Process circulation tracking management method and system based on production data
By using feature screening, scheduling rule mining and deep reinforcement learning models in production scheduling, the problem of data feature redundancy is solved, the accuracy and efficiency of production scheduling are improved, and the optimal flow of production tasks and resource optimization are achieved.
Patent Information
- Application Number
- CN202411931455.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-26
- Publication Date
- 2025-04-18
AI Technical Summary
When combining industrial big data with production scheduling, the existing technology failed to effectively eliminate data feature redundancy, resulting in a decrease in the generalization ability of scheduling rules and limiting the improvement of production efficiency.
By obtaining production tasks and real-time production data, using feature screening models to extract important production features, mining the model for generating scheduling sequences, and optimizing production strategies through deep reinforcement learning models to achieve dynamic adjustment of production optimal strategies.
Effectively filter redundant features, extract highly relevant important production features, improve the accuracy and reliability of scheduling rules, realize the optimal order flow of production tasks, reduce resource waste and task conflicts, and improve process execution efficiency.
Smart Images
Figure CN119849902A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of process tracking management, and in particular relates to a process flow tracking management method and system based on production data. Background Art
[0002] The production management problem is one of the most important issues in the modern industrial production process. Its goal is to reasonably allocate production resources in the workshop so that the jobs to be processed can be processed efficiently on available machines and improve production efficiency. Intelligent scheduling, as a key means to achieve this goal, has become a hot spot and difficulty in research and application. The traditional job shop scheduling method only allows one machine to be used for each process, which is difficult to adapt to the complex and changeable modern production scenarios. The flexible job shop scheduling method provides multiple available machines for each process and allows different machines to have different processing times, providing a more adaptive direction for solving complex scheduling problems. Under the framework of intelligent manufacturing, flexible job shop scheduling not only needs to optimize scheduling rules, but also needs to cope with the challenges brought by industrial big data. The data generated by flexible job shop scheduling is experiencing explosive growth, which provides new opportunities for systematic analysis and intelligent process tracking management, and also puts forward higher requirements for scheduling optimization methods.
[0003] Under the big framework of intelligent manufacturing, it is of great significance to study data-driven production scheduling. At present, more and more methods try to combine industrial big data with scheduling optimization problems. A dynamic hybrid flow shop scheduling method that completes production attributes and corresponding optimal scheduling rules through BP neural network; a dynamic scheduling rule algorithm that combines classification algorithm with production attributes of semiconductor production lines through feature selection algorithm; a discrete shop scheduling method that combines multi-objective Jaya optimization algorithm with Tent mapping. None of the above methods consider the problem of redundancy of production data features. Feature redundancy will lead to a decrease in the generalization ability of scheduling rules, limiting the improvement of production efficiency. Therefore, it is urgent to study how to effectively eliminate data feature redundancy, mine key features and optimize scheduling rules, so as to minimize the maximum completion time and improve production efficiency. Summary of the invention
[0004] In order to solve the above problems existing in the prior art, the present invention provides a process flow tracking management method and system based on production data.
[0005] The purpose of the present invention can be achieved through the following technical solutions:
[0006] A process flow tracking management method based on production data, comprising:
[0007] Acquire production tasks, real-time production data, and production features, and obtain important production features through a feature screening model based on the production features and the production tasks;
[0008] The production tasks and the important production features are stored to obtain a scheduling rule mining data set, and a scheduling order is obtained through a scheduling rule mining model according to the scheduling rule mining data set;
[0009] Adjust the execution order of the production tasks according to the scheduling order to obtain a sequence production task;
[0010] The optimal production strategy is obtained through a tracking scheduling model according to the sequence production tasks and the real-time production data.
