggering conditions that are adapted to the current intervention target and constraints, and clarifies the intervention action type, judgment index, adjusted threshold, and effective execution requirements corresponding to each triggering condition, forming a set of currently applicable triggering conditions.

[0031] The thresholds, coefficients, ratios, and adjustment rules mentioned are all examples and can be preset or adaptively adjusted according to the experiment type, experiment object, device capability, and historical execution data, and do not constitute a limitation.

[0032] The process involves matching real-time acquired experimental status data with each trigger condition in the currently applicable trigger condition set. Specifically, this includes: matching real-time experimental status data with the current trigger condition set; classifying experimental status data and trigger conditions according to intervention action type; and then comparing pre-processed experimental status data with each corresponding trigger condition. For numerical indicators, such as sensor parameters like temperature, pressure, and concentration, the numerical values ​​in the experimental status data are directly compared with the adjusted threshold of the trigger condition to determine whether the value reaches or exceeds the threshold. For feature-based indicators, such as image features like the shape and color changes of the experimental object, image recognition is used to determine whether the value reaches or exceeds the threshold. The system extracts feature values ​​using an algorithm and calculates the similarity between the feature values ​​and the trigger feature threshold. The similarity calculation formula is the number of matching features divided by the total number of features. When the similarity reaches the preset similarity threshold, the corresponding trigger condition is determined to be met. For status-based judgment indicators, such as process execution failure, human-machine collaboration requirements, and equipment failure, the system directly judges whether the actual status of the indicator is consistent with the status description of the trigger condition. The system will verify whether the effective execution requirements of each trigger condition are met. If the effective requirements are not met, even if the indicator matches, it is determined to be unmet. After all trigger conditions are compared, the system records the satisfied trigger conditions and the corresponding judgment basis, thus completing the matching judgment.

[0033] When experimental state data meets a certain trigger condition, a corresponding candidate intervention action is generated from a predefined set of intervention action types based on the intervention action type corresponding to that trigger condition, and the basic attributes of the intervention action are determined. The correspondence between preset trigger conditions and intervention action types includes: when sensor parameters, image features, or execution deviations in the experimental state data meet parameter adjustment trigger conditions, a parameter adjustment action is generated, and the type of parameter to be adjusted is determined. Specifically, this includes: generating candidate intervention actions based on the met trigger conditions and determining their basic attributes; the system first retrieves the preset correspondence table between trigger conditions and intervention action types, matches the corresponding intervention action type based on the recorded met trigger conditions, and if the met trigger condition is a parameter adjustment trigger condition... The system generates parameter adjustment actions as candidate intervention actions from predefined intervention action types. Based on the specific experimental state data indicators that trigger the action, the system accurately determines the basic attributes of the intervention action, i.e., the type of parameter to be adjusted: if triggered by experimental environment sensor parameters such as temperature, pressure, and humidity, the parameter to be adjusted is the corresponding environmental parameter; if triggered by image features such as particle size, morphology, and aggregation degree of the experimental object, the parameter to be adjusted is determined based on the experimental influence parameters associated with the image features, such as abnormal particle size, which is associated with parameters such as stirring speed and reaction temperature; if triggered by execution deviation, such as the deviation between the actual operating parameters and the set value, the parameter to be adjusted is the corresponding actuator operating parameter, thus completing the determination of the candidate intervention actions and basic attributes for parameter adjustment.

[0034] When the stage change trend and risk boundary in the experimental state data approach the condition switching trigger condition, a condition switching action is generated, and the type of condition to be switched is determined. When the composition concentration and morphological color changes in the experimental state data meet the substance introduction or removal trigger condition, a substance introduction or removal action is generated, and the substance type and operation type are determined. When the process event and human-machine collaboration requirement in the experimental state data meet the process control trigger condition, an experimental process control action is generated, and the control type is determined. Specifically, to complete the generation of candidate intervention actions and determination of basic attributes corresponding to the other three types of trigger conditions, the system performs operations in sequence according to the trigger condition type. If the stage change trend in the experimental state data, such as the transition from the reaction change stage to the stable stage, or the approach of the risk boundary, such as the parameter approaching the safety warning threshold, meets the condition switching trigger condition, the system generates a condition switching action as a candidate intervention action, and determines the type of condition to be switched according to the specific trigger indicators. The stage change trend triggers experimental process conditions such as atmosphere, illumination, and temperature control mode. In some implementations, when the execution constraint boundary approaches, actions to reduce execution intensity can be generated. The system generates conditions for delaying or switching execution modes. If the composition concentration in the experimental status data (e.g., reactant concentration too low, product concentration too high), or changes in morphology and color (e.g., precipitation, color abrupt change) meet the trigger conditions for substance introduction or removal, the system generates substance introduction or removal actions as candidate intervention actions. The type of substance and operation type are determined based on specific trigger indicators. Composition concentration triggers the corresponding reactant, product, or reaction regulator; morphology and color change triggers the corresponding solvent, eluent, or separating agent; and operation types are determined based on actual experimental needs, such as feeding, replenishing liquid, filtration, discharge, or replacement. If process events in the experimental status data (e.g., critical node arrival, process execution failure, action timeout), or human-machine collaboration needs (e.g., requiring manual sampling or confirmation) meet the process control trigger conditions, the system generates experimental process control actions as candidate intervention actions. The control type is determined based on specific trigger indicators. Process event triggers include step advancement, pause, termination, and re-execution; human-machine collaboration needs trigger process waiting or continuation after manual confirmation. This completes the determination of all candidate intervention actions and their basic attributes that meet the trigger conditions.

[0035] In this embodiment of the invention, by collecting and standardizing the data of four types of experimental states in real time, the data source for generating experimental intervention actions is ensured to be authentic, effective, and consistent, thus avoiding deviations in action generation caused by data distortion from the source. By dynamically adjusting the preset triggering conditions and defining the execution boundaries in combination with the intervention goals and constraints, the triggering conditions are highly adapted to the actual experimental needs, overcoming the limitations of traditional fixed triggering conditions that cannot adapt to different experimental scenarios. Through precise matching judgment by category and index, efficient and accurate comparison between experimental state data and triggering conditions is achieved, improving the accuracy of action triggering judgment. By determining the basic attributes of actions based on the correspondence between preset triggering conditions and intervention action types, combined with specific triggering indicators, the automatic and dynamic generation of candidate intervention actions is realized, ensuring a high degree of matching between intervention actions and real-time experimental states. This overcomes the shortcomings of traditional technologies where intervention actions are pre-set and cannot be dynamically generated according to experimental states, while also eliminating excessive reliance on human experience and avoiding the blindness of manual setting. This lays a solid foundation of actions and data for the precise execution of intervention actions, improving the scientific, dynamic, and accurate nature of experimental intervention action generation.

