Artificial intelligence-based industrial production line dynamic scheduling method and system

Abstract

The application discloses an industrial production line dynamic scheduling method and system based on artificial intelligence, relates to the technical field of intelligent manufacturing and industrial intelligent scheduling, and comprises the following steps: collecting real-time state data and performing preprocessing, taking the maximization of production capacity, the minimization of delivery time and the minimization of energy consumption as optimization targets, constructing a multi-objective reinforcement learning model, generating scheduling priority data, adjusting the job allocation of each process node in combination with current equipment load threshold and production bottleneck node information, triggering a rescheduling mechanism according to the scheduling priority data when an abnormal condition is detected, updating the job order and resource allocation result, synchronously feeding back the updated job allocation result and execution effect to the multi-objective reinforcement learning model, and iteratively optimizing the multi-objective reinforcement learning model through a priority experience replay mechanism to realize the continuous optimization of the scheduling strategy. The application combines multi-objective reinforcement learning and dynamic scheduling, and improves the learning efficiency and optimization effect.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent manufacturing and industrial intelligent scheduling, in particular to an industrial production line dynamic scheduling method and system based on artificial intelligence. BACKGROUND

[0002] Traditional scheduling methods based on fixed rules have been difficult to meet the comprehensive requirements of high efficiency, low energy consumption and fast delivery in dynamic environments. In order to improve overall productivity and resource utilization efficiency, the industry has gradually introduced artificial intelligence, data-driven optimization and adaptive learning mechanisms, hoping to achieve intelligent scheduling and dynamic coordination of resources in industrial production lines through real-time sensing, intelligent decision-making and feedback optimization. However, most current scheduling systems still mainly rely on static rule bases or local optimization strategies, lack comprehensive consideration of multi-objective performance indicators, and cannot adapt to complex situations such as device state fluctuations, production bottleneck evolution and sudden abnormal events. In particular, there is still a lot of room for improvement in how to dynamically reconstruct scheduling strategies and improve the response speed and adaptive ability of algorithms to real-time changing environments. In addition, current systems generally fail to achieve continuous evolution of scheduling optimization strategies, and cannot iteratively optimize decision-making models based on feedback information, resulting in limited scheduling effectiveness and insufficient robustness. In view of the above problems, it is urgent to propose an industrial production line dynamic scheduling method and system based on artificial intelligence to realize adaptive and efficient allocation of multi-process node tasks, thereby improving the running performance and intelligence level of the production system in complex dynamic environments.

[0003] CN114548840B discloses a production scheduling method and system based on artificial intelligence, which sorts order requests by obtaining user location information and target time, and introduces user levels for secondary sorting of regional tables to generate a task queue. Although this scheme builds a scheduling model for customer response and improves the personalization and speed of order processing, it mainly focuses on the judgment of external order priority and does not fully consider real-time device status, resource bottlenecks and energy consumption targets in the production process, so it is still insufficient in realizing dynamic coordination and optimal resource allocation within the production process.

[0004] CN115469612B discloses a dynamic scheduling method for job workshops, which realizes dynamic scheduling of process switching and resource allocation between machine tools through normalized priority rules, enhancing the accuracy of process handling. However, this method relies on a pre-set rule base and lacks adaptive model learning ability, making it unable to realize real-time optimization and rapid response when the production environment changes dynamically (such as sudden equipment failure or bottleneck process migration), and the scheduling strategy update requires human intervention, lacking self-evolution ability. SUMMARY

[0005] In view of the following problems existing in the existing industrial production scheduling technology: 1. Lack of comprehensive optimization mechanism for multiple targets such as production capacity, delivery cycle and energy consumption, the scheduling effect is limited; 2. The scheduling algorithm is generally static and lacks adaptability, which is difficult to cope with real-time changing working conditions; 3. Lack of effective scheduling feedback mechanism and model iteration ability, which cannot realize the continuous optimization of scheduling strategy, the present application is proposed.

[0006] Therefore, the problem to be solved by the present application is how to construct an intelligent scheduling method that can dynamically perceive the equipment state and order progress, has multi-target adaptive optimization capability, can quickly reschedule in abnormal state, and can continuously iterate and optimize the scheduling strategy, so as to realize the dynamic balance between maximum production capacity, shortest delivery time and lowest energy consumption, and comprehensively improve the operation efficiency and intelligent decision-making level of the industrial production line.

[0007] To solve the above technical problems, the present application provides the following technical solutions:

[0008] In a first aspect, the present application provides an artificial intelligence-based dynamic scheduling method for an industrial production line, which comprises,

[0009] Collecting real-time state data of each process node in the industrial production line and preprocessing the real-time state data;

[0010] Based on the preprocessed real-time state data, a multi-target reinforcement learning model is constructed with the optimization targets of maximum production capacity, shortest delivery time and lowest energy consumption, and scheduling priority data is generated;

[0011] Based on the scheduling priority data, the job allocation of each process node is adjusted in combination with the current device load threshold and production bottleneck node information;

[0012] When an abnormal condition is detected, a rescheduling mechanism is triggered according to the scheduling priority data to update the job order and resource allocation result;

[0013] The updated job allocation result and execution effect are fed back to the multi-target reinforcement learning model, and the multi-target reinforcement learning model is iteratively optimized through a priority experience replay mechanism to realize continuous optimization of the scheduling strategy.

