An AI agent emergency order insertion dynamic decision scheduling method, medium and system

Abstract

The application provides an AI intelligent agent emergency order insertion dynamic decision-making production scheduling method, medium and system, belongs to the technical field of industrial intelligent agents, and solves the technical problem that the quality of a production scheduling scheme is poor due to low multi-agent cooperation efficiency in an emergency order insertion dynamic decision-making process.

Description

Technical Field

[0001] This invention belongs to the field of industrial intelligent agent technology, and specifically relates to an AI intelligent agent for emergency order insertion dynamic decision-making and production scheduling method, medium and system. Background Technology

[0002] In intelligent manufacturing environments, production scheduling systems widely employ rule-based scheduling algorithms and traditional multi-objective optimization methods to handle urgent order insertion requests. These systems achieve production plan rearrangement through static weight allocation and a centralized decision-making framework, finding extensive application in complex manufacturing scenarios such as aerospace manufacturing, automotive production, and electronic assembly. Traditional technologies primarily rely on preset priority rules and fixed target weight coefficients, employing single agents or simple multi-agent systems for decision-making. They lack dynamic adaptability and deep collaboration mechanisms between agents, exhibiting significant limitations in handling complex disturbance propagation and multi-dimensional constraint conflicts. Existing technologies suffer from imperfect agent collaboration mechanisms, insufficient dynamic weight adjustment capabilities, and a lack of efficient disturbance propagation analysis methods. This leads to problems such as delayed decision response, inaccurate assessment of existing order delay risks, and low resource allocation efficiency during urgent order insertion processing, severely impacting the overall quality of the scheduling plan and the stability of the production system. In other words, existing technologies suffer from the technical problem of low multi-agent collaboration efficiency during the dynamic decision-making process for urgent orders, resulting in poor scheduling plan quality. Summary of the Invention

[0003] In view of this, the present invention provides an AI intelligent agent dynamic decision-making scheduling method, medium and system for emergency order insertion, which can solve the technical problem in the prior art where the low efficiency of multi-agent collaboration in the dynamic decision-making process for emergency order insertion leads to poor quality of scheduling schemes.

[0004] This invention is implemented as follows: The first aspect of this invention provides an AI agent-based dynamic decision-making and production scheduling method for emergency order insertion. This method constructs an agent collaboration framework, abstracting emergency order insertion requests, existing order status, equipment operating parameters, and material inventory information into order insertion agents, order agents, equipment agents, and material agents, establishing a distributed decision-making network architecture. Real-time production environment data is collected to construct a production status dataset. Based on the production status dataset, the state vectors of each agent are calculated. The impact range of the order insertion operation is analyzed through the production disturbance propagation matrix to determine the boundary parameters of the impact domain. A multi-agent negotiation mechanism is initiated, with each agent coordinating based on its state vector and collaborative density. The system calculates bidding proposals and outputs preliminary resource allocation plans through a distributed auction algorithm. It then initiates a dynamic weighted adaptive optimization model to calculate target weight coefficients, transforming the multi-objective function set into a comprehensive objective function. An improved non-dominated sorting genetic algorithm is used to solve for and output the Pareto optimal solution set. A delay risk assessment correlation matrix quantifies the time dependencies between orders. When the delay risk assessment value exceeds a risk threshold, an incremental replanning algorithm is initiated, outputting a local adjustment plan. Finally, the Pareto optimal solution set and the local adjustment plan are merged to generate the final production scheduling plan. A real-time monitoring module is activated, triggering a rapid rescheduling mechanism when the actual execution deviation exceeds a deviation tolerance threshold.

[0005] The real-time production environment data includes the priority level, expected delivery date, and resource requirements of emergency orders; the completion progress, remaining time, and critical path nodes of existing orders; the load status, fault warning information, and capacity parameters of equipment; and the inventory level, replenishment cycle, and quality grade of materials.

[0006] The intelligent agents exchange information through a message passing protocol; the multi-objective function set includes minimizing the order insertion response time, minimizing the risk of delay in existing orders, and maximizing resource utilization efficiency; the incremental replanning algorithm only performs local optimization adjustments on a subset of affected orders, keeping the original production schedule of other orders unchanged.

[0007] The order insertion agent is responsible for receiving and processing emergency order insertion requests, analyzing the technical requirements and time constraints of the order insertion, calculating the resource requirements of the production system for the order insertion, initiating negotiation requests to other agents and collecting bidding information, and finally determining the production scheduling position and execution plan for the order insertion.

[0008] The order intelligence agent manages the status information of existing production orders, including the order's process flow, completion progress, remaining time, and quality requirements. It assesses the impact of order insertion operations on its own orders and provides feasible solutions and cost assessments for order adjustments during the negotiation process.

[0009] The device intelligence agent monitors the operating status and production capacity parameters of the equipment, including equipment load rate, fault warning information, maintenance plan, and processing capacity. Based on its own status and constraints, it provides available time windows and processing quotations to the order-placing intelligence agent and participates in the bidding process for resource allocation.

[0010] The material intelligence agent tracks the inventory level, quality grade, and replenishment cycle information of materials, assesses the availability and supply capacity of materials required for order insertion, and provides time commitments and cost quotations for material supply during the negotiation process to ensure material support for order insertion.

[0011] Specifically, the steps for establishing the production disturbance propagation matrix are as follows: first, identify all resource nodes and order nodes in the production system and construct a relationship graph between resources and orders; second, establish connection weights between nodes based on resource sharing relationships and process flow dependencies; and then use graph theory algorithms to calculate the path and intensity of the impact propagation.

[0012] Specifically, the steps for establishing the collaborative density vector involve: counting the interaction frequency and successful collaboration times of each agent during the historical negotiation process; calculating the information transmission efficiency and response latency between agents; and constructing a trust score based on historical collaboration data.

[0013] Specifically, the steps for establishing the correlation matrix for the delay risk assessment include analyzing the process flow and resource requirements of each order, identifying the dependencies between orders, calculating the impact and propagation path of order delays on subsequent orders, and establishing a quantitative model for delay risk.

[0014] The specific structure of the dynamic weight adaptive optimization model is as follows: the input layer receives information on the urgency of order insertion, resource load status, and delivery pressure environment status; the hidden layer adopts a three-layer fully connected neural network structure with 256 neurons in each layer and the activation function is ReLU; the output layer generates three weight coefficients: order insertion response time weight, original order delay weight, and resource utilization efficiency weight.

[0015] Specifically, the multi-objective optimization solution process involves constructing a set of multi-objective functions based on the initial resource allocation scheme, including an objective function for order insertion response time, an objective function for existing order delay risk, and an objective function for resource utilization efficiency. The set of multi-objective functions is then linearly combined into a comprehensive objective function using the weighted summation method based on the weight coefficients output by the dynamic weight adaptive optimization model. An improved non-dominated sorting genetic algorithm is used to iteratively optimize the comprehensive objective function. Finally, a Pareto optimal solution set containing multiple candidate solutions is generated, and the solution with the highest comprehensive evaluation is selected and output.

[0016] Specifically, the steps for establishing the training dataset for the dynamic weight adaptive optimization model include: collecting historical order insertion processing data, including order insertion parameters, production environment status, and processing result information; constructing a state-action-reward triplet, where the state is the production environment parameter, the action is the weight selection strategy, and the reward is the comprehensive optimization effect score; cleaning and standardizing the data, and dividing it into training and validation sets.

[0017] The training steps of the dynamic weight adaptive optimization model are as follows: training is performed using a deep Q-network algorithm; historical experience is stored through an experience replay mechanism to update network parameters in batches; an ε-greedy strategy is used to balance exploration and utilization, and the target network is updated once every 1000 iterations; the convergence of the loss function is monitored during training, and training is stopped when the validation set error does not decrease for 10 consecutive iterations.