[0036] In a preferred embodiment of the present invention, step 3 above may include: The system acquires candidate intervention actions and their basic attributes. These attributes include the intervention action type, the parameter type to be adjusted, the condition type to be switched, the type of substance to be introduced or removed, and the control type to be executed. Specifically, the system extracts all generated candidate intervention actions and their corresponding judgment data from the real-time data cache of the intervention action generation module. These actions are then categorized and organized according to intervention action types: parameter adjustment, condition switching, substance introduction or removal, and experimental process control. The system extracts the core information of each candidate intervention action, clearly labeling the intervention action type and the corresponding basic attributes: parameter adjustment actions are labeled with the parameter type to be adjusted; condition switching actions are labeled with the condition type to be switched; substance introduction or removal actions are labeled with the type of substance to be introduced or removed and the operation type; and experimental process control actions are labeled with the control type to be executed.

[0037] After information extraction is completed, the system performs a completeness check on the basic attributes of each candidate intervention action, removes invalid candidate actions with missing attributes or ambiguous labels, organizes the verified candidate intervention actions and their complete basic attributes into a standardized dataset, and stores it in the dedicated data area of ​​the action parameter configuration module, thus completing the acquisition of candidate intervention actions and their basic attributes.

[0038] Based on the current real-time experimental status data and the intervention constraints included in the experimental intervention decision, the execution parameter range for each candidate intervention action is determined. Specifically, this includes: determining the execution parameter range for each candidate intervention action based on the current real-time experimental status data and the intervention constraints in the experimental intervention decision; the system performs quantitative calculations for different types of candidate intervention actions, combining two types of core data, and delineates the upper and lower boundaries of each execution parameter. All calculation results must meet the intervention constraints of safety thresholds, equipment capability ranges, actuator limitations, and time windows. Parameters exceeding the constraint range are directly corrected according to the constraint boundaries. For parameter adjustment actions... First, extract the real-time status value of the parameter to be adjusted. Then, combine this with the minimum and maximum adjustable values ​​of the actuator corresponding to that parameter within the equipment's capability range to calculate the initial range of the execution parameter. The lower limit of the initial adjustment range is the real-time status value minus the minimum adjustment amount of the actuator, and the upper limit is the real-time status value plus the maximum adjustment amount of the actuator. These are then corrected using a safety threshold. If the initial upper limit exceeds the safety threshold for that parameter, the upper limit is corrected to the safety threshold; if the initial lower limit is lower than the safety threshold, the lower limit is corrected to the safety threshold. This finalizes the execution parameter range for the action. For actions involving the introduction or removal of substances, the range is determined based on the real-time composition and concentration of the experimental object. Based on the temperature data and the total volume of the experimental system, the basic dose range of the required substances is calculated. The lower limit of the basic introduction dose is determined by multiplying the experimental system volume by (target concentration minus real-time concentration) and then adjusting it according to a preset correction ratio. The upper limit of the basic introduction dose is determined by multiplying the experimental system volume by (target concentration minus real-time concentration) and then adjusting it within the allowable range. For substance removal, the basic range is calculated based on the actual content of the product to be separated or the content of excess reactants. This is then combined with the rated range correction of the feeding and filtration equipment of the actuator to finally determine the range of execution parameters such as dose and rate. For condition switching actions, the switchable conditions are defined according to the type of condition to be switched and the equipment capacity range. For example, atmosphere switching is defined as the concentration range of inert gas and reactive gas that the equipment can achieve, and temperature control mode switching is defined as the temperature range of constant temperature and programmed temperature rise supported by the equipment. At the same time, the effective execution period of the switching action is set in conjunction with the time window as the range of execution time parameters. For experimental process control actions, the time range of process advancement, pause and termination is defined according to the process execution progress and time window. Combined with the linkage rules of the actuator, the range of executable control operations is defined. The range of execution parameters can be determined by rule calculation, empirical model, historical sample fitting, simulation evaluation or a combination thereof to ensure no equipment conflict.

[0039] For each candidate intervention action, its execution time, execution order, execution intensity, or execution range is determined: execution time includes the triggering time, duration, or delay time of the action; execution order includes the sequence, parallel execution relationship, or conditional dependency relationship between multiple intervention actions; execution intensity includes the amplitude, rate, acceleration, or frequency of parameter adjustment; execution range includes the dose, rate, concentration, or spatial distribution range of the introduced substance. Specifically, to determine the execution time, execution order, execution intensity, or execution range for each candidate intervention action, based on the defined execution parameter range, combined with the real-time state change trend and the experimental intervention target, the execution parameters of each dimension are precisely quantified and determined. In determining the execution time, the action triggering moment is set to the current moment when the system determines in real time that the experimental state data meets the triggering conditions. The action duration is calculated based on the execution intensity and the range of execution parameters. For example, the duration of parameter adjustment is the determined adjustment range divided by the rated adjustment rate of the actuator. The action delay time is set according to the equipment linkage requirements. If multiple devices need to be executed collaboratively, a reasonable delay time is set according to the order of device response, and all time parameters must fall within the time window of the intervention constraint conditions. In determining the execution order, the system first analyzes the correlation and dependency between each candidate intervention action. Actions without direct correlation and without mutual influence are set as parallel execution relationships. Actions with sequential dependencies, such as adjusting the temperature before introducing the substance, are set as sequential execution relationships. Actions that need to meet specific experimental conditions before execution are set as conditional dependencies. At the same time, the execution priority of each action is marked. In determining the execution intensity, for parameter adjustment actions, the execution intensity is set as follows: The execution intensity is defined as the adjustment range and adjustment rate. The adjustment range is determined by the deviation between the target parameter value and the real-time parameter value (target parameter value minus real-time parameter value). The adjustment rate is the rated adjustment rate of the actuator, determined based on the actuator's capability range, target convergence requirements, and overshoot suppression requirements to avoid parameter adjustment overshoot. For other actions, the corresponding intensity index is determined based on the experimental intervention target. For example, the intensity of condition switching actions is determined by the response rate of equipment switching. In determining the execution range, for actions involving the introduction or removal of substances, the intermediate value is selected from the defined range as the base value. For example, the introduced dose is taken as the average value of the dose range, and then fine-tuned based on the real-time morphology and performance data of the experimental object. At the same time, the spatial distribution range is clearly defined. For example, the stirring reaction system is set to uniform feeding throughout the entire area, and the local reaction system is set to feeding in a designated area. For other actions, the corresponding execution range boundary is clearly defined according to the experimental requirements to ensure that the execution range is within the constraints.