[0014] As a preferred scheme of the artificial intelligence-based dynamic scheduling method for an industrial production line, wherein: the updated job allocation result and execution effect are fed back to the multi-target reinforcement learning model, and the multi-target reinforcement learning model is iteratively optimized through a priority experience replay mechanism to realize continuous optimization of the scheduling strategy, comprising:

[0015] Record the updated scheduling scheme execution process of each process node, productivity index data, delivery time index data and energy consumption index data, and establish an execution effect evaluation data set;

[0016] Compare the execution effect evaluation data set with the expected index data of the job allocation scheme, calculate the deviation value of each index, and generate scheduling effect deviation data;

[0017] Based on the scheduling effect deviation data, mark the priority score of the state transition sample in the multi-objective reinforcement learning model;

[0018] Store the marked state transition sample into a priority experience pool, and sort the samples according to the priority score;

[0019] Extract the sample group with the top 60% priority score from the priority experience pool to form a training batch, input the value evaluation network, calculate the time series difference error, and update the network weight parameter;

[0020] Based on the time series difference error, adjust the gradient of the decision probability distribution of the strategy generation network, and generate the optimized network model parameter;

[0021] Update the optimized network model parameter to the multi-objective reinforcement learning model to complete the policy iteration.

[0022] As a preferred scheme of the industrial production line dynamic scheduling method based on artificial intelligence, when an abnormal condition is detected, a rescheduling mechanism is triggered according to the scheduling priority data to update the job order and resource allocation result, including:

[0023] Continuously monitor the equipment running state data of each process node, and when the temperature value exceeds the preset upper limit of temperature or the vibration frequency value exceeds the normal range, mark it as an abnormal process node;

[0024] Calculate the remaining job amount and the estimated recovery time of the abnormal process node, and generate abnormal influence evaluation data;

[0025] Based on the abnormal influence evaluation data, extract the job allocation scheme of the upstream three nodes and the downstream three nodes of the abnormal process node, and establish an emergency response buffer;

[0026] Read the scheduling priority data, and transfer the to-be-processed order of the abnormal process node to the available node in the emergency response buffer according to the descending order of priority;

[0027] Statistical the remaining productivity and material storage capacity of the available node, and generate a node resource state table;

[0028] According to the node resource state table, the to-be-processed orders are distributed to the available nodes according to a preset proportion to generate a new job sequence table;

[0029] The expected completion time of each order in the new job sequence table is calculated, and when the expected completion time exceeds the delivery deadline, the corresponding order is marked as an urgent order;

[0030] The urgent order is inserted into the front end position of the current job queue, and the processing sequence of other non-urgent orders is postponed, and an updated scheduling scheme is output.

[0031] As a preferred scheme of the industrial production line dynamic scheduling method based on artificial intelligence, the generation method of the job allocation scheme is,

[0032] Read the scheduling priority data to obtain the load rate value of each process node, wherein the load rate value is equal to the ratio of the current job quantity to the maximum production capacity of the equipment;

[0033] Set a load threshold parameter, and when the load rate value is greater than a first threshold value, the corresponding process node is marked as a high-load node;

[0034] Calculate the preceding task completion rate and the subsequent node idle rate of each process node, and mark the process node with a preceding task completion rate less than a second threshold value or a subsequent node idle rate greater than a third threshold value as a production bottleneck node;

[0035] For the job tasks of the high-load node, select a same-type node with a load rate value less than a fourth threshold value as a standby node according to the order of the scheduling priority data;

[0036] Calculate the material transmission distance from the standby node to the high-load node, and select the standby node with the shortest material transmission distance as the task transfer target node;

[0037] For the production bottleneck node, extract the job storage amount of the upstream node and the production capacity margin of the downstream node to generate a node balance parameter;

[0038] According to the node balance parameter, the job tasks of the production bottleneck node are distributed to three adjacent process nodes according to a preset proportion;

[0039] Generate a job allocation scheme, wherein the job allocation scheme includes task transfer path and allocation proportion data.

[0040] As a preferred scheme of the industrial production line dynamic scheduling method based on artificial intelligence, the generation method of the scheduling priority data is,

[0041] read a normalized state matrix from a production state database, extract equipment running state data, order progress information data and job load condition data in the standardized data matrix, and construct a state vector;

[0042] establish a multi-objective reinforcement learning model, wherein the multi-objective reinforcement learning model adopts a double network structure, including a value evaluation network and a policy generation network;

[0043] based on the state vector, calculate a scheduling priority score of each process node through the value evaluation network, wherein the scheduling priority score includes a production capacity score, a delivery time score and an energy efficiency score;

[0044] input the scheduling priority score into the policy generation network to generate a scheduling priority value of each process node;

[0045] arrange the scheduling priority value in descending order to generate a scheduling priority data.

[0046] As a preferred scheme of the artificial intelligence-based industrial production line dynamic scheduling method, wherein the multi-objective reinforcement learning model is established by,

[0047] construct a state space matrix and an action space matrix, wherein the state space matrix includes process node numbers, equipment running state data, order progress information data and job load condition data; the action space matrix includes scheduling priority options of process nodes;

[0048] based on the state space matrix and the action space matrix, construct a reward function, wherein the reward function includes a production capacity reward item, a delivery reward item and an energy consumption reward item;

[0049] build a value evaluation network and a policy generation network according to the reward function;

[0050] set an experience replay buffer, and store state transition sequences, action selection results and obtained reward values in the experience replay buffer;

[0051] update the parameters of the value evaluation network using a time difference algorithm, wherein the time difference algorithm calculates the error between the target value and the predicted value based on the Bellman equation;

[0052] update the parameters of the policy generation network using a policy gradient algorithm, wherein the policy gradient algorithm optimizes the action selection strategy by maximizing the expected long-term cumulative reward;

[0053] synchronize the parameters of the value evaluation network and the policy generation network using a soft update method to reduce the shock of the training process;

[0054] An experience priority sampling mechanism is set to preferentially select samples with high timing difference errors to train the multi-objective reinforcement learning model.

