Laboratory agent construction method and construction system based on artificial intelligence

Abstract

The invention discloses a laboratory intelligent agent construction method and system based on artificial intelligence, and relates to the technical field of intelligent laboratories. In order to solve the problems of scattered laboratory data information, low scientific and technological resource matching efficiency, lack of a unified AI driving platform and the like in the prior art, a core agent is constructed by introducing machine learning or an artificial intelligence architecture, and digital access and resource pooling are carried out on scientific research total factors such as infrastructures, software and hardware, human property and the like. The intelligent agent is based on semantic alignment and inference rules, dynamic intelligent configuration of resources such as field observation, experimental analysis and computational simulation is achieved, and finally a full-life-cycle intelligent management and scientific research environment integrating equipment perception, intelligent control and business application is formed. According to the method, resource islands are effectively broken, the resource utilization efficiency is improved, a novel scientific research normal form of'big data + AI + model 'is promoted, and systematic support is provided for refined and automatic management of laboratories.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of laboratory automation, in particular to a laboratory intelligent agent construction method and system based on artificial intelligence. BACKGROUND

[0002] In modern scientific research and industrial development, automated laboratory systems have played a crucial role. They can perform high-throughput experimental processes around the clock without interruption by precisely controlling robotic arms, pipetting stations, analytical instruments and other execution units, greatly improving scientific research efficiency and data repeatability. The core logic of current mainstream laboratory control systems is usually based on pre-set, static programmed scripts or workflows. However, the intelligence level of such systems is relatively limited. They are essentially obedient instruction executors and lack effective response capabilities for undefined abnormal situations in programs or logic, especially for potential risks caused by gradual device wear and tear, minor environmental changes or indirect coupling between multiple tasks. They basically have no ability to predict and avoid, and their error correction mechanism is usually triggered passively after the fault actually occurs. SUMMARY

[0003] In view of the above existing problems, the present application is proposed.

[0004] Therefore, the present application provides a laboratory intelligent agent construction method based on artificial intelligence, which solves the problems of lack of effective response capabilities for undefined abnormal situations in programs or logic and lack of ability to predict and avoid.

[0005] To solve the above technical problems, the present application provides the following technical solutions:

[0006] In a first aspect, the present application provides a laboratory intelligent agent construction method based on artificial intelligence, which includes the following steps:

[0007] Step 1: Construct a semantic model that integrates real-time state data of laboratory equipment and common operation and operation knowledge between laboratory equipment;

[0008] Step 2: Based on the semantic model, analyze the received scientific research target according to the semantic model based on the semantic model, and output a certain result as an optimal execution plan containing specific operations, resource allocation and execution timing;

[0009] Step 3: According to the optimal execution plan, generate and distribute control instructions to drive the automated execution units in the laboratory to work cooperatively;

[0010] Step 4: During the execution, the actual state feedback of the automation execution unit is continuously monitored, and when a preset condition deviation between the actual state and the optimal execution plan is detected, an instant re-planning and dynamic adjustment of the optimal execution plan is automatically triggered.

[0011] As a preferred scheme of the laboratory intelligent agent construction method based on artificial intelligence, wherein: the semantic model is constructed by fusing real-time state data of laboratory equipment and conventional operation and operation common sense between laboratory equipment, and the specific steps are as follows:

[0012] Through a standardized data interface layer, real-time state data of current attributes of various equipment in the laboratory is continuously collected, and a set of structured digital rules are constructed in advance to clearly specify how all physical equipment cooperate, the order of operation, and the formalized structure of their respective capability limits and safety red lines. Then, through a dynamic association mechanism, the real-time state data is bound to the corresponding device expression in the formalized knowledge structure as the attribute value of the device, so as to form a computable model that integrates the real-time state of the laboratory equipment and the judgment logic.

[0013] As a preferred scheme of the laboratory intelligent agent construction method based on artificial intelligence, wherein: based on the semantic model, the received scientific research task target is analyzed according to the semantic model, and a certain result is output, which is an optimal execution plan containing specific operations, resource allocation and execution timing, and the specific steps are as follows:

[0014] First, the programmed framework in the semantic model is constructed, and through the constructed programmed framework, the scientific research task target is decomposed into an ordered operation sequence composed of multiple operations with logical association programs. Then, the operations in the ordered operation sequence are matched and assigned a specific execution device that is currently available and has adaptive capability by querying the real-time attributes of entities in the laboratory in the semantic model. Finally, in the candidate execution path formed by all resource assignment combinations, at least one preset optimization index is used for evaluation and screening to determine the optimal execution plan.