[0040] In some implementations, the execution effect corresponding to the determined execution parameters can be estimated based on historical execution data, rule models, experience models, simulation models, or knowledge bases to obtain pre-performance results. The execution parameters are then optimized and adjusted based on these results, and the final combination of execution parameters is selected as the execution parameters. Specifically, the execution intensity or execution range is determined in numerical range form and is subject to safety thresholds and equipment capability limits in the intervention constraints. This includes: determining the final execution parameters after optimization and ensuring that the execution intensity or execution range is presented in numerical range form throughout the process, and is strictly limited by safety thresholds and equipment capability limits; the system retrieves historical intervention data of similar experiments stored in the knowledge base, experimental state change data corresponding to different execution parameters, the correlation model between intervention actions and experimental effects, and relevant rules for experimental safety management as the basis for effect pre-performance; the execution parameters of each determined candidate intervention action are input into the experimental effect pre-performance model in the knowledge base; the model simulates the changes in various dimensions such as experimental object representation, experimental environment parameters, and device operating status after the intervention action is executed according to the parameters, and outputs pre-performance results, including the degree of achievement of the intervention target, experimental status, etc. The system evaluates key indicators such as the rate of change of state, equipment operating load, and experimental risk level. Next, it sets evaluation thresholds for the pre-simulation results: the intervention target achievement rate must be no less than 85%, the equipment operating load no more than 90%, and the experimental risk level must be low. If the pre-simulation results do not meet the evaluation thresholds, the system optimizes and adjusts the execution parameters. If the target achievement rate is insufficient, the execution intensity is increased by multiplying the original intensity by 1.1, and the increased intensity does not exceed the defined range of execution parameters. If the equipment operating load is too high, parallel actions are changed to sequential execution to reduce the equipment load per time period. If the experimental risk level is too high, the execution intensity will be reduced by 0.8 times the original execution intensity, and the execution range will be adjusted to within the safety threshold. If the simulation results reach the evaluation threshold, the system will retain the current execution parameters. The system integrates the execution parameters of all candidate intervention actions, and organizes the execution time, execution order, execution intensity, and execution range of each action into a standardized parameter combination in the form of numerical range and clear correlation. This standardized parameter combination is then stored in the system data center and transmitted to the experimental intervention execution module to complete the entire execution parameter configuration process.

[0041] In this embodiment of the invention, the accuracy and completeness of the data source for the execution parameter configuration are ensured by classifying and verifying the candidate intervention actions and their basic attributes, thus avoiding parameter configuration deviations caused by missing attributes. By combining real-time experimental status data and intervention constraints to quantify different types of candidate actions, a scientific range of execution parameters is defined, ensuring that the execution parameters meet experimental safety requirements and equipment operating capabilities from the outset, thus mitigating experimental risks and equipment malfunctions caused by parameters exceeding boundaries. Furthermore, by precisely determining the execution time, sequence, intensity, and range across multiple dimensions, the execution parameters are highly aligned with the real-time experimental status and intervention objectives. This approach addresses the issues of vague intervention parameters and lack of clear execution dimensions in traditional techniques. By using knowledge-based experimental effect simulation and parameter optimization, the setting of execution parameters is based on historical experimental data and related models, eliminating excessive reliance on human experience and effectively avoiding poor intervention results due to unreasonable parameters. Furthermore, multiple optimizations ensure optimal execution effects and minimal risks for the intervention actions. Overall, it achieves scientific, precise, and safe configuration of experimental intervention action execution parameters, laying a solid parameter foundation for the accurate execution of experimental intervention actions and improving the automation and intelligence level of experimental intervention parameter configuration.

[0042] In a preferred embodiment of the present invention, step 3 above may include: The process involves obtaining the determined experimental intervention actions and their corresponding execution parameters. These parameters include execution time, execution order, execution intensity, and execution range. Specifically, the experimental intervention execution module retrieves all optimized and finalized experimental intervention actions and their associated complete execution parameters from the system data center. These parameters are then categorized and archived according to intervention action type. An independent execution parameter file is created for each intervention action, containing the intervention action type, basic attributes, and all execution parameters such as execution time, execution order, execution intensity, and execution range. Simultaneously, the retrieved parameters are validated. This validation includes verifying whether the execution parameters are within the safety thresholds and equipment capabilities defined by the intervention constraints, and whether the execution time falls within the valid time window. Invalid intervention actions are removed. The validated experimental intervention actions and their corresponding valid execution parameters are integrated into a standardized execution list and stored in the local data area of ​​the experimental intervention execution module, thus completing the acquisition of the determined experimental intervention actions and their execution parameters.

[0043] Based on the execution order in the execution parameters, the system determines the execution relationship between multiple intervention actions. This relationship includes single action execution, sequential execution of multiple actions, or parallel execution of multiple actions. Specifically, this involves: determining the execution relationship between multiple intervention actions based on the execution order in the execution parameters; the system extracts the execution order parameters of all intervention actions in the execution list; and parses the execution priority, dependency identifier, and execution time node of each action one by one. If the execution list contains only a single experimental intervention action, it is directly determined to be a single action execution relationship; if it contains multiple experimental intervention actions, actions with sequential dependency identifiers are first identified, such as those marked as needing to be performed in action XX. After the intervention is completed, the order of the actions is determined according to the dependencies, and multiple actions are judged to be executed sequentially. The order of execution nodes is also clarified. Then, actions without any dependency markers and with the same execution time are identified as multiple actions executed in parallel, and the synchronous triggering conditions for parallel execution are clarified. For actions with conditional dependency markers, they are marked as conditional parallel / sequential relationships executed after the XX experimental state is met, and included in the corresponding execution relationship category. Finally, the system classifies and labels the execution relationships of all intervention actions to form an execution relationship judgment table, which clarifies the execution method and related actions of each action, and completes the judgment of the execution relationship.

[0044] For each intervention action to be executed, a corresponding execution instruction is generated based on the action type and execution parameters. The execution instruction includes an action identifier, an execution timestamp, and an execution intensity value or execution range value. Specifically, this includes: generating a corresponding execution instruction for each intervention action to be executed; the system generates a unique action identifier for each verified intervention action based on the execution list and execution relationship judgment table; the action identifier includes core information such as the intervention action type, execution priority, and execution agency code to ensure the uniqueness and identifiability of the identifier; extracting the execution time parameter of the action and converting the action triggering time into a system-unified timestamp to accurately mark the execution start time node; and extracting the execution intensity or execution range parameter and converting it into a numerical execution intensity value or execution range value according to the parameter receiving specifications of the execution agency. For example, the execution intensity value of stirring speed adjustment is marked as a specific speed value, and the execution range value of material introduction is marked as a specific dosage and rate value. The system integrates unique action identifiers, execution timestamps, execution intensity values ​​or execution range values, as well as auxiliary information such as action type and execution relationship, and generates standardized execution instructions according to the instruction receiving format of the actuator. Each intervention action to be executed corresponds to a unique execution instruction. The instruction format is uniformly adapted to the identification requirements of different types of actuators such as mechanical, electronic, and hybrid actuators, thus completing the generation of execution instructions.