[0011] Preferably, obtaining important production features through a feature screening model according to the production features and the production tasks includes:
[0012] S101: obtaining a production task feature data set by storing the production features and the production tasks, and acquiring a scheduling sample set;
[0013] S102: Initializing production feature weights, and obtaining a normalized scheduling sample set by normalizing the scheduling sample set, wherein the normalized scheduling sample set carries a sampling times label;
[0014] S103: Obtain scheduling samples, and classify the normalized scheduling sample set through k-means according to the scheduling samples to obtain the nearest similar scheduling samples and the nearest heterogeneous scheduling samples;
[0015] S104: obtaining an initial updated production feature weight through production weight update calculation according to the scheduling sample, the most recent similar scheduling sample, and the most recent heterogeneous scheduling sample;
[0016] S105: Preset the number of sample samplings, and determine the updated feature weights based on the number of sample samplings and the sampling number labels;
[0017] The sampling times judgment includes judging whether the sampling times of the sample and the sampling times label are equal, if yes, the initial updated production feature weight is the updated feature weight;
[0018] No, the updated feature weight is obtained by repeating the step S103 and the step S104 according to the initial updated production feature weight until the number of sample samplings and the sampling number label are equal;
[0019] S106: Obtaining the key production feature by sorting the production features according to the updated production feature weights.
[0020] Preferably, obtaining the scheduling sequence through a scheduling rule mining model according to the scheduling rule mining data set includes:
[0021] According to the scheduling rule mining data set, a scheduling rule mining training set and a scheduling rule mining test set are obtained by dividing;
[0022] Preset attribute categories, obtain an attribute mining training set by dividing the scheduling rule mining training set and the attribute categories, and obtain a Gini index by calculating the Gini index based on the attribute mining training set;
[0023] Obtaining a rule mining tree by sorting recursively according to the Gini index;
[0024] Obtain a scheduling rule mining test subset, and obtain a scheduling result through the rule mining tree according to the scheduling rule mining test set; obtain a scheduling subset result through the rule mining tree according to the scheduling rule mining test subset;
[0025] The scheduling order is obtained through scheduling judgment according to the scheduling result and the scheduling subset result.
[0026] Preferably, the sorting recursion includes obtaining the minimum attribute by sorting according to the Gini index, obtaining the minimum attribute scheduling rule mining data as the root node according to the minimum attribute; obtaining the remaining attribute samples as leaf nodes under the root node, and continuing to recursively obtain the rule mining tree in sequence according to the principle of minimizing the attribute value.
[0027] Preferably, the scheduling judgment includes judging whether the scheduling result and the scheduling subset result are the same, and if so, the scheduling result is the scheduling order;
[0028] If not, the rule mining tree is pruned to obtain a pruned rule mining tree, and step S204 and step S205 are repeated according to the pruned rule mining tree until the scheduling result is the same as the scheduling subset result.
[0029] Preferably, obtaining the optimal production strategy through a tracking scheduling model according to the sequence production tasks and the real-time production data includes:
[0030] S401: Preset scheduling rules and reward rewards, use the real-time production data as the state space, and use the scheduling rules as the action space;
[0031] S402: The tracking scheduling model includes a decision strategy network, a learning strategy network, a target value network, and an experience buffer. The decision strategy network, the learning strategy network, the target value network, and the experience buffer are initialized to obtain an initialized decision strategy network, an initialized learning strategy network, an initialized target value network, and an initialized experience buffer.
