A Method for Constructing a Dynamic Scheduling Simulation Model for Cigarette Production Process Based on Artificial Intelligence Technology
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-18
- Publication Date
- 2026-08-14
AI Technical Summary
[0005]本发明的目的在于克服现有技术中的不足，现有的卷烟生产排产过程中智能化程度低、对生产状态变化响应不及时、排产模型与生产仿真过程缺乏持续在线学习优化能力的技术问题
[0040]本发明通过获取卷烟生产过程中的生产状态数据，以获得排产策略，并通过仿真推演获取对应的方案评估结果，对排产策略进行更新，从而能够实现卷烟生产过程中的动态排产与在线优化；相较于现有依赖人工经验或者固定规则的排产方式，本发明能够根据实时生产状态动态调整排产方案，提高卷烟生产排产的动态响应能力。
Smart Images
Figure CN122573290A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a method for constructing a dynamic scheduling simulation model of cigarette production process based on artificial intelligence technology, belonging to the field of cigarette production. Background Technology
[0002] As the scale of cigarette production continues to expand and market demand becomes increasingly diversified, the complexity of production scheduling in the cigarette production process continues to rise. Cigarette production typically involves multiple production stages, such as tobacco processing, cigarette making and packaging, and warehousing and logistics. These stages are closely interconnected and are also affected by various factors, including order delivery cycles, equipment operating status, material inventory, and capacity changes.
[0003] Most existing cigarette production scheduling methods rely on manual planning or automatic scheduling based on fixed rules. Manual planning mainly depends on the planner's experience, and adjustments and arrangements are made manually based on order priority, equipment availability, and production experience. When faced with dynamic situations such as order fluctuations, equipment failures, or emergency orders, it is easy to lead to reduced equipment utilization, order delays, and waste of production resources.
[0004] Some cigarette manufacturers use automated production scheduling systems based on operations research models. These systems allocate production resources by constructing mathematical optimization models to improve production efficiency. However, such scheduling systems are usually based on static constraints and require extensive simplification of the actual production process, making it difficult to accurately reflect the dynamic changes in the cigarette production site. Summary of the Invention
[0005] The purpose of this invention is to overcome the shortcomings of the existing technology, namely, the low level of intelligence in the existing cigarette production scheduling process, the untimely response to changes in production status, and the lack of continuous online learning and optimization capabilities in the scheduling model and production simulation process.
[0006] To solve the above-mentioned technical problems, the present invention is implemented using the following technical solution:
[0007] A method for constructing a dynamic scheduling simulation model for cigarette production based on artificial intelligence technology is provided, including the following steps:
[0008] Acquire production status data during the cigarette production process and construct a production status representation based on the production status data;
[0009] A production scheduling decision model is established based on the production status representation, and the production scheduling decision model is trained by combining historical production data to obtain a production scheduling strategy.
[0010] By combining real-time updated production status representations, candidate production scheduling schemes are generated using scheduling strategies.
[0011] Simulations were performed on candidate production scheduling schemes to obtain corresponding scheme evaluation results;
[0012] Based on the obtained scheme evaluation results, the target production scheduling scheme is determined, and the data is fed back to the production scheduling decision model to update the production scheduling strategy.
[0013] Furthermore, the production status representation includes order status, equipment status, inventory status, and capacity forecast status.
[0014] Furthermore, the production state representation is described through a state space;
[0015] The expression for the state space is:
[0016] ;
[0017] in, Let be the set of orders to be produced at the current time t. This represents the state of the device at time t. Inventory status. Production capacity forecast data for the next N time windows.
[0018] Furthermore, the production scheduling decision model generates corresponding production scheduling actions based on the production status representation. These actions are described through an action space, the expression of which is:
[0019] ;
[0020] in, For order i to be produced, For the allocatable machine j, Let i be the planned start time for order i on machine j.
[0021] Furthermore, the production scheduling decision model evaluates the production scheduling results using a reward function, the expression of which is:
[0022] ;
[0023] in, is the final feedback score obtained by the model after taking action at time t. w1, w2, w3, and w4 are dynamically adjustable weights. YieldReward is the reward for matching the actual output with the target output. ChangeoverPenalty is the penalty for downtime and cost caused by production changeover. DelayPenalty is the penalty for order delivery delay. EnergyPenalty is the penalty for energy consumption per unit output exceeding the standard.