[0045] The execution instructions are sent to the corresponding actuators to drive them to complete the experimental intervention actions. The actuators include mechanical actuators, electrically controlled actuators, or hybrid actuators. Specifically, this includes: sending execution instructions to the corresponding actuators and driving them to complete the experimental intervention actions; parsing the actuator code in each execution instruction; classifying the execution instructions according to the code and sending them to the corresponding mechanical actuators, electrically controlled actuators, or hybrid actuators; mechanical actuators receiving mechanical action instructions such as feeding and filtration; electrically controlled actuators receiving parameter adjustment instructions such as temperature control and speed adjustment; and hybrid actuators receiving composite instructions requiring mechanical and electrical coordination. After receiving the execution instructions, the actuators first parse and verify the instructions to confirm the completeness of the instruction information. Once the parameters meet their own operating specifications, the execution mechanism initiates the action execution according to the execution timestamp, execution intensity value, or execution range value in the instruction: For single-action execution relationships, the execution mechanism completes the corresponding action independently according to the instruction; for sequential execution relationships, each execution mechanism triggers the action in sequence, and after the previous action is completed, it sends a completion receipt to the system, and the system then sends an execution trigger signal to the execution mechanism of the subsequent action; for parallel execution relationships, each execution mechanism triggers the action synchronously according to a unified timestamp to ensure that the actions are carried out in parallel. After all execution mechanisms complete the actions as required by the instruction, they all send an action execution completion receipt to the experimental intervention execution module. The receipt contains information such as the action execution duration, actual execution parameters, and equipment operating status, thus completing the driving execution of the experimental intervention action.

[0046] During execution, constraint verification is performed to ensure that the execution parameters fall within the allowable execution boundaries. When an execution parameter is detected to exceed the safety threshold, execution interruption or parameter correction is triggered. Specifically, this includes: the safety monitoring module establishing a real-time data interaction channel with each execution mechanism; collecting actual execution parameters, real-time status parameters of the experimental environment, and operating status parameters of the experimental device at a frequency that meets real-time requirements throughout the entire process of the experimental intervention; the safety monitoring module continuously compares the collected actual parameters with the safety threshold in the intervention constraints to determine whether the actual parameters are within the safety threshold range; if the actual parameters are within the allowable execution boundaries, the current execution action is maintained, and monitoring continues; if the actual parameters deviate from the allowable execution boundaries, graded processing is performed according to the degree of deviation: when the deviation is slight, parameter correction is triggered, and a parameter correction command is sent to the corresponding execution mechanism to adjust the execution parameters to within the safety threshold range; when the deviation is severe, execution interruption is triggered, and an interruption command is sent to the corresponding execution mechanism to stop the execution of the current intervention action, and a safety warning message is sent to the system backend. The safety warning message includes the parameter type exceeding the threshold, the actual value, and the safety threshold. Through the above methods, safety monitoring and anomaly handling of the execution process are completed.

[0047] Based on the human confirmation requirement in experimental intervention decision-making, a human confirmation request is generated before executing the intervention action. Execution continues after receiving the human confirmation instruction, or execution parameters are adjusted based on human feedback before execution. Specifically, this includes: completing human-machine collaborative execution operations based on the human confirmation requirement. After the experimental intervention execution module generates the execution instruction, the human-machine interaction module parses the human confirmation requirement identifier in the experimental intervention decision-making. If this identifier is not enabled, the system directly sends the execution instruction to the corresponding execution agency without human intervention. If the identifier is enabled, the human-machine interaction module immediately generates a human confirmation request, which fully displays core information such as the type of intervention action to be executed, execution parameters, execution time, execution agency, and intervention target. This request is pushed to the experimenter through the experimental operation terminal, and the experimenter receives the confirmation request. Afterwards, the user can choose to send a manual confirmation instruction or provide feedback on parameter adjustments. If a manual confirmation instruction is received, the human-computer interaction module transmits the instruction to the experimental intervention execution module, which then sends an execution instruction to the corresponding execution mechanism to initiate the intervention action. If parameter adjustment feedback is received from the experimenter, the human-computer interaction module converts the feedback into a standardized parameter adjustment instruction and transmits it to the experimental intervention execution module. The module then corrects the original execution parameters based on the feedback. The corrected parameters need to be verified again to ensure they meet the intervention constraints. Once the verification is successful, an execution instruction is regenerated and sent to the corresponding execution mechanism for execution. If the corrected parameters do not meet the constraints, the module sends a notification to the experimenter indicating that the parameter adjustment is invalid and requests reconfirmation, thus completing the intervention action execution operation based on the manual confirmation requirement.

[0048] In this embodiment of the invention, by retrieving, verifying, and standardizing the determined experimental intervention actions and execution parameters, the accuracy and effectiveness of the data source in the execution process are ensured, thus avoiding execution errors caused by invalid parameters from the source. Through precise analysis of the execution sequence and scientific determination of the execution relationships, the execution methods of multiple intervention actions are clarified, solving the problems of chaotic action execution and poor coordination in traditional execution processes, and improving the orderly execution of intervention actions. By generating standardized execution instructions with unique identifiers, the execution instructions are accurately adapted to different types of execution mechanisms, ensuring the identifiability of the instructions and the accuracy of execution. By classifying and sending instructions according to the execution mechanism code and driving execution, multiple execution methods such as single, sequential, and parallel execution are implemented, adapting to complex experiments. Diverse intervention needs; through high-frequency safety monitoring and graded anomaly handling during the execution process, full-process safety control of experimental intervention execution was achieved, timely avoiding experimental safety risks and equipment failures caused by parameter exceeding thresholds, and improving the safety of the execution process; by setting up a human-machine collaboration link for manual confirmation requests, the combination of intelligent execution and manual control was realized, preserving the decision-making initiative of experimental personnel, optimizing execution parameters based on human feedback, and making experimental intervention execution more in line with actual experimental needs. Overall, the standardized, precise, and safe control of experimental intervention actions from instruction generation to implementation was achieved, improving the automation and collaboration level of experimental intervention execution, and effectively solving the defects of traditional technologies such as the disconnect between execution control and experimental status, lack of effective monitoring of the execution process, and lack of human-machine collaboration.

[0049] In a preferred embodiment of the present invention, step 4 above may include: After the experimental intervention is completed, the changes in the experimental state are acquired in real time. These changes include changes in the experimental object's characterization parameters, process and environmental parameters, device and actuator status, experimental stage and trend changes, and safety and risk-related changes. Specifically, the execution feedback module, upon receiving the completion receipt from each actuator, immediately triggers a secondary data acquisition process for multi-dimensional experimental state data. The data acquisition targets five categories: changes in the experimental object's characterization parameters, process and environmental parameters, device and actuator status, experimental stage and trend changes, and safety and risk-related changes. The acquisition method and time are continuous periods after the action is completed to ensure the capture of the true state changes after execution. The experimental object's characterization parameters are obtained by collecting morphological and performance data before and after intervention using online imaging, microscopic detection, and other equipment, and calculating the difference. The data is processed as follows: Changes in process and environmental parameters are collected from real-time parameters after intervention by various sensors, and the amount and rate of change are calculated by combining these with pre-intervention parameters; Changes in device and actuator status are retrieved from the actuator control system, including post-action operating parameters and execution feedback, and the degree of action completion and actual parameter deviation are statistically analyzed; Changes in experimental phases and trends are identified by the experimental phase identification module based on post-intervention status data, analyzing the trends in confidence level and target achievement; Changes related to safety and risk are assessed by the safety monitoring module, evaluating the risk level and safety margin changes after intervention. All collected status change data are transmitted to the data processing module for noise reduction and normalization preprocessing. After removing abnormal data, the data is categorized into five major parameter classes, and the absolute and relative changes of each parameter are calculated. Finally, the data is integrated into a standardized experimental status change dataset, stored in the dedicated data area of ​​the execution feedback module, thus completing the acquisition of post-execution experimental status changes.