[0032] S403: storing the sequence production task into a list of jobs to be scheduled, obtaining a current state, obtaining an action probability distribution through the initialization decision strategy network according to the current state, and obtaining an execution action through probability sampling according to the action probability distribution;
[0033] S404: Preset a new job adding tag, and determine whether the new job adding tag is 1. If yes, add the new job to the list of jobs to be scheduled to obtain an updated list of jobs to be scheduled; if no, execute step S405;
[0034] S405: Obtain a reward through the reward calculation according to the current state and the execution action, and update the state to obtain the next state;
[0035] S406: Calculating an expected reward through the target value network according to the current state and the execution action;
[0036] S407: storing the current state, the executed action, the reward, the action probability distribution, and the next state into the initialized experience buffer to obtain a stored experience buffer;
[0037] S408: Acquire a motion trajectory, and obtain the same batch of data through the motion trajectory according to the storage experience buffer;
[0038] S409: Obtaining a learning action probability distribution through the initialized learning strategy network according to the same batch of data;
[0039] S410: Obtaining an importance sampling factor through importance sampling calculation according to the learning action probability distribution and the action probability distribution;
[0040] S411: Obtaining a loss function through strategy network loss calculation according to the importance sampling factor;
[0041] S412: updating the decision strategy network parameters by using the Adam algorithm according to the loss function to obtain an updated decision strategy network, wherein the updated decision strategy network carries a training times label;
[0042] S413: Preset the number of iterations, and obtain the optimal production strategy through iterative judgment according to the number of iterations and the training number label.
[0043] Preferably, the iteration judgment includes judging whether the number of iterations and the training number labels are equal, if yes, obtaining the optimal production strategy; if no, updating the decision strategy network by repeating steps S403 to S412 until the number of iterations and the training number labels are equal.
[0044] A process flow tracking management system based on production data, including a feature screening module, a rule mining module, a scheduling sequence module, and a tracking management module, including:
[0045] The feature screening module is used to obtain production tasks, real-time production data, and production features, and obtain important production features through a feature screening model based on the production features and the production tasks;
[0046] The rule mining module is used to obtain a scheduling rule mining data set by storing the production tasks and the important production features, and obtain a scheduling sequence through a scheduling rule mining model according to the scheduling rule mining data set;
[0047] The scheduling sequence module is used to adjust the execution sequence of the production tasks according to the scheduling sequence to obtain a sequence production task;
[0048] The tracking management module is used to obtain the optimal production strategy through a tracking scheduling model according to the sequence production tasks and the real-time production data.
[0049] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the process flow tracking and management method based on production data is implemented.
[0050] A storage medium containing computer executable instructions, which are used to execute the above-mentioned process flow tracking management method based on production data when executed by a computer processor.
[0051] The beneficial effects of the present invention are:
[0052] (1) Through normalization and weight updating methods, production features are screened, some redundant features are filtered out, and important production features that are highly relevant to production tasks are extracted, thereby improving the accuracy of feature data and making subsequent scheduling and optimization more accurate.
[0053] (2) By using the Gini index recursive sorting to generate a rule mining tree, the rule mining tree is pruned and optimized to improve the accuracy and reliability of the rules. Finally, the scheduling order of tasks is generated to ensure that tasks are circulated in the optimal order, reduce resource waste and task conflicts, improve process execution efficiency, and minimize the maximum completion time.
[0054] (3) Through the deep reinforcement learning model, we design appropriate state space, action space and rewards, continuously adjust and optimize the scheduling strategy, and realize dynamic adjustment of production strategy according to real-time production data, ensuring that the production system can still operate efficiently in a complex environment with multiple tasks and multiple resources, minimizing the maximum completion time and improving production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.
[0056] Figure 1 The figure is a flow chart of a process flow tracking management method based on production data of the present invention. DETAILED DESCRIPTION
[0057] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.
[0058] See also Figure 1 , a process flow tracking management method based on production data, comprising:
[0059] S1: Obtain production tasks, real-time production data, and production features, and obtain important production features through a feature screening model based on the production features and the production tasks;
[0060] S2: obtaining a scheduling rule mining data set by storing the production tasks and the important production features, and obtaining a scheduling sequence through a scheduling rule mining model according to the scheduling rule mining data set;
[0061] S3: adjusting the execution order of the production tasks according to the scheduling order to obtain a sequence production task;
[0062] S4: Obtaining the optimal production strategy through a tracking scheduling model according to the sequence production tasks and the real-time production data.