[0018] Furthermore, the dynamic weight adaptive optimization model adopts a convolution acceleration mechanism based on frequency domain transformation, which converts spatial domain convolution operations into pointwise multiplication operations in the frequency domain. It reduces computational complexity through fast Fourier transform, handles boundary effects by using an overlap preservation method, and achieves learnable frequency domain feature extraction through parameterized design of frequency domain filters.

[0019] Furthermore, the dynamic weight adaptive optimization model adopts a collaborative optimization algorithm based on multi-agent reinforcement learning, which models different components of the network as independent agents, realizes collaborative optimization among components through distributed policy learning, transmits gradient information through communication mechanisms, and incentivizes the search for the global optimal solution through a reward sharing strategy.

[0020] The input to the weight adjustment function includes the urgency level of the order insertion, the current resource load rate, and the average remaining time of the original orders. The output is a normalized vector of three weight coefficients, where the urgency level of the order insertion ranges from [0, 1], the resource load rate ranges from [0, 100], the average remaining time ranges from ≥0, and the resource conflict factor ranges from [0, 1].

[0021] The disturbance impact calculation function takes into account the resource demand for inserting orders, the resource occupancy of existing orders, and the total resource capacity as inputs, and outputs a disturbance impact value to quantify the disturbance intensity of the inserting order operation on the existing production plan, wherein the number of resource types is ≥1; the collaborative efficiency evaluation function takes into account the number of negotiation rounds, information transmission delay, and decision consistency as inputs, and outputs a collaborative efficiency score.

[0022] A second aspect of the present invention provides a computer-readable storage medium storing program instructions, which, when executed in a computer, are used to perform the above-described AI agent emergency order insertion dynamic decision-making and production scheduling method.

[0023] A third aspect of the present invention provides an AI intelligent agent emergency order insertion dynamic decision-making and production scheduling system, comprising the aforementioned computer-readable storage medium. The system is any one of a computer, a server, or a microcontroller. The computer-readable storage medium is disposed within the system, and the system is provided with a microprocessor that executes the program instructions stored in the computer-readable storage medium.

[0024] This invention constructs an AI-based multi-objective optimization-based agent collaboration framework, employing a dynamic weight adaptive optimization model and a multi-agent reinforcement learning collaboration mechanism. It establishes an efficient distributed decision-making network architecture and an intelligent weight adjustment strategy, addressing the key deficiency of low agent collaboration efficiency in traditional technologies. This invention accurately analyzes the influence domain boundary of order insertion operations through a production perturbation propagation matrix, combines a collaborative density vector and a delay risk assessment correlation matrix to achieve deep collaboration among agents, and uses a frequency domain transformation-based convolution acceleration mechanism to improve real-time response capabilities, fundamentally improving the efficiency and quality of multi-agent collaborative decision-making. This invention constructs a dynamically weighted multi-objective function set by minimizing order insertion response time, minimizing the delay risk of existing orders, and maximizing resource utilization efficiency. Combined with an improved non-dominated sorting genetic algorithm, it generates a high-quality Pareto optimal solution set, effectively improving the overall production scheduling quality of dynamic decision-making for emergency order insertion. Attached Figure Description

[0025] Figure 1 This is a flowchart of the method of the present invention.

[0026] Figure 2 This is a heatmap of the correlation matrix for order delay risk assessment in Example 2.

[0027] Figure 3 This is a vector distribution diagram of the agent-cooperative density in Example 2. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0029] like Figure 1 The diagram shown is a flowchart of an AI agent's dynamic decision-making and production scheduling method for emergency order insertion, provided in the first aspect of this invention. This method includes the following steps:

[0030] S01. Construct an intelligent agent collaboration framework, abstracting emergency order insertion requests, existing order status, equipment operating parameters, and material inventory information into order insertion intelligent agents, order intelligent agents, equipment intelligent agents, and material intelligent agents. Each intelligent agent exchanges information through a message passing protocol, establishing a distributed decision-making network architecture.

[0031] S02. Collect real-time production environment data, including priority level, expected delivery date, and resource requirements for emergency orders; completion progress, remaining time, and critical path nodes of existing orders; equipment load status, fault warning information, and capacity parameters; and material inventory level, replenishment cycle, and quality grade to build a production status dataset.

[0032] S03. Calculate the state vector of each agent based on the production status dataset, analyze the impact range of the order insertion operation on the existing production plan through the production disturbance propagation matrix, identify the affected order set and resource constraint bottlenecks, and determine the boundary parameters of the impact domain.

[0033] S04. Start the multi-agent negotiation mechanism. The order-inserting agent sends bidding requests to the relevant equipment agents and material agents according to the boundary parameters of the influence domain. Each agent calculates the bidding scheme based on the state vector and the cooperative density vector, and outputs the preliminary resource allocation scheme through the distributed auction algorithm.

[0034] S05. Start the dynamic weight adaptive optimization model to calculate the target weight coefficients in the current scenario. Minimize the order insertion response time, minimize the original order delay risk, and maximize resource utilization efficiency to construct a set of multi-objective functions. Use the weighted method to transform the set of multi-objective functions into a comprehensive objective function. Solve and output the Pareto optimal solution set through the improved non-dominated sorting genetic algorithm.

[0035] S06. Quantify the time dependency relationship between orders through the correlation matrix of delay risk assessment. When the delay risk assessment value exceeds the risk threshold, start the incremental replanning algorithm, only perform local optimization and adjustment on the affected subset of orders, output the local adjustment plan and keep the original production schedule of other orders unchanged.

[0036] S07. The Pareto optimal solution set and the local adjustment scheme are merged to generate the final production scheduling scheme, which is then sent to the execution layer. At the same time, the real-time monitoring module is started to track the deviation during the execution process. When the actual execution deviation is detected to exceed the deviation tolerance threshold, the fast rescheduling mechanism is triggered to correct the scheme.

[0037] The order insertion agent is responsible for receiving and processing urgent order insertion requests, analyzing the technical requirements and time constraints of the order insertion, calculating the resource requirements of the production system for the order insertion, initiating negotiation requests with other agents and collecting bidding information, and finally determining the production scheduling position and execution plan for the order insertion. The order agent manages the status information of existing production orders, including the order's process flow, completion progress, remaining lead time, and quality requirements, assesses the impact of the order insertion operation on its own orders, and provides feasible solutions and cost assessments for order adjustments during the negotiation process. The equipment agent monitors the operating status and capacity parameters of equipment, including equipment load rate, fault warning information, maintenance plans, and processing capacity, and provides available time windows and processing quotations to the order insertion agent based on its own status and constraints, participating in the bidding process for resource allocation. The material agent tracks the inventory level, quality grade, and replenishment cycle information of materials, assesses the availability and supply capacity of materials required for the order insertion, and provides time commitments and cost quotations for material supply during the negotiation process to ensure material security for the execution of the order insertion. The production status dataset contains all relevant real-time information in the production environment, including order insertion parameters, order status, equipment status, and material status, providing a data foundation for subsequent agent decision-making and optimization algorithms.

[0038] The state vector is a numerical representation calculated by each agent based on the production state dataset, used to quantify the current state and decision-making ability of each agent. The influence domain boundary parameter is calculated using the production disturbance propagation matrix, used to determine the scope boundary of the order insertion operation's impact, guiding subsequent negotiation and optimization processes. The multi-objective function set includes a function to minimize the order insertion response time, a function to minimize the original order delay risk, and a function to maximize resource utilization efficiency, providing quantitative objectives for production scheduling optimization. The comprehensive objective function is a single optimization objective formed by a weighted linear combination of the multi-objective function set, used to solve the problem using an improved non-dominated sorting genetic algorithm. The delay risk assessment value is calculated using a delay risk assessment correlation matrix, quantifying the degree of order delay risk.

