Low-energy-consumption chemical experiment ai intelligent scheduling system and method fusing agv feedback

Abstract

The application discloses a kind of low-energy consumption chemical experiment AI intelligent scheduling system and method of fusion AGV feedback, system includes experiment task analysis and modeling module, multi-source heterogeneous sensing and data fusion module, intelligent scheduling and dispatch kernel module, interactive scheduling confirmation and visualization module and instruction issue and dynamic feedback execution module.The application realizes the deep coupling of chemical process logic and AGV dynamic feedback by directly incorporating AGV physical operating characteristics into scheduling constraints;Through the construction of improved NSGA-III multi-objective optimization model with total completion time, AGV total transportation energy consumption and chemical timeliness satisfaction as the target, and the introduction of dynamic penalty mechanism based on health degree, the efficiency, energy efficiency and timeliness are balanced organically;Through the double-layer architecture of global optimization and local rescheduling and closed-loop dynamic feedback execution mechanism, the system can automatically trigger local rapid replanning when facing AGV blockage, low battery and other abnormalities.

Description

Technical Field

[0001] This invention relates to the fields of laboratory automation and artificial intelligence technology, specifically to a low-energy chemical experiment AI intelligent scheduling system and method that integrates AGV feedback. Background Technology

[0002] With the rapid development of artificial intelligence and robotics, automated equipment is increasingly being used in chemical laboratories, gradually replacing traditional manual operations from sample preparation and reaction control to product detection, significantly improving experimental efficiency and reproducibility. However, current automated laboratory management systems still face many technical bottlenecks, hindering their overall effectiveness.

[0003] In laboratory scheduling, traditional scheduling systems typically treat AGVs and other logistics equipment as on-demand, subordinate tools, failing to incorporate their physical operational characteristics (such as location, speed, power consumption, and path congestion) as core constraints into the master scheduling model. This scheduling approach, which prioritizes process over logistics, makes it difficult to accurately implement the generated experimental plans in the physical world. It easily leads to resource waste, such as experiments waiting for AGVs or AGVs running idly, resulting in low experimental efficiency and increased energy consumption.

[0004] Meanwhile, experimental execution control and material transportation control are usually managed by two independent systems, lacking an effective coordination mechanism. This prevents deep coupling between chemical process logic and AGV physical operation efficiency, further exacerbating the problem of insufficient resource utilization.

[0005] Chemical experiments are inherently complex, involve numerous steps, and are highly time-sensitive. For example, operations such as heating, stirring, adding, and detecting often have strict time dependencies, and some intermediate products have specific chemical time windows, requiring them to proceed to the next step within a specified timeframe; otherwise, the experiment will fail. Existing scheduling methods mostly employ static rules or single-priority sorting, which struggle to handle the complex timing constraints of multi-task concurrent scenarios and fail to meet the timeliness requirements of experimental processes. Furthermore, when the experimental process undergoes dynamic changes, such as the insertion of urgent tasks, equipment failures, or diversification of experimental scale, existing systems generally lack flexible rescheduling capabilities, making it difficult to adjust and optimize in real time according to actual conditions.

[0006] To address the aforementioned issues, published patent applications have proposed some solutions. For example, CN120975421A discloses a method and apparatus for scheduling experimental tasks in intelligent laboratories. This method obtains an experimental task pool, determines priority weights based on task sensitivity levels and estimated time consumption, and generates a sequence of operation instructions to control an experimental robot to execute experimental tasks. However, this patent still has shortcomings in the dynamic priority evaluation stage, failing to comprehensively consider real-time changes in multi-dimensional factors such as task urgency and resource availability. CN120996400A proposes a dynamic scheduling method for intelligent laboratories, determining experimental process information by searching a calculation formula when tasks have the same priority level and share equipment. However, this patent still lacks in experimental data management and analysis, lacking a real-time monitoring and feedback mechanism for experimental results and equipment status, making it difficult to support closed-loop dynamic scheduling.

[0007] In summary, the existing technology has the following main drawbacks: 1. Disconnect between process and logistics: Traditional scheduling systems treat AGVs as on-demand tools and do not incorporate their physical operating characteristics into the main scheduling constraints. This makes it difficult to implement scheduling solutions in the physical world, and it is easy to have AGVs running empty, such as for experiments, resulting in resource waste and increased energy consumption.

[0008] 2. Lack of coordination mechanism: Experiment execution and material transportation are controlled by two independent systems, lacking effective coordination. This makes it impossible to achieve deep coupling between chemical process logic and AGV physical operation efficiency, affecting the overall efficiency and resource utilization of the laboratory.

[0009] 3. Insufficient timeliness guarantee: Chemical experiments involve multiple time-sensitive operations. Traditional scheduling methods are difficult to cope with the complex timing constraints in multi-task concurrent scenarios and cannot meet the timeliness requirements of experiments.

[0010] 4. Poor dynamic adaptability: The existing system lacks flexibility and intelligence, making it difficult to adapt to dynamic changes in experimental procedures and diverse experimental scales, and unable to make real-time adjustments and optimizations according to actual conditions.

[0011] 5. One-sided priority assessment: The existing solution has shortcomings in the dynamic priority assessment of experimental tasks. It lacks comprehensive consideration of multiple factors such as task urgency and resource availability, making it difficult to achieve efficient task scheduling.

[0012] Therefore, there is an urgent need for an intelligent scheduling system and method that can deeply couple chemical process logic with AGV physical operating efficiency and possess multi-objective optimization and dynamic adaptive capabilities, in order to achieve efficient, energy-saving, and intelligent operation of chemical experiments. This invention is proposed to solve the aforementioned technical problems. Summary of the Invention

[0013] To address the aforementioned technical problems in related technologies, this invention proposes a low-energy chemical experiment AI intelligent scheduling system and method that integrates AGV feedback, which can overcome the above-mentioned shortcomings of the prior art.