[0050] The acquired experimental state changes are compared and analyzed with the intervention targets in the experimental intervention decision to determine whether the intervention effect has achieved the expected results. The intervention effect includes target achievement, convergence speed, stability, or risk level change. Specifically, the effect analysis module extracts the preset intervention targets and corresponding quantitative indicators from the experimental intervention decision, and extracts the state change parameters matching the intervention targets from the dataset. It then performs quantitative calculations on the four intervention effect indicators: target achievement, convergence speed, stability, and risk level change. Target achievement is calculated using the formula: Target Achievement = (Actual Indicator Value After Intervention - Indicator Value Before Intervention) ÷ (Intervention Target Indicator Value - Indicator Value Before Intervention) × 100%. A result of 100% indicates that the target has been fully achieved, and a result below 100% indicates that it has not been achieved. Convergence speed is the amount of change per unit time by which the experimental indicator approaches the target value after the intervention, calculated using the formula: Convergence Speed ​​= (Indicator Value After Intervention - Indicator Value Before Intervention) ÷ (Indicator Value After Intervention) × 100%. The monitoring duration is defined as follows: stability is the fluctuation range of the core experimental indicators within the preset monitoring period after intervention, calculated by the formula: stability = (maximum value of indicator during the monitoring period - minimum value of indicator) ÷ average value of indicator during the monitoring period × 100%. The smaller the value, the more stable the indicator. The risk level change is the difference between the risk level score after intervention and the risk level score before intervention. The risk level score is comprehensively evaluated by the safety monitoring module based on indicators such as the degree of approach to the safety threshold and the operating status of the equipment. The full score is 10 points, and the higher the score, the greater the risk. After completing the quantitative calculation, the system retrieves the preset intervention effect evaluation thresholds: target achievement rate not less than 80%, convergence speed not less than the preset convergence threshold, stability not higher than 10%, and risk level change not greater than 0. If all four indicators meet the evaluation thresholds, the intervention effect is judged to have reached the expected level. If any indicator is not met, or the risk level change is greater than 0, i.e., the risk level increases, the intervention effect is judged to have not reached the expected level. The effect analysis module integrates the judgment results and the quantitative calculation data of each indicator into an intervention effect analysis report and transmits it to the strategy adjustment module.

[0051] When the experimental state changes as expected, the current intervention strategy is maintained, and the post-implementation state changes are used as input state data for the intervention action generation. When the experimental state changes deviate from expectations or the risk level increases, the type, execution parameters, or execution order of the intervention action are automatically adjusted. Adjustments include updating the action type, adjusting the execution intensity range, modifying the execution time, or changing the action execution order. Specifically, this includes: executing the corresponding strategy processing based on the intervention effect judgment result. The strategy adjustment module first parses the judgment result in the intervention effect analysis report: if the intervention effect is judged to have met expectations, the system will maintain the current intervention strategy without any adjustments, and simultaneously synchronize the acquired post-implementation experimental state change dataset to the system data center as... Real-time input status data during the generation of new experimental intervention actions provides the latest experimental status basis for action generation. If it is determined that the intervention effect has not met expectations or the risk level has increased, the system will initiate an automatic adjustment process for the intervention strategy. Based on the specific indicators that have not met expectations, the type, execution parameters, or execution order of the intervention actions will be adjusted accordingly. If the expected results are not met due to insufficient goal achievement, the execution intensity in the original execution parameters will be adjusted. When the intervention effect does not meet expectations, the type, execution parameters, and / or execution order of the experimental intervention actions will be adjusted according to the type of failure to meet expectations. The adjustments may include increasing or decreasing the execution intensity, modifying the execution time, adjusting the order of actions, replacing the type of intervention action, or adding auxiliary intervention actions.

[0052] The adjusted intervention actions and their execution parameters are fed back to trigger a new round of intervention action generation and execution control process, forming a closed-loop control. The execution results of the experimental intervention actions, changes in experimental status, and corresponding adjustment records are stored in a knowledge base for optimizing the intervention action generation and execution control strategy. Specifically, this includes: achieving closed-loop control and completing the knowledge base storage of experimental data; the strategy adjustment module feeding back the verified new intervention strategy scheme, i.e., the adjusted intervention actions and their execution parameters, to the experimental intervention action generation module in real time, triggering a new round of experimental intervention action generation and execution control process. This completes the entire process from re-executing action generation, parameter determination, instruction generation, and execution, forming a closed-loop control of experimental intervention from execution to feedback and then to adjustment. The execution feedback module will record the results of this experimental intervention. The entire process of data is systematically organized, including three parts: the execution results of experimental intervention actions, covering intervention action type, final execution parameters, execution duration, execution mechanism, and execution relationship; experimental state change data, covering quantitative data such as the original values, changes, and rates of change of five major categories of parameters before and after the intervention; and the adjustment records of intervention strategies, covering the reasons for adjustment, the specific content of the adjustment, the comparison of parameters before and after the adjustment, and the basis for the adjustment. All organized data is classified and archived according to experimental type and intervention action type, generating standardized experimental data archives. The relevant execution results, state changes, and adjustment records are saved as execution feedback data for subsequent intervention action generation and execution parameter optimization, providing real and effective historical data basis for the optimization of experimental intervention action generation and execution control strategies.

[0053] In this embodiment of the invention, by comprehensively and standardizedly collecting and preprocessing five categories of experimental state changes after the implementation of the intervention, the data source for evaluating the intervention effect is ensured to be authentic, accurate, and comprehensive. This avoids misjudgment of the effect due to incomplete state perception at the data level and solves the problem of untimely and incomplete capture of experimental state changes in traditional technologies. By quantifying the four core indicators of the intervention effect and setting clear evaluation thresholds, a scientific and objective judgment of the intervention effect is achieved, overcoming the limitations of relying on human experience to judge the intervention effect in traditional technologies and making the effect judgment more quantitatively based. By implementing differentiated strategy processing based on the intervention effect judgment results, targeted and formulaic parameter adjustments are made for situations where expectations are not met, realizing the dynamic nature of the intervention strategy. The optimization effectively addresses the shortcomings of traditional techniques where intervention actions, once executed, cannot be adjusted and are difficult to adapt to changes in experimental states. By triggering a new round of processes through feedback of the adjusted intervention strategy, a complete closed-loop control system is constructed, from execution to feedback, adjustment, and execution. This allows the intervention to continuously align with the real-time experimental state, improving its effectiveness and accuracy. By organizing, archiving, and storing the entire intervention process data in a knowledge base, effective data accumulation and retention are achieved, providing rich historical data support for intervention strategy optimization. This enables the system to have self-learning and self-optimization capabilities, gradually improving the intelligence level of the generation and execution control of intervention actions. It also provides a reference for intervention operations in similar experiments, effectively enhancing the overall adaptability of the experimental system.