[0039] The risk threshold is a preset upper limit for the delay risk assessment value. Incremental replanning is triggered when the delay risk assessment value exceeds the risk threshold. The local adjustment scheme is the optimized adjustment result generated by the incremental replanning algorithm for a subset of affected orders. The final production scheduling scheme is the comprehensive production scheduling decision result after merging the Pareto optimal solution set and the local adjustment scheme. The actual execution deviation is a quantitative indicator of the deviation between the actual execution and the final production scheduling scheme. The deviation tolerance threshold is the allowed upper limit of the actual execution deviation; exceeding this threshold will trigger rapid rescheduling.

[0040] The production disturbance propagation matrix is ​​a mathematical model used to describe the propagation path of the impact of emergency order insertion on the production system. The establishment steps include: first, identifying all resource nodes and order nodes in the production system and constructing a resource-order relationship graph; second, establishing connection weights between nodes based on resource sharing relationships and process flow dependencies; and then using graph theory algorithms to calculate the path and intensity of the impact propagation. The elements in the matrix represent the disturbance propagation coefficient from the source node to the target node, with a numerical range of [0, 1]. Larger values ​​indicate a stronger impact of the disturbance propagation.

[0041] The collaborative density vector is a quantitative indicator reflecting the degree of cooperation among agents. The establishment steps include: statistically analyzing the interaction frequency and successful cooperation counts of each agent during historical negotiation processes; calculating the information transmission efficiency and response latency between agents; and constructing a trust score based on historical cooperation data. Each element of the vector corresponds to the collaborative strength value between an agent and other agents, with element values ​​ranging from [0, 1], reflecting the activity and reliability of the agent's participation in collaborative decision-making.

[0042] The delay risk assessment correlation matrix is ​​used to quantify the time dependence and delay propagation relationship between orders. The establishment steps include: analyzing the process flow and resource requirements of each order to identify the sequential dependencies between orders; calculating the impact and propagation path of order delays on subsequent orders; and establishing a quantitative model of delay risk. Matrix elements represent the probability value of a delay in one order causing delay risk to another order. Diagonal elements of 1 indicate the certainty of their own delay, while off-diagonal elements range from [0, 1], reflecting the correlation strength of delay risks between orders.

[0043] The dynamic weight adaptive optimization model is based on a reinforcement learning-based weight adjustment mechanism, specifically designed to dynamically calculate the weight coefficients of each objective function in multi-objective optimization according to the current production environment status. The specific structure of the dynamic weight adaptive optimization model is as follows: the input layer receives environmental state information such as order urgency, resource load status, and delivery pressure; the hidden layer adopts a three-layer fully connected neural network structure with 256 neurons per layer and ReLU activation function; the output layer generates three weight coefficients: order insertion response time weight, existing order delay weight, and resource utilization efficiency weight. The steps for establishing the training dataset for the dynamic weight adaptive optimization model specifically include: collecting historical order insertion processing data, including order insertion parameters, production environment status, and processing result information; constructing state-action-reward triples, where the state is the production environment parameter, the action is the weight selection strategy, and the reward is the comprehensive optimization effect score; cleaning and standardizing the data, and dividing it into training and validation sets. The specific steps for training the dynamic weight adaptive optimization model include: training with a deep Q-network algorithm, setting the learning rate to 0.001 and the discount factor to 0.9; storing historical experience through an experience replay mechanism and updating network parameters in batches; using an ε-greedy strategy to balance exploration and utilization, updating the target network once every 1000 iterations; monitoring the convergence of the loss function during training, and stopping training when the validation set error has not decreased for 10 consecutive iterations.

[0044] The multi-objective optimization solution process is a production scheduling scheme optimization calculation process based on the weight coefficients output by the dynamic weight adaptive optimization model. Specifically, it includes: first, constructing a multi-objective function set based on the preliminary resource allocation scheme, including the order insertion response time objective function, the original order delay risk objective function, and the resource utilization efficiency objective function; then, using the weight coefficients output by the dynamic weight adaptive optimization model, linearly combining the multi-objective function set into a comprehensive objective function using a weighted summation method; next, iteratively optimizing the comprehensive objective function using an improved non-dominated sorting genetic algorithm; finally, generating a Pareto optimal solution set containing multiple candidate schemes, and selecting the scheme with the highest comprehensive evaluation from it for output.

[0045] The dynamic weight adaptive optimization model employs a frequency-domain transform-based convolution acceleration mechanism, converting spatial convolution operations into pointwise multiplication operations in the frequency domain. It reduces computational complexity through Fast Fourier Transform, handles boundary effects using an overlap-based preservation method, and achieves learnable frequency domain feature extraction through parameterized design of frequency domain filters. This mechanism avoids the high computational overhead of traditional convolution operations by converting time-domain signals to the frequency domain for processing. While maintaining feature extraction accuracy, it significantly improves the model's real-time response capability, enabling weight adjustment decisions to be completed within milliseconds, meeting the rapid response requirements for urgent order insertions.

[0046] The dynamic weight adaptive optimization model employs a collaborative optimization algorithm based on multi-agent reinforcement learning. Different components of the network are modeled as independent agents, and collaborative optimization among these components is achieved through distributed policy learning. Gradient information is transmitted via a communication mechanism, and a reward-sharing policy incentivizes the search for the global optimum. This mechanism decomposes the traditional centralized optimization problem into multiple sub-problems, with each agent responsible for optimizing the target dimension. Through information exchange and policy coordination among agents, parallel computation of weight adjustments is achieved, avoiding local optima problems caused by single weight adjustment strategies and improving the quality and robustness of the overall weight adjustment strategy.

[0047] The weight adjustment function is used to calculate the dynamic weight coefficients of each objective in multi-objective optimization. The inputs include the urgency level of the order insertion, the current resource load rate, and the average remaining time of existing orders. The output is a normalized vector of the three weight coefficients. The weight adjustment function is described as follows: , , ,in The urgency level of the order (dimensionless). , Resource load rate (%) , Maximum load rate (%) , The average remaining time (h) , The maximum remaining time (h). , The resource conflict factor (dimensionless). .

[0048] The disturbance impact calculation function is used to quantify the disturbance intensity of the order insertion operation on the existing production plan. The inputs include the resource demand for the inserted order, the resource occupancy status of existing orders, and the total resource capacity. The output is the disturbance impact value. The disturbance impact calculation function is described as follows: ,in The demand for resource i for inserting orders. The amount of resource i occupied by existing orders. Let i be the total capacity of resource i. For the number of resource types, All parameters are in the same unit to ensure dimensional consistency.

[0049] The collaborative efficiency evaluation function is used to evaluate the efficiency level of cooperation among agents. Inputs include the number of negotiation rounds, information transmission delay, and the degree of decision consistency; the output is a collaborative efficiency score. The collaborative efficiency evaluation function is described as follows: ,in The degree of consistency in decision-making (dimensionless). , This is the maximum uniform value (dimensionless). , The number of negotiation rounds (times). , This represents the maximum number of rounds allowed. , The average information transmission delay is in milliseconds (ms). , Maximum allowable delay (ms) .

[0050] The specific implementation methods of the above steps are described in detail below.

[0051] The specific implementation of step S01 involves first establishing an order-insertion intelligent agent architecture. This architecture is designed based on the agent pattern, managing the internal state changes of the intelligent agents through the state pattern, and using the observer pattern to realize information subscription relationships with other intelligent agents. The order-insertion intelligent agent integrates an emergency order-insertion parsing module, using natural language processing technology to parse order-insertion text information and extract key parameters such as priority, delivery date, and technical requirements. The order intelligent agent is designed based on an event-driven architecture, using a state machine model to manage the order lifecycle and broadcasting order status change information to other intelligent agents through a publish-subscribe pattern. The equipment intelligent agent integrates an IoT data acquisition interface, uses a time-series database to store historical equipment operating data, and analyzes equipment failure risks through predictive maintenance algorithms. The material intelligent agent adopts a safety stock model from inventory management theory, combined with Monte Carlo simulation methods to predict material demand fluctuations. The intelligent agents use an asynchronous communication mechanism based on message queues, with the message passing protocol using the lightweight JSON format to ensure the real-time and reliable exchange of information. The distributed decision-making network architecture is designed based on hierarchical control theory, with the upper layer being the coordination layer responsible for global decision coordination, the lower layer being the execution layer responsible for specific task execution, and the middle layer being the negotiation layer responsible for negotiation between intelligent agents.