[0055] As a preferred scheme of the industrial production line dynamic scheduling method based on artificial intelligence, the value evaluation network adopts a three-layer structure including an input layer, a hidden layer and an output layer; the input layer receives a state space matrix; the output layer generates a state value score; the strategy generation network adopts a four-layer structure including an input layer, two hidden layers and an output layer, and the input layer receives the state value score; and the output layer generates an action selection probability.

[0056] In a second aspect, the present application provides an industrial production line dynamic scheduling system based on artificial intelligence, which comprises:

[0057] The acquisition module is configured to transmit real-time state data of each process node in the industrial production line to the central data processing unit for preprocessing.

[0058] The construction module is configured to construct a multi-objective reinforcement learning model based on the preprocessed real-time state data, with the optimization objectives of maximizing production capacity, minimizing delivery time and minimizing energy consumption, and generate scheduling priority data.

[0059] The adjustment module is configured to adjust the job allocation of each process node based on the scheduling priority data and in combination with the current device load threshold and production bottleneck node information.

[0060] The update module is configured to trigger a rescheduling mechanism to update the job order and resource allocation result according to the scheduling priority data when an abnormal condition is detected.

[0061] The optimization module is configured to synchronize the updated job allocation result and execution effect to the multi-objective reinforcement learning model, and iteratively optimize the multi-objective reinforcement learning model through a priority experience replay mechanism to realize continuous optimization of the scheduling strategy.

[0062] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements any step of the above-mentioned industrial production line dynamic scheduling method based on artificial intelligence when executing the computer program.

[0063] In a fourth aspect, the present application provides a computer readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any step of the above-mentioned industrial production line dynamic scheduling method based on artificial intelligence.

[0064] Compared with the prior art, the application has the beneficial effects that: by constructing a multi-target reinforcement learning model with a double-network structure, the capacity, delivery time and energy consumption are unified in the same framework for optimization, overcoming the problem that the traditional method is difficult to balance multiple competitive targets, at the same time, the separation structure of the value evaluation network and the strategy generation network used significantly improves the training stability and convergence speed of the model, avoiding the oscillation in the strategy optimization process; through the dynamic job distribution mechanism combining the load threshold and the bottleneck node information, the balanced distribution of production load is realized on the basis of ensuring the scheduling priority, reducing the material turnover loss, and through the calculation of the node balance parameter, the production bottleneck is effectively relieved, and the overall production efficiency is improved; by establishing a multi-link emergency response mechanism, combining the buffer capacity of upstream and downstream nodes, an efficient abnormal processing flow is formed, an order reallocation method based on a preset proportion is used, and an urgent order priority insertion mechanism is introduced, which not only ensures the production continuity under abnormal conditions, but also maximizes the influence of abnormal events on the delivery time; through the priority experience replay mechanism, the execution effect is evaluated and fed back in real time, a closed-loop continuous optimization system is established, the priority score labeling method based on the scheduling effect deviation data improves the utilization efficiency of samples with important reference value, speeds up the optimization speed of the model, and makes the scheduling strategy continuously adapt to the dynamic changes of the production environment. BRIEF DESCRIPTION OF DRAWINGS

[0065] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor. Among them:

[0066] Figure 1 Flowchart of the dynamic scheduling method for the industrial production line based on artificial intelligence;

[0067] Figure 2 Flowchart of the generation of scheduling priority data for the dynamic scheduling method for the industrial production line based on artificial intelligence. DETAILED DESCRIPTION

[0068] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0069] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that the present application can be practiced without the specific details set forth in this description. In other instances, well-known methods, procedures, components, and circuits have not been described in detail so as not to unnecessarily obscure aspects of the present application.

[0070] Secondly, the "one embodiment" or "an embodiment" referred to herein means containing a specific feature, structure, or characteristic in at least one implementation of the present application. The "in one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an embodiment that is independent of or mutually exclusive with other embodiments.

[0071] The present application is described in detail below in conjunction with the schematic drawings, and in the detailed description of the embodiments of the present application, the sectional view of the device structure is partially enlarged without the general proportion for the convenience of illustration, and the schematic drawings are only examples, which should not limit the scope of protection of the present application herein. In addition, three-dimensional spatial dimensions including length, width and depth should be included in actual production.

[0072] Meanwhile, in the description of the present application, it should be noted that the terms "upper, lower, inner and outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the devices or elements referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first, second or third" are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0073] In the present application, unless otherwise explicitly specified and limited, the terms "mounting, connecting, and connecting" should be understood broadly, for example: it can be fixedly connected, detachably connected or integrally connected; it can also be mechanically connected, electrically connected or directly connected, it can also be indirectly connected through an intermediate medium, or it can be the internal communication of two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0074] Embodiment 1

[0075] Reference Figures 1-2 For the first embodiment of the present application, the embodiment provides an artificial intelligence-based industrial production line dynamic scheduling method, comprising:

[0076] S1: Collecting real-time state data of each process node in the industrial production line, and pre-processing the real-time state data.