[0063] Specifically, the production data includes production equipment data, product process data, basic resource data, equipment operation data, production process data, and production system attribute data.
[0064] In step S1, the feature screening model includes:
[0065] S101: obtaining a production task feature data set by storing the production feature and the production task, and obtaining a scheduling sample set by extracting the production task feature data set;
[0066] S102: Initializing production feature weights, and obtaining a normalized scheduling sample set by normalizing the scheduling sample set, wherein the normalized scheduling sample set carries a sampling times label;
[0067] S103: Obtain scheduling samples, and classify the normalized scheduling sample set through k-means according to the scheduling samples to obtain the nearest similar scheduling samples and the nearest heterogeneous scheduling samples;
[0068] S104: obtaining an initial updated production feature weight through production weight update calculation according to the scheduling sample, the most recent similar scheduling sample, and the most recent heterogeneous scheduling sample;
[0069] The production weight update calculation expression is:
[0070]
[0071] Wherein, W(A′) is the initial updated production feature weight, W(A) represents the production feature weight after initialization, Represents the scheduling sample x i and x g The difference in production characteristic A, Represents the scheduling sample x i The value of production feature A, is the most recent similar scheduling sample, m is the number of samplings, j represents the jth scheduling sample, k represents the number of similar scheduling samples, the number of similar scheduling samples is equal to the number of heterogeneous scheduling samples, p(l) represents the proportion of the lth type scheduling sample to the total number of scheduling samples, l represents the lth type scheduling sample, type(x i ) represents a dispatch rule, which is a first-come, first-served rule. represents the recent heterogeneous scheduling sample, Represents the scheduling sample x g The value on the production feature A, max(A) represents the maximum value operation, min(A) represents the minimum value operation, and A represents the production feature;
[0072] S105: Preset the number of sample samplings, and determine the updated feature weights based on the number of sample samplings and the sampling number labels;
[0073] The sampling times judgment includes judging whether the sampling times of the sample and the sampling times label are equal, if yes, the initial updated production feature weight is the updated feature weight;
[0074] No, the updated feature weight is obtained by repeating the step S103 and the step S104 according to the initial updated production feature weight until the number of sample samplings and the sampling number label are equal;
[0075] S106: Obtaining the key production feature by sorting the production features according to the updated production feature weights.
[0076] In this embodiment, there are 16 production characteristics, including the average utilization rate of all machines in the next stage corresponding to the current scheduling point, the standard deviation of the utilization rate of all machines in the next stage corresponding to the current scheduling point, the average utilization rate of all machines in the current scheduling point, the standard deviation of the utilization rate of all machines in the current scheduling point, the stage of the current scheduling point, the number of idle machines in the docking stage of the current scheduling point, the minimum processing time of the workpiece to be processed in the buffer where the current scheduling point is located, the maximum processing time of the workpiece to be processed in the buffer where the current scheduling point is located, the average processing time of the workpiece to be processed in the buffer where the current scheduling point is located, the standard deviation of the processing time of the workpiece to be processed in the buffer where the current scheduling point is located, the average completion rate of all workpieces at the current scheduling point, the standard deviation of the completion rate of all workpieces at the current scheduling point, the completion rate of the workpiece corresponding to the previous processing procedure of the idle machine at the current scheduling point, the number of workpieces to be processed in the buffer of the stage where the current scheduling point is located, the average value of the distance to delivery period of the buffer where the current scheduling point is located, and the standard deviation of the distance to delivery period of the buffer where the current scheduling point is located. Initialize the production feature weight to 0, update the production feature weight through the feature screening model, obtain the sorted production feature weight by sorting according to the updated production feature weight, and take the top 10 production features of the sorted production feature weight to perform scheduling rule mining.
