Industrial park multi-target collaborative optimization scheduling system and method based on artificial intelligence

Abstract

The invention provides an industrial park multi-target collaborative optimization scheduling method and system based on artificial intelligence, relates to the technical field of intelligent scheduling of industrial parks, and aims to establish a digital twinborn model and a dynamic topological graph, perform resource flow relationship modeling by using a graph neural network, and perform resource demand analysis in combination with a multi-model prediction framework. A multi-agent reinforcement learning algorithm is adopted to generate a scheduling strategy, migration from a global strategy to a local decision module is realized through a strategy distillation technology, robustness of the scheduling strategy is simulated and verified in a digital twin model, the strategy is adjusted according to a simulation result, and finally efficient scheduling of park resources is realized. The method effectively improves the intelligentization and collaboration level of the operation of the industrial park, reduces the energy consumption and carbon emission, enhances the ability of the park to cope with complex scenes, and is suitable for the efficient management of the modern industrial park.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent scheduling in industrial parks, and more specifically, to a multi-objective collaborative optimization scheduling system and method for industrial parks based on artificial intelligence. Background Art

[0002] In the context of the rapid development of digitalization and intelligentization today, the management and operation of industrial parks face many challenges. Traditional scheduling systems and methods often struggle to meet complex and changing production requirements, especially in aspects such as resource optimization, energy management, equipment maintenance, and logistics scheduling. In addition, with the continuous expansion of the scale of industrial parks and the increasing complexity of operations, achieving multi-objective collaborative optimization has become an urgent problem to be solved.

[0003] In recent years, the rapid development of artificial intelligence technology has provided new ideas and solutions for the optimization scheduling of industrial parks. By introducing advanced technologies such as machine learning, deep learning, graph neural networks, and multi-agent reinforcement learning, it is possible to achieve real-time collection and analysis of multi-source data such as the operating status of equipment, energy consumption, and logistics information within industrial parks. For example, using digital twin technology to build a virtual model of an industrial park can realize dynamic simulation and optimization of resource flow relationships. At the same time, an artificial intelligence-based prediction model can accurately predict energy demand and equipment failure probabilities, thereby formulating reasonable scheduling strategies in advance. In addition, the multi-agent reinforcement learning algorithm can generate efficient scheduling strategies, and through a centralized training and distributed execution framework, achieve an organic combination of global optimization and local decision-making. These technologies have improved the resource utilization rate and production efficiency of industrial parks, and also significantly reduced energy consumption and operating costs.

[0004] The prior art has problems such as low scheduling efficiency, low resource utilization rate, and high energy consumption in industrial parks. Summary of the Invention

[0005] In order to overcome the problems in the prior art such as low scheduling efficiency, low resource utilization rate, and high energy consumption in industrial parks, the present invention designs a multi-objective collaborative optimization scheduling system and method for industrial parks based on artificial intelligence, which can effectively solve the above technical problems.

[0006] To solve the above technical problems, the technical solution of the present invention is as follows:

[0007] A multi-objective collaborative optimization scheduling method for an industrial park based on artificial intelligence, comprising the following steps:

[0008] Construct a digital twin model and generate a dynamic topology graph, use Internet of Things technology to collect the operating status of equipment, energy consumption, and logistics information within the industrial park, and construct a digital twin model and a dynamic topology graph based on the collected data. The nodes of the dynamic topology graph include entities such as factories, warehouses, and energy stations, and the edges represent the resource flow relationships between the nodes;

[0009] Perform relationship modeling using a graph neural network, define node and edge features, calculate the correlation weights between nodes through an attention mechanism, and use the graph neural network to learn the resource flow characteristics between nodes in the industrial park, outputting an optimized logistics path plan and energy allocation weights;

[0010] Conduct resource demand analysis based on a multi-model prediction framework, use a random forest model to predict energy demand fluctuations, adopt a support vector machine algorithm to predict equipment failure probabilities, and use a decision tree algorithm to classify and prioritize resource demands;

[0011] Adopt a multi-agent reinforcement learning algorithm to generate a scheduling strategy, define a global reward function, generate a joint action strategy through a centralized training - distributed execution framework, and use policy distillation technology to transfer the global policy to the local decision-making modules of each agent;

[0012] Simulate and verify the robustness of the scheduling strategy in the digital twin model, adjust the strategy according to the simulation results, generate the final scheduling instructions and issue them to the park execution terminals.