[0014] To achieve the above-mentioned technical objectives, the technical solution of the present invention is implemented as follows: A low-energy chemical experiment AI intelligent scheduling system integrating AGV feedback includes: The experimental task parsing and modeling module is used to decompose the standard operating procedures of chemical experiments into a sequence of atomic tasks with time-dependent and chemical time-dependent constraints, and to define key attributes for each atomic task, including equipment requirements, execution time, bill of materials, and chemical time-dependent window. The multi-source heterogeneous sensing and data fusion module is used to collect and fuse multi-source heterogeneous data in real time to form a unified real-time system status view. The multi-source heterogeneous data includes experimental equipment status information, automated guided vehicle operation status information, and laboratory environment and path information. The intelligent scheduling and scheduling kernel module is used to construct and solve a multi-objective optimization model based on the real-time status view of the system and the atomic task sequence, with the objectives of minimizing the total completion time of the experiment, minimizing the total transportation energy consumption of the automated guided vehicle, and maximizing the chemical timeliness satisfaction, to generate a Pareto optimal scheduling scheme. An interactive scheduling confirmation and visualization module is used to visually display the scheduling plan, allowing users to adjust and confirm the plan; and The instruction issuance and dynamic feedback execution module is used to decompose the confirmed scheduling plan into operation instructions and issue them to the experimental equipment and automated guided vehicles. During the execution process, it triggers a local rescheduling mechanism based on real-time feedback.

[0015] Furthermore, the intelligent scheduling and scheduling kernel module adopts an improved non-dominated sorting genetic algorithm for multi-objective optimization, and constructs a multi-objective cost function that includes the total transportation energy consumption of the automated guided vehicle, the total experimental time, and the satisfaction of chemical timeliness. By introducing a dynamic penalty mechanism, the optimization weights are adjusted in real time according to the health of the automated guided vehicle.

[0016] Furthermore, the multi-source heterogeneous sensing and data fusion module includes: The experimental equipment sensing unit is used to collect the busy / idle status, fault information and operating parameters of each experimental device in real time. The automated guided vehicle (AGV) operation status monitoring unit is used to acquire high-precision coordinates, instantaneous speed, current load, state of charge, and health status of key components for each AGV in real time; and The environment and path perception unit is used to receive traffic control information and identify dynamic obstacles or congested areas.

[0017] Furthermore, the objective function of the multi-objective optimization model includes: The primary objective is to minimize the total completion time. The second objective is to minimize total transportation energy consumption, which is calculated based on an automated guided vehicle (AGV) energy consumption model, which is expressed as follows: E total =Σ(d i *L i *R i )+Σ(s j *W j ), Where, d i Let L be the distance L travels continuously in the i-th segment of a transportation task. i R is the dynamic load energy consumption factor. i s is the path resistance coefficient. j For the j-th start-stop event that occurs during the entire mission of the automated guided vehicle, W j This refers to the start-stop loss factor. The third objective is to maximize the satisfaction of chemical timeliness requirements.

[0018] Furthermore, the dynamic load energy consumption factor L i Defined as L i =1+α*(m load / m rated ), where m load m is the actual load of the automated guided vehicle during the i-th segment of its journey. rated The rated maximum load of the automated guided vehicle is α, where α is the load energy consumption coefficient; the path resistance coefficient R is... i Defined as R i =R base +β*(1-H agv ), where R base H is the reference road surface resistance coefficient. agv The current overall health status of the automated guided vehicle is represented by β, where β is the health status influencing factor; the start-stop loss factor W... j Defined as W j =W base *(1+γ*T motor / T max ), where W base Based on the start-stop energy consumption, T motor T represents the real-time temperature of the motor of the automated guided vehicle at the current moment. max γ is the rated maximum operating temperature of the motor, and γ is the temperature influence coefficient.

[0019] Furthermore, the intelligent scheduling and dispatching kernel module adopts a centralized architecture deployed on a central server. The central server uniformly issues task instructions to all experimental equipment and automated guided vehicles, and uniformly calculates the path planning of the automated guided vehicles. The multi-source heterogeneous sensing and data fusion module also includes an LSTM time series prediction model, which is used to predict the power consumption of the automated guided vehicles and the task completion time of the equipment within a preset time in the future, and uses the prediction data as the input of the multi-objective optimization model.

[0020] Furthermore, the intelligent scheduling and dispatching kernel module adopts a distributed architecture that coordinates cloud, edge, and terminal. The cloud-based central server is responsible for task parsing and global scheduling, and publishes the generated task instructions to all automated guided vehicles (AGVs). The edge computing unit is mounted on each AGV and has a built-in lightweight deep reinforcement learning model for local real-time decision-making. Each AGV acts as an edge device and autonomously bids for transportation tasks based on its own state and the deep reinforcement learning model. The cloud-based central server then assigns the task to the optimal AGV based on the bidding results.

[0021] Furthermore, in the instruction issuance and dynamic feedback execution module, when the automated guided vehicle is detected to be blocked or its battery level is lower than a preset threshold, or when the deviation between real-time monitoring data and predicted data exceeds a preset threshold, a local rescheduling mechanism is triggered to lock the affected task set, call the multi-objective optimization model to perform local rapid replanning, and dynamically update the optimized instructions to the corresponding devices.