[0015] As a preferred scheme of the laboratory intelligent agent construction method based on artificial intelligence, wherein: the control instructions are generated and distributed according to the optimal execution plan to drive the automation execution unit in the laboratory to perform collaborative work, and the specific steps are as follows:

[0016] By performing the state feedback of the execution unit and the expected state, the deviation event is identified by continuous comparison, and after the deviation event is identified, the actual changes in the laboratory caused by the deviation event are updated in the semantic model in real time, and based on the updated semantic model and the current actual system state, a process oriented to complete the original scientific research task target is re-planned to generate a new execution plan.

[0017] As a preferred scheme of the laboratory intelligent agent construction method based on artificial intelligence, wherein: during the execution, the actual state feedback of the automatic execution unit is continuously monitored, and when the deviation between the actual state and the optimal execution plan under the preset condition is detected, the optimal execution plan is automatically triggered for real-time re-planning and dynamic adjustment, and the specific steps are as follows:

[0018] The semantic model is a unified knowledge structure, in which the real-time state of each laboratory entity is adapted to a set of pre-set logical rules about how entities interact, in what order to act, and what safety restrictions to follow, and the consequences of all decisions are deduced and calculated, and;

[0019] The attributes of each device itself are defined, the fixed spatial position layout, logical subordination and collaborative functions realized when they are combined together are defined, and the formal description of the executable operation of each entity in the laboratory is also defined, which includes the system state conditions required to perform the operation and the expected changes to the system state after the operation is completed.

[0020] The operations in the optimal execution plan are parsed and converted into the action step list corresponding to the specified execution unit, and based on the semantic model, the problem of collision between two or more automated execution units using the same resources at the same time is avoided, and the collaborative scheduling strategy is implemented. The synchronization mechanism is synchronized by issuing instructions and state execution feedback to ensure accurate execution of the entire plan according to the predetermined timing, and after the deviation between the actual system state and the optimal execution plan is identified by continuous monitoring, the real changes in the laboratory caused by the deviation are synchronized to the semantic model, the updated semantic model is used as the starting condition for re-planning, and the original task target that has not changed is used as the end point for re-planning to generate a new adjusted execution plan to replace the original plan.

[0021] In a second aspect, the present application provides a laboratory agent construction system based on artificial intelligence, characterized in that a unified knowledge model module is configured to integrate ontology rules for defining static attributes and physical constraints of laboratory entities, real-time sensing data for representing instantaneous states of the laboratory entities, and predictive logic for describing causal correlations between operations and potential execution biases;

[0022] a forward-looking planning algorithm module connected to the unified knowledge model module and configured to generate at least one candidate execution path based on a received scientific research task objective, and to deduce the candidate execution path using the predictive logic to predict a probability distribution of bias events occurring in the future execution of the path, thereby outputting a quantitative risk profile containing the probability distribution;

[0023] a decision optimization module configured to receive the quantitative risk profile and determine an optimal execution plan from the at least one candidate execution path according to a preset strategy that combines execution efficiency and the risk profile;

[0024] and a closed-loop execution and model adaptation module configured to generate control instructions to drive automated equipment in the laboratory according to the optimal execution plan, and to continuously monitor actual execution states fed back by the automated equipment, and to automatically correct the predictive logic in the unified knowledge model module when there is a discrepancy between the actual execution states and the prediction results of the forward-looking planning algorithm module, so as to reduce future prediction biases.

[0025] In a third aspect, as a preferred solution of the laboratory agent construction system of the present application, wherein: the forward-looking planning algorithm module is further configured to: map the candidate execution path into an ordered sequence composed of multiple operation units, and for each operation unit in the sequence, extract a set of mutually exclusive potential results including at least one success result and one or more bias results from the predictive logic of the unified knowledge model module, wherein each potential result is assigned a corresponding occurrence probability;

[0026] and by iteratively performing probability deduction on the sequence, randomly determine the execution result of each operation unit according to the occurrence probability in each deduction, and integrate the statistical results of multiple deductions to generate the quantitative risk profile, which at least represents the overall execution success expectation value of the candidate execution path and the specific operation unit with the highest bias risk.

[0027] In a fourth aspect, as a preferred solution of the laboratory intelligent agent construction system, the closed-loop execution and model self-adaptation module is further configured to: parse the optimal execution plan into a set of underlying control instructions that ensure operation synchronization and resource exclusivity among the plurality of automated devices, and generate a deviation event record containing snapshots of the system state before and after the deviation when the actual execution state deviates from any possible result predicted by the forward-looking planning algorithm module, and then use the deviation event record as feedback to adjust the predictive logic in the unified knowledge model module to integrate the unforeseen causal relationships revealed by the deviation event into future predictions.