[0013] Preferably, in the step of performing relationship modeling using a graph neural network, the nodes include equipment type, equipment status, energy consumption rate, and logistics throughput, and the edges include resource flow type, resource flow rate, and resource flow priority;

[0014] The attention mechanism determines the correlation weights between the nodes by calculating the Euclidean distance and cosine similarity between the nodes, and the correlation weights are used to adjust the edge weights of the graph neural network.

[0015] Preferably, in the step of conducting resource demand analysis based on a multi-model prediction framework, it further includes:

[0016] Use a long short-term memory network to model the time series data of the equipment operating state to predict the potential trend of equipment failures, and fuse the prediction results with the prediction results of the support vector machine algorithm to generate a comprehensive equipment failure probability prediction value;

[0017] The priority sorting logic of the decision tree algorithm includes: dividing the resource emergency level according to the equipment failure probability threshold, and combining the predicted value of energy demand fluctuations to generate a dynamic weight table for resource scheduling, and its weight calculation formula is:

[0018] W = α·P fault +β·|D predicted -D current |

[0019] where α and β are adjustable parameters, P fault is the failure probability, Dpredicted and D current are the predicted demand and the current demand respectively.

[0020] Preferably, the generation of the joint action policy through the centralized training-distributed execution framework and the migration of the global policy to the local decision-making modules of each agent using the policy distillation technology include:

[0021] Define the global reward function as a multi-objective weighted sum:

[0022] R global = λ 1 ·R cost + λ 2 ·R energy + λ 3 ·R carbon

[0023] where λ 1 , λ 2 , λ 3 are dynamically adjusted weight coefficients; R global is the global reward, R cost is the reward related to cost, R energy is the reward related to energy consumption efficiency, R carbon is the reward related to carbon emissions;

[0024] Adopt the centralized training-distributed execution framework, share the local observation information of the agents during the training phase, and generate the joint action policy;

[0025] Migrate the global policy to the local decision-making modules of each agent through the policy distillation technology.

[0026] Preferably, in the step of simulating and verifying the robustness of the scheduling policy in the digital twin model, it further includes:

[0027] Conduct random perturbation tests on the scheduling policy through the Monte Carlo simulation method, evaluate the adaptability of the scheduling policy under different fault scenarios and demand fluctuations, and optimize and adjust the scheduling policy according to the test results;

[0028] In the step of generating the final scheduling instruction and sending it to the park execution terminal, it further includes:

[0029] Transmit the scheduling instruction to the execution terminal in real time through the industrial Internet of Things platform, and conduct real-time monitoring and adjustment of the execution situation of the scheduling instruction through the feedback mechanism of the execution terminal.

[0030] An industrial park multi-objective collaborative optimization scheduling system based on artificial intelligence, including:

[0031] A data acquisition module, which is used to collect the operating status of equipment, energy consumption and logistics information in the industrial park by using Internet of Things technology;

[0032] A digital twin construction module, which is used to construct a digital twin model and a dynamic topology map based on the collected data;

[0033] A relationship modeling module, which is used to perform relationship modeling by using a graph neural network and output an optimized logistics path plan and energy allocation weights;

[0034] A resource demand analysis module, which is used to perform resource demand analysis based on a multi-model prediction framework;

[0035] A scheduling strategy generation module, which is used to generate a scheduling strategy by using a multi-agent reinforcement learning algorithm;

[0036] A verification and adjustment module, which is used to simulate and verify the robustness of the scheduling strategy in the digital twin model, adjust the strategy according to the simulation results, and generate a final scheduling instruction.

[0037] An electronic device, including a processor, a memory, and a computer program stored on the memory and executable on the processor. When the computer program is executed by the processor, the steps of the above-mentioned multi-objective collaborative optimization scheduling method for an industrial park based on artificial intelligence are implemented.