[0022] A low-energy chemical experiment AI intelligent scheduling method integrating AGV feedback, applied to any of the systems described above, includes the following steps: S1 Experimental Task Analysis and Modeling: Read the standard operating procedures of chemical experiments, use natural language processing technology to decompose them into a sequence of atomic tasks with time dependence and chemical time constraints, define key attributes for each atomic task and store them in the task pool. S2 Real-time perception and fusion of multi-source heterogeneous data: Real-time acquisition of experimental equipment status information, automated guided vehicle operation status information, and laboratory environment and path information, and time alignment and spatial synchronization of multi-source heterogeneous data to form a unified real-time system status view; S3 Constructs and solves a multi-objective optimization scheduling model: Based on the real-time state view and task pool of the system, a multi-objective optimization model is constructed with the objectives of minimizing the total completion time of the experiment, minimizing the total transportation energy consumption of the automated guided vehicle, and maximizing the satisfaction of chemical timeliness. An improved non-dominated sorting genetic algorithm is used to solve the model and generate a set of Pareto optimal scheduling schemes. S4 Interactive Scheduling Scheme Confirmation and Adjustment: The generated Pareto optimal scheduling scheme is visualized in the form of a Gantt chart on a graphical interface, allowing users to adjust the scheme and select the final execution scheme after user confirmation. S5 command issuance and closed-loop dynamic feedback execution: The final solution is decomposed into specific operation commands, which are issued to the experimental equipment controller and the automated guided vehicle control system respectively. During the execution process, the execution status is continuously monitored. When an abnormal event occurs, a local rescheduling mechanism is triggered, and the multi-objective optimization model is called to perform local rapid replanning until all experimental tasks are completed.

[0023] Furthermore, in step S3, the total transportation energy consumption of the automated guided vehicle is calculated based on the automated guided vehicle energy consumption model, which is expressed as: E total =Σ(d i *L i *R i )+Σ(s j *W j ), Where, d i Let L be the distance L travels continuously in the i-th segment of a transportation task. i R is the dynamic load energy consumption factor. i s is the path resistance coefficient. j For the j-th start-stop event that occurs during the entire mission of the automated guided vehicle, W j The start-stop loss factor; the dynamic load energy consumption factor L i Defined as L i =1+α*(m load / m rated ), where m load m is the actual load of the automated guided vehicle during the i-th segment of its journey. rated The rated maximum load of the automated guided vehicle is α, where α is the load energy consumption coefficient; the path resistance coefficient R is... i Defined as R i =R base +β*(1-H agv ), where R base H is the reference road surface resistance coefficient. agv The current overall health status of the automated guided vehicle is represented by β, where β is the health status influencing factor; the start-stop loss factor W... j Defined as W j =W base *(1+γ*T motor / T max ), where W base Based on the start-stop energy consumption, T motor T represents the real-time temperature of the motor of the automated guided vehicle at the current moment.max γ is the rated maximum operating temperature of the motor, and γ is the temperature influence coefficient.

[0024] The beneficial effects of this invention are as follows: By directly incorporating the physical operating characteristics of AGVs (position, power, load) into scheduling constraints, this invention achieves deep coupling between chemical process logic and AGV dynamic feedback, thereby enabling the scheduling scheme to be accurately implemented in the physical world and avoiding resource waste caused by AGVs waiting for experiments or AGVs running idly. By constructing an improved NSGA-III multi-objective optimization model with the objectives of total completion time, total AGV transportation energy consumption, and chemical timeliness satisfaction, and introducing a dynamic penalty mechanism based on health, an organic balance is achieved between experimental efficiency, energy efficiency, and timeliness. Through a two-layer architecture of global optimization and local rescheduling, as well as a closed-loop dynamic feedback execution mechanism, the system can automatically trigger local rapid replanning when faced with anomalies such as AGV blockage and low power, significantly improving the system's robustness and dynamic adaptability. Through a refined AGV energy consumption model (including dynamic load factor, path resistance coefficient, and start-stop loss factor) and a visual interactive interface, energy consumption calculation is accurate and controllable, and the scheduling scheme is intuitive and adjustable, ultimately achieving low-energy, high-efficiency, and intelligent operation of chemical experiments. Detailed Implementation

[0025] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the scope of protection of the present invention.

[0026] A low-energy chemical experiment AI intelligent scheduling system integrating AGV feedback, as described in an embodiment of the present invention, includes: The experimental task parsing and modeling module is used to decompose the standard operating procedures of chemical experiments into a sequence of atomic tasks with time-dependent and chemical time-dependent constraints, and to define key attributes for each atomic task, including equipment requirements, execution time, bill of materials, and chemical time-dependent window. The multi-source heterogeneous sensing and data fusion module is used to collect and fuse multi-source heterogeneous data in real time to form a unified real-time system status view. The multi-source heterogeneous data includes experimental equipment status information, automated guided vehicle operation status information, and laboratory environment and path information. The intelligent scheduling and scheduling kernel module is used to construct and solve a multi-objective optimization model based on the real-time status view of the system and the atomic task sequence, with the objectives of minimizing the total completion time of the experiment, minimizing the total transportation energy consumption of the automated guided vehicle, and maximizing the chemical timeliness satisfaction, to generate a Pareto optimal scheduling scheme. An interactive scheduling confirmation and visualization module is used to visually display the scheduling plan, allowing users to adjust and confirm the plan; and The instruction issuance and dynamic feedback execution module is used to decompose the confirmed scheduling plan into operation instructions and issue them to the experimental equipment and automated guided vehicles. During the execution process, it triggers a local rescheduling mechanism based on real-time feedback.

[0027] Preferably, the intelligent scheduling and scheduling kernel module uses an improved non-dominated sorting genetic algorithm for multi-objective optimization and constructs a multi-objective cost function that includes the total transportation energy consumption of the automated guided vehicle, the total experimental time, and the satisfaction of chemical timeliness. By introducing a dynamic penalty mechanism, the optimization weights are adjusted in real time according to the health of the automated guided vehicle.

[0028] Preferably, the multi-source heterogeneous sensing and data fusion module includes: The experimental equipment sensing unit is used to collect the busy / idle status, fault information and operating parameters of each experimental device in real time. The automated guided vehicle (AGV) operation status monitoring unit is used to acquire high-precision coordinates, instantaneous speed, current load, state of charge, and health status of key components for each AGV in real time; and The environment and path perception unit is used to receive traffic control information and identify dynamic obstacles or congested areas.