[0077] In an optional implementation, the real-time state data includes device running state data, order progress information data, and job load condition data.

[0078] S1.1: Install data acquisition sensors at each process node in the industrial production line.

[0079] It should be noted that data acquisition sensors include equipment operation sensors, workpiece tracking sensors, and load monitoring sensors.

[0080] In an optional implementation, temperature, rotation speed, current, and vibration frequency values ​​at each process node are collected by equipment operation sensors to obtain equipment operation status data; the number of completed workpieces, workpiece processing time, and workpiece waiting time at each process node are collected by workpiece tracking sensors to obtain order progress information data; and the amount of material accumulation, work-in-process quantity, and equipment idle time at each process node are collected by load monitoring sensors to obtain work load data.

[0081] S1.2: Integrate equipment operating status data, order progress information data, and work load data into a status data packet, and transmit it to the data preprocessing unit via industrial Ethernet.

[0082] In an optional implementation, the data preprocessing unit performs abnormal data detection and removal on the state data packets to generate a normalized state matrix.

[0083] In an optional implementation, anomaly detection is performed on the temperature, current, and vibration frequency values ​​in the status data packet based on the 3σ criterion. If the current value exceeds the rated value by ±15%, the bearing temperature value exceeds the threshold, or the vibration frequency value exceeds the safe range for three consecutive samples, it is marked as abnormal data and removed, generating a post-cleaning status dataset. For the post-cleaning status dataset, linear interpolation is performed on the number of completed workpieces, workpiece processing time, and workpiece waiting time based on the collection timestamp to generate a synchronized status data sequence. Fast Fourier Transform is performed on the equipment vibration amplitude in the synchronized status data sequence to extract the energy proportion of the 0-100Hz frequency band as the vibration feature quantity; PCA algorithm is used to reduce the dimensionality of the temperature, current, and vibration frequency values ​​to generate low-dimensional feature vectors; the low-dimensional feature vectors, order progress information data, and work load data are mapped to the [0,1] interval using Min-Max normalization to generate a normalized status matrix.

[0084] S1.3: Store the normalized state matrix into the production state database according to the preset sampling time interval.

[0085] S2: Based on the preprocessed real-time status data, a multi-objective reinforcement learning model is constructed with the optimization objectives of maximizing capacity, minimizing delivery time, and minimizing energy consumption to generate scheduling priority data.

[0086] S2.1: Read the normalized state matrix from the production state database, extract the equipment running state data, order progress information data and job load data in the standardized data matrix, and construct a state vector.

[0087] S2.2: Establish a multi-objective reinforcement learning model, wherein the multi-objective reinforcement learning model adopts a double network structure, including a value evaluation network and a policy generation network.

[0088] In an optional embodiment, the value evaluation network adopts a three-layer structure, including an input layer, a hidden layer and an output layer; the input layer receives the state space matrix; the output layer generates a state value score; the policy generation network adopts a four-layer structure, including an input layer, two hidden layers and an output layer, the input layer receives the state value score; and the output layer generates an action selection probability.

[0089] Specifically, it includes:

[0090] S2.2.1: Construct a state space matrix and an action space matrix, wherein the state space matrix includes process node number, equipment running state data, order progress information data and job load data; and the action space matrix includes scheduling priority options of process nodes;

[0091] S2.2.2: Based on the state space matrix and the action space matrix, construct a reward function, wherein the reward function includes a capacity reward item, a delivery reward item and an energy consumption reward item;

[0092] Preferably, the specific formula of the reward function is as follows:

[0093] ;

[0094] Wherein, is the reward function value of action a taken under state s, is the capacity reward weight coefficient, is the delivery time reward weight coefficient, is the energy consumption reward weight coefficient, is the total number of process nodes, is the actual capacity of the i-th node, is the maximum capacity limit value, is the total number of orders, is the actual completion time of the j-th order, is the delivery deadline of the j-th order, is the current energy consumption value, is the maximum energy consumption limit value.

[0095] It should be noted that the reward function The value range of the scheduling priority score is [0, 1], wherein 0 represents the worst scheduling state, and 1 represents the optimal scheduling state; the capacity reward term adopts simple normalization, and the value range is [0, 1]; the delivery reward term adopts exponential function mapping, and the value range is [0, 1]; and the energy consumption reward term adopts linear normalization, and the value range is [0, 1].

[0096] S2.2.3: Building a value evaluation network and a policy generation network according to a reward function;

[0097] S2.2.4: Setting an experience replay buffer, and storing a state transition sequence, an action selection result and an obtained reward value into the experience replay buffer;

[0098] S2.2.5: Updating parameters of the value evaluation network by using a time difference algorithm, wherein the time difference algorithm calculates an error between a target value and a predicted value based on a Bellman equation;

[0099] S2.2.6: Updating parameters of the policy generation network by using a policy gradient algorithm, wherein the policy gradient algorithm optimizes an action selection policy by maximizing an expectation of a long-term cumulative reward;

[0100] S2.2.7: Synchronizing parameters of the value evaluation network and the policy generation network by using a soft update method, so as to reduce oscillation in a training process;

[0101] S2.2.8: Setting an experience priority sampling mechanism, and preferentially selecting a sample with a high time difference error to train the multi-objective reinforcement learning model.