[0024] Furthermore, the training of the production scheduling decision model using historical production data includes:
[0025] Historical production data is input into the production scheduling decision model, and corresponding production scheduling actions are generated.
[0026] Obtain the reward results corresponding to the production scheduling action;
[0027] The parameters of the production scheduling decision model are updated based on the reward results to obtain an initial production scheduling strategy.
[0028] Furthermore, the strategy for obtaining the initial production schedule includes:
[0029] Simulate scheduling actions in a virtual environment, observe changes in the state space, calculate scores based on the reward function, and obtain the basic strategy.
[0030] The strategy function Represented as:
[0031] ;
[0032] Among them, the production scheduling decision model obtains its basic strategy through training. , the current state Mapping to Actions ; In the state Estimate the reward for the next action 'a'.
[0033] Furthermore, the basic strategy undergoes policy adjustment and incremental learning based on the acquired data. The policy adjustment formula is as follows:
[0034] ;
[0035] in, The optimized policy function, To find the policy parameter π that minimizes the subsequent expression, This is the expected value calculated under the historical data distribution D. For the current decision π(s) and expert decision The absolute error between them The action to be manually selected in state s;
[0036] The formula for incremental learning is:
[0037] ;
[0038] in, These are the model's internal weight parameters. To control the hyperparameter of the step size for parameter updates, The derivative of the loss function with respect to the parameters. The TD-Error is calculated for all samples (s, a, r, s') in the experience pool B, where B is a buffer that stores newly generated production data.
[0039] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:
[0040] This invention acquires production status data during cigarette production to obtain a production scheduling strategy, and updates the production scheduling strategy by obtaining corresponding scheme evaluation results through simulation and deduction. This enables dynamic production scheduling and online optimization in the cigarette production process. Compared with existing production scheduling methods that rely on manual experience or fixed rules, this invention can dynamically adjust the production scheduling scheme according to the real-time production status, thereby improving the dynamic response capability of cigarette production scheduling. Attached Figure Description
[0041] Figure 1 The diagram shown is an architecture diagram of a dynamic scheduling simulation model for cigarette production based on artificial intelligence technology provided in an embodiment of the present invention. Detailed Implementation
[0042] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments and specific features in the embodiments are detailed descriptions of the technical solution of the present application, rather than limitations thereof. In the absence of conflict, the technical features in the embodiments and embodiments can be combined with each other.
[0043] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.
[0044] Example 1:
[0045] This embodiment provides a method for constructing a dynamic scheduling simulation model for cigarette production based on artificial intelligence technology, including the following steps:
[0046] Acquire production status data during the cigarette production process and construct a production status representation based on the production status data;
[0047] A production scheduling decision model is established based on the production status representation, and the production scheduling decision model is trained by combining historical production data to obtain a production scheduling strategy.
[0048] By combining real-time updated production status representations, candidate production scheduling schemes are generated using scheduling strategies.
[0049] Simulations were performed on candidate production scheduling schemes to obtain corresponding scheme evaluation results;
[0050] Based on the obtained scheme evaluation results, the target production scheduling scheme is determined, and the data is fed back to the production scheduling decision model to update the production scheduling strategy.
[0051] Specifically, the construction method is deployed in an intelligent production scheduling system consisting of a perception and simulation layer, a decision and optimization layer, and an interaction and execution layer. The perception and simulation layer is used to acquire real-time production data during the cigarette production process and construct a digital twin simulation environment. The decision and optimization layer is used to generate production scheduling decision schemes based on artificial intelligence algorithms. The interaction and execution layer is used to implement the issuance and execution of the production scheduling schemes and the feedback of the execution, thereby forming a closed-loop self-learning mechanism.
[0052] The production status data includes order status, equipment status, inventory status, and capacity forecast status. Order status represents the product specifications, quantity, delivery cycle, and priority information of current orders awaiting production. Equipment status represents the current operating status, downtime status, fault status, and production load of production equipment. Inventory status represents the quantity of work-in-process inventory, material reserves, and buffer status. Capacity forecast status represents changes in production line capacity within a preset future timeframe. The system performs normalization, data cleaning, and time-series correlation processing on the acquired production status data to construct a production status representation corresponding to the current production environment.
[0053] In this embodiment, the production state is characterized by a state space, the expression of which is:
[0054] ;
[0055] in, Let be the set of orders to be produced at the current time t. This represents the state of the device at time t. Inventory status. This provides capacity forecast data for N future time windows.