[0038] A computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned multi-objective collaborative optimization scheduling method for an industrial park based on artificial intelligence are implemented.

[0039] Compared with the prior art, the beneficial effects of the present invention are as follows: By constructing a digital twin model and a dynamic topology map, it can reflect the operating status of the industrial park in real time and accurately, provide data support for scheduling decisions, improve the accuracy and real-time performance of scheduling. Using graph neural networks for relationship modeling to capture the complex resource flow characteristics in the industrial park, optimize the logistics path and energy distribution, improve resource utilization efficiency, and reduce operating costs. Based on the resource demand analysis of the multi-model prediction framework, it can accurately predict the energy demand fluctuations and the probability of equipment failures. Through the decision tree algorithm for priority sorting, it ensures the efficient allocation of key resources and the emergency response ability. The scheduling strategy generated by the multi-agent reinforcement learning algorithm, through the centralized training-distributed execution framework and policy distillation technology, realizes the coordination of global policies and local decisions, enhances the adaptability and flexibility of the scheduling strategy. Finally, by simulating and verifying the robustness of the scheduling strategy in the digital twin model and combining with the Monte Carlo simulation method for optimization and adjustment, the present invention ensures the stability and reliability of the scheduling strategy in different scenarios. At the same time, through the industrial Internet of Things platform, the real-time transmission and monitoring of scheduling instructions are realized, improving the overall operating efficiency and management level of the industrial park. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only exemplary, and for those of ordinary skill in the art, without creative efforts, other implementation drawings can be obtained based on the provided drawings.

[0041] Figure 1 FIG. [X] is a structural diagram of a multi-objective collaborative optimization scheduling system for an industrial park based on artificial intelligence;

[0042] Figure 2 FIG. [Y] is a flowchart of the steps of a multi-objective collaborative optimization scheduling method for an industrial park based on artificial intelligence. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] The drawings are only for exemplary illustration and should not be construed as a limitation of this patent;

[0044] To better illustrate this embodiment, some components in the drawings will be omitted, enlarged or reduced, which do not represent the actual size of the product;

[0045] For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0046] The technical solutions of the present invention will be further described below with reference to the drawings and embodiments.

[0047] Embodiment 1

[0048] An artificial intelligence-based multi-objective collaborative optimization scheduling method for industrial parks, as Figure 1 shown, includes the following steps:

[0049] Construct a digital twin model and generate a dynamic topology map. Use Internet of Things technology to collect the operating status of equipment, energy consumption, and logistics information in the industrial park, and construct a digital twin model and a dynamic topology map based on the collected data. The nodes of the dynamic topology map include factory, warehouse, and energy station entities, and the edges represent the resource flow relationships between the nodes;

[0050] Deploy a large number of Internet of Things sensors in the park, covering equipment operation status monitoring sensors, such as motor speed, temperature, pressure sensors, etc., energy consumption metering instruments, electricity, gas, water resource meters, etc., and logistics tracking devices, RFID tags, GPS locators, etc.

[0051] The sensors collect data in real time and transmit it to the park data center. Based on the collected data, the data center uses modeling software to construct a digital twin model, accurately reproducing the appearance, layout, and internal equipment structure of each factory, warehouse, and energy station entity in the park. At the same time, according to the actual resource flow relationships between the entities, a dynamic topology map is generated. For example, Factory A transports a certain amount of finished products to Warehouse B every day, and Energy Station C provides electricity and heat for Factory A and Warehouse B. These relationships are presented in the dynamic topology map by connecting the corresponding nodes with directed edges, and the thickness of the edges can intuitively reflect the magnitude of the resource flow.

[0052] Use graph neural networks for relationship modeling, define node and edge features, calculate the correlation weights between nodes through the attention mechanism, and use the graph neural networks to learn the resource flow characteristics between nodes in the industrial park, and output an optimized logistics path plan and energy allocation weights;

[0053] Define node and edge features. In terms of node features, the equipment types cover machine tools, cranes, conveyor belts, etc., the equipment status is divided into normal operation, standby, fault repair, etc., the energy consumption rate is calculated based on the real-time energy consumption data of different equipment, and the logistics throughput statistics the amount of goods entering and leaving each node per unit time; in terms of edge features, the resource flow types include raw material transportation, finished product distribution, energy transmission, etc., the resource flow rate is calculated based on the real-time flow data collected by the Internet of Things, and the resource flow priority is comprehensively set according to the urgency of production tasks, the demand for energy supply stability, etc.