[0102] S2.3: Calculating a scheduling priority score of each process node by using the value evaluation network based on a state vector, wherein the scheduling priority score includes a capacity score, a delivery time score and an energy efficiency score.

[0103] S2.4: Inputting the scheduling priority score into the policy generation network to generate a scheduling priority value of each process node.

[0104] Preferably, a specific formula of the scheduling priority value is as follows:

[0105] ;

[0106] wherein, is the scheduling priority value of the action a in the state s, is a weight coefficient of a state-action value, is a weight coefficient of load balancing, is a weight coefficient of waiting time, is an output value of the value evaluation network in the state s, is a maximum operation in all selectable actions a', the current job load of the process node corresponding to action a, the maximum load threshold set for the system, the current waiting time of the process corresponding to action a, the maximum waiting time allowed by the system, the natural exponential function.

[0107] For example, taking an electronic product assembly line as an example, when the value evaluation network outputs the value scores of three optional actions under state s as: a1 (prioritize processing A process) = 0.87, a2 (prioritize processing B process) = 0.92, and a3 (prioritize processing C process) = 0.78; the strategy generation network will combine real-time load data when calculating: assuming that the current load of B process is 45 pieces (threshold 50 pieces) and the waiting time is 120 minutes (threshold 180 minutes), the scheduling priority value formula can be obtained as 0.903; at the same time, a1 = 0.812 and a3 = 0.704 are calculated, so the scheduling priority of B process is set to the highest (0.903), and the generated processing order is B→A→C.

[0108] S2.5: Arrange the scheduling priority values in descending order to generate scheduling priority data.

[0109] S3: Based on the scheduling priority data, combine the current device load threshold and production bottleneck node information to adjust the job allocation of each process node.

[0110] S3.1: Read the scheduling priority data to obtain the load rate value of each process node, wherein the load rate value is equal to the ratio of the current job load to the maximum production capacity of the device.

[0111] Preferably, the specific formula of the load rate value is as follows:

[0112] ;

[0113] wherein, is the load rate value of the bth process node, is the time load weight coefficient, is the queue load weight coefficient, is the cache load weight coefficient, is the number of jobs currently being processed, is the processing time of the ith job, is the complexity coefficient of the ith job, is the standard time period of node b, is the processing capacity coefficient of node b, is the current queue length of node b, is the maximum queue capacity of node b, current available buffer amount of the node b, total buffer capacity of the node n.

[0114] S3.2: Set the load threshold parameter, when the load rate value is greater than the first threshold value, the corresponding process node is marked as a high load node.

[0115] It should be noted that the load threshold parameter includes a first threshold value, a second threshold value, a third threshold value and a fourth threshold value; the first threshold value is set based on the historical maximum load rate statistical value of the equipment; the second threshold value is set based on the average completion rate of the previous task; the third threshold value is set based on the idle rate distribution of the subsequent node; and the fourth threshold value is set based on the average capacity utilization rate of the same type of node.

[0116] S3.3: Calculate the previous task completion rate and the subsequent node idle rate of each process node, and mark the process node with a previous task completion rate less than the second threshold value or a subsequent node idle rate greater than the third threshold value as a production bottleneck node.

[0117] S3.4: For the job task of the high load node, select the same type of node with a load rate value less than the fourth threshold value as a backup node according to the order of the scheduling priority data.

[0118] S3.5: Calculate the material transfer distance from the backup node to the high load node, and select the backup node with the shortest material transfer distance as the task transfer target node.

[0119] Preferably, the specific formula of the material transfer distance is as follows:

[0120] ;

[0121] wherein, the material transfer distance from station i to station j, and the coordinate position of station i, and the coordinate position of station j, the path coefficient, the obstacle bypass coefficient, the number of obstacles on the transfer path, the bypass indicator (0 or 1) of the kth obstacle, the additional bypass distance caused by the kth obstacle.

[0122] S3.6: For the production bottleneck node, extract the job storage amount of the upstream node and the capacity surplus of the downstream node to generate a node balance parameter.

[0123] S3.7: According to the node balance parameter, the job task of the production bottleneck node is distributed to three adjacent process nodes according to a preset proportion.

[0124] For example, the typical processing procedure of a bottleneck node in an injection molding workshop is as follows: it is detected that the upstream module of the injection molding machine (node X) has accumulated 200 pieces (normal value < 50), and the idle rate of the downstream spraying process is 40%; the system calculates the node balance parameters and finds that the upstream storage capacity / the downstream surplus = 200 / (500x0.4) = 1.0 (warning value > 0.8); according to the preset "30%-40%-30%" distribution rule, the current injection molding task is divided into three parts: 30% is transferred to the standby injection molding machine Y (transmission distance 15 meters, path coefficient 1.2) closest to the same workshop, 40% is allocated to the original node X for continuous processing, and the remaining 30% is dispatched to the injection molding machine Z across the workshop (obstacle coefficient 1.5); real-time monitoring of the load rate of each node after adjustment: X node from 95% to 58%, Y node from 30% to 65%, Z node from 45% to 72%, overall production capacity increased by 32% while avoiding single node overload.

[0125] S3.8: generating a job allocation scheme, wherein the job allocation scheme includes task transfer path and allocation ratio data.

[0126] S4: when an abnormal condition is detected, triggering a rescheduling mechanism according to the scheduling priority data to update the job order and resource allocation results.