[0028] The present application has the following advantages:

[0029] The present application realizes the intelligent leap from "passive execution" to "active decision-making": it can autonomously understand scientific research goals and plan the optimal path in advance, greatly reducing the dependence on manual programming, while actively avoiding potential risks; it gives the system strong robustness and fault self-healing ability: in the face of execution deviation, the system can quickly re-plan through closed-loop to autonomously adjust the strategy to ensure task completion, improving the success rate and stability of the automated process; it realizes the global optimization and efficient collaboration of laboratory resources: through unified knowledge model for global scheduling, it effectively avoids resource conflicts and motion collisions, maximizes equipment utilization, thereby significantly shortens the experimental period and improves the overall operation throughput. BRIEF DESCRIPTION OF DRAWINGS

[0030] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0031] Figure 1 Flow chart for the laboratory intelligent agent construction method based on artificial intelligence.

[0032] Figure 2 Flow chart for constructing semantic model.

[0033] Figure 3 Flow chart for generating optimal plan.

[0034] Figure 4 Flow chart for closed-loop execution and dynamic re-planning. DETAILED DESCRIPTION

[0035] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0036] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be appreciated that the present application can be practiced in a variety of ways beyond the specific embodiments described herein without departing from the present application. Thus, the present application is not to be limited by the precise details set forth herein.

[0037] Second, as used herein, the term "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one implementation of the present application. The appearances of the phrase "in one embodiment" or "an embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments.

[0038] Embodiment 1

[0039] The present application proposes an artificial intelligence-based laboratory agent construction method, comprising the following steps:

[0040] Step 1: Construct a semantic model that integrates real-time state data of laboratory equipment and common operation and operation common sense between laboratory equipment. Through a standardized data interface layer, continuously collect real-time state data of current attributes of various equipment in the laboratory, and pre-construct a set of structured digital rules to clearly specify how all physical equipment cooperate, the order of operation, and the form structure of their respective capability limits and safety red lines. Then, through a dynamic association mechanism, bind the real-time state data as the attribute value of the equipment to the corresponding equipment expression in the formalized knowledge structure in real time, thereby forming a computable model that integrates the instantaneous state of laboratory equipment and judgment logic.

[0041] The construction of the model includes the following steps:

[0042] First step: Define entity categories and attribute framework.

[0043] A structured entity classification system is established in advance to standardize the description of all objects in the laboratory. The system includes at least top-level categories such as "equipment", "consumables", and "reagents", and can be further subdivided as needed. Define the inherent static attributes (such as model, physical size, maximum range) and variable dynamic attributes (such as current location, temperature, and possession) of each category.

[0044] Second step: Instantiate specific entities and construct static topology.

[0045] According to the actual physical configuration of the laboratory, the class defined in the first step is instantiated as a specific entity node. For example, a unique entity node named "pipette 1" is created for a physical pipetting workstation, and it is given static attribute values. At the same time, these nodes are connected by creating directed edges with types to solidify the static relationships between them, including at least:

[0046] Spatial relationship: describes the position or relative layout of entities in the coordinate system.

[0047] Subordinate relationship: describes the component relationship between entities, such as a rotor node subordinate to a centrifuge node.

[0048] Step 3: Formally define the executable operations of the entity.

[0049] For device entity nodes with action capabilities (such as pipetting workstations, robotic arms), one or more formal operation descriptions are added. Each operation description is a data structure that explicitly contains the following three core elements:

[0050] Operation name: a unique identifier for identifying the operation, such as Extract.

[0051] Preconditions: one or more logical conditions that must be met by the dynamic attributes of the relevant entity nodes before the operation is executed, to specify "how entities interact, in what order actions are taken, and what safety restrictions are followed". For example, the preconditions for the robotic arm grasping operation may include "the robotic arm gripper is empty" and "the target object is within reach", which are both sequence rules and safety rules.

[0052] Expected effect: the deterministic change to the dynamic attribute values of one or more entity nodes after the operation is successfully executed.

[0053] Step 4: Establish a dynamic data association mechanism.

[0054] Deploy a dynamic association service that connects to all relevant physical sensors and actuators in the laboratory through a standardized data interface. The function of this service is to continuously listen to real-time state data from the interface, and according to the pre-set mapping rules, update the received data values to the dynamic attributes of the corresponding entity nodes in the model in real time and accurately.

[0055] By executing the above four steps, a complete, dynamic and computable and queryable laboratory semantic model is constructed, providing the necessary data and logical basis for the subsequent steps of the invention.