[0029] Furthermore, the objective function of the multi-objective optimization model includes: The primary objective is to minimize the total completion time. The second objective is to minimize total transportation energy consumption, which is calculated based on an automated guided vehicle (AGV) energy consumption model, which is expressed as follows: E total =Σ(d i *L i *R i )+Σ(s j *W j ), Where, d i Let L be the distance L travels continuously in the i-th segment of a transportation task. i R is the dynamic load energy consumption factor. i s is the path resistance coefficient. j For the j-th start-stop event that occurs during the entire mission of the automated guided vehicle, W j This refers to the start-stop loss factor. The third objective is to maximize the satisfaction of chemical timeliness requirements.

[0030] Preferably, the dynamic load energy consumption factor L i Defined as L i =1+α*(mload / m rated ), where m load m is the actual load of the automated guided vehicle during the i-th segment of its journey. rated The rated maximum load of the automated guided vehicle is α, where α is the load energy consumption coefficient; the path resistance coefficient R is... i Defined as R i =R base +β*(1-H agv ), where R base H is the reference road surface resistance coefficient. agv The current overall health status of the automated guided vehicle is represented by β, where β is the health status influencing factor; the start-stop loss factor W... j Defined as W j =W base *(1+γ*T motor / T max ), where W base Based on the start-stop energy consumption, T motor T represents the real-time temperature of the motor of the automated guided vehicle at the current moment. max γ is the rated maximum operating temperature of the motor, and γ is the temperature influence coefficient.

[0031] Preferably, the intelligent scheduling and dispatching kernel module adopts a centralized architecture deployed on a central server. The central server uniformly issues task instructions to all experimental equipment and automated guided vehicles, and uniformly calculates the path planning of the automated guided vehicles. The multi-source heterogeneous sensing and data fusion module also includes an LSTM-based time series prediction model, which is used to predict the power consumption of the automated guided vehicles and the task completion time of the equipment within a preset time in the future, and uses the prediction data as the input of the multi-objective optimization model.

[0032] Preferably, the intelligent scheduling and dispatching kernel module adopts a distributed architecture of cloud-edge-device collaboration; wherein, the cloud central server is responsible for task parsing and global scheduling, and publishes the generated task instructions to all automated guided vehicles; the edge computing unit is mounted on each automated guided vehicle, with a built-in lightweight deep reinforcement learning model, used to realize local real-time decision-making of the automated guided vehicle; each automated guided vehicle, as an edge device, autonomously bids for transportation tasks based on its own state and the deep reinforcement learning model, and the cloud central server assigns the task to the best automated guided vehicle according to the bidding results.

[0033] Preferably, in the instruction issuance and dynamic feedback execution module, when the automated guided vehicle is detected to be blocked or the battery level is lower than a preset threshold, or when the deviation between the real-time monitoring data and the predicted data exceeds a preset threshold, a local rescheduling mechanism is triggered to lock the affected task set, call the multi-objective optimization model to perform local rapid replanning, and dynamically update the optimized instructions to the corresponding devices.

[0034] A low-energy chemical experiment AI intelligent scheduling method integrating AGV feedback, applied to any of the systems described above, includes the following steps: S1 Experimental Task Analysis and Modeling: Read the standard operating procedures of chemical experiments, use natural language processing technology to decompose them into a sequence of atomic tasks with time dependence and chemical time constraints, define key attributes for each atomic task and store them in the task pool. S2 Real-time perception and fusion of multi-source heterogeneous data: Real-time acquisition of experimental equipment status information, automated guided vehicle operation status information, and laboratory environment and path information, and time alignment and spatial synchronization of multi-source heterogeneous data to form a unified real-time system status view; S3 Constructs and solves a multi-objective optimization scheduling model: Based on the real-time state view and task pool of the system, a multi-objective optimization model is constructed with the objectives of minimizing the total completion time of the experiment, minimizing the total transportation energy consumption of the automated guided vehicle, and maximizing the satisfaction of chemical timeliness. An improved non-dominated sorting genetic algorithm is used to solve the model and generate a set of Pareto optimal scheduling schemes. S4 Interactive Scheduling Scheme Confirmation and Adjustment: The generated Pareto optimal scheduling scheme is visualized in the form of a Gantt chart on a graphical interface, allowing users to adjust the scheme and select the final execution scheme after user confirmation. S5 command issuance and closed-loop dynamic feedback execution: The final solution is decomposed into specific operation commands, which are issued to the experimental equipment controller and the automated guided vehicle control system respectively. During the execution process, the execution status is continuously monitored. When an abnormal event occurs, a local rescheduling mechanism is triggered, and the multi-objective optimization model is called to perform local rapid replanning until all experimental tasks are completed.

[0035] Preferably, in step S3, the total transportation energy consumption of the automated guided vehicle is calculated based on the automated guided vehicle energy consumption model, which is expressed as: E total =Σ(d i *L i *R i )+Σ(s j *W j ), Where, d i Let L be the distance L travels continuously in the i-th segment of a transportation task. i R is the dynamic load energy consumption factor. i s is the path resistance coefficient. j For the j-th start-stop event that occurs during the entire mission of the automated guided vehicle, W j The start-stop loss factor; the dynamic load energy consumption factor L i Defined as Li =1+α*(m load / m rated ), where m load m is the actual load of the automated guided vehicle during the i-th segment of its journey. rated The rated maximum load of the automated guided vehicle is α, where α is the load energy consumption coefficient; the path resistance coefficient R is... i Defined as R i =R base +β*(1-H agv ), where R base H is the reference road surface resistance coefficient. agv The current overall health status of the automated guided vehicle is represented by β, where β is the health status influencing factor; the start-stop loss factor W... j Defined as W j =W base *(1+γ*T motor / T max ), where W base Based on the start-stop energy consumption, T motor T represents the real-time temperature of the motor of the automated guided vehicle at the current moment. max γ represents the rated maximum operating temperature of the motor, and γ is the temperature influence coefficient. To facilitate understanding of the above technical solution of the present invention, the following detailed description is provided through specific usage examples.