[0127] S4.1: continuously monitor the equipment operating state data of each process node, and when the temperature value exceeds the preset upper limit of temperature or the vibration frequency value exceeds the normal range, mark it as an abnormal process node.

[0128] S4.2: calculate the remaining job quantity and estimated recovery time of the abnormal process node, and generate abnormal influence evaluation data.

[0129] Preferably, the related formulas of the remaining job quantity and the estimated recovery time are as follows:

[0130] ;

[0131] wherein, is the estimated recovery time, is the number of remaining job types, is the unit work quantity coefficient of the i-th type of job, is the remaining number of the i-th type of job, is the current processing capacity, is the continuous working time, is the maximum continuous working time limit.

[0132] S4.3: based on the abnormal influence evaluation data, extracting the job allocation scheme of the upstream three-level nodes and the downstream three-level nodes of the abnormal process node, and establishing an emergency response buffer zone.

[0133] S4.4: Read the scheduling priority data, and transfer the abnormal process nodes of the orders to be processed to the available nodes in the emergency response buffer according to the descending order of priority.

[0134] S4.5: Calculate the remaining capacity and material storage capacity of the available nodes, and generate a node resource status table.

[0135] S4.6: Based on the node resource status table, distribute the orders to be processed to the available nodes according to the preset proportion, and generate a new job order table.

[0136] For example, when a numerical control milling machine suddenly overheats and alarms, the system executes the following emergency procedures:

[0137] The remaining processing capacity includes 3 types of workpieces (20 pieces of type A x 1.2 coefficient, 15 pieces of type B x 0.8 coefficient, and 8 pieces of type C x 1.5 coefficient), and the device cooling requires 90 minutes. The total recovery time is calculated as ((20 x 1.2) + (15 x 0.8) + (8 x 1.5)) / (60 pieces / hour) + 90 / 60 = 2.33 hours.

[0138] Activate the emergency buffer and select the upstream 3 turning centers and the downstream 2 grinding machines as available nodes;

[0139] According to the remaining capacity of each node (turning center 1 remaining 35 pieces, turning center 2 remaining 28 pieces,...), distribute type A workpieces to turning center 1 according to the proportion (35 / (35+28)=55.6%);

[0140] When recalculating the delivery time, it is found that type B workpieces will be delayed by 2 hours, so they are marked as urgent orders and inserted at the beginning of the queue of grinding machine 1, finally reducing the overall order delay by 63%.

[0141] S4.7: Calculate the estimated completion time of each order in the new job order table. When the estimated completion time exceeds the delivery deadline, mark the corresponding order as an urgent order.

[0142] S4.8: Insert the urgent order at the front of the current job queue, and adjust the processing sequence of other non-urgent orders, and output the updated scheduling scheme.

[0143] S5: Synchronize the updated job allocation results and execution effects to the multi-objective reinforcement learning model, and iteratively optimize the multi-objective reinforcement learning model through the priority experience replay mechanism, to realize the continuous optimization of the scheduling strategy.

[0144] S5.1: Record the capacity index data, delivery time index data and energy consumption index data of each process node in the execution process of the updated scheduling scheme, and establish an execution effect evaluation data set.

[0145] S5.2: Compare the execution effect evaluation data set with the expected index data of the job allocation scheme, calculate the deviation value of each index, and generate scheduling effect deviation data.

[0146] S5.3: Based on the scheduling effect deviation data, mark the priority score of the state transition sample in the multi-objective reinforcement learning model.

[0147] S5.4: Store the marked state transition sample into the priority experience pool, and sort the sample according to the priority score.

[0148] S5.5: Extract the sample group with the top 60% priority score from the priority experience pool to form a training batch, input the value evaluation network, calculate the time difference error, and update the network weight parameter.

[0149] S5.5: Based on the time difference error, adjust the gradient of the decision probability distribution of the strategy generation network, and generate the optimized network model parameter.

[0150] S5.6: Update the optimized network model parameter to the multi-objective reinforcement learning model, and complete the strategy iteration.

[0151] In summary, the multi-objective reinforcement learning model with double network structure is constructed, the capacity, delivery time and energy consumption are optimized in the same framework, the problem of balancing multiple competitive targets in traditional methods is overcome, the separation structure of the value evaluation network and the strategy generation network is adopted, the training stability and convergence speed of the model are significantly improved, and the shock in the strategy optimization process is avoided; Through the dynamic job allocation mechanism combined with the load threshold and the bottleneck node information, the balanced distribution of production load is realized on the basis of ensuring the scheduling priority, the material turnover loss is reduced, and through the calculation of the node balance parameter, the production bottleneck is effectively relieved, and the overall production efficiency is improved; Through the establishment of the multi-level linkage emergency response mechanism, combined with the buffer capacity of upstream and downstream nodes, an efficient abnormal processing flow is formed, an order reallocation method based on a preset proportion is adopted, and an urgent order priority insertion mechanism is introduced, which not only ensures the production continuity under abnormal conditions, but also maximizes the influence of abnormal events on delivery time; Through the priority experience replay mechanism, the execution effect is evaluated and fed back in real time, a closed-loop continuous optimization system is established, the priority score marking method based on the scheduling effect deviation data improves the utilization efficiency of samples with important reference value, speeds up the optimization speed of the model, and makes the scheduling strategy adapt to the dynamic changes of the production environment.

[0152] Embodiment 2

[0153] The embodiment also provides an artificial intelligence-based industrial production line dynamic scheduling system, comprising:

[0154] The acquisition module is configured to transmit real-time state data of each process node in the industrial production line to the central data processing unit for preprocessing.