[0056] Step 2: Based on the semantic model, analyze the received scientific research task target according to the semantic model, and output the corresponding result, which is an optimal execution plan containing specific operations, resource allocation, and execution timing. First, build a programmed framework in the semantic model, and then decompose the scientific research task target into an ordered operation sequence composed of multiple operations with logical association programs through the programmed framework. Then, match and assign a specific execution device that is currently available and has appropriate capabilities to the operation in the ordered operation sequence by querying the real-time attributes of entities in the laboratory in the semantic model. Finally, among all the candidate execution paths formed by the resource assignment combinations, evaluate and filter according to at least one preset optimization index to determine the optimal execution plan.

[0057] characterized in that the preset optimization index is a cost function defined as

[0058] The expression is:

[0059]

[0060] where P represents a complete candidate execution path; T(P) is the total time consumption of execution path P; C(P) is the resource consumption cost of execution path P, including reagents, consumables, etc.; and R(P) is the estimated failure risk of execution path P. , , are preset weight coefficients corresponding to time, cost, and risk, respectively, and The specific values are determined by different experimental scenarios and determine different priority strategies.

[0061] The construction of this programmed framework can be divided into the following six core steps:

[0062] Step 1: Establish a device driver layer and define a unified interface. Use an object-oriented programming language to define a base class and implement specific drivers. For each physical device in the laboratory, create a specific class that inherits from the above base class. Create a driver manager. When the system starts, it will instantiate all the specific driver classes required according to the configuration file and register them in a dictionary with the device unique identifier in the semantic model as the key and the corresponding driver object instance as the value.

[0063] Step 2: Formally define the planning domain by creating a domain static file to define the general rules in the entire system.

[0064] Third step: implement the task planner module, choose a ready-made, PDDL-supported open-source or commercial planning solver, and write a wrapper program responsible for interacting with the planning solver.

[0065] Fourth step: build the instruction dispatcher and executor module, create an execution loop, traverse each item in the plan list generated by the task planner in order, for each atomic operation in the list, the executor parses the operation name (aspirate), device name (Pipettor1), and parameters (0.5ml), uses the device name to find the corresponding device driver object in the "driver manager" created in the first step, calls the corresponding method on the object, after calling the driver, the executor will block and wait until the driver returns a "complete" or "failed" signal, then continue to execute the next step in the plan.

[0066] Fifth step: implement the state monitor and deviation analysis module to establish a feedback mechanism, subscribe to changes in key dynamic attributes, after the "executor" executes each step, actively query the semantic model to get the latest state updated by the sensor, for each operation, the module compares the actual state (empirical data based on physical measurement) and the expected state (prediction based on logical model), when the deviation exceeds the preset threshold, the module will immediately publish a "deviation event" with detailed deviation information.

[0067] Sixth step: build the dynamic re-planning controller module, the goal is to give the system the intelligence to handle exceptions. This module is an event listener that specifically listens for "deviation events" published by the "state monitor", when a deviation event is received, the controller immediately performs the following operations: a. send a "pause" instruction to the "instruction executor" to stop the current plan execution, b. trigger the "task planner module" and set the current semantic model state containing the deviation as the new initial state, and the original final target as the goal, request a re-planning, c. when the planner returns a new plan, the controller hands over the new plan to the "instruction executor" and sends a "resume" instruction to start execution from the new plan.

[0068] Through the above six steps, a modular, scalable, closed-loop feedback and dynamic error correction capable programmatic framework is built.

[0069] ​Step 3: According to the optimal execution plan, generate and distribute control instructions to drive the collaborative work of the automation execution units in the laboratory. Through the state feedback of the execution units and the expected state, continuously compare to identify deviation events. After identifying the deviation event, update the actual changes in the laboratory caused by the deviation event to the semantic model in real time, and based on the updated semantic model and the current actual system state, re-plan a process oriented to complete the original scientific research task target to generate a new execution plan.

[0070] The feature is that the deviation event is identified by a state deviation metric The expression is:

[0071] Where:

[0072] is the state vector collected from the sensor, describing the current actual state of the system;

[0073] is the expected state vector of the system at the current time, derived according to the optimal execution plan;

[0074] ||| represents the vector norm calculation;

[0075] iff When and only when > ε, a deviation event is identified and triggers the subsequent re-planning process.

[0076] Where the specific value of the vector Reference:

[0077] State vector is a "digital snapshot" containing the state of key equipment and items in the laboratory, which is a list (i.e. vector) composed of multiple numerical values directly from sensors or derived by plans.