[0054] The Euclidean distance and cosine similarity between nodes are calculated using the attention mechanism. For two adjacent factory nodes, if they have a high similarity in equipment type and a similar trend in energy consumption rate fluctuations, a high correlation weight is obtained through complex calculations. This weight is used to dynamically adjust the edge weights of the graph neural network, enabling it to focus more on the relationships between closely related nodes when learning the characteristics of resource flow. Finally, an optimized logistics path is output, such as planning a distribution route that avoids traffic congestion sections and reduces transportation distance, as well as a reasonable energy allocation weight to ensure that key production links are supplied with energy first.

[0055] Resource demand analysis is carried out based on a multi-model prediction framework. The random forest model is used to predict energy demand fluctuations, the support vector machine algorithm is adopted to predict the probability of equipment failure, and the decision tree algorithm is utilized to classify and prioritize resource demands.

[0056] Energy demand prediction: Using the random forest model, input multi-dimensional features such as historical energy consumption data, the current day's weather conditions, factors affecting energy-consuming equipment such as air conditioners, the factory production plan, and energy consumption differences of different products to predict energy demand fluctuations in the next few hours or even days. If a factory plans to increase production of high-energy-consuming products and the temperature suddenly rises on the same day, the model can accurately predict a significant increase in energy demand.

[0057] Equipment failure prediction: Adopting the support vector machine algorithm, combining historical data of equipment operation parameters, maintenance records, etc., to predict the probability of equipment failure. At the same time, the long short-term memory network is used to model the time series data of the equipment operation state to capture the trend of equipment performance degradation. For example, if the vibration amplitude of a key machine tool has been gradually increasing and the temperature has been continuously rising recently, the long short-term memory network predicts an increased risk of its failure, and the prediction results of the two are fused to generate a comprehensive prediction value of the equipment failure probability.

[0058] Classification and prioritization of resource demands: Using the decision tree algorithm, the emergency level of resources is divided into high, medium, and low according to the set threshold of the equipment failure probability. Combining the predicted value of energy demand fluctuations, a dynamic weight table for resource scheduling is generated according to the given weight calculation formula. When the failure probability of a core equipment exceeds the threshold and the energy demand is tight, the corresponding resource emergency level is high, and maintenance personnel, spare equipment, and sufficient energy are preferentially allocated during scheduling.

[0059] The multi-agent reinforcement learning algorithm is used to generate a scheduling strategy. A global reward function is defined, a joint action strategy is generated through a centralized training - distributed execution framework, and the global strategy is migrated to the local decision-making modules of each agent using policy distillation technology.

[0060] Define the global reward function. According to the park operation goals, determine the global reward function in the form of a multi-objective weighted sum. Since the cost control pressure is high in the current stage, set a high reward weight related to cost. For example, for every 1% reduction in cost, the reward increases by 10 points; when the energy consumption efficiency is significantly improved, give reward points related to energy consumption. For example, for every 5% reduction in energy consumption, the reward increases by 8 points; if the carbon emissions are lower than the standard value, give carbon emission-related rewards in proportion. By dynamically adjusting the weight coefficients, guide the scheduling strategy to develop in the optimal direction.

[0061] Centralized training - distributed execution framework. During the training stage, each intelligent agent in the park, such as the intelligent agent responsible for logistics scheduling, the energy management and control intelligent agent, the equipment maintenance intelligent agent, etc., shares local observation information. For example, the logistics intelligent agent summarizes information such as road congestion conditions and vehicle positions in transit, and the energy intelligent agent summarizes information such as energy reserves and real-time power generation power, and jointly participates in training to generate a joint action strategy. For example, in the face of a sudden peak in orders and a tight energy supply, the joint strategy coordinates the logistics vehicles to give priority to delivering emergency orders, the energy intelligent agent reasonably allocates energy to ensure key production, and the equipment maintenance intelligent agent conducts pre-inspections on easily faulty equipment in advance.