[0056] The state space enables a unified digital representation of orders, equipment, inventory, and future production capacity in the cigarette production site, serving as the input basis for the production scheduling decision model. After constructing the production state representation, the system further establishes a production scheduling decision model, which is constructed using a deep reinforcement learning model and generates corresponding production scheduling actions based on the current production state representation.
[0057] The production scheduling action is described using an action space, the expression of which is:
[0058] ;
[0059] in, For order i to be produced, For the allocatable machine j, Let i be the planned start time for order i on machine j.
[0060] The production scheduling decision model matches orders with corresponding production equipment and determines the corresponding start time based on the current order status, equipment status, and inventory status, thereby forming a production scheduling action. To enable the production scheduling decision model to autonomously learn the optimal production scheduling strategy, this embodiment further constructs a reward function to evaluate the production scheduling results. The reward function expression is as follows:
[0061] ;
[0062] in, is the final feedback score obtained by the model after taking action at time t. w1, w2, w3, and w4 are dynamically adjustable weights. YieldReward is the reward for matching the actual output with the target output. ChangeoverPenalty is the penalty for downtime and cost caused by production changeover. DelayPenalty is the penalty for order delivery delay. EnergyPenalty is the penalty for energy consumption per unit output exceeding the standard.
[0063] The system uses a reward function to quantitatively evaluate the execution results of different production scheduling actions, thereby guiding the production scheduling decision model towards optimization towards high output, low latency, low changeover, and low energy consumption.
[0064] In this embodiment, the system trains the production scheduling decision model by combining historical production data, including inputting historical production data into the production scheduling decision model to generate corresponding production scheduling actions; obtaining the reward results corresponding to the production scheduling actions; and updating the parameters of the production scheduling decision model according to the reward results to obtain an initial production scheduling strategy.
[0065] Historical production data includes historical order data, historical equipment operation data, historical inventory data, and historical production execution result data. The system simulates the production execution process under different scheduling actions using historical data and continuously updates the model parameters based on reward results, thereby enabling the scheduling decision model to learn the scheduling patterns in the cigarette production process.
[0066] The acquisition of the initial production scheduling strategy includes simulating scheduling actions in a virtual environment, observing changes in the state space, calculating a score based on a reward function, and acquiring the basic strategy; the virtual environment is a digital twin simulation environment used to perform high-fidelity simulation of the operating status of the cigarette production line.
[0067] Specifically, the policy function Represented as:
[0068] ;
[0069] Among them, the production scheduling decision model obtains the basic strategy through training. and the current state Mapping to Actions ; In the state Estimate the reward for the next action 'a'.
[0070] The system calculates the reward estimates for multiple candidate actions using a basic strategy and selects the action with the highest reward estimate as the current production scheduling action. After completing the initial production scheduling strategy training, the system acquires the updated production status representation in real time during the actual cigarette production process. It then uses the production scheduling strategy to generate multiple candidate production scheduling schemes in combination with the real-time updated production status representation. Subsequently, the system performs simulation and deduction on the candidate production scheduling schemes to obtain the corresponding scheme evaluation results.
[0071] The simulation process is completed through a digital twin simulation environment to simulate the cigarette production process within a preset time range in the future. The system performs statistical analysis on equipment utilization, order delivery rate, production cost, changeover frequency and energy consumption. Based on the obtained scheme evaluation results, the system determines the target production scheduling scheme and feeds back the production data after execution, the manual planner's adjustment data and the scheme evaluation data to the production scheduling decision model to update the production scheduling strategy.
[0072] The basic strategy is adjusted and incrementally learned based on the acquired data. The strategy adjustment formula is as follows:
[0073] ;
[0074] in, The optimized policy function, To find the policy parameter π that minimizes the subsequent expression, This is the expected value calculated under the historical data distribution D. For the current decision π(s) and expert decision The absolute error between them The action to be manually selected in state s;
[0075] The formula for incremental learning is:
[0076] ;
[0077] in, These are the model's internal weight parameters. To control the hyperparameter of the step size for parameter updates, The derivative of the loss function with respect to the parameters. The TD-Error is calculated for all samples (s, a, r, s') in the experience pool B, where B is a buffer that stores newly generated production data.