[0054] In a preferred embodiment of the present invention, step 5 above may include: The system acquires parameters representing changes in the experimental object, including changes in structure or morphology, color or optical characteristics, location or spatial distribution, and object state. Changes in structure or morphology include size, morphological type, surface texture, boundary features, degree of aggregation or dispersion, and phase or crystal form. Changes in color or optical characteristics include color space characteristics, brightness or contrast, reflection or transmission characteristics, and fluorescence or spectral characteristics. Changes in location or spatial distribution include location migration, distribution uniformity, regional proportion, and sedimentation or drift trends. Changes in object state include appearance or disappearance, generation or consumption, aggregation or deaggregation, precipitation or dissolution, and the appearance of intermediate stages. Specifically, the system acquires these parameters using online imaging equipment, microscopic detection devices, spectrometers, particle size analyzers, and other specialized detection equipment to collect full-dimensional characterization data of the experimental object before and after the intervention. It then performs quantitative calculations and state determination for each of the four types of parameters: changes in structure or morphology, changes in color or optical characteristics, changes in location or spatial distribution, and changes in object state. For structural or morphological changes, the size change is calculated as the size after intervention minus the size before intervention. The degree of aggregation or dispersion is represented by the difference in the ratio of the number of aggregated particles to the total number of particles. Changes in crystal form are determined by the matching degree of characteristic diffraction peaks. Simultaneously, visual changes in morphology type, surface texture, and boundary features are recorded. For color or optical feature changes, the Lab values ​​of color space features are extracted to calculate the difference before and after intervention. Brightness or contrast is represented by the rate of change of grayscale values, calculated as (grayscale value after intervention - grayscale value before intervention) ÷ grayscale value before intervention × 100%. Simultaneously, peak changes in reflection / transmission and fluorescence / spectral features are collected. For position or space... The distribution changes are represented by the coordinate difference of the centroid of the experimental object, and the uniformity of distribution is determined by the coefficient of variation of the parameter, which is the standard deviation of the data divided by the mean × 100%. The area proportion is the difference in the area of ​​the target area before and after intervention. At the same time, sedimentation or drift trends are identified. For changes in the state of the object, quantitative data and qualitative judgment are combined. Precipitation / dissolution is represented by the volume proportion of precipitates or dissolves, and generation / consumption is represented by the change in substance concentration. At the same time, state changes such as polymerization / depolymerization and the appearance / disappearance of intermediate stages are directly determined. The system integrates the quantitative data of the four types of parameters with the state judgment results to form a standardized set of parameters representing changes in the experimental object.

[0055] The system acquires changes in process and environmental parameters, including changes in process parameter trajectories and disturbance response characteristics. Changes in process parameter trajectories include the amount, rate, fluctuation, and stability of changes in temperature, pressure, humidity, atmosphere or gas concentration, pH, conductivity, and viscosity. Disturbance response characteristics include the response delay time after intervention, the time required to reach steady state, and the magnitude of overshoot or undershoot. Specifically, the system acquires real-time parameters of the experimental process and environment after intervention using sensors such as temperature, humidity, pressure, gas concentration, pH, conductivity, and viscosity at preset time intervals, forming time-series data. Then, it calculates quantitative indicators for changes in process parameter trajectories and disturbance response characteristics. For changes in process parameter trajectories, the change in each parameter is first calculated as the real-time parameter after intervention minus the baseline parameter before intervention. The rate of change is the change divided by the data acquisition time interval. The volatility is the standard deviation of the parameter within the preset monitoring window divided by the mean × 100%. The stability is 1 minus the volatility. The closer the value is to 1, the more stable the parameter is. At the same time, the complete time series trajectory of core parameters such as temperature and pressure is recorded. For disturbance response characteristics, the response delay time is the duration from the completion of the intervention command to the start of parameter change. The time required to reach steady state is the duration from the start of parameter change to volatility ≤ 5%. The overshoot amplitude is the absolute difference between the peak value and the target value during the parameter change process. The undershoot amplitude is the absolute difference between the trough value and the target value during the parameter change process. After all indicators have been accurately timed and quantified, they are integrated into a set of process and environmental parameter changes.

[0056] The system acquires changes in the status of devices and actuators, including changes in execution status and feedback, and deviations in actual parameters. Changes in execution status and feedback include action completion rate, execution time, exception codes, timeouts, failure reasons, and number of retries. Actual parameter deviations include the deviation between planned and actual parameters and how this deviation changes over time. Specifically, the system acquires changes in the status of devices and actuators by retrieving full operational data in real time from the control systems of various actuators such as stirring, temperature control, and feeding after the intervention action. It then statistically analyzes changes in execution status and feedback, calculates actual parameter deviations, and for changes in execution status and feedback, the action completion rate is calculated by dividing the actual number of execution steps by... The execution steps are calculated by multiplying the planned number of steps by 100%, and the execution time, system-generated exception code type, number of timeouts, specific reasons for action failures, and number of equipment retries are accurately recorded to form a complete record of the execution status. For actual parameter deviations, the actual parameter deviation at a single point in time is first calculated as the actual execution parameter at that point in time minus the planned execution parameter. Then, the deviation values ​​at different time points are collected to fit the deviation change curve over time. At the same time, the average deviation and the maximum deviation are calculated. The average deviation is the arithmetic mean of the absolute values ​​of the deviations at each time point, and the maximum deviation is the maximum value of the absolute values ​​of the deviations at each time point, fully presenting the dynamic change characteristics of the parameter deviations. Finally, these are integrated into a set of device and actuator status changes.