[0062] Application of policy distillation technology. After training, use policy distillation technology to transfer the global policy to the local decision-making module of each intelligent agent. Taking the logistics intelligent agent as an example, localize the optimal distribution path selection strategy learned from the joint strategy under different road conditions and order demands, so that it can make quick decisions based on local real-time information during actual operation without frequent interaction with other intelligent agents.

[0063] Simulate and verify the robustness of the scheduling strategy in the digital twin model, and adjust the strategy according to the simulation results to generate the final scheduling instruction and send it to the park execution terminal.

[0064] Use the Monte Carlo simulation method to conduct a large number of random perturbation tests on the generated scheduling strategy, simulate different scenarios of sudden equipment failures, such as the sudden shutdown of core production equipment, and situations of large fluctuations in energy demand, such as the need for internal emergency energy allocation in the park due to an external power grid failure. Simulate the operation of the scheduling strategy multiple times, observe the changes in key indicators such as the overall production, logistics, and energy supply in the park, evaluate its adaptability, and if it is found that a certain strategy causes serious logistics congestion and production stagnation in a specific failure scenario, mark the problem in a timely manner.

[0065] Instruction Issuance and Monitoring Adjustment: The final optimized scheduling instructions are transmitted in real time to each execution terminal in the park through the industrial Internet of Things platform, including the control systems of factory automated production lines, logistics vehicle scheduling terminals, energy supply equipment control cabinets, etc. The execution terminals perform operations according to the instructions and, through built-in feedback mechanisms, such as sensors transmitting execution results and equipment operating status updates, etc., monitor the execution of the scheduling instructions in real time. When execution deviations are found, such as logistics vehicles deviating from the predetermined route or equipment not adjusting power as required, the adjustment mechanism is immediately triggered, and the control center re-optimizes the scheduling or remotely intervenes to correct it.

[0066] In the step of using the graph neural network for relationship modeling, the nodes include equipment type, equipment status, energy consumption rate, and logistics throughput, and the edges include resource flow type, resource flow rate, and resource flow priority;

[0067] The attention mechanism determines the association weight between the nodes by calculating the Euclidean distance and cosine similarity between the nodes, and the association weight is used to adjust the edge weight of the graph neural network.

[0068] In the step of performing resource demand analysis based on the multi-model prediction framework, it further includes:

[0069] Using a long short-term memory network to model the time series data of the equipment operating status to predict the potential trend of equipment failures, and fusing the prediction results with the prediction results of the support vector machine algorithm to generate a comprehensive equipment failure probability prediction value;

[0070] The priority sorting logic of the decision tree algorithm includes: dividing the resource emergency level according to the equipment failure probability threshold, and combining the predicted value of energy demand fluctuation to generate a dynamic weight table for resource scheduling, and its weight calculation method is:

[0071] W = α·P fault +β·|D predicted -D current |

[0072] Where α and β are adjustable parameters, P fault is the failure probability, D predicted and D current are the predicted demand and the current demand respectively.

[0073] The step of generating a joint action policy through the centralized training - distributed execution framework and using policy distillation technology to migrate the global policy to the local decision-making modules of each agent includes:

[0074] Define the global reward function as a multi-objective weighted sum:

[0075] R global = λ1 ·R cost + λ 2 ·R energy + λ 3 ·R carbon

[0076] where λ 1 ,λ 2 ,λ 3 are dynamically adjusted weight coefficients; R global is the global reward, R cost is the reward related to cost, R energy is the reward related to energy consumption efficiency, R carbon is the reward related to carbon emissions;

[0077] Adopt a centralized training - distributed execution framework, share the local observation information of the agents during the training phase, and generate a joint action policy;

[0078] Transfer the global policy to the local decision - making modules of each agent through policy distillation technology.