[0036] In practical use, the low-energy chemical experiment AI intelligent scheduling system integrating AGV feedback according to the present invention includes an experimental task analysis and modeling module, a multi-source heterogeneous perception and data fusion module, an intelligent scheduling and scheduling kernel module, an interactive scheduling confirmation and visualization module, and an instruction issuance and dynamic feedback execution module.

[0037] The Experiment Task Parsing and Modeling module is used to decompose the Standard Operating Procedure (SOP) for chemical experiments into a sequence of atomic tasks with time dependencies and chemical time constraints, and to define key attributes for each atomic task, including equipment requirements, execution time, bill of materials, and chemical time constraints.

[0038] The multi-source heterogeneous sensing and data fusion module is responsible for real-time acquisition of experimental equipment status information, AGV operation status monitoring, and environmental and path perception, achieving effective fusion of multi-source heterogeneous data. Specifically, the experimental equipment sensing unit collects real-time data on the busy / idle status, fault information, and operating parameters of each experimental device; the automated guided vehicle (AGV) operation status monitoring unit acquires real-time high-precision coordinates, instantaneous speed, current load, charge status, and key component health of each AGV; and the environmental and path perception unit receives traffic control information and identifies dynamic obstacles or congested areas.

[0039] The intelligent scheduling and operation kernel module, as the core of the system, employs an improved non-dominated sorting genetic algorithm (NSGA-III) for multi-objective optimization, constructing a multi-objective cost function that includes AGV energy consumption, total experiment duration, and chemical timeliness satisfaction. A dynamic penalty mechanism is introduced to adjust the optimization weights in real time based on the AGV's health status. Specifically, this includes: Decision module: Based on deep learning or reinforcement learning algorithms, a multi-level task scheduling model is constructed to realize intelligent allocation and optimized scheduling of experimental tasks.

[0040] Sensing module: Real-time access to information such as the location, remaining power, current load, and operating speed of the laboratory AGV, and real-time monitoring of the experimental environment and equipment status through sensors and cameras.

[0041] Execution module: Based on the optimization results, it issues instructions to the experimental equipment and AGVs, and monitors the execution process in real time. When the AGV becomes blocked or its battery is too low, it immediately initiates a local rescheduling mechanism to optimize the execution time and path of subsequent tasks.

[0042] Multi-objective optimization model construction: Objective 1 is to minimize the total completion time; Objective 2 is to minimize the total transportation energy consumption, calculated based on the AGV energy consumption model; Objective 3 is to maximize the chemical timeliness satisfaction. Constraints include resource constraints, time constraints, AGV physical constraints, and strong chemical constraints.

[0043] The interactive scheduling confirmation and visualization module provides a user-friendly graphical interface that intuitively displays scheduling plans and allows users to confirm, modify, and optimize these plans. This module includes: Visualization: The optimal scheduling scheme generated by the AI ​​kernel is displayed in the form of a Gantt chart, clearly showing the working sequence of each device and the transportation task time sequence of each AGV.

[0044] Human-computer interaction: Users can adjust the scheduling plan by dragging, clicking and other methods. The system evaluates the feasibility of the adjustment in real time and provides prompts.

[0045] Confirmation Mechanism: After user confirmation, the system will decompose the generated instructions and send them to the experimental equipment controller and the AGV robot control system respectively, and begin executing the experimental task.

[0046] Closed-loop dynamic adjustment mechanism: Real-time monitoring of the execution process. When an AGV experiences unexpected blockage or insufficient power, the system automatically triggers local rescheduling, calling the AI ​​kernel to optimize the execution time and AGV path of affected subsequent tasks, ensuring the continuity and reliability of the overall solution.

[0047] This system achieves refined AGV energy efficiency management by deeply coupling chemical process logic with AGV physical operation characteristics, improving the system's robustness and dynamic adaptability, thereby significantly improving the efficiency and resource utilization of chemical experiments.

[0048] A low-energy chemical experiment AI-based intelligent scheduling method integrating AGV feedback includes the following steps: Step 1: Experimental Task Analysis and Modeling: Read the Standard Operating Procedure (SOP) for a chemical experiment and use natural language processing (NLP) to decompose it into a sequence of atomic tasks with time dependencies and chemical time constraints. Define key attributes for each atomic task, including equipment requirements, execution duration, bill of materials, and chemical time constraints, and store them in the task pool.

[0049] Example: Read the SOP document "Synthesis of Aspirin" and parse it into atomic tasks: T1 (Weigh salicylic acid), T2 (Weigh acetic anhydride), T3 (Dissolve salicylic acid), T4 (Mix and react), T5 (Cool and crystallize), T6 (Filter), T7 (Wash), T8 (Dry). After T4 (Mix and react), T5 (Cool and crystallize) must begin within 30 minutes (the aging window); otherwise, the product will be ineffective.

[0050] Step 2: Real-time perception and fusion of multi-source heterogeneous data: Real-time acquisition of experimental equipment status information, AGV operation status information, and laboratory environment and path information, and time alignment and spatial synchronization of multi-source heterogeneous data to form a unified real-time system status view.

[0051] Example: The status of "Rotary Evaporator R-1" is read as "Busy" at a frequency of 1Hz via OPC-UA protocol. The real-time coordinates (12.5, 8.3, 0°), current battery level of 85%, and load of 0kg of AGV-03 are obtained via MQTT protocol. A temporary obstacle is identified 5 meters in front of AGV-03 by the vision system.

[0052] Step 3: Construct and solve the multi-objective optimization scheduling model: Based on the real-time state view and task pool of the system, construct a multi-objective optimization model with the objectives of minimizing the total completion time of the experiment, minimizing the total transportation energy consumption of the AGV, and maximizing the satisfaction of chemical timeliness. Then, use the improved non-dominated sorting genetic algorithm (NSGA-III) to solve the model and generate a set of Pareto optimal scheduling schemes.