[0077] In step S2, the scheduling rule mining model includes:
[0078] S201: obtaining a scheduling rule mining training set and a scheduling rule mining test set by dividing the scheduling rule mining data set;
[0079] S202: Preset attribute categories, obtain an attribute mining training set by dividing the scheduling rule mining training set and the attribute categories, and obtain a Gini index by calculating the Gini index based on the attribute mining training set;
[0080] The Gini index calculation expression is:
[0081]
[0082] Where Gini(U) is the Gini index, k represents the kth attribute category, K represents the number of attribute categories, |U| represents the number of samples in the scheduling rule mining training set, |I k | represents samples belonging to the kth attribute category;
[0083] S203: Obtain a rule mining tree by recursively sorting according to the Gini index;
[0084] The recursive sorting includes obtaining the minimum attribute by sorting according to the Gini index, obtaining the minimum attribute scheduling rule mining data as the root node according to the minimum attribute; obtaining the remaining attribute samples as leaf nodes under the root node, and continuing to recursively obtain the rule mining tree in turn according to the principle of minimizing the attribute value;
[0085] S204: Acquire a scheduling rule mining test subset, and obtain a scheduling result through the rule mining tree according to the scheduling rule mining test set; obtain a scheduling subset result through the rule mining tree according to the scheduling rule mining test subset;
[0086] S205: Obtaining the scheduling order through scheduling judgment according to the scheduling result and the scheduling subset result;
[0087] The scheduling judgment includes determining whether the scheduling result and the scheduling subset result are the same. If yes, the scheduling result is the scheduling order; if not, the rule mining tree is pruned to obtain a pruned rule mining tree, and steps S204 and S205 are repeated according to the pruned rule mining tree until the scheduling result and the scheduling subset result are the same.
[0088] In this embodiment, the dispatch rule mining data set is divided into the ratio of 3:7 to obtain the dispatch rule mining training set and the dispatch rule mining test set.
[0089] In step S4, the tracking scheduling model is based on the PPO algorithm, which specifically includes:
[0090] S401: Preset scheduling rules and reward rewards, use the real-time production data as the state space, and use the scheduling rules as the action space;
[0091] The return reward is expressed as:
[0092] r t =((Dr e (t)-Dr e (t+1))+(U ave (t+1)-U ave (t)))*α,
[0093] Among them, r t represents the reward, (Dr e (t)-Dr e (t+1)) represents the reduction in delay from time t to t+1, (U ave (t+1)-U ave (t)) represents the increase in average machine utilization from time t to t+1, and α is the coefficient;
[0094] S402: The tracking scheduling model includes a decision strategy network, a learning strategy network, a target value network, and an experience buffer. The decision strategy network, the learning strategy network, the target value network, and the experience buffer are initialized to obtain an initialized decision strategy network, an initialized learning strategy network, an initialized target value network, and an initialized experience buffer.
[0095] S403: storing the sequence production task into a list of jobs to be scheduled, obtaining a current state, obtaining an action probability distribution through the initialization decision strategy network according to the current state, and obtaining an execution action through probability sampling according to the action probability distribution;
[0096] S404: Preset a new job adding tag, and determine whether the new job adding tag is 1. If yes, add the new job to the list of jobs to be scheduled to obtain an updated list of jobs to be scheduled; if no, execute step S405;
[0097] S405: Obtain a reward through the reward calculation according to the current state and the execution action, and update the state to obtain the next state;
[0098] S406: Calculating an expected reward through the target value network according to the current state and the execution action;
[0099] S407: storing the current state, the executed action, the reward, the action probability distribution, and the next state into the initialized experience buffer to obtain a stored experience buffer;
[0100] S408: Acquire a motion trajectory, and obtain the same batch of data through the motion trajectory according to the storage experience buffer;
[0101] S409: Obtaining a learning action probability distribution through the initialized learning strategy network according to the same batch of data;