[0155] The construction module is configured to construct a multi-objective reinforcement learning model based on the preprocessed real-time state data, with the optimization objectives of maximum production capacity, shortest delivery time and lowest energy consumption, and generate scheduling priority data.

[0156] The adjustment module is configured to adjust the job allocation of each process node based on the scheduling priority data, in combination with the current device load threshold and production bottleneck node information.

[0157] The update module is configured to trigger a rescheduling mechanism to update the job order and resource allocation result according to the scheduling priority data when an abnormal condition is detected.

[0158] The optimization module is configured to synchronously feed back the updated job allocation result and execution effect to the multi-objective reinforcement learning model, and iteratively optimize the multi-objective reinforcement learning model through a priority experience replay mechanism, so as to realize continuous optimization of the scheduling strategy.

[0159] The embodiment also provides an electronic device, which comprises a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The communication interface of the computer device is configured to perform wired or wireless communication with an external terminal. The wireless communication can be achieved through WIFI, a carrier network, NFC (Near Field Communication) or other technologies. The computer program is executed by the processor to implement a multi-task edge computing resource scheduling method. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device. In addition, the computer device can be connected with an external keyboard, touchpad or mouse.

[0160] The embodiment also provides a computer readable storage medium having a computer program stored thereon, which is executed by a processor to implement the method proposed in the above embodiment.

[0161] The storage medium proposed in the embodiment belongs to the same inventive concept as the method proposed in the above embodiment, and the technical details not described in the embodiment can be referred to the above embodiment, and the embodiment has the same beneficial effects as the above embodiment.

[0162] Through the above description of the embodiments, those skilled in the art can clearly understand that the present application can be realized by means of software and necessary general hardware, and of course can also be realized by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a floppy disk, a read-only memory (ROM), a random access memory (RAM), a FLASH memory, a hard disk, or an optical disc, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods of the embodiments of the present application.

[0163] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application rather than limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the present application, and all of them should be covered in the scope of the claims of the present application.

[0164] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program codes. The solutions in the embodiments of the present application can be implemented in various computer languages.

[0165] The present application is described with reference to flowcharts and / or block diagrams according to the methods, devices (systems), and computer program products of the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be realized by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the computer or other programmable data processing devices produce a device for implementing the functions described in the flowcharts and / or block diagrams. Figure 1one or more processes and / or blocks Figure 1 an apparatus for performing the functions specified in the flowchart

[0166] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flowchart Figure 1 one or more processes and / or blocks Figure 1 one or more blocks or steps

[0167] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions that are executed on the computer or other programmable apparatus provide steps for implementing the flowchart Figure 1 one or more processes and / or blocks Figure 1 one or more blocks or steps

[0168] While the preferred embodiments of the application have been described, additional variations and modifications can be made to the embodiments by those skilled in the art once they learn of the basic inventive concepts. Therefore, the appended claims are intended to cover all such modifications and variations as fall within the true spirit and scope of the application.

[0169] Obviously, numerous modifications and variations of the present application are possible in light of the above teachings. It is therefore to be understood that within the scope of the appended claims and their equivalents, the application can be practiced otherwise than as specifically described.

Claims

1. An artificial intelligence-based dynamic scheduling method for an industrial production line, characterized in that: The application relates to a method for generating a job allocation scheme for an industrial production line. Real-time state data of each process node in the industrial production line is collected and preprocessed; Based on the preprocessed real-time state data, a multi-objective reinforcement learning model is constructed to maximize production capacity, minimize delivery time, and minimize energy consumption, and scheduling priority data is generated; Based on the scheduling priority data, the current device load threshold and production bottleneck node information are combined to adjust the job allocation of each process node; When an abnormal condition is detected, the scheduling priority data is used to trigger a rescheduling mechanism to update the job order and resource allocation results; The updated job allocation results and execution effects are fed back to the multi-objective reinforcement learning model, and the multi-objective reinforcement learning model is iteratively optimized through a priority experience replay mechanism to continuously optimize the scheduling strategy; The production capacity index data, delivery time index data, and energy consumption index data of each process node during the execution of the updated scheduling scheme are recorded to establish an execution effect evaluation dataset; The execution effect evaluation dataset is compared with the expected index data of the job allocation scheme to calculate the deviation values of each index, and scheduling effect deviation data is generated; Based on the scheduling effect deviation data, priority scores are labeled for state transition samples in the multi-objective reinforcement learning model; The labeled state transition samples are stored in a priority experience pool, and the samples are sorted according to the priority scores; Samples with priority scores in the top 60% are extracted from the priority experience pool to form a training batch, which is input into a value evaluation network to calculate the time difference error and update the network weight parameters; Based on the time difference error, the decision probability distribution of the strategy generation network is adjusted by gradient, and the optimized network model parameters are generated; The optimized network model parameters are updated to the multi-objective reinforcement learning model to complete policy iteration; The device running state data of each process node is continuously monitored, and when the temperature value exceeds the preset upper limit or the vibration frequency value exceeds the normal range, the abnormal process node is marked; The residual job amount and the estimated recovery time of the abnormal process node are calculated to generate abnormal influence evaluation data; Based on the abnormal influence evaluation data, the job allocation scheme of the upstream three nodes and the downstream three nodes of the abnormal process node is extracted to establish an emergency response buffer; The scheduling priority data is read, and the orders to be processed of the abnormal process node are transferred to the available nodes in the emergency response buffer in descending order of priority; The residual production capacity and material storage capacity of the available nodes are counted to generate a node resource state table; Based on the node resource state table, the orders to be processed are allocated to the available nodes according to a preset proportion to generate a new job order table; The estimated completion time of each order in the new job order table is calculated, and when the estimated completion time exceeds the delivery deadline, the corresponding order is marked as an urgent order; The urgent order is inserted into the front end of the current job queue, and the processing order of other non-urgent orders is extended, and an updated scheduling scheme is output.