[0078] A typical state vector contains the following information:

[0079]

[0080] The generation and control instructions are constructed with the following steps:

[0081] Step 1: Reception and dependency analysis of the execution plan

[0082] The cooperative job execution engine receives the execution plan generated by the task planner module, which is essentially an ordered list of atomic operations. The engine first performs dependency analysis on the plan, analyzing the preconditions and effects of each atomic operation to construct a directed acyclic graph in which nodes represent each atomic operation and directed edges represent logical or physical dependencies between operations. If the execution of operation B must follow the completion of operation A, then there is a directed edge from node A to node B. This step converts the linear list of instructions into a data structure that clearly expresses parallelism possibilities and execution constraints.

[0083] Second step: Resource static allocation and locking

[0084] Before execution begins, the engine performs resource requirement analysis on each operation node in the directed acyclic graph to identify all resources that each operation needs to exclusively use during execution, including but not limited to physical resources (such as specific pipetting channels, robotic arms, centrifuges), logical resources (such as a specific microplate or reagent bottle), and spatial resources (such as critical paths that a robotic arm will pass through). The engine maintains a resource manager internally to track the real-time status of all resources (e.g., free or occupied).

[0085] Third step: Event-based real-time scheduling

[0086] The engine uses an event-driven real-time scheduler to determine the execution timing of operations. The scheduler continuously monitors the directed acyclic graph for all nodes with an "in-degree" of zero, which represent operations that have no pre-requisites and can currently be executed. For each executable candidate operation, the scheduler queries the resource manager to determine if all required resources are in the "free" state. If resources are available, the scheduler immediately marks the operation as "to-be-executed" and distributes it to the dedicated instruction queue of the corresponding physical device, while locking its required resources through the resource manager. If resources are occupied, the operation continues to wait.

[0087] Fourth step: Instruction translation and underlying distribution

[0088] The dedicated instruction queue of each physical device is handled by a device communication agent. When an atomic operation enters the queue, the agent calls the device driver abstraction layer pre-built in the programmed framework. The agent looks up and calls the corresponding specific device driver based on the device identifier in the operation instruction. The driver is responsible for translating the abstract instruction into underlying control instructions (such as TCP / IP packets, serial port commands, or API calls to specific software development kits (SDKs)) that the target physical device can recognize and execute. After translation, the generated underlying control instructions are immediately distributed to the target physical device through the corresponding communication channel.

[0089] Step 5: Synchronous monitoring of execution status

[0090] After the instruction is dispatched, the device communication agent enters a waiting and monitoring state to confirm whether the operation is successfully completed, in the following ways: receiving the explicit "operation completed" or "operation failed" response signal returned by the device controller; receiving the active notification of the device controller through the pre-registered callback function; or indirectly confirming that the operation has produced the expected effect by periodically querying the state values related to the operation in the semantic model updated by the physical sensor.

[0091] Step 6: Status update and dependency release

[0092] When an operation is confirmed to be successfully completed, the device communication agent reports "task completed" to the scheduler, which then removes the completed node and all its outgoing edges in the directed acyclic graph, and notifies the resource manager to release all resources occupied by the operation. The release of resources and the removal of nodes in the graph make the "in-degree" of other nodes that originally depended on this operation become zero, thereby becoming new executable candidate operations for the scheduler to schedule in the next round of loop.

[0093] Through the above six-step loop, the collaborative job execution engine of the present application can efficiently and safely decompose a complex experimental process into multiple automated devices for parallel execution, achieving truly unattended collaborative job.

[0094] Step 4: During the execution process, the actual status feedback of the automated execution unit is continuously monitored, and when the deviation between the actual status and the optimal execution plan under the preset condition is detected, the optimal execution plan is automatically triggered for immediate re-planning and dynamic adjustment. The constructed semantic model is a unified knowledge structure, in which the constantly changing instantaneous state of each laboratory entity is adapted to a set of pre-set logical rules about how entities interact, what order to act in, and what safety restrictions to follow, and the consequences of all decisions are deduced and calculated, and;

[0095] The definition of the properties of each device itself, the fixed spatial position layout between them, the logical subordination, and the collaborative function realized when they are combined together, and the formal description of the executable operation of each entity in the laboratory, which includes the system state conditions required to perform the operation and the expected changes to the system state after the operation is completed, the operation in the optimal execution plan is parsed and converted into the action step list of the corresponding execution unit, and the collaborative scheduling strategy is implemented based on the semantic model to avoid the problem of collision between two or more automated execution units using the same resources at the same time. The synchronization mechanism is synchronized by issuing instructions and state execution feedback to ensure that the entire plan is executed accurately according to the predetermined timing.