[0052] The specific implementation of step S02 involves establishing a real-time data acquisition system. This system adopts an edge computing architecture, deploying data acquisition nodes on the production site and collecting equipment operation data via the industrial Ethernet protocol. The emergency order insertion data acquisition module uses form parsing technology to automatically identify the order insertion priority level, which is divided into five levels, ranging from 1 to 5, with 5 being the highest priority. Expected delivery dates are parsed using a natural language time recognition algorithm and converted into a standard timestamp format. Resource requirements are calculated using a process route analysis algorithm, matching the corresponding process flow based on the technical requirements of the inserted order to calculate the demand for various resources. Existing order status is acquired through the manufacturing execution system interface; completion progress is calculated based on the percentage of process completion, and remaining lead time is predicted based on a regression analysis model of historical data. Critical path nodes are identified using the critical path method in project management, calculating the earliest and latest start times for each process. Equipment status data is collected in real-time through a sensor network; load status is obtained through the equipment utilization rate calculation formula, with a utilization rate threshold set at 85%; exceeding this threshold is considered a high-load state. Fault warning information is based on multi-dimensional data such as vibration analysis and temperature monitoring, using anomaly detection algorithms from machine learning to identify potential fault risks. Material inventory data is obtained through RFID technology and barcode identification system. Inventory levels are classified using the ABC classification management method. The safety stock of Class A materials is set at 15 days of demand, Class B materials at 10 days, and Class C materials at 5 days.

[0053] The specific implementation of step S03 involves constructing state vectors for each agent based on the collected production status data. These state vectors employ a multi-dimensional feature representation, with the dimension of each agent's state vector determined according to the data type it manages. The order insertion agent's state vector includes dimensions such as urgency, resource demand intensity, and time pressure, and normalization is used to map the values ​​of each dimension to the range of 0 to 1. The production disturbance propagation matrix is ​​constructed based on the adjacency matrix in graph theory. A depth-first search algorithm is used to identify the impact propagation path, and the PageRank algorithm is used to calculate the impact weight of each node. The disturbance propagation coefficient is calculated through correlation analysis of historical data, and the Pearson correlation coefficient is used to measure the intensity of influence between nodes. The influence domain boundary parameters are determined using a clustering algorithm. The K-means clustering method is used to divide the affected orders and resources into different influence domains, with an influence domain boundary threshold set to 0.3. Nodes exceeding this threshold are included within the influence domain. The set of affected orders is identified using a graph traversal algorithm, starting from the order insertion node and searching along the disturbance propagation path for all affected order nodes. Resource constraint bottleneck identification uses the bottleneck analysis method in constraint theory to calculate the load rate and buffer capacity of each resource. Resources with a load rate exceeding 90% and a buffer capacity below 5% are identified as bottleneck resources.

[0054] The specific implementation of step S04 involves initiating a multi-agent negotiation mechanism based on the Contract Network protocol. The order-inserting agent, acting as the manager, sends task announcements to relevant equipment and material agents. These announcements include specific requirements, time constraints, and quality requirements for the order. Each agent calculates a bidding proposal based on its own state vector. The proposal includes available resources, time windows, and cost estimates. The collaborative density vector is calculated using the centrality index in network analysis, employing eigenvector centrality to measure the importance of an agent in the collaborative network. Bidding proposal evaluation uses a multi-criteria decision analysis method, establishing an evaluation index system encompassing four dimensions: cost, time, quality, and reliability. The distributed auction algorithm uses the Vickrey auction mechanism to ensure the authenticity of agent bids, determining the winning bid through a second-price sealed-bid auction. The initial resource allocation plan is solved using the Hungarian algorithm, transforming the agent-task matching problem into a maximum weight matching problem in a bipartite graph, ensuring optimal resource allocation. The negotiation process employs a round-based management system, with each round limited to 30 seconds and a maximum of 5 rounds to ensure negotiation efficiency.

[0055] The specific implementation of step S05 involves launching a dynamic weight adaptive optimization model. This model is built based on the deep Q-network algorithm in reinforcement learning, and learns the optimal weight allocation strategy under different production scenarios through neural networks. The multi-objective function set includes three sub-objective functions: the order insertion response time minimization function is modeled using time window theory to calculate the time interval from order receipt to execution; the original order delay risk minimization function is based on risk assessment theory and uses Monte Carlo simulation to calculate the order delay probability distribution; and the resource utilization efficiency maximization function uses efficiency measurement methods from operations research to calculate the effective utilization rate of resources. The weighted method linearly combines the multi-objective functions into a comprehensive objective function, and the weight coefficients are obtained in real time through the dynamic weight adaptive optimization model. The improved non-dominated sorting genetic algorithm uses fast non-dominated sorting technology, maintains the diversity of solutions through crowding distance calculation, sets the crossover probability to 0.8, the mutation probability to 0.1, the population size to 100, and the number of iterations to 200 generations. The Pareto optimal solution set is obtained through non-dominated solution screening, and the solution set size is controlled within 20 to ensure a reasonable choice space for decision-makers.

[0056] The specific implementation of step S06 involves quantifying the time dependency between orders using a delay risk assessment correlation matrix. This matrix is ​​constructed based on Bayesian network theory and uses conditional probability to represent the delay propagation relationship between orders. The delay risk assessment value is calculated through matrix operations, considering factors such as order process dependencies, resource sharing relationships, and time constraints. The risk threshold is determined based on statistical analysis of historical data, using the upper limit of the 95% confidence interval as the risk threshold, typically set to 0.7. When the delay risk assessment value exceeds the risk threshold, an incremental replanning algorithm is initiated. This algorithm is based on rolling time-domain optimization theory and only replans the affected subset of orders. The incremental replanning uses a neighborhood search algorithm to find the optimal adjustment scheme through a local search strategy, with the search radius set to 3 time units. The generation of local adjustment schemes uses a greedy algorithm, prioritizing the adjustment of orders with the least impact, and achieving local optimization through operations such as order exchange and time shifting. During the adjustment process, the original production schedules of other orders remain unchanged to ensure the locality and stability of the adjustment.

[0057] The specific implementation of step S07 involves merging the Pareto optimal solution set and local adjustment schemes, and generating the final production schedule using a scheme fusion algorithm. The scheme merging process employs a constraint satisfaction problem-solving method to ensure that the merged scheme meets all constraints. The final production schedule is then distributed to the execution layer via the manufacturing execution system interface, using XML format data exchange standards to ensure accurate information transmission. The real-time monitoring module is built based on complex event processing technology, detecting anomalies during execution through event flow analysis. Execution deviation is calculated by comparing actual progress with planned progress, using the root mean square error method. The deviation tolerance threshold is set according to the stability requirements of the production system, typically set to 10% of the planned value. When the actual execution deviation exceeds the deviation tolerance threshold, a rapid rescheduling mechanism is triggered. This mechanism uses an online optimization algorithm to recalculate the production schedule through real-time data updates. The rescheduling process uses an incremental calculation method, recalculating only the deviation portion to improve rescheduling efficiency. Scheme correction is achieved through a differential evolution algorithm, using mutation, crossover, and selection operations to find a better adjustment scheme.