[0079] In the step of simulating and verifying the robustness of the scheduling policy in the digital twin model, it further includes:

[0080] Conduct random perturbation tests on the scheduling policy through the Monte Carlo simulation method, evaluate the adaptability of the scheduling policy under different fault scenarios and demand fluctuations, and optimize and adjust the scheduling policy according to the test results;

[0081] In the step of generating the final scheduling instruction and sending it to the park execution terminal, it further includes:

[0082] Transmit the scheduling instruction to the execution terminal in real - time through the industrial Internet of Things platform, and monitor and adjust the execution situation of the scheduling instruction in real - time through the feedback mechanism of the execution terminal.

[0083] Embodiment 2

[0084] An industrial park multi - objective collaborative optimization scheduling system based on artificial intelligence, as Figure 2 shown, includes:

[0085] A data acquisition module, used to collect the operation status of equipment, energy consumption situation, and logistics information in the industrial park by using Internet of Things technology;

[0086] A digital twin construction module, used to construct a digital twin model and a dynamic topology map based on the collected data;

[0087] A relationship modeling module, used to perform relationship modeling by using graph neural networks and output an optimized logistics path plan and energy allocation weights;

[0088] A resource requirement analysis module for performing resource requirement analysis based on a multi-model prediction framework;

[0089] A scheduling strategy generation module for generating a scheduling strategy by using a multi-agent reinforcement learning algorithm;

[0090] A verification and adjustment module for simulating and verifying the robustness of the scheduling strategy in the digital twin model, and adjusting the strategy according to the simulation results to generate a final scheduling instruction.

[0091] An electronic device includes a processor, a memory, and a computer program stored on the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of the above-mentioned multi-objective collaborative optimization scheduling method for an industrial park based on artificial intelligence.

[0092] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, it implements the steps of the above-mentioned multi-objective collaborative optimization scheduling method for an industrial park based on artificial intelligence.

[0093] In a specific implementation, various Internet of Things sensors are deployed. High-precision vibration and temperature sensors are installed on equipment such as chip mounters and reflow soldering machines in the electronics manufacturing workshop to monitor the operation status of the equipment in real time and determine whether the equipment is operating normally and whether there are potential fault risks; at equipment such as ultra-low temperature refrigerators and centrifuges in the biomedical R & D laboratory, sensors for energy consumption monitoring and operation duration recording are installed to master the details of energy consumption; in terms of logistics, RFID readers and tags are fully laid on the entrance and exit of the park, warehouse shelves, and transport vehicles to accurately track the flow trajectories of raw materials, semi-finished products, and finished products. These sensors transmit data to the park data center through a low-power Bluetooth or Wi-Fi network.

[0094] Based on the massive real-time data collected, the digital twin construction module uses advanced modeling tools to build a digital twin model of the park. Based on the geographical information of the park, it three-dimensionally presents the appearance and internal layout of each factory building, warehousing facility, and energy supply station. For the electronics factory, the model is detailed to the layout of each process equipment on the chip packaging production line; the biomedical factory accurately restores key areas such as cell culture rooms and purification workshops. At the same time, the dynamic topology map reflects the resource flow in real time, such as the raw material transportation path from the chemical raw material warehouse to the new material synthesis workshop, and the power distribution link from the energy station to each high-energy-consuming workshop, and is updated in real time as the production process progresses.

[0095] The relationship modeling module activates the graph neural network and defines that the node features include the device process types, such as the SMT process in electronics manufacturing, the fermentation process equipment in biopharmaceuticals, the device health status, the energy consumption rate, and the logistics processing capacity; the edge features cover the resource flow categories, flow rates, and priorities. Through the attention mechanism that calculates the Euclidean distance and cosine similarity between nodes, the connection of key nodes is strengthened. For example, high correlation weights are assigned to the device chain nodes that ensure the production of core products. Finally, an optimized logistics route is output, avoiding the peak human flow areas in the park and shortening the transportation time; the energy is reasonably allocated to ensure stable power supply for key process links.