[0096] Embodiment 2

[0097] For the second embodiment of the present application, the embodiment provides a use scenario of a laboratory intelligent agent construction method based on artificial intelligence, including the following steps:

[0098] Intuitive part: In which the user needs to automatically complete the preparation of a batch of soil samples, the preparation of a suspension, the determination of pH value and the colorimetric determination of nitrate content.

[0099] 1. Task initiation and intelligent planning, the user issues a high-level instruction: "Prepare the suspension in P1 plate for soil samples in positions A1 to A8, and complete the pH value and nitrate content analysis in P2 plate". The task planner module receives the instruction, queries the semantic model to obtain the initial state of the environment, and generates a sequence of atomic operations containing dozens of steps (execution plan), such as adding buffer, transferring soil powder, oscillating mixing, transferring supernatant, measuring pH, adding color developing agent, colorimetric determination, etc.

[0100] 2. Instruction distribution and collaborative work, the execution plan is sent to the collaborative work execution engine. The engine constructs its directed acyclic graph (DAG) and excavates "pipette adds buffer" and "robot arm grabs sample tube" as parallel operations, and distributes instructions to different devices at the same time to realize multi-device collaborative work.

[0101] 3. Unexpected detection and dynamic re-planning, when executing the "suck 2 soil powder" step, the air pressure sensor of the pipette detects abnormal air suction resistance due to sample clumping.

[0102] Deviation detection: The state monitoring and deviation analysis module compares the real-time air pressure reading with the expected range and finds a serious deviation, immediately determines that an "material transfer failure" accident has occurred, and issues a deviation alarm.

[0103] Dynamic re-planning: The dynamic re-planning controller module receives the alarm and immediately suspends all devices. It re-submits the current semantic model state containing the new information "Sample A2 transfer failed" to the task planner. The planner generates a new execution plan containing a corrective step based on this new state, e.g.: move to waste station, discard contaminated tip, get new tip, send alarm to user and suspend processing task for sample 2, skip and continue processing subsequent samples.

[0104] 4. New plan execution and task completion engine resumes execution based on the new plan containing the corrective step and successfully completes analysis of all other samples. Upon task completion, the system reports to the user: "Analysis task completed, sample A2 failed to process successfully due to clogging, please check."

[0105] System processing part: where,

[0106] Step 1, initial information reception and model definition,

[0107] The system's internal model contains the expectation for this operation, which is pipette aspirating liquid, the key monitoring parameter is pressure P. The initial information received by the system: real-time pressure measurement from the associated pressure sensor: .

[0108] System's internal expectation model ( ): the expected normal range of pressure associated with this operation is defined as: .

[0109] Step 2, call deviation cost function ,

[0110] To quantify the degree of deviation of the measurement value from the normal range, the system calls a predefined deviation cost function ( ), denoted as . The purpose of this function is to calculate the penalty cost of the measurement value P exceeding its normal range . In this embodiment, the function is defined as follows:

[0111] Formula 1: Deviation cost function If If If

[0112] The function outputs a non-negative value, the larger the value, the more serious the deviation from the normal range.

[0113] Step 3: calculate the deviation cost value, the system substitutes the received initial information into the deviation cost function for calculation.

[0114] Input: P = 5.8 kPa, = 2.5 kPa.

[0115] Judgment: 5.8 kPa > 2.5 kPa, P condition is met.

[0116] Calculation: (5.8) = 5.8 kPa - 2.5 kPa = 3.3 kPa.

[0117] Intermediate Result: The deviation value is calculated to be 3.3 kPa.

[0118] Step 4: Calculate the state deviation metric and identify.

[0119] To make the deviation comparable and eliminate the dimension effect, the system further calculates a normalized state deviation metric (S ), denoted as S .

[0120] Formula 2: State deviation metric

[0121]

[0122] The system then performs the following calculation:

[0123] Input: = 3.3 kPa (from Step 3), = 2.5 kPa - 1.5 kPa = 1.0 kPa.

[0124] Calculate = 3.3 kPa / 1.0 kPa = 3.3.

[0125] After calculating , the system compares it with the preset state identification threshold to map the continuous metric value to the discrete state category.

[0126] State identification rule:

[0127] If = 0, identify as state normal (N ).

[0128] If 0 ≤ , identify as state warning (W ).

[0129] If > , identify as state abnormal (A ).

[0130] preset threshold value:

[0131] critical abnormality threshold value = 2.0.

[0132] identification process:

[0133] The calculated value of 3.3 is compared with the threshold value of 2.0. The judgment 3.3 > 2.0 is true.

[0134] The identification result:

[0135] The system identifies the current state as: state abnormality (3.3 > 2.0).

[0136] ). ).