[0102] S410: Obtaining an importance sampling factor through importance sampling calculation according to the learning action probability distribution and the action probability distribution;
[0103] The importance sampling calculation expression is:
[0104]
[0105] Among them, ratio(θ) is the importance sampling factor, p θ (a t |s t ) is the learning action probability distribution, is the action probability distribution;
[0106] S411: Obtaining a loss function through strategy network loss calculation according to the importance sampling factor;
[0107] The strategy network loss function calculation expression is:
[0108]
[0109] Among them, loss a represents the loss function, s t Indicates the current state, a t represents the execution action, min represents the minimization operation, ratio represents the importance sampling factor, A t is the advantage function, clip is the limit strategy update amplitude function, ε is the greedy coefficient, δ t is the motion trajectory, γ and λ are discount factors, T is the number of time steps, t represents the tth time step, r t represents the reward, v(s t+1 ) represents the expected return of the next state, v(s t ) represents the expected reward of the current state;
[0110] S412: updating the decision strategy network parameters by using the Adam algorithm according to the loss function to obtain an updated decision strategy network, wherein the updated decision strategy network carries a training times label;
[0111] S413: Preset the number of iterations, and obtain the optimal production strategy through iterative judgment according to the number of iterations and the training number label;
[0112] The iteration judgment includes judging whether the number of iterations and the training number labels are equal. If yes, the optimal production strategy is obtained; if not, the decision strategy network is updated by repeating steps S403 to S412 until the number of iterations and the training number labels are equal.
[0113] Specifically, the dispatching rules include 6 composite dispatching rules, including:
[0114] S401-1: Determine whether there is a delayed job. If yes, select a job with the largest expected delay time from the set of jobs that can be scheduled; if no, select a job with the smallest slack time that can be allocated to each process;
[0115] Select the machine with the earliest processing start time among the available machines;
[0116] S401-2: Determine whether there is a delayed job. If yes, select a job with the largest expected delay time from the set of jobs that can be scheduled; if no, select a job with the smallest ratio of slack time to remaining processing time to complete it first;
[0117] Select the machine with the earliest processing start time among the available machines;
[0118] S401-3: Select a job whose expected delay time multiplied by the job inverse completion rate is the largest from the set of jobs that can be scheduled;
[0119] S401-4: Determine whether there is a delayed job. If yes, select a job with the largest product of expected delay time and job inverse completion rate from the set of allowed scheduling jobs; if no, select a job with the smallest ratio of slack time to remaining processing time and complete it first;
[0120] Select the machine with the earliest processing start time among the available machines;
[0121] S401-5: Determine whether there is a delayed job. If yes, select a job with the longest expected delay time from the set of jobs that can be scheduled; if no, select the job with the shortest remaining time after completion;
[0122] Select the machine with the earliest processing start time among the available machines;
[0123] S401-6: Randomly select a job from the set of jobs that are allowed to be scheduled; select the machine with the earliest start processing time from the available machines.
[0124] In this embodiment, the problem model is: 5000 workpieces are processed on 200 different equipment sets. Each workpiece has a set of 10 processes, and each process has 5 equipment to choose from. The processing time of the process also varies depending on the equipment selected, and there are sequence constraints between processes. Given the processing time of each process, determine the processing order and processing time of all workpieces on the equipment. At the same time, scheduling should follow the scheduling constraints;
[0125] The scheduling constraints include:
[0126] One process can only be processed on one optional equipment;
[0127] Two processes on the same process route cannot be processed at the same time;
[0128] The same equipment can only process one process at a time;
[0129] The processing cannot be interrupted;
[0130] Different workpieces have the same processing priority;
[0131] The processes of the same workpiece have sequence constraints, but there are no sequence constraints between different workpieces.