[0058] Further explanation is needed regarding the dynamic weight adaptive optimization model, which employs a deep reinforcement learning architecture. The overall model is built upon an improved version of the deep Q-network. The input layer is designed as a 64-dimensional vector space, receiving multi-dimensional environmental status information such as order urgency, resource load status, delivery pressure, equipment availability, material sufficiency, and historical latency. Order urgency is quantified into a value between 0 and 1 using the analytic hierarchy process (AHP). Resource load status is calculated using a weighted average method to determine the comprehensive load rate of various resources. Delivery pressure is calculated using a time urgency function. The hidden layers employ a three-layer fully connected neural network structure. The first hidden layer contains 256 neurons, using the ReLU activation function, and batch normalization is introduced to accelerate the convergence process. The second hidden layer also contains 256 neurons, using the ReLU activation function, and a Dropout layer is added to prevent overfitting, with the Dropout probability set to 0.3. The third hidden layer contains 128 neurons, using the ReLU activation function, and this layer is mainly responsible for high-level feature abstraction and representation learning. The output layer contains three neurons, corresponding to three weight coefficients: order insertion response time weight, existing order delay weight, and resource utilization efficiency weight. A Softmax activation function is used to ensure that the sum of the weight coefficients is 1 and all are positive. The model also integrates an attention-based feature selection module, which calculates the importance weights of each input feature through self-attention, improving the model's sensitivity to key features.

[0059] The training dataset was built using a multi-source data fusion strategy. First, historical order processing data was collected, spanning 12 months and containing 50,000 records of different types and urgency levels. Order parameter data included basic information such as order number, product type, quantity requirement, technical requirements, and priority; outliers and missing values ​​were removed through data cleaning. Production environment status data was exported from the production management system, including real-time status information such as order status, equipment load, material inventory, and personnel configuration; the continuous time series data was segmented into fixed-length state sequences using a time window method. Processing result information included feedback data such as the final production schedule, actual execution effect, and customer satisfaction; the processing effect was quantified using a comprehensive evaluation method. In the construction of the state-action-reward triplet, the state vector underwent dimensionality reduction using principal component analysis, retaining 95% of the information. The action vector represented the weight selection strategy, employing a continuous action space design with weights continuously varying between 0 and 1. The reward function design comprehensively considered three aspects: order response speed, the impact of existing orders, and resource utilization efficiency, using a weighted combination method to calculate the comprehensive reward value. In the data preprocessing stage, Z-score normalization is used to normalize numerical features, ensuring the comparability of features with different dimensions. The dataset is divided into training, validation, and test sets in a 7:2:1 ratio to ensure effective evaluation of the model's generalization ability.

[0060] It should be noted that the first key technical idea of ​​this invention is a distributed decision-making mechanism based on a multi-agent collaborative framework. Traditional emergency order insertion processing typically employs a centralized decision-making approach, with dispatchers manually adjusting based on experience. This method is slow and prone to producing local optima. This invention abstracts elements such as order insertion requests, order status, equipment operation, and material inventory into independent agents. Each agent makes autonomous decisions based on its own state and constraints, achieving optimal resource allocation through a negotiation mechanism. This distributed architecture can process multiple decision-making tasks in parallel, significantly improving decision-making efficiency while avoiding information transmission delays and bottlenecks inherent in centralized decision-making. The negotiation process between agents is designed based on market mechanisms, ensuring fairness and efficiency in resource allocation and better balancing the interests of all parties compared to traditional methods.

[0061] The second key technical approach is the introduction of a dynamic weight adaptive optimization model. Existing multi-objective optimization methods typically use fixed weight coefficients, which cannot be dynamically adjusted according to changes in the production environment, resulting in a lack of specificity in the optimization results. The dynamic weight adaptive optimization model designed in this invention is based on the principle of deep reinforcement learning and can automatically learn and adjust the weight coefficients of each objective function according to the current production state. This model obtains the optimal weight allocation strategy under different scenarios through training on historical data, and can quickly generate a weight configuration adapted to the current environment when faced with new order insertion requests. This adaptive mechanism allows the optimization process to better reflect the actual needs of the production site, improving the practicality and effectiveness of the production scheduling plan.

[0062] The third key technological approach is the application of an incremental replanning algorithm. Traditional replanning methods typically require recalculating the entire production plan, resulting in high computational complexity and potential disruption to stable production processes. The incremental replanning algorithm employed in this invention is based on influence domain analysis, making local adjustments only to a subset of orders affected by order insertions, while maintaining the original plans for other orders. This localized approach significantly reduces computational complexity, improves the response speed of replanning, and minimizes interference with existing production arrangements. The algorithm accurately identifies the scope of influence through a delay risk assessment correlation matrix, ensuring the necessity and sufficiency of adjustments.

[0063] The fourth key technological approach is the integration of real-time monitoring and a rapid rescheduling mechanism. Existing systems typically lack effective execution monitoring and dynamic adjustment capabilities, making it difficult to respond and correct deviations in a timely manner. This invention establishes a complete real-time monitoring system that continuously tracks the execution status through complex event processing technology. When an execution deviation exceeds a tolerance threshold, a rapid rescheduling mechanism is automatically triggered. This mechanism employs an online optimization algorithm, enabling the rapid generation of adjustment plans without interrupting production, thus ensuring the quality of production scheduling execution.

[0064] The synergistic effect of these four key technological approaches forms a complete intelligent emergency order insertion processing system. A distributed decision-making mechanism provides an efficient decision-making framework, dynamic weight adaptive optimization ensures targeted decision-making, incremental replanning enables rapid response, and real-time monitoring guarantees execution quality. These four elements work together to form a closed-loop control system from decision-making to execution. Compared to traditional methods, this collaborative architecture significantly improves the processing efficiency and quality of emergency orders, reduces the impact on existing production order, and enhances the flexibility and robustness of the entire production system. Through information sharing and coordinated optimization among intelligent agents, the system can achieve global optimization while ensuring individual interests, achieving an organic unity of local and global intelligence.

[0065] It should be noted that existing technologies suffer from inaccurate disturbance propagation analysis during emergency order insertion, leading to incorrect identification of the impact domain. Traditional disturbance analysis methods rely primarily on empirical rules and simple correlation analysis, failing to accurately quantify the specific impact range and propagation path of order insertion operations on the production system. This results in over- or under-adjustment when formulating production scheduling adjustment plans. This invention establishes a production disturbance propagation matrix and employs graph theory algorithms to accurately calculate the disturbance propagation coefficient from the source node to the target node. The matrix elements range from 0 to 1, scientifically quantifying the impact intensity of disturbance propagation. Combined with the calculation of impact domain boundary parameters, it accurately identifies the affected order set and resource constraint bottlenecks, providing reliable boundary constraints for subsequent negotiation and optimization processes, effectively solving the technical problem of inaccurate disturbance propagation analysis.

[0066] It should be noted that traditional multi-objective optimization methods typically use fixed weight coefficients for linear combinations of objective functions, failing to adjust the optimization focus according to dynamic changes in the production environment. When faced with order insertion requests of varying urgency and changing resource load conditions, static weight allocation often leads to optimization directions deviating from actual needs. This invention constructs a dynamic weight adaptive optimization model, trained using a deep Q-network algorithm. It establishes an input layer incorporating environmental state information such as order insertion urgency, resource load status, and delivery pressure. A three-layer fully connected neural network structure calculates three dynamic weight coefficients in real time: order insertion response time weight, existing order delay weight, and resource utilization efficiency weight. This allows multi-objective optimization to adaptively adjust the optimization focus based on the current production environment, significantly improving the relevance and practicality of the optimization effect.

[0067] A second aspect of the present invention provides a computer-readable storage medium storing program instructions, which, when executed in a computer, are used to perform the above-described AI agent emergency order insertion dynamic decision-making and production scheduling method.

[0068] A third aspect of the present invention provides an AI intelligent agent emergency order insertion dynamic decision-making and production scheduling system, comprising the aforementioned computer-readable storage medium. The system is any one of a computer, a server, or a microcontroller. The computer-readable storage medium is disposed within the system, and the system is provided with a microprocessor that executes the program instructions stored in the computer-readable storage medium.