[0096] Using a multi-model prediction framework, the random forest model combines historical energy consumption data and weather forecasts. Temperature affects the heat dissipation energy consumption of electronic devices, humidity is related to the energy consumption for the preservation of biopharmaceuticals, and the enterprise production plan predicts the energy demand fluctuations to plan energy allocation in advance. The support vector machine algorithm refers to the past failure data of the device and the changes in real-time operation parameters to predict the failure probability; the long short-term memory network tracks the decline trend of device performance, and fuses the results of the two to accurately give early warnings. The decision tree algorithm divides the resource urgency according to the failure probability threshold (such as setting 0.7 as high risk), and generates a dynamic weight table in combination with the energy demand prediction to give priority to ensuring the maintenance of high-risk devices and the energy supply for key production.

[0097] Define the global reward function and adjust the weights according to the park development strategy. If the current focus is on green development, the reward coefficient related to carbon emissions is increased; if it is at the peak order delivery period, the weights of production efficiency and cost control are increased. Adopt a centralized training - distributed execution framework. Agents such as logistics, energy, and equipment maintenance share local information such as workshop congestion, energy reserves, and equipment inspection results, and jointly train to generate strategies to cope with complex scenarios. In case of a sudden power failure, coordinate the logistics to suspend non-urgent deliveries, the energy agent switches to the backup power supply to ensure key workshops, and the equipment maintenance agent checks for potential hazards. Then, localize the global strategy through policy distillation, and each agent makes quick decisions based on local real-time information.

[0098] In the digital twin model, use Monte Carlo simulation to introduce disturbances such as random device failures, energy supply fluctuations, and temporary order changes, and test the scheduling strategy multiple times. Observe indicators such as production continuity, energy waste, and logistics delays to evaluate the robustness. If it is found that a certain strategy causes the production line to stop waiting for materials during multiple device failures, optimize the material allocation and standby equipment activation rules accordingly. Finally, generate accurate scheduling instructions, which are sent to the factory automation system, logistics scheduling terminal, and energy control cabinet through the industrial Internet of Things platform to achieve real-time and efficient execution.

[0099] The park control center is equipped with high-performance electronic devices. The processor adopts a multi-core architecture and has powerful computing capabilities, enabling it to quickly process complex algorithms. The memory includes high-speed memory for temporarily storing real-time data, and a large-capacity hard disk for storing historical data, model parameters, and program codes. After the computer program is loaded into the memory, it is executed by the processor, driving each module according to the established scheduling method steps to ensure the smooth operation of the system and real-time response to the dynamic changes in the park.

[0100] A highly reliable solid-state drive is selected as the computer-readable storage medium to store the complete scheduling system program. It has fast read and write characteristics and can efficiently complete tasks whether it is loading the program during system startup or updating model parameters and optimizing algorithms during operation, ensuring that the system can quickly recover and stably support the intelligent operation of the park even in the event of sudden situations such as power outages and device restarts.

[0101] The same or similar reference numerals correspond to the same or similar components.

[0102] The terms describing the positional relationship in the drawings are for illustrative purposes only and should not be construed as a limitation of this patent.

[0103] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention and are not limitations on the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to enumerate all the implementation manners here. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the claims of the present invention.

Claims

1. The multi-objective collaborative optimization scheduling method for industrial parks based on artificial intelligence is characterized by: The following steps are involved: Build a digital twin model and generate a dynamic topology map. Use the Internet of Things technology to collect the equipment operation status, energy consumption and logistics information in the industrial park, and build a digital twin model and a dynamic topology map based on the collected data. The nodes of the dynamic topology map include factories, warehouses and energy station entities, and the edges represent the resource flow relationship between nodes. Use graph neural networks to model relationships, define node and edge features, calculate the association weights between nodes through the attention mechanism, and use the graph neural network to learn the resource flow characteristics between nodes in the industrial park, and output logistics path optimization solutions and energy allocation weights; Perform resource demand analysis based on a multi-model forecasting framework, use a random forest model to predict energy demand fluctuations, use a support vector machine algorithm to predict equipment failure probability, and use a decision tree algorithm to classify and prioritize resource demands; A multi-agent reinforcement learning algorithm is used to generate scheduling strategies, define global reward functions, generate joint action strategies through a centralized training-distributed execution framework, and use policy distillation technology to migrate global strategies to local decision modules of each agent. The robustness of the scheduling strategy is simulated and verified in the digital twin model, and the strategy is adjusted according to the simulation results. The final scheduling instructions are generated and sent to the park execution terminal.