[0132] A process flow tracking management system based on production data, including a feature screening module, a rule mining module, a scheduling sequence module, and a tracking management module, including:
[0133] The feature screening module is used to obtain production tasks, real-time production data, and production features, and obtain important production features through a feature screening model based on the production features and the production tasks;
[0134] The rule mining module is used to obtain a scheduling rule mining data set by storing the production tasks and the important production features, and obtain a scheduling sequence through a scheduling rule mining model according to the scheduling rule mining data set;
[0135] The scheduling sequence module is used to adjust the execution sequence of the production tasks according to the scheduling sequence to obtain a sequence production task;
[0136] The tracking management module is used to obtain the optimal production strategy through a tracking scheduling model according to the sequence production tasks and the real-time production data.
[0137] The computer storage medium of the embodiment of the present invention may adopt any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device.
[0138] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.
[0139] The program code included on the computer readable medium can be transmitted with any appropriate medium, including but not limited to wireless, electric wire, optical cable, RF, etc., or any suitable combination of the above. The computer program code for performing the operation of the present invention can be written in one or more programming languages or their combinations, and the programming language includes object-oriented programming languages-such as Java, Smalltalk, C++, and also includes conventional procedural programming languages-such as "C" language or similar programming languages. The program code can be executed completely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on the remote computer, or completely on the remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).
[0140] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments according to the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of the technical solution of the present invention.
Claims
1. A process flow tracking management method based on production data, characterized in that: The following steps are involved: Acquire production tasks, real-time production data, and production features, and obtain important production features through a feature screening model based on the production features and the production tasks; The production tasks and the important production features are stored to obtain a scheduling rule mining data set, and a scheduling order is obtained through a scheduling rule mining model according to the scheduling rule mining data set; Adjust the execution order of the production tasks according to the scheduling order to obtain a sequence production task; The optimal production strategy is obtained through a tracking scheduling model according to the sequence production tasks and the real-time production data.
2. The process flow tracking management method based on production data according to claim 1 is characterized in that: The important production features obtained by the feature screening model according to the production features and the production tasks include: S101: obtaining a production task feature data set by storing the production features and the production tasks, and acquiring a scheduling sample set; S102: Initializing production feature weights, and obtaining a normalized scheduling sample set by normalizing the scheduling sample set, wherein the normalized scheduling sample set carries a sampling times label; S103: Obtain scheduling samples, and classify the normalized scheduling sample set through k-means according to the scheduling samples to obtain the nearest similar scheduling samples and the nearest heterogeneous scheduling samples; S104: obtaining an initial updated production feature weight through production weight update calculation according to the scheduling sample, the most recent similar scheduling sample, and the most recent heterogeneous scheduling sample; S105: Preset the number of sample samplings, and determine the updated feature weights based on the number of sample samplings and the sampling number labels; The sampling times judgment includes judging whether the sampling times of the sample and the sampling times label are equal, if yes, the initial updated production feature weight is the updated feature weight; No, the updated feature weight is obtained by repeating the step S103 and the step S104 according to the initial updated production feature weight until the number of sample samplings and the sampling number label are equal; S106: Obtaining the key production feature by sorting the production features according to the updated production feature weights.
3. The process flow tracking management method based on production data according to claim 1 is characterized in that: The step of obtaining the scheduling sequence by using the scheduling rule mining model according to the scheduling rule mining data set comprises: According to the scheduling rule mining data set, a scheduling rule mining training set and a scheduling rule mining test set are obtained by dividing; Preset attribute categories, obtain an attribute mining training set by dividing the scheduling rule mining training set and the attribute categories, and obtain a Gini index by calculating the Gini index based on the attribute mining training set; Obtaining a rule mining tree by sorting recursively according to the Gini index; Obtain a scheduling rule mining test subset, and obtain a scheduling result through the rule mining tree according to the scheduling rule mining test set; obtain a scheduling subset result through the rule mining tree according to the scheduling rule mining test subset; The scheduling order is obtained through scheduling judgment according to the scheduling result and the scheduling subset result.
4. The process flow tracking management method based on production data according to claim 3 is characterized in that: The sorting recursion includes obtaining the minimum attribute by sorting according to the Gini index, obtaining the minimum attribute scheduling rule mining data as the root node according to the minimum attribute; obtaining the remaining attribute samples as leaf nodes under the root node, and continuing to recursively obtain the rule mining tree in sequence according to the principle of minimizing the attribute value.