[0137] Embodiment 3

[0138] The present application also provides an artificial intelligence laboratory agent construction system, and comprises the following mutually coordinated modules:

[0139] A unified knowledge model module configured to integrate ontology rules for defining static attributes and physical constraints of laboratory entities, real-time sensing data for representing instantaneous states of the laboratory entities, and predictive logic for describing causal relationships between operations and potential execution biases;

[0140] A forward-looking planning algorithm module connected with the unified knowledge model module and configured to generate at least one candidate execution path based on a received scientific research task target, and to deduce the candidate execution path using the predictive logic to predict a probability distribution of bias events occurring in future execution of the path, thereby outputting a quantitative risk profile containing the probability distribution;

[0141] A decision optimization module configured to receive the quantitative risk profile and determine an optimal execution plan from the at least one candidate execution path according to a preset strategy that combines execution efficiency and the risk profile;

[0142] And a closed-loop execution and model self-adaptation module configured to generate control instructions to drive automated equipment in the laboratory according to the optimal execution plan, and to continuously monitor actual execution states fed back by the automated equipment, and to automatically correct the predictive logic in the unified knowledge model module when there is a discrepancy between the actual execution states and the prediction results of the forward-looking planning algorithm module, so as to reduce future prediction biases.

[0143] ​The embodiment also provides a computer device suitable for the laboratory intelligent agent construction method and construction system based on artificial intelligence, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to realize the laboratory intelligent agent construction method and construction system based on artificial intelligence proposed in the above embodiment.

[0144] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, an operator network, NFC (Near Field Communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0145] The embodiment also provides a storage medium having a computer program stored thereon, the program being executed by a processor to realize the laboratory intelligent agent construction method and construction system based on artificial intelligence proposed in the above embodiment. The storage medium can be realized by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk or an optical disk.

[0146] To sum up, the application realizes the intelligent leap from "passive execution" to "active decision": it can independently understand the scientific research target and plan the optimal path in advance, greatly reducing the dependence on manual programming, and can actively avoid potential risks; it gives the system strong robustness and fault self-healing ability: in the face of execution deviation, the system can quickly re-plan through a closed loop to adjust the strategy independently to ensure task completion, improving the success rate and stability of the automated process; it realizes the global optimization and efficient collaboration of laboratory resources: through unified global scheduling of the knowledge model, it effectively avoids resource conflicts and motion collisions, maximizes equipment utilization, significantly shortens the experimental period, and improves the overall operation throughput.

[0147] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application, which should be covered by the scope of the claims of the present application.

Claims

1. A method for constructing an intelligent laboratory agent based on artificial intelligence, wherein the method establishes a system for autonomous planning and closed-loop adaptive control of laboratory research processes, characterized in that, Includes the following steps: Step 1: Construct a semantic model that integrates real-time status data of laboratory equipment with common operational and common sense among laboratory equipment; Step 2: Based on the semantic model, analyze the received scientific research task objectives according to the semantic model, and output a certain result, which is an optimal execution plan that includes specific operations, resource allocation and execution sequence. Step 3: Based on the optimal execution plan, generate and distribute control commands to drive the automated execution units in the laboratory to perform collaborative operations; Step 4: During execution, continuously monitor the actual status feedback of the automated execution unit. When a deviation between the actual status and the optimal execution plan is detected under preset conditions, automatically trigger the immediate replanning and dynamic adjustment of the optimal execution plan.

2. The method for constructing an artificial intelligence-based laboratory intelligent agent according to claim 1, characterized in that, Step 1, in constructing the semantic model, specifically includes: continuously collecting real-time status data of various equipment attributes within the laboratory through a standardized data interface layer; pre-constructing a set of rigorously structured digital rules to clearly define how all physical equipment collaborates, the order of operations, and their respective capability limits and safety red lines—a formal knowledge structure; and then, through a dynamic association mechanism, binding the real-time status data as equipment attribute values ​​to the corresponding equipment descriptions in the formal knowledge structure in real time, thereby forming a computable model that integrates the real-time status of laboratory equipment with judgment logic.

3. The method for constructing an artificial intelligence-based laboratory intelligent agent as described in claim 1, characterized in that, Step 2 specifically includes: First, constructing a procedural framework in the semantic model. Through the constructed procedural framework, the scientific research task objective is decomposed into an ordered sequence of operations. This ordered sequence of operations consists of multiple logically related operations. Then, for the operations in the ordered sequence, the real-time attributes of entities in the laboratory in the semantic model are queried to match them, and a currently available and capable specific execution device is assigned. Finally, among the candidate execution paths formed by all resource assignment combinations, the optimal execution plan is determined by evaluation and screening based on at least one preset optimization index.

