Textile production scheduling method and system based on deep learning
Through the optimization method of textile production scheduling models and genetic algorithms based on deep learning, the problem of inefficiency of traditional textile production scheduling methods is solved, and efficient and reliable textile production scheduling decisions and production plans are achieved.
Patent Information
- Application Number
- CN202411931281.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-26
- Publication Date
- 2025-05-16
AI Technical Summary
The existing textile production scheduling methods rely on manual experience, are inefficient, prone to errors, and are difficult to adapt to production changes. How to build an accurate textile production scheduling model and achieve intelligent decision-making.
A deep learning-based method is adopted to create a textile production scheduling model through convolutional neural networks, recurrent neural networks, long-term and short-term memory networks and feature fusion networks, and combine genetic algorithms and modbus protocols to achieve processing and optimization of real-time data.
The quality and efficiency of textile production scheduling are improved, and the generated production scheduling plan is more reliable, and can quickly adapt to production changes without manual operation, which significantly improves the overall efficiency of textile production.
Smart Images
Figure CN120013120A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of textile production technology, and in particular to a textile production scheduling method and system based on deep learning. Background Art
[0002] Before textile production, production scheduling is required to ensure the smooth progress of textile production and to ensure the quality and quantity of textile products. Traditionally, textile scheduling relies on manual experience, which has disadvantages such as low efficiency, prone to errors, and difficulty in adapting to production changes. With the development of big data and artificial intelligence technology, it has become possible to use big data in the textile production process for intelligent scheduling. However, how to effectively integrate and utilize these big data, build an accurate textile scheduling model, and realize intelligent scheduling decisions is an important technical challenge facing the current textile industry.
[0003] Therefore, how to provide a textile production scheduling method and system based on deep learning to improve the quality and efficiency of textile production scheduling has become a technical problem that needs to be solved urgently. Summary of the invention
[0004] The technical problem to be solved by the present invention is to provide a textile production scheduling method and system based on deep learning, so as to improve the quality and efficiency of textile production scheduling.
[0005] In a first aspect, the present invention provides a textile production scheduling method based on deep learning, comprising the following