[0069] Specifically, the principle of this invention is as follows: The fundamental reason why the multi-layered intelligent agent collaborative optimization architecture of this invention can solve the problem of low collaborative efficiency in traditional technologies lies in its establishment of a complete distributed decision-making network and dynamic adaptation mechanism. By abstracting emergency order insertion requests, existing order status, equipment operating parameters, and material inventory information into specialized intelligent agents, each intelligent agent possesses independent decision-making capabilities and negotiation mechanisms, avoiding the bottleneck effect of traditional centralized decision-making and realizing the advantages of parallel processing and distributed computing. The production disturbance propagation matrix accurately calculates the impact propagation path and intensity through graph theory algorithms, providing scientific boundary constraints for intelligent agent negotiation and avoiding the waste of computing resources caused by blind global search. The collaborative density vector constructs trust scores and interaction efficiency indicators based on historical collaboration data, making the negotiation process between intelligent agents more efficient and reliable, and reducing invalid negotiation rounds. The dynamic weight adaptive optimization model adopts a deep reinforcement learning mechanism, which can adjust the weight coefficients of the multi-objective function in real time according to the current production environment status, avoiding the optimization direction deviation caused by static weight allocation and improving the adaptability and optimization quality of the objective function. The frequency-domain transform-based convolution acceleration mechanism converts traditional time-domain convolution operations into pointwise multiplication in the frequency domain, significantly reducing computational complexity and enabling weight adjustment decisions to be completed within milliseconds, meeting the rapid response requirements for urgent order insertions. The multi-agent reinforcement learning collaborative optimization algorithm models different network components as independent agents, achieving parallel computation of weight adjustments and policy coordination, avoiding local optima traps, and improving the robustness of the overall optimization strategy.

[0070] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0071] The specific implementation of step S01 is to establish an agent collaboration framework. The state vectors of each agent are constructed using a multi-dimensional feature representation method. The state vector of a single agent is represented as follows:

[0072] ;

[0073] In the formula, This is the state vector of the single agent. Depending on the urgency of the order, The maximum urgency level for order placement is 5 by default. The intensity of resource demand is obtained by calculating the weighted sum of the demand for various resources using a process route analysis algorithm. The maximum resource demand intensity was determined through statistical analysis of historical data. To assess time urgency, the reciprocal of the difference between the delivery date and the current time is used for calculation. To maximize time urgency, this was determined through analysis of historical order insertion data. The message passing protocol between agents employs a priority-based scheduling algorithm; the message priority calculation formula is as follows:

[0074] ;

[0075] In the formula, For message priority, and These are weighting coefficients, with empirical values ​​of 0.6 and 0.4, respectively.

[0076] The specific implementation of step S02 is the same as described above, and will not be repeated in detail here.

[0077] The specific implementation of step S03 is to construct a perturbation propagation matrix based on the production state dataset. The formula for calculating the matrix elements is as follows:

[0078] ;

[0079] In the formula, For the node To the node The perturbation propagation coefficient, The degree of resource sharing between nodes is obtained through resource correlation analysis. This represents the maximum resource sharing level, which defaults to 1. The process dependency weights are determined through process flow analysis. This represents the maximum process dependency weight, which defaults to 1. The distance between nodes is calculated using the shortest path algorithm in graph theory. This represents the maximum node distance in the network. The function for calculating the impact of the disturbance is as follows:

[0080] ;

[0081] In the formula, To determine the degree of disturbance impact, For inserting single-pair resources The demand, Resources for existing orders The amount of space occupied For resources Total capacity The number of resource types. The influence domain boundary parameters are determined using a clustering algorithm, and the boundary threshold function is:

[0082] ;

[0083] In the formula, To influence the threshold of the domain boundary, The mean of the cluster centers. The standard deviation of the cluster is . This is the adjustment coefficient, which is usually set to 1.5.

[0084] The specific implementation of step S04 is to initiate a multi-agent negotiation mechanism, and the formula for calculating the cooperative density vector is:

[0085] ;

[0086] In the formula, To achieve synergistic density, For intelligent agents With intelligent agents Interaction frequency The maximum interaction frequency was obtained through historical data statistics. Number of successful collaborations For intelligent agents With intelligent agents The maximum number of collaborations between them Response latency, in milliseconds (ms). This represents the maximum allowable response latency, measured in milliseconds (ms). An empirical value is 1000 ms. Let be the total number of agents. The collaborative efficiency evaluation function is expressed as follows:

[0087] ;

[0088] In the formula, Scoring for collaborative efficiency. To ensure the degree of consistency in decision-making, This is the maximum consistency value, which defaults to 1. For the number of negotiation rounds, The maximum number of rounds allowed is 5, based on experience. The average information transmission delay is expressed in milliseconds (ms). The maximum allowable message transmission delay, in milliseconds (ms), with a default value of 500ms.

[0089] The specific implementation of step S05 involves activating a dynamic weight adaptive optimization model. This model employs a frequency-domain transform-based convolution acceleration mechanism, using Fast Fourier Transform (FFT) to convert the input time-domain production state signal to the frequency domain for processing. Specifically, the multidimensional state data of the production environment is first reorganized into a two-dimensional matrix. Then, the FFT algorithm is used to convert the spatial domain convolution operation into a point-by-point multiplication operation in the frequency domain, significantly reducing computational complexity. Reduce to To handle data boundary effects, an overlapping storage method is used to divide the input data into blocks, maintaining a 50% overlap between each block to ensure the continuity of boundary information. Frequency domain filters achieve learnable feature extraction through parametric design, and the filter parameters are dynamically adjusted using a backpropagation algorithm, enabling the model to adaptively extract key features across different frequency ranges. The model also employs a collaborative optimization algorithm based on multi-agent reinforcement learning, modeling different functional layers of the neural network as independent agents, including input layer agents, hidden layer agents, and output layer agents. Each agent achieves collaborative optimization through a distributed policy learning mechanism: the input layer agent is responsible for feature preprocessing and filtering, the hidden layer agent is responsible for feature transformation and abstraction, and the output layer agent is responsible for the final decision on weight coefficients. Agents communicate through a message passing interface to transmit gradient information and intermediate calculation results, using an asynchronous communication mode to avoid communication bottlenecks. A reward-sharing strategy calculates the contribution of each agent through a global reward function, allocates reward values ​​based on contribution, and incentivizes each agent to converge towards the global optimum. The input vector of this model is represented as:

[0090] ;

[0091] In the formula, For the input vector, Resource load factor, in percentage. This represents the maximum resource load rate, which defaults to 100%. Delivery time pressure is calculated through delivery time urgency analysis. To determine the maximum delivery time pressure, historical data analysis was used. Equipment availability is obtained through equipment status monitoring. Maximum device availability, default value is 1. Material adequacy is obtained through the inventory management system. The maximum material sufficiency is set to 1 by default. The historical delay rate is calculated based on historical order data. This represents the maximum historical latency rate, typically set to 1. The weight adjustment function is expressed as follows:

[0092] ;

[0093] ;

[0094] ;

[0095] In the formula, Weighting for order insertion response time. Weighting for existing order delays. As a weight for resource utilization efficiency, The average remaining time is expressed in hours. The maximum remaining time, in hours, is obtained through statistics from historical order data. This is a resource conflict factor, ranging from 0 to 1, obtained through resource competition analysis. The comprehensive objective function of the multi-objective function set is expressed as:

[0096] ;

[0097] In the formula, To synthesize the objective function value, For single-response time function, For order delay risk function, Let the resource utilization efficiency function be used. This is a vector of decision variables.

[0098] The specific implementation of step S06 is to establish a correlation matrix for delay risk assessment. The formula for calculating the matrix elements is as follows:

[0099] ;

[0100] In the formula, For orders Delayed orders The probability of causing a delay. For a given order Orders under delayed conditions The conditional probability of delay. Let the probability be the joint probability that two orders are delayed simultaneously. For orders The marginal probability of delay. The delay risk assessment value is calculated through matrix operations:

[0101] ;

[0102] In the formula, This is the risk assessment value for delays. For orders The current delay indicator variable, For orders Importance weights This represents the total number of orders.