[0053] Step 4: Interactive Scheduling Scheme Confirmation and Adjustment: The generated Pareto optimal scheduling scheme is visualized in a Gantt chart format on a graphical interface. The system allows users to adjust the scheme by dragging and clicking, and evaluates the feasibility of the adjustments in real time. After user confirmation, the final execution scheme is selected.

[0054] Example: An operator notices on the Gantt chart that AGV-02 is being overloaded with tasks, potentially causing it to run out of power. The operator then manually drags a non-urgent transport task from AGV-02 to AGV-04. The system calculates in real-time and displays "Adjustment feasible, total time increased by 5 minutes, AGV energy efficiency improved by 8%." The adjustment takes effect after operator confirmation.

[0055] Step 5: Command Issuance and Closed-Loop Dynamic Feedback Execution: The final solution is decomposed into specific operation commands, which are issued to the experimental equipment controller and AGV control system respectively. During execution, the execution status is continuously monitored. When an abnormal event occurs, a local rescheduling mechanism is immediately triggered, the affected task set is locked, the multi-objective optimization model is invoked for local rapid replanning, and the optimized commands are dynamically updated to the corresponding equipment until all experimental tasks are completed.

[0056] Example: During material transport for task T5 (cooling and crystallization), AGV-03 reports that its battery level is below 15% (trigger event). The system immediately locks AGV-03's current task and all subsequent tasks that depend on it for transport, initiating a partial rescheduling. The AI ​​kernel quickly calculates and transfers AGV-03's remaining task (T5 transport) to AGV-01, and instructs AGV-03 to automatically return to the charging station. The updated instructions are sent to AGV-01 and AGV-03 in real time, and the experiment continues.

[0057] In this embodiment, the intelligent scheduling and scheduling kernel module is deployed on a central server and uses an improved NSGA-III algorithm for global multi-objective optimization. Specifically: 1. Predictive Scheduling Mechanism: The multi-source heterogeneous sensing and data fusion module not only collects real-time data but also predicts the AGV's power consumption and task completion time within the next 15 minutes based on an LSTM time series prediction model. This predicted data serves as input to the optimization model in step 3, enabling the scheduling scheme to be forward-looking.

[0058] 2. Centralized Task Allocation: All task instructions for experimental equipment and AGVs are uniformly issued by a central server. AGVs only act as execution terminals and do not have local decision-making capabilities. AGV path planning is uniformly calculated by the central server based on a global static map and real-time congestion information.

[0059] 3. Rescheduling trigger conditions: When the deviation between real-time monitoring data and predicted data exceeds a preset threshold (e.g., the actual power consumption of the AGV is 10% faster than the prediction), or when a device failure occurs, the central server immediately interrupts the current optimization, locks the affected task window, and initiates local rescheduling.

[0060] 4. Energy consumption model application: Strictly follow formula E total=Σ(d i *L i *R i )+Σ(s j *W j ) calculate, where L i R i W j The parameters are periodically calibrated and updated by the central server through offline experiments. The physical meaning and calculation method of each parameter are defined as follows: (1)d i (meters): The distance traveled by the AGV in the i-th continuous segment during this transportation task. This is achieved by dividing the AGV's travel path into multiple continuous segments based on intersections or changes in road conditions and then summing them up.

[0061] (2) L i (Dimensionless): Dynamic load energy consumption factor. Defined as L i =1+α*(m load / m rated ), m load (kg): The actual load of the AGV during the i-th segment of travel; m rated (kg): Rated maximum load capacity of the AGV; α (dimensionless): Load energy consumption coefficient, obtained through experimental calibration. For example, on a standard test track, measurements were taken under no-load and full-load conditions (m²). rated Energy consumption E for traveling the same distance d) empty and E full Then α=(E full / E empty )-1. This coefficient characterizes the sensitivity of load to energy consumption.

[0062] (3) R i (Dimensionless): Path resistance coefficient. Defined as R i =R base +β*(1-H agv R base (Dimensionless): Reference road surface resistance coefficient, characterizing the resistance of a flat, hard road surface, usually set to 1.0; H agv (Value range 0~1): The current overall health of the AGV, which is evaluated in real time by the multi-source sensing module. The health is calculated based on the weighted average of the status of key components, such as H. agv =w1*(1-T motor / T max )+w2*(1-Wear rate ), where T motor T represents the real-time temperature of the motor. max Wear is the rated maximum temperature of the motor. rateβ represents the wheel wear rate, w1 and w2 are weighting coefficients; β (dimensionless): health factor, which represents the degree of influence of decreased health on increased driving resistance, and is determined experimentally based on the mechanical characteristics of AGV.

[0063] (4)s j (time): The j-th start-stop event that occurs during the entire task. A start-stop event is defined as the complete process of the AGV going from speed > 0.1 m / s to speed < 0.1 m / s and then back to speed > 0.1 m / s.

[0064] (5) W j (Joules / cycle): Start-stop loss factor. Defined as W j =W base *(1+γ*T motor / T max W base (Joules): Baseline start-stop energy consumption, i.e., the energy consumed to complete one standard start-stop cycle under normal motor temperature (e.g., 25°C), determined experimentally; T motor (°C): Real-time motor temperature of the AGV at the current moment; T max (°C): The rated maximum operating temperature of the motor; γ (dimensionless): Temperature influence coefficient, which indicates the degree of influence of motor temperature rise on additional energy loss (such as increased internal resistance) during start-up and shutdown, and is calibrated through motor bench test.

[0065] 5. Communication Protocol: The central server and lower-level devices use the OPC-UA protocol for data acquisition and command issuance to ensure data interoperability and communication security.