[0057] The system acquires experimental stage and trend-related changes, including stage-related changes and trend or convergence changes. Stage-related changes include changes in stage identification confidence, changes in the degree of satisfaction of stage boundary criteria, and the appearance or disappearance of stage switching trigger events. Trend or convergence changes include increases or decreases in goal achievement, changes in convergence speed, changes in information gain, and changes in uncertainty. Specifically, the system acquires experimental stage and trend-related changes by having the experimental stage identification module perform quantitative calculations of stage-related changes and dynamic analysis of trend or convergence changes based on full-dimensional experimental state data after intervention. For phase-related changes, the change in phase identification confidence is first calculated as the post-intervention phase identification confidence minus the pre-intervention confidence. The degree of satisfaction of phase boundary criteria is calculated as the number of phase boundary criteria indicators satisfied divided by the total number of criteria indicators × 100%. Simultaneously, the occurrence or disappearance of phase switching trigger events is identified through data, and the specific indicator basis for event triggering is recorded. For trend or convergence changes, the increase or decrease in target achievement is calculated as the post-intervention target achievement minus the pre-intervention target achievement. The convergence speed is calculated as the change in target achievement divided by the monitoring duration after intervention. The information gain is calculated as the information entropy of the experimental state after intervention minus the information entropy before intervention. Uncertainty is represented by 1 minus the phase identification confidence; the larger the value, the higher the uncertainty in the determination of the experimental phase. After all quantitative indicators are calculated, they are integrated with the trend analysis results to form a set of experimental phase and trend-type changes.

[0058] The system acquires safety and risk-related changes, including changes in risk level or safety margin, and changes in safety linkage status. Changes in risk level or safety margin include increases or decreases in risk level, changes in the degree of approach to safety thresholds, and changes in hazardous situation indicators. Changes in safety linkage status include whether safety actions such as warnings, suspensions, or restrictions are triggered, along with their triggering reasons and effective scope. Specifically, the system acquires safety and risk-related changes by combining the safety monitoring module with comprehensive status change data of the experimental object, environment, and equipment to quantitatively calculate changes in risk level or safety margin. Simultaneously, it statistically analyzes changes in safety linkage status. For changes in risk level or safety margin, each risk factor in the experiment is first scored according to a preset scoring system, and the risk level is calculated by weighting the scores. The maximum score is set to [missing information]. The risk level is calculated as follows: 10 points. The higher the score, the greater the risk. The risk level change is calculated by subtracting the risk level before intervention from the risk level after intervention, and then calculating the safety margin of each core parameter. The safety margin is calculated as (safety threshold - actual parameter value) ÷ safety threshold × 100%. The degree of approach to the safety threshold is 1 minus the safety margin. The closer the value is to 1, the closer it is to the safety threshold. The danger status indicator is the sum of the scores of each risk factor multiplied by their corresponding weights. The sum of the weights of each risk factor is 1. For changes in the safety linkage status, the system directly determines whether the intervention triggers safety actions such as warnings, suspension, or restriction of execution. At the same time, it accurately records the triggering reasons of the safety actions, the experimental parameters or implementing agencies involved, and the time and space range of the effect. Finally, the quantitative data and status judgment results are integrated into a set of safety and risk-related changes.

[0059] Experimental state changes are categorized into absolute state values, relative changes, rates of change or trends, window statistical features, and event-type changes. Relative changes include the difference or ratio before and after intervention. Rates of change or trends include first or second-order differences, slopes, and volatility. Window statistical features include mean, variance, peak value, and steady-state determination. Event-type changes include occurrence, disappearance, boundary crossing, regression, and abrupt change. Specifically, all experimental state change data are expressed in a unified format. The system's data processing module extracts the raw data and quantitative indicators of all five types of experimental state changes, performs standardization transformation and expression according to five forms: absolute state values, relative changes, rates of change or trends, window statistical features, and event-type changes. Real-time values ​​of all parameters after intervention are extracted as absolute state values, serving as the basis for data expression. Next, relative changes are calculated, with the difference being the parameter value after intervention minus the parameter value before intervention, and the ratio... The value is calculated by dividing the parameter value after intervention by the parameter value before intervention, and the ratio is discarded when the parameter value is 0. Then, the rate of change or trend is analyzed. The first-order difference is used to represent the parameter difference between adjacent time points, and the second-order difference is used to represent the adjacent difference of the first-order difference. The slope of the fitted curve is used to represent the overall trend of parameter change, and the volatility index of each parameter is retained. Next, the window statistical characteristics within the preset time window are calculated, including the arithmetic mean, variance, and peak value of the parameters. At the same time, the steady state is determined according to the standard of volatility ≤5% within the window. If the standard is met, it is marked as steady state; otherwise, it is marked as non-steady state. All qualitative state changes are labeled as event-type changes, including types such as appearance, disappearance, boundary crossing, regression, and mutation. At the same time, corresponding quantitative data are matched as supporting evidence. The system encodes and classifies all parameters in five forms to form a standardized, comparable, and analyzable experimental state change expression dataset.

[0060] In this embodiment of the invention, the experimental state changes are collected and quantified in a comprehensive and detailed manner across five dimensions: experimental object characterization, process and environment, device and actuator, experimental stage and trend, and safety and risk. Specific numerical calculations or state determinations are performed for the sub-parameters of each dimension. This overcomes the shortcomings of traditional technologies, such as incomplete capture of experimental state changes, low quantification, and the inability to make only intuitive qualitative judgments. This makes the feedback data on experimental state changes more accurate and specific. Simultaneously, by setting five unified expression formats, all state change data are standardized and transformed, solving the problems of chaotic experimental state data formats, inconsistent dimensions, and difficulty in comparison, analysis, and subsequent utilization in traditional technologies. This provides a unified analytical benchmark for the feedback data, ultimately achieving comprehensiveness, quantification, and standardization of experimental state change feedback data. This provides reliable quantitative data support for the scientific determination of intervention effects and the targeted adjustment of intervention strategies, effectively ensuring the scientificity and accuracy of experimental intervention closed-loop control. Furthermore, the standardized data format makes it easier to store experimental data in a knowledge base for correlation analysis and accumulation, laying a solid data foundation for system self-learning and continuous optimization of intervention control strategies.

[0061] like Figure 2 As shown, embodiments of the present invention also provide an artificial intelligence-based experimental intervention action generation and execution control system, comprising: The experimental intervention decision acquisition module is used to acquire experimental intervention decisions, which are derived from artificial intelligence decision-making systems or human input, and are used to indicate the need for intervention in the experimental process. The intervention action generation module is used to automatically generate experimental intervention actions that match the experimental status data based on experimental intervention decisions and in conjunction with real-time experimental status data. The experimental status data includes the morphological or performance data of the experimental object, experimental environment parameter data, experimental device operating parameter data, and experimental process status data. The experimental intervention actions are generated from predefined intervention action types according to the matching of the experimental status data type with preset conditions or target trajectories. The predefined intervention action types include parameter adjustment actions, condition switching actions, substance introduction or removal actions, and experimental process control actions. The execution parameter determination module is used to process experimental intervention actions and determine corresponding execution parameters for each experimental intervention action to obtain the determined execution parameters; the execution parameters include execution time, execution order, execution intensity or execution range; The experimental intervention execution module is used to drive the execution mechanism to complete the corresponding experimental intervention actions according to the determined execution parameters; The execution feedback module is used to obtain the changes in experimental status after the execution of experimental intervention actions, and to use the changes in experimental status to adjust and optimize the intervention actions. The changes in experimental status include changes in the parameters representing the experimental object, changes in process and environmental parameters, changes in the status of the device and actuator, changes in experimental stages and trends, and changes related to safety and risks.