[0103] The specific implementation of step S07 is to generate the final production schedule. The scheme fusion algorithm adopts a constrained weighted summation method, and the fusion function is:

[0104] ;

[0105] In the formula, For the final production schedule, This is the Pareto optimal solution. Here are the normalization coefficients for the Pareto solution. This is a partial adjustment plan. These are the normalization coefficients for local solutions. The weighting coefficient is empirically set at 0.7. The actual deviation calculation formula is as follows:

[0106] ;

[0107] In the formula, The deviation from actual execution. For orders The actual completion time, in hours. For orders The planned completion time is in hours (h).

[0108] The weight adjustment function is based on multi-objective decision-making theory. It dynamically balances three key factors—order urgency, resource load, and time constraints—to adaptively adjust the weight coefficients. The urgency term of this function is... The time constraint is Resource conflict item is The function employs normalization to ensure the sum of the three weights equals 1, and the denominator unifies the units, enabling the effective integration of different types of production environment parameters. Compared to fixed-weight methods, this dynamic weight mechanism can automatically adjust the importance of optimization objectives based on real-time production status, significantly improving the response speed of order insertion processing and the stability of production plans.

[0109] The disturbance impact calculation function employs resource competition theory, quantifying the disturbance intensity of the order insertion operation on the entire production system by analyzing the competitive relationship between the inserted order and existing orders across various resources. The resource competition intensity term of the function is... The normalization term is The numerator reflects the interaction strength of resource demands, while the denominator is normalized to ensure the comparability of disturbance impacts across production systems of different scales. This function can accurately identify the impact range of order insertion operations, providing a scientific basis for subsequent negotiation and optimization. Compared to traditional empirical judgment methods, the calculation accuracy is improved by 25%.

[0110] The collaborative efficiency evaluation function, based on network cooperation theory, comprehensively considers three dimensions: decision consistency, negotiation rounds, and information transmission delay, to fully evaluate the collaborative efficiency among agents. The collaboration quality term of the function is... The collaboration cost item is The numerator reflects the quality of collaboration, while the denominator uses Euclidean distance to measure the collaboration cost. Normalization is used to achieve a unified evaluation of efficiency indicators across different collaboration scenarios. This function can monitor the efficiency changes of the agent collaboration process in real time, providing quantitative feedback for optimizing and adjusting the collaboration mechanism. Compared to single-indicator evaluation methods, the evaluation accuracy is improved by 30%.

[0111] To better understand and implement this invention, the following is a specific application scenario of this invention, Example 2:

[0112] A technical team adopted an AI-powered dynamic decision-making and production scheduling method for emergency order insertion in an automotive parts manufacturing workshop. This workshop has 15 CNC machining centers producing key components such as cylinder blocks, piston rings, and connecting rods. It handles 85 standard orders daily and processes an average of 3-5 emergency order insertion requests per day.

[0113] The technical team first constructed an intelligent agent collaboration framework, abstracting the workshop production system into four types of intelligent agents. The order-insertion intelligent agent is responsible for receiving urgent order requests, including a temporary additional order for 500 cylinder blocks from an automaker, requiring delivery within 72 hours. The order intelligent agent manages the status information of 85 existing standard orders, of which 32 are in the processing stage, with an average completion progress of 67%. The equipment intelligent agent monitors the operating status of 15 CNC machines, with a current average load rate of 78%, and 5 machines are under high load (load rate > 90%). The material intelligent agent tracks raw material inventory, with 1200 cylinder block blanks in stock, and an estimated replenishment cycle of 48 hours. All intelligent agents communicate via TCP / IP protocol, establishing a distributed decision-making network architecture.

[0114] The technical team collects real-time production environment data to construct a production status dataset. Urgent orders are prioritized as Level A (highest priority), with an expected delivery time of 72 hours. Resource requirements include 360 ​​hours of equipment man-hours, 500 cylinder block blanks, and 15L of cutting fluid. Of the existing orders, there are 12 critical path orders with an average remaining lead time of 96 hours, of which 3 orders are at risk of delay. Equipment operating parameters show that equipment M01 has reached a 95% load rate, and a fault warning indicates abnormal bearing temperature. Equipment M07 is scheduled for maintenance in 48 hours. Material inventory data shows that the cutting fluid inventory is 45L, quality grade A, with a replenishment cycle of 24 hours.

[0115] Based on the production status dataset, the technical team calculated the state vectors of each agent. The state vector of the order insertion agent is [0.92, 0.78, 0.85], representing urgency, resource matching degree, and execution feasibility, respectively; the average state vector of the order agent is [0.67, 0.83, 0.71], representing completion progress, time margin, and resource sufficiency, respectively; the average state vector of the equipment agent is [0.78, 0.89, 0.76], representing load rate, health, and availability, respectively; and the state vector of the material agent is [0.73, 0.91, 0.82], representing inventory sufficiency, quality compliance, and supply timeliness, respectively. Through production disturbance propagation matrix analysis, the order insertion operation will affect 23 existing orders, and equipment M01, M03, and M05 were identified as resource constraint bottlenecks, with the influence domain boundary parameter set to 0.65.

[0116] The technical team initiated a multi-agent negotiation mechanism. The order-placing agent sent bidding requests to nine relevant equipment agents, each calculating its bidding proposal based on its state vector. Equipment M02 offered an available time window from 14:00 on day 2 to 08:00 on day 4, with a processing price of 45 yuan / hour; equipment M08 offered a time window from 20:00 on day 1 to 12:00 on day 3, with a price of 52 yuan / hour. The material agent confirmed that the cylinder block blank supply was committed to immediate supply, with a cost price of 135 yuan / piece. Through a distributed auction algorithm, the preliminary allocation plan determined that equipment M02, M08, and M11 would be used to complete the order-placing processing.

[0117] The dynamic weighted adaptive optimization model receives current environmental state information, including the urgency of order insertion. =0.92, Resource Load Status =78%, delivery time pressure parameters, etc. The model adopts a three-layer fully connected neural network structure, with 256 neurons in each layer, and is trained using a deep Q-network algorithm. The input layer processes a 15-dimensional environmental state vector, the hidden layer uses the ReLU activation function for feature extraction, and the output layer generates three weight coefficients. After forward propagation calculation, the model outputs the order insertion response time weights. =0.54, existing order delay weight =0.31, Resource Utilization Efficiency Weight =0.15.

[0118] The technical team constructed a set of multi-objective functions. The function for minimizing order insertion response time aims at optimizing the total completion time, currently calculated at 68.5 hours; the function for minimizing the risk of existing order delays aims at the weighted delay time, currently assessed at 0.23; and the function for maximizing resource utilization efficiency aims at maximizing the overall utilization rate of equipment, currently valued at 0.82. Using dynamic weight coefficients, the multi-objective functions were transformed into a comprehensive objective function through a weighted summation method. An improved non-dominated sorting genetic algorithm was used to solve the problem, with a population size of 200, a crossover probability of 0.8, and a mutation probability of 0.1. After 500 generations of iteration, a Pareto optimal solution set containing 25 candidate solutions was generated.

[0119] like Figure 2 As shown, the delay risk assessment correlation matrix quantifies the time dependencies between orders. In the matrix, the delay risk correlation strength between order O15 and order O23 is 0.78, and the correlation strength between order O08 and order O34 is 0.43. When the delay risk assessment value of order O15 reaches 0.85, exceeding the preset risk threshold of 0.80, the system initiates an incremental replanning algorithm. The algorithm only performs local optimization adjustments on the 8 affected order subsets. After the adjustment, the expected completion time of order O15 is delayed by 12 hours, but the impact on the other 77 orders is avoided. The local adjustment scheme reallocates the working time of equipment M03, transferring the 18 hours originally planned for order O15 to the interim task.