[0066] In this embodiment, the intelligent scheduling and dispatch kernel module adopts a cloud-edge-device collaborative architecture. Specifically: 1. Real-time edge response: Each AGV is equipped with an edge computing unit and has a built-in lightweight deep reinforcement learning (DRL) model. This model is trained by a central cloud server and then distributed to the edge, enabling the AGV to make local real-time decisions, such as dynamic obstacle avoidance and last 50-meter path optimization.

[0067] 2. Distributed Task Bidding: The central server (cloud) is responsible for task parsing and global scheduling, but the generated task instructions are no longer directly assigned to specific AGVs. Instead, they are published to all AGVs (endpoints). AGVs bid autonomously for transportation tasks based on their own status (battery level, location, health) and the DRL model. The central server makes a comprehensive evaluation based on the bidding results (such as lowest energy consumption commitment and fastest response time) and awards the task to the best AGV.

[0068] 3. Rescheduling Triggering and Execution: Local Anomaly Handling (End): When the AGV encounters a temporary blockage, the edge computing unit initiates local path replanning without reporting to the central server, achieving millisecond-level response; Global Task Redistribution (Cloud): When the AGV malfunctions or reaches critical power, the AGV actively releases the task to the central server, which then re-issues the task for bidding, allowing other healthy AGVs to take over the completion.

[0069] 4. Energy Consumption Model Application: When the AGV calculates its own energy consumption locally, it uses a simplified version of the energy consumption model E. local =Σ(d i *L i *R i The central server uses the complete E (Engineering) standard for global scheduling and task evaluation. total The model evaluates the energy efficiency of each AGV's bidding proposal.

[0070] 5. Communication Protocol: Cloud-edge communication adopts the MQTT protocol to achieve lightweight, low-bandwidth task publishing and status reporting; edge-end (AGV internal) communication adopts the CAN bus to ensure the real-time performance and reliability of control commands.

[0071] In summary, by incorporating the physical operating characteristics (position, power, and load) of AGVs directly into scheduling constraints, deep coupling between chemical process logic and AGV dynamic feedback is achieved. This ensures that the scheduling scheme is accurately implemented in the physical world, avoiding resource waste caused by AGVs waiting for experiments or AGVs running idly. By constructing an improved NSGA-III multi-objective optimization model with total completion time, total AGV transportation energy consumption, and chemical timeliness satisfaction as objectives, and introducing a dynamic penalty mechanism based on health, an organic balance is achieved between experimental efficiency, energy efficiency, and timeliness. Through a two-layer architecture of global optimization and local rescheduling, as well as a closed-loop dynamic feedback execution mechanism, the system can automatically trigger rapid local replanning when faced with anomalies such as AGV blockage or low power, significantly improving the system's robustness and dynamic adaptability. Through a refined AGV energy consumption model (including dynamic load factor, path resistance coefficient, and start-stop loss factor) and a visual interactive interface, energy consumption calculation is accurate and controllable, and the scheduling scheme is intuitive and adjustable, ultimately achieving low-energy, high-efficiency, and intelligent operation of chemical experiments.

[0072] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A low-energy chemical experiment AI intelligent scheduling system integrating AGV feedback, characterized in that, include: The experimental task parsing and modeling module is used to decompose the standard operating procedures of chemical experiments into a sequence of atomic tasks with time-dependent and chemical time-dependent constraints, and to define key attributes for each atomic task, including equipment requirements, execution time, bill of materials, and chemical time-dependent window. The multi-source heterogeneous sensing and data fusion module is used to collect and fuse multi-source heterogeneous data in real time to form a unified real-time system status view. The multi-source heterogeneous data includes experimental equipment status information, automated guided vehicle operation status information, and laboratory environment and path information. The intelligent scheduling and scheduling kernel module is used to construct and solve a multi-objective optimization model based on the real-time status view of the system and the atomic task sequence, with the objectives of minimizing the total completion time of the experiment, minimizing the total transportation energy consumption of the automated guided vehicle, and maximizing the chemical timeliness satisfaction, to generate a Pareto optimal scheduling scheme. The interactive scheduling confirmation and visualization module is used to visualize the scheduling scheme and allows users to adjust and confirm the scheme. as well as The instruction issuance and dynamic feedback execution module is used to decompose the confirmed scheduling plan into operation instructions and issue them to the experimental equipment and automated guided vehicles. During the execution process, it triggers a local rescheduling mechanism based on real-time feedback.

2. The low-energy chemical experiment AI intelligent scheduling system integrating AGV feedback as described in claim 1, characterized in that, The intelligent scheduling and scheduling kernel module uses an improved non-dominated sorting genetic algorithm for multi-objective optimization and constructs a multi-objective cost function that includes the total transportation energy consumption of the automated guided vehicle, the total experimental time, and the satisfaction of chemical timeliness. By introducing a dynamic penalty mechanism, the optimization weights are adjusted in real time according to the health of the automated guided vehicle.

3. The low-energy chemical experiment AI intelligent scheduling system integrating AGV feedback as described in claim 1, characterized in that, The multi-source heterogeneous sensing and data fusion module includes: The experimental equipment sensing unit is used to collect the busy / idle status, fault information and operating parameters of each experimental device in real time. The automated guided vehicle (AGV) operation status monitoring unit is used to acquire high-precision coordinates, instantaneous speed, current load, state of charge, and health status of key components for each AGV in real time; and The environment and path perception unit is used to receive traffic control information and identify dynamic obstacles or congested areas.

4. The low-energy chemical experiment AI intelligent scheduling system integrating AGV feedback as described in claim 1, characterized in that, The objective function of the multi-objective optimization model includes: The primary objective is to minimize the total completion time. The second objective is to minimize total transportation energy consumption, which is calculated based on an automated guided vehicle (AGV) energy consumption model, which is expressed as follows: E total =Σ(d i *L i *R i )+Σ(s j *W j ), Where, d i Let L be the distance L travels continuously in the i-th segment of a transportation task. i R is the dynamic load energy consumption factor. i s is the path resistance coefficient. j For the j-th start-stop event that occurs during the entire mission of the automated guided vehicle, W j This refers to the start-stop loss factor. The third objective is to maximize the satisfaction of chemical timeliness requirements.