2. The multi-objective collaborative optimization scheduling method for industrial parks based on artificial intelligence according to claim 1 is characterized in that: In the step of using a graph neural network to perform relationship modeling, the nodes include equipment type, equipment status, energy consumption rate, and logistics throughput, and the edges include resource flow type, resource flow rate, and resource flow priority; The attention mechanism determines the association weights between the nodes by calculating the Euclidean distance and cosine similarity between the nodes, and the association weights are used to adjust the edge weights of the graph neural network.

3. The multi-objective collaborative optimization scheduling method for industrial parks based on artificial intelligence according to claim 1 is characterized in that: The step of performing resource demand analysis based on the multi-model prediction framework also includes: Use long short-term memory networks to model the time series data of equipment operating status to predict the potential trend of equipment failure, and fuse the prediction results with the prediction results of the support vector machine algorithm to generate a comprehensive prediction value of equipment failure probability; The priority sorting logic of the decision tree algorithm includes: dividing the resource emergency level according to the equipment failure probability threshold, and combining the energy demand fluctuation prediction value to generate a dynamic weight table for resource scheduling, and the weight calculation method is: W=α·P fault +β·|D predicted -D current | Among them, α, β are adjustable parameters, P fault is the failure probability, D predicted and D current They are forecast demand and current demand respectively.

4. The multi-objective collaborative optimization scheduling method for industrial parks based on artificial intelligence according to claim 1 is characterized in that: The method of generating a joint action strategy through a centralized training-distributed execution framework and migrating the global strategy to the local decision-making module of each agent using the strategy distillation technology includes: Define the global reward function as a weighted sum of multiple objectives: R global =λ1·R cost +λ2·R energy +λ3·R carbon Among them, λ1, λ2, λ3 are dynamically adjusted weight coefficients; R global is the global reward, R cost is the reward related to the cost, R energy Rewards related to energy efficiency, R carbon incentives tied to carbon emissions; A centralized training-distributed execution framework is adopted to share the local observation information of the intelligent agent during the training phase and generate a joint action strategy; The global strategy is transferred to the local decision-making module of each agent through policy distillation technology.

5. The multi-objective collaborative optimization scheduling method for industrial parks based on artificial intelligence according to claim 1 is characterized in that: The step of simulating and verifying the robustness of the scheduling strategy in the digital twin model also includes: Conduct random perturbation tests on the scheduling strategy through Monte Carlo simulation method to evaluate the adaptability of the scheduling strategy under different failure scenarios and demand fluctuations, and optimize and adjust the scheduling strategy according to the test results; The step of generating the final dispatch instruction and sending it to the park execution terminal also includes: The scheduling instructions are transmitted to the execution terminal in real time through the industrial Internet of Things platform, and the execution of the scheduling instructions is monitored and adjusted in real time through the feedback mechanism of the execution terminal.

6. An industrial park multi-objective collaborative optimization scheduling system based on artificial intelligence is implemented by the method described in any one of claims 1 to 5, characterized in that: include: Data collection module, used to collect equipment operation status, energy consumption and logistics information in the industrial park using IoT technology; Digital twin construction module, used to build digital twin models and dynamic topology maps based on collected data; The relationship modeling module is used to use graph neural networks to perform relationship modeling and output logistics path optimization solutions and energy allocation weights; Resource demand analysis module, used to perform resource demand analysis based on a multi-model prediction framework; The scheduling strategy generation module is used to generate scheduling strategies using multi-agent reinforcement learning algorithms; The verification and adjustment module is used to simulate and verify the robustness of the scheduling strategy in the digital twin model, adjust the strategy according to the simulation results, and generate the final scheduling instructions.

7. An electronic device, characterized in that: It includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, the steps of the multi-objective collaborative optimization scheduling method for an industrial park based on artificial intelligence described in claims 1-5 are implemented.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the multi-objective collaborative optimization scheduling method for an industrial park based on artificial intelligence as described in claims 1-5.

Images