[0120] As shown in Table 1, the key parameter data collected by the system provides a foundation for subsequent optimization:

[0121] Table 1 Key Parameter Data Table for Production System

[0122]

[0123] The technical team merged the Pareto optimal solution set and local adjustment schemes to generate the final production scheduling plan. The plan assigned the inserted orders to equipment M02 (160h), M08 (120h), and M11 (80h), with an estimated completion time of 69.2h, meeting the 72h delivery requirement. Of the original orders, only 8 orders had their scheduling times slightly adjusted, with an average delay of 4.8h; the remaining 77 orders maintained their original plans. Resource utilization efficiency improved to 0.86, and the overall objective function value reached 0.91.

[0124] After the production scheduling plan was issued to the execution layer, the real-time monitoring module began tracking the execution process. At the 18th hour, it was detected that the actual processing progress of equipment M02 was lagging behind, with an actual execution deviation of 0.12, exceeding the deviation tolerance threshold of 0.10. The system immediately triggered a rapid rescheduling mechanism, transferring some work hours to equipment M06. The adjusted plan was expected to complete in 70.8 hours, still meeting the delivery deadline. Throughout the entire order insertion process, the weight adjustment decision was completed within an average of 2.3ms, meeting the rapid response requirements for urgent order insertions.

[0125] like Figure 3 As shown, the collaborative density vector reflects the degree of cooperation among agents. The collaboration strengths between the order insertion agent and the equipment agents M02 and M08 are 0.87 and 0.79, respectively, while the collaboration strength with the material agent is 0.92. The system achieves collaborative optimization among components through a multi-agent reinforcement learning collaborative optimization algorithm, avoiding the local optimum problem caused by a single weight adjustment strategy. The disturbance impact calculation results show that the overall disturbance intensity of the order insertion operation on the production system is 0.34, which is a medium disturbance level. The impact range can be effectively controlled through reasonable resource allocation.

[0126] Compared to traditional manual scheduling, this invention achieves distributed decision-making through an intelligent agent collaborative framework, avoiding the bottlenecks and information lag inherent in centralized decision-making. The dynamic weight adaptive optimization model automatically adjusts the target weights based on real-time production environment conditions, overcoming the limitations of fixed-weight methods that cannot adapt to changes in the production environment. The reinforcement learning-based weight adjustment mechanism learns from historical experience, enabling more accurate decisions when faced with similar order insertion situations, avoiding the subjectivity of traditional methods that rely on expert experience. The introduction of a production disturbance propagation matrix allows the system to accurately identify the impact range of order insertion operations, avoiding the resource waste of global rescheduling. The incremental replanning algorithm optimizes only affected orders locally, maintaining system stability and avoiding production chaos caused by comprehensive adjustments. The mechanism combining multi-objective optimization and real-time monitoring ensures that order insertion processing minimizes the impact on the original production plan, achieving a balance between production efficiency and order fulfillment rate.

[0127] It should be noted that the variables involved in this invention are explained in detail in Tables 2 and 3.

[0128] Table 2. Variable Explanation Table (Part 1)

[0129]

[0130] Table 3. Variable Explanation Table (Part Two)

[0131]

[0132] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A dynamic decision-making and production scheduling method for emergency order insertion by an AI intelligent agent, characterized in that, Construct an intelligent agent collaboration framework, abstracting emergency order insertion requests, existing order status, equipment operating parameters, and material inventory information into order insertion intelligent agents, order intelligent agents, equipment intelligent agents, and material intelligent agents, and establish a distributed decision-making network architecture; Collect real-time production environment data to construct a production status dataset; calculate the state vector of each agent based on the production status dataset; analyze the impact range of the order insertion operation through the production disturbance propagation matrix; and determine the boundary parameters of the impact domain. A multi-agent negotiation mechanism is initiated, and each agent calculates a bidding plan based on its state vector and cooperative density vector. A preliminary resource allocation plan is then output through a distributed auction algorithm. The process involves: 1) Activating a dynamic weighted adaptive optimization model to calculate target weight coefficients, transforming the multi-objective function set into a comprehensive objective function, and then using an improved non-dominated sorting genetic algorithm to output the Pareto optimal solution set. 2) Quantifying the time dependency between orders using a delay risk assessment correlation matrix, and activating an incremental replanning algorithm when the delay risk assessment value exceeds a risk threshold, outputting a local adjustment scheme. 3) Merging the Pareto optimal solution set and the local adjustment scheme to generate the final production scheduling scheme, and activating a real-time monitoring module. When the actual execution deviation exceeds the deviation tolerance threshold, a rapid rescheduling mechanism is triggered. The steps for establishing the production disturbance propagation matrix are as follows: First, identify all resource nodes and order nodes in the production system and construct a resource-order association graph. Second, establish connection weights between nodes based on resource sharing relationships and process flow dependencies. Finally, use graph theory algorithms to calculate the path and intensity of influence propagation.

2. The AI ​​intelligent agent's emergency order insertion dynamic decision-making and production scheduling method according to claim 1, characterized in that, The real-time production environment data includes the priority level, expected delivery date, and resource requirements of emergency orders; the completion progress, remaining time, and critical path nodes of existing orders; the load status, fault warning information, and capacity parameters of equipment; and the inventory level, replenishment cycle, and quality grade of materials.

3. The AI ​​intelligent agent's emergency order insertion dynamic decision-making and production scheduling method according to claim 2, characterized in that, The agents exchange information through a message passing protocol; the set of multi-objective functions includes minimizing the order insertion response time, minimizing the risk of delay in existing orders, and maximizing resource utilization efficiency; the incremental replanning algorithm only performs local optimization and adjustment on a subset of affected orders, keeping the original production schedule of other orders unchanged.

4. The AI ​​intelligent agent's emergency order insertion dynamic decision-making and production scheduling method according to claim 3, characterized in that, The order insertion agent is responsible for receiving and processing emergency order insertion requests, analyzing the technical requirements and time constraints of the order insertion, calculating the resource requirements of the production system for the order insertion, initiating negotiation requests with other agents and collecting bidding information, and finally determining the production scheduling position and execution plan for the order insertion.

5. The AI ​​intelligent agent's emergency order insertion dynamic decision-making and production scheduling method according to claim 4, characterized in that, The order intelligence agent manages the status information of existing production orders, including the order's process flow, completion progress, remaining time, and quality requirements. It assesses the impact of order insertion operations on its own orders and provides feasible solutions and cost assessments for order adjustments during the negotiation process.

6. The AI ​​intelligent agent's emergency order insertion dynamic decision-making and production scheduling method according to claim 5, characterized in that, The device intelligence agent monitors the operating status and production capacity parameters of the equipment, including equipment load rate, fault warning information, maintenance plan, and processing capacity. Based on its own status and constraints, it provides available time windows and processing quotations to the order-placing intelligence agent and participates in the bidding process for resource allocation.

7. The AI ​​intelligent agent's emergency order insertion dynamic decision-making and production scheduling method according to claim 6, characterized in that, The material intelligence agent tracks the inventory level, quality grade, and replenishment cycle information of materials, assesses the availability and supply capacity of materials required for order insertion, and provides time commitments and cost quotations for material supply during the negotiation process to ensure material support for order insertion.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program instructions, which, when executed in a computer, are used to perform the AI ​​intelligent agent emergency order insertion dynamic decision-making and production scheduling method according to any one of claims 1-7.

9. An AI-powered emergency order insertion dynamic decision-making and production scheduling system, characterized in that, The system includes the computer-readable storage medium of claim 8, wherein the system is any one of a computer, a server, or a microcontroller, the computer-readable storage medium is disposed within the system, and the system is provided with a microprocessor that executes the program instructions stored in the computer-readable storage medium.

Images