[0078] The system learns from the historical decision-making behavior of human planners to fine-tune the basic strategy, thereby improving the matching degree between the production scheduling strategy and actual production experience. The system continuously incrementally learns the production scheduling decision model, enabling the model to adapt to dynamic production environments such as equipment aging, process adjustments, new product introductions, and order changes, thus achieving continuous optimization of the dynamic production scheduling capability in the cigarette production process.
[0079] Example 2:
[0080] This embodiment also provides a dynamic production scheduling simulation model for cigarette production based on artificial intelligence technology, which is constructed using the method provided in Embodiment 1.
[0081] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A method for constructing a dynamic scheduling simulation model for cigarette production process based on artificial intelligence technology, characterized in that: Includes the following steps: Acquire production status data during the cigarette production process and construct a production status representation based on the production status data; A production scheduling decision model is established based on the production status representation, and the production scheduling decision model is trained by combining historical production data to obtain a production scheduling strategy. By combining real-time updated production status representations, candidate production scheduling schemes are generated using scheduling strategies. Simulations were performed on candidate production scheduling schemes to obtain corresponding scheme evaluation results; Based on the obtained scheme evaluation results, the target production scheduling scheme is determined, and the data is fed back to the production scheduling decision model to update the production scheduling strategy.
2. The method for constructing a dynamic scheduling simulation model for cigarette production based on artificial intelligence technology according to claim 1, characterized in that, The production status representation includes order status, equipment status, inventory status, and capacity forecast status.
3. The method for constructing a dynamic scheduling simulation model of cigarette production process based on artificial intelligence technology according to claim 1, characterized in that, The production state representation is described through a state space; The expression for the state space is: ； in, Let be the set of orders to be produced at the current time t. This represents the state of the device at time t. Inventory status. Production capacity forecast data for the next N time windows.
4. The method for constructing a dynamic scheduling simulation model for cigarette production based on artificial intelligence technology according to claim 1, characterized in that, The production scheduling decision model generates corresponding production scheduling actions based on the production status representation. These actions are described through an action space, the expression of which is: ； in, For order i to be produced, For the allocatable machine j, Let i be the planned start time for order i on machine j.
5. The method for constructing a dynamic scheduling simulation model for cigarette production based on artificial intelligence technology according to claim 1, characterized in that, The production scheduling decision model evaluates the production scheduling results using a reward function, the expression of which is: ； in, is the final feedback score obtained by the model after taking action at time t. w1, w2, w3, and w4 are dynamically adjustable weights. YieldReward is the reward for matching the actual output with the target output. ChangeoverPenalty is the penalty for downtime and cost caused by production changeover. DelayPenalty is the penalty for order delivery delay. EnergyPenalty is the penalty for energy consumption per unit output exceeding the standard.
6. The method for constructing a dynamic scheduling simulation model for cigarette production based on artificial intelligence technology according to claim 5, characterized in that, The training of the production scheduling decision model by combining historical production data includes: Historical production data is input into the production scheduling decision model, and corresponding production scheduling actions are generated. Obtain the reward results corresponding to the production scheduling action; The parameters of the production scheduling decision model are updated based on the reward results to obtain an initial production scheduling strategy.
7. The method for constructing a dynamic scheduling simulation model for cigarette production based on artificial intelligence technology according to claim 6, characterized in that, The strategy for obtaining the initial production schedule includes: Simulate scheduling actions in a virtual environment, observe changes in the state space, calculate scores based on the reward function, and obtain the basic strategy. The strategy function Represented as: ； Among them, the production scheduling decision model obtains the basic strategy through training. , the current state Mapping to Actions ; In the state Estimate the reward for the next action 'a'.
8. The method for constructing a dynamic scheduling simulation model for cigarette production based on artificial intelligence technology according to claim 7, characterized in that, The basic strategy is adjusted and incrementally learned based on the acquired data. The strategy adjustment formula is as follows: ； in, The optimized policy function, To find the policy parameter π that minimizes the subsequent expression, This is the expected value calculated under the historical data distribution D. For the current decision π(s) and expert decision The absolute error between them The action to be manually selected in state s; The formula for incremental learning is: ； in, These are the model's internal weight parameters. To control the hyperparameter of the step size for parameter updates, The derivative of the loss function with respect to the parameters. The TD-Error is calculated for all samples (s, a, r, s') in the experience pool B, where B is a buffer that stores newly generated production data.