5. The process flow tracking management method based on production data according to claim 3 is characterized in that: The scheduling judgment includes judging whether the scheduling result and the scheduling subset result are the same, and if yes, the scheduling result is the scheduling order; If not, the rule mining tree is pruned to obtain a pruned rule mining tree, and step S204 and step S205 are repeated according to the pruned rule mining tree until the scheduling result is the same as the scheduling subset result.
6. The process flow tracking management method based on production data according to claim 1 is characterized in that: The obtaining of the optimal production strategy by tracking the scheduling model according to the sequence production tasks and the real-time production data includes: S401: Preset scheduling rules and reward rewards, use the real-time production data as the state space, and use the scheduling rules as the action space; S402: The tracking scheduling model includes a decision strategy network, a learning strategy network, a target value network, and an experience buffer. The decision strategy network, the learning strategy network, the target value network, and the experience buffer are initialized to obtain an initialized decision strategy network, an initialized learning strategy network, an initialized target value network, and an initialized experience buffer. S403: storing the sequence production task into a list of jobs to be scheduled, obtaining a current state, obtaining an action probability distribution through the initialization decision strategy network according to the current state, and obtaining an execution action through probability sampling according to the action probability distribution; S404: Preset a new job adding tag, and determine whether the new job adding tag is 1. If yes, add the new job to the list of jobs to be scheduled to obtain an updated list of jobs to be scheduled; if no, execute step S405; S405: Obtaining a reward through the reward calculation according to the current state and the execution action, and updating the state to obtain the next state; S406: Calculating an expected reward through the target value network according to the current state and the execution action; S407: storing the current state, the executed action, the reward, the action probability distribution, and the next state into the initialized experience buffer to obtain a stored experience buffer; S408: Acquire a motion trajectory, and obtain the same batch of data through the motion trajectory according to the storage experience buffer; S409: Obtaining a learning action probability distribution through the initialized learning strategy network according to the same batch of data; S410: Obtaining an importance sampling factor through importance sampling calculation according to the learning action probability distribution and the action probability distribution; S411: Obtaining a loss function through strategy network loss calculation according to the importance sampling factor; S412: updating the decision strategy network parameters by using the Adam algorithm according to the loss function to obtain an updated decision strategy network, wherein the updated decision strategy network carries a training times label; S413: Preset the number of iterations, and obtain the optimal production strategy through iterative judgment according to the number of iterations and the training number label.
7. The process flow tracking management method based on production data according to claim 6 is characterized in that: The iteration judgment includes judging whether the iteration number and the training number label are equal, and if yes, obtaining the optimal production strategy; If not, the update decision strategy network is repeated from step S403 to step S412 until the number of iterations and the number of training labels are equal.
8. A process flow tracking management system based on production data, It includes a feature screening module, a rule mining module, a scheduling sequence module, and a tracking management module, and is characterized by including: The feature screening module is used to obtain production tasks, real-time production data, and production features, and obtain important production features through a feature screening model based on the production features and the production tasks; The rule mining module is used to obtain a scheduling rule mining data set by storing the production tasks and the important production features, and obtain a scheduling sequence through a scheduling rule mining model according to the scheduling rule mining data set; The scheduling sequence module is used to adjust the execution sequence of the production tasks according to the scheduling sequence to obtain a sequence production task; The tracking management module is used to obtain the optimal production strategy through a tracking scheduling model according to the sequence production tasks and the real-time production data.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the process flow tracking management method based on production data as described in any one of claims 1-7 is implemented.
10. A storage medium containing computer executable instructions, characterized in that: The computer executable instructions are used to execute the process flow tracking management method based on production data as described in any one of claims 1-7 when executed by a computer processor.