5. The low-energy chemical experiment AI intelligent scheduling system integrating AGV feedback as described in claim 4, characterized in that, The dynamic load energy consumption factor L i Defined as L i =1+α*(m load / m rated ), where m load m is the actual load of the automated guided vehicle during the i-th segment of its journey. rated The rated maximum load of the automated guided vehicle is α, where α is the load energy consumption coefficient; the path resistance coefficient R is... i Defined as R i =R base +β*(1-H agv ), where R base H is the reference road surface resistance coefficient. agv β represents the current overall health status of the automated guided vehicle, and β is the health status influencing factor; the start-stop loss factor W j Defined as W j =W base *(1+γ*T motor / T max ), where W base Based on the start-stop energy consumption, T motor T represents the real-time temperature of the motor of the automated guided vehicle at the current moment. max γ is the rated maximum operating temperature of the motor, and γ is the temperature influence coefficient.

6. The low-energy chemical experiment AI intelligent scheduling system integrating AGV feedback as described in claim 1, characterized in that, The intelligent scheduling and dispatching kernel module adopts a centralized architecture and is deployed on a central server. The central server uniformly issues task instructions to all experimental equipment and automated guided vehicles, and uniformly calculates the path planning of automated guided vehicles. The multi-source heterogeneous sensing and data fusion module also includes an LSTM time series prediction model, which is used to predict the power consumption of automated guided vehicles and the task completion time of equipment within a preset time in the future, and uses the prediction data as the input of the multi-objective optimization model.

7. The low-energy chemical experiment AI intelligent scheduling system integrating AGV feedback as described in claim 1, characterized in that, The intelligent scheduling and dispatching kernel module adopts a distributed architecture that coordinates cloud, edge, and terminal. The cloud central server is responsible for task parsing and global scheduling, and publishes the generated task instructions to all automated guided vehicles (AGVs). The edge computing unit is mounted on each AGV and has a built-in lightweight deep reinforcement learning model to realize local real-time decision-making for the AGVs. Each AGV acts as an edge device and bids autonomously for transportation tasks based on its own state and the deep reinforcement learning model. The cloud central server assigns the task to the best AGV based on the bidding results.

8. The low-energy chemical experiment AI intelligent scheduling system integrating AGV feedback as described in claim 1, characterized in that, In the instruction issuance and dynamic feedback execution module, when the automated guided vehicle is detected to be blocked or its battery level is lower than a preset threshold, or when the deviation between real-time monitoring data and predicted data exceeds a preset threshold, a local rescheduling mechanism is triggered to lock the affected task set, call the multi-objective optimization model to perform local rapid replanning, and dynamically update the optimized instructions to the corresponding devices.

9. A low-energy chemical experiment AI intelligent scheduling method integrating AGV feedback, applied to the system according to any one of claims 1 to 8, characterized in that, Includes the following steps: S1 Experimental Task Analysis and Modeling: Read the standard operating procedures of chemical experiments, use natural language processing technology to decompose them into a sequence of atomic tasks with time dependence and chemical time constraints, define key attributes for each atomic task and store them in the task pool. S2 Real-time perception and fusion of multi-source heterogeneous data: Real-time acquisition of experimental equipment status information, automated guided vehicle operation status information, and laboratory environment and path information, and time alignment and spatial synchronization of multi-source heterogeneous data to form a unified real-time system status view; S3 Constructs and solves a multi-objective optimization scheduling model: Based on the real-time state view and task pool of the system, a multi-objective optimization model is constructed with the objectives of minimizing the total completion time of the experiment, minimizing the total transportation energy consumption of the automated guided vehicle, and maximizing the satisfaction of chemical timeliness. An improved non-dominated sorting genetic algorithm is used to solve the model and generate a set of Pareto optimal scheduling schemes. S4 Interactive Scheduling Scheme Confirmation and Adjustment: The generated Pareto optimal scheduling scheme is visualized in the form of a Gantt chart on a graphical interface, allowing users to adjust the scheme and select the final execution scheme after user confirmation. S5 command issuance and closed-loop dynamic feedback execution: The final solution is decomposed into specific operation commands, which are issued to the experimental equipment controller and the automated guided vehicle control system respectively. During the execution process, the execution status is continuously monitored. When an abnormal event occurs, a local rescheduling mechanism is triggered, and the multi-objective optimization model is called to perform local rapid replanning until all experimental tasks are completed.

10. The low-energy chemical experiment AI intelligent scheduling method integrating AGV feedback as described in claim 9, characterized in that, In step S3, the total transportation energy consumption of the automated guided vehicle is calculated based on the automated guided vehicle energy consumption model, which is expressed as follows: E total =Σ(d i *L i *R i )+Σ(s j *W j ), Where, d i Let L be the distance L travels continuously in the i-th segment of a transportation task. i R is the dynamic load energy consumption factor. i s is the path resistance coefficient. j For the j-th start-stop event that occurs during the entire mission of the automated guided vehicle, W j The start-stop loss factor; the dynamic load energy consumption factor L i Defined as L i =1+α*(m load / m rated ), where m load m is the actual load of the automated guided vehicle during the i-th segment of its journey. rated The rated maximum load of the automated guided vehicle is α, where α is the load energy consumption coefficient; the path resistance coefficient R is... i Defined as R i =R base +β*(1-H agv ), where R base H is the reference road surface resistance coefficient. agv β represents the current overall health status of the automated guided vehicle, and β is the health status influencing factor; the start-stop loss factor W j Defined as W j =W base *(1+γ*T motor / T max ), where W base Based on the start-stop energy consumption, T motor T represents the real-time temperature of the motor of the automated guided vehicle at the current moment. max γ is the rated maximum operating temperature of the motor, and γ is the temperature influence coefficient.