[0062] The thresholds, coefficients, ratios, time windows, sampling periods, evaluation criteria, and correction rules described in this article can all be preset or adaptively adjusted according to the experiment type, experimental object, device capabilities, and historical execution data, and do not constitute limitations.

[0063] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for generating and controlling the execution of experimental intervention actions based on artificial intelligence, characterized in that, The method includes: Obtain experimental intervention decisions, which are derived from artificial intelligence decision-making systems or human input, to indicate the need for intervention in the experimental process; Based on experimental intervention decisions and combined with real-time experimental status data, experimental intervention actions that match the experimental status data are automatically generated. The experimental status data includes the morphology or performance data of the experimental object, experimental environment parameter data, experimental device operating parameter data, and experimental process status data. The experimental intervention actions are generated from predefined intervention action types based on the matching of the experimental status data type with preset conditions or target trajectories. The predefined intervention action types include parameter adjustment actions, condition switching actions, substance introduction or removal actions, and experimental process control actions. The experimental intervention actions are processed, and corresponding execution parameters are determined for each experimental intervention action to obtain the determined execution parameters; the execution parameters include execution time, execution order, execution intensity or execution range; Based on the determined execution parameters, drive the actuator to complete the corresponding experimental intervention action; The experimental state changes after the implementation of the experimental intervention are obtained and used to adjust and optimize the intervention. The experimental state changes include changes in the parameters representing the experimental subjects, changes in process and environmental parameters, changes in the state of the devices and actuators, changes in experimental stages and trends, and changes related to safety and risks.

2. The method according to claim 1, characterized in that, The experimental intervention decision originates from at least one of the following: upstream artificial intelligence decision-making module, path planning module, stage identification module, risk control module, or manual input.

3. The method for generating and controlling experimental intervention actions based on artificial intelligence according to claim 2, characterized in that, Based on experimental intervention decisions and combined with real-time experimental status data, experimental intervention actions are automatically generated that match the experimental status data. The experimental status data includes morphological or performance data of the experimental object, experimental environment parameter data, experimental apparatus operating parameter data, and experimental process status data, including: Real-time acquisition of experimental status data, including morphological or performance data of experimental objects, experimental environment parameter data, experimental device operating parameter data, and experimental process status data; Based on the intervention objectives and constraints included in the experimental intervention decision, the preset triggering conditions are dynamically adjusted to determine the currently applicable set of triggering conditions; The experimental status data acquired in real time is matched and judged against each trigger condition in the set of currently applicable trigger conditions; When the experimental state data meets a certain triggering condition, a corresponding candidate intervention action is generated from the predefined intervention action types according to the intervention action type corresponding to the triggering condition, and the basic attributes of the intervention action are determined. The correspondence between the preset triggering condition and the intervention action type includes: when the sensor parameters, image features or execution deviations in the experimental state data meet the parameter adjustment triggering condition, a parameter adjustment action is generated, and the type of parameter to be adjusted is determined. When the stage change trend and risk boundary in the experimental status data approach the condition switching trigger condition, a condition switching action is generated, and the type of condition to be switched is determined; when the composition concentration and morphological color changes in the experimental status data meet the substance introduction or removal trigger condition, a substance introduction or removal action is generated, and the substance type and operation type are determined; when the process events and human-machine collaboration requirements in the experimental status data meet the process control trigger condition, an experimental process control action is generated, and the control type is determined.

4. The method according to claim 3, characterized in that, Determining execution parameters for the candidate intervention action includes: Obtain candidate intervention actions and their basic attributes; The execution parameter range for each candidate intervention action is determined based on real-time experimental status data and intervention constraints. Within the range of execution parameters, determine the execution time, execution order, execution intensity, and / or execution range of each candidate intervention action.

5. The method according to claim 4, characterized in that, In some implementations, the method also includes estimating the execution effect corresponding to the candidate execution parameters based on historical execution data, rule models, experience models, simulation models, or knowledge bases, and optimizing the execution parameters based on the estimation results.

6. The method according to claim 4 or 5, characterized in that, Perform experimental intervention actions according to the execution parameters, including: Determine the execution relationship between multiple experimental intervention actions based on the execution order; Generate execution instructions based on action type and execution parameters; The execution command is sent to the corresponding execution mechanism to complete the experimental intervention.

7. The method according to claim 6, characterized in that, During the execution of experimental intervention actions, constraint verification is performed to ensure that the execution parameters fall within the allowable execution boundaries. When the actual execution is detected to deviate from the allowable execution boundaries, parameter correction, execution suspension, or re-issuance of execution instructions are performed.

8. The method according to claim 6 or 7, characterized in that, In some implementations, a manual confirmation request is generated before the experimental intervention is performed, and the intervention is executed or the execution parameters are adjusted based on the manual confirmation result.

9. The method according to any one of claims 6 to 8, characterized in that, Also includes: Obtain changes in experimental status after the implementation of experimental intervention actions; The changes in the experimental state are compared and analyzed with the intervention target to determine whether the intervention effect has achieved the expected results. When the intervention effect does not meet expectations, adjust the type of experimental intervention, execution parameters, and / or execution order; The adjusted experimental intervention actions and / or execution parameters are fed back to the intervention action generation and execution control process to form a closed-loop control.

10. An artificial intelligence-based experimental intervention action generation and execution control system, wherein the system implements the method as described in any one of claims 1 to 7, characterized in that, include: The experimental intervention decision acquisition module is used to acquire experimental intervention decisions, which are derived from artificial intelligence decision-making systems or human input, and are used to indicate the need for intervention in the experimental process. The intervention action generation module is used to automatically generate experimental intervention actions that match the experimental status data based on experimental intervention decisions and in conjunction with real-time experimental status data. The experimental status data includes the morphological or performance data of the experimental object, experimental environment parameter data, experimental device operating parameter data, and experimental process status data. The experimental intervention actions are generated from predefined intervention action types according to the matching of the experimental status data type with preset conditions. The predefined intervention action types include parameter adjustment actions, condition switching actions, substance introduction or removal actions, and experimental process control actions. The execution parameter determination module is used to process experimental intervention actions and determine corresponding execution parameters for each experimental intervention action to obtain the determined execution parameters; the execution parameters include execution time, execution order, execution intensity or execution range; The experimental intervention execution module is used to drive the execution mechanism to complete the corresponding experimental intervention actions according to the determined execution parameters; The execution feedback module is used to obtain the changes in experimental status after the execution of experimental intervention actions, and to use the changes in experimental status to adjust and optimize the intervention actions. The changes in experimental status include changes in the parameters representing the experimental object, changes in process and environmental parameters, changes in the status of the device and actuator, changes in experimental stages and trends, and changes related to safety and risks.

Images