2. The artificial intelligence-based industrial production line dynamic scheduling method of claim 1, wherein: The method for generating a job allocation scheme is reading scheduling priority data, obtaining load rate values of each process node, wherein the load rate value is equal to the ratio of current job quantity and maximum production capacity of the equipment; setting a load threshold parameter, when the load rate value is greater than a first threshold value, marking the corresponding process node as a high-load node; calculating the preceding task completion rate and the subsequent node idle rate of each process node, marking the process node with the preceding task completion rate less than a second threshold value or the subsequent node idle rate greater than a third threshold value as a production bottleneck node; selecting a same-type node with a load rate value less than a fourth threshold value as a backup node for the job task of the high-load node according to the order of the scheduling priority data; calculating the material transmission distance from the backup node to the high-load node, and selecting the backup node with the shortest material transmission distance as a task transfer target node; for the production bottleneck node, extracting the job storage quantity of the upstream node and the production capacity margin of the downstream node to generate a node balance parameter; distributing the job task of the production bottleneck node to three adjacent process nodes according to a preset proportion based on the node balance parameter; generating a job distribution scheme, wherein the job distribution scheme includes task transfer path and distribution proportion data.

3. The artificial intelligence-based industrial production line dynamic scheduling method of claim 2, wherein: The generation method of the scheduling priority data is, reading a normalized state matrix from a production state database, extracting equipment running state data, order progress information data and job load condition data in the standardized data matrix to construct a state vector; establishing a multi-objective reinforcement learning model, wherein the multi-objective reinforcement learning model adopts a double network structure including a value evaluation network and a policy generation network; calculating scheduling priority scores of each process node through the value evaluation network based on the state vector, wherein the scheduling priority scores include production capacity scores, delivery time scores and energy efficiency scores; inputting the scheduling priority scores into the policy generation network to generate scheduling priority values of each process node; arranging the scheduling priority values in descending order to generate scheduling priority data.

4. The artificial intelligence-based industrial production line dynamic scheduling method of claim 3, wherein: The establishment method of the multi-objective reinforcement learning model is, constructing a state space matrix and an action space matrix, wherein the state space matrix includes process node numbers, equipment running state data, order progress information data and job load condition data; the action space matrix includes scheduling priority options of the process nodes; constructing a reward function based on the state space matrix and the action space matrix, wherein the reward function includes production capacity reward items, delivery reward items and energy consumption reward items; building a value evaluation network and a policy generation network according to the reward function; setting an experience replay buffer, and storing state transition sequences, action selection results and obtained reward values in the experience replay buffer; updating parameters of the value evaluation network using a time difference algorithm, wherein the time difference algorithm calculates the error between target value and predicted value based on Bellman equation; updating parameters of the policy generation network using a policy gradient algorithm, wherein the policy gradient algorithm optimizes the action selection strategy by maximizing the expected value of long-term cumulative reward. The parameters of the value evaluation network and the policy generation network are synchronized in a soft update manner to reduce the shock in the training process. An experience priority sampling mechanism is set to preferentially select samples with high time difference errors to train the multi-objective reinforcement learning model.

5. The artificial intelligence-based industrial production line dynamic scheduling method of claim 4, wherein: The value evaluation network adopts a three-layer structure including an input layer, a hidden layer and an output layer; the input layer receives a state space matrix; the output layer generates a state value score; the policy generation network adopts a four-layer structure including an input layer, two hidden layers and an output layer, and the input layer receives the state value score; and the output layer generates an action selection probability.

6. An artificial intelligence-based dynamic scheduling system for an industrial production line, based on the artificial intelligence-based dynamic scheduling method for an industrial production line according to any one of claims 1 to 5, characterized in that: The method comprises the following steps: The acquisition module is configured to transmit real-time state data of each process node in the industrial production line to a central data processing unit for preprocessing; The construction module is configured to construct a multi-objective reinforcement learning model based on the preprocessed real-time state data, with the optimization objectives of maximizing production capacity, minimizing delivery time and minimizing energy consumption, and to generate scheduling priority data; The adjustment module is configured to adjust the job allocation of each process node based on the scheduling priority data, in combination with the current device load threshold and production bottleneck node information; The update module is configured to trigger a rescheduling mechanism to update the job order and resource allocation result according to the scheduling priority data when an abnormal condition is detected; The optimization module is configured to synchronously feed back the updated job allocation result and execution effect to the multi-objective reinforcement learning model, and to iteratively optimize the multi-objective reinforcement learning model through a priority experience replay mechanism, so as to realize continuous optimization of the scheduling strategy. 7.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The computer program is executed by the processor to realize the steps of the artificial intelligence-based industrial production line dynamic scheduling method according to any one of claims 1-5.

8. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the steps of the artificial intelligence-based industrial production line dynamic scheduling method according to any one of claims 1-5.

Images