4. The method for constructing an artificial intelligence-based laboratory intelligent agent as described in claim 1, characterized in that, Step 3 specifically includes: continuously comparing the status feedback of the execution unit with the expected status to identify deviation events; after identifying a deviation event, immediately updating the actual changes in the laboratory caused by the deviation event to the semantic model; and based on the updated semantic model and the current actual system status, replanning a process oriented towards completing the original scientific research task goal to generate a new execution plan.

5. The method for constructing an artificial intelligence-based laboratory intelligent agent as described in claim 1, characterized in that, The semantic model constructed in step 4 is a unified knowledge structure. This structure adapts the constantly changing real-time state of each laboratory entity to a set of pre-defined logical rules about how entities interact, in what order they act, and what safety restrictions they follow. It also deduces and calculates the consequences of all decisions. Each device has its own attributes, and its fixed spatial layout, logical dependencies, and collaborative functions that can only be achieved when they are combined are defined. Furthermore, a formal description of the operations that can be performed on each entity in the laboratory is defined. This description includes the system state conditions that must be met to perform the operation, as well as the expected changes to the system state after the operation is completed.

6. The method for constructing an artificial intelligence-based laboratory intelligent agent as described in claim 3, characterized in that, Step 4 specifically includes: parsing the operations in the optimal execution plan and converting them into a list of action steps corresponding to their designated execution units; at the same time, implementing a coordinated scheduling strategy based on a semantic model to avoid collisions between two or more automated execution units using the same resources at the same time; and using a synchronization mechanism to confirm synchronization through issuing instructions and status execution feedback to ensure that the entire plan is executed accurately according to the predetermined sequence.

7. The method for constructing an artificial intelligence-based laboratory intelligent agent as described in claim 6, characterized in that, After continuously monitoring and identifying the deviation between the actual system state and the optimal execution plan, the actual changes in the laboratory caused by the deviation are first synchronized to the semantic model. The updated semantic model is used as the starting condition for replanning, and the unchanged original task objective is used as the endpoint of replanning, to generate a new, adjusted execution plan to replace the original plan.

8. A laboratory intelligent agent construction system, based on the artificial intelligence-based laboratory intelligent agent construction method according to any one of claims 1 to 7, characterized in that, The unified knowledge model module is configured to integrate ontological rules for defining the static properties and physical constraints of laboratory entities, real-time sensor data for characterizing the instantaneous state of said laboratory entities, and predictive logic for describing causal relationships and potential execution deviations between operations. The forward-looking planning calculation module, which is connected to the unified knowledge model module, is configured to generate at least one candidate execution path based on a received scientific research task objective, and use the predictive logic to deduce the candidate execution path to predict the probability distribution of deviation events that will occur during the future execution of the path, thereby outputting a quantitative risk profile containing the probability distribution. A decision optimization module is configured to receive the quantified risk profile and determine an optimal execution plan from the at least one candidate execution path based on a preset strategy that integrates execution efficiency and the risk profile. And a closed-loop execution and model adaptation module, which is configured to generate control instructions based on the optimal execution plan to drive automated equipment in the laboratory, continuously monitor the actual execution status fed back by the automated equipment, and automatically correct the predictive logic in the unified knowledge model module when there is an inconsistency between the actual execution status and the prediction results of the forward planning calculation module, so as to reduce the deviation of future predictions.

9. The laboratory intelligent agent construction system as described in claim 8, characterized in that, The forward planning calculus module is further configured to: map the candidate execution path into an ordered sequence consisting of multiple operation units, and for each operation unit in the sequence, extract a set of mutually exclusive potential results from the predictive logic of the unified knowledge model module, including at least one successful result and one or more deviation results, wherein each potential result is assigned a corresponding probability of occurrence. Furthermore, by performing iterative probabilistic deduction on the sequence, the execution result of each operation unit is randomly determined based on the occurrence probability in each deduction, and the statistical results of multiple deductions are combined to generate the quantitative risk profile, which at least characterizes the overall expected value of the execution success of the candidate execution path and the specific operation unit that causes the highest risk of deviation.

10. The laboratory intelligent agent construction system as described in claim 9, characterized in that, The closed-loop execution and model adaptation module is further configured to: parse the optimal execution plan into a set of underlying control instructions that ensure operational synchronization and resource exclusivity among multiple automated devices; and when the actual execution state deviates from any possible result predicted by the forward planning calculation module, generate a deviation event record containing snapshots of the system state before and after the deviation; and then use the deviation event record as feedback to adjust the predictive logic in the unified knowledge model module to integrate the unforeseen causal relationship revealed by the deviation event into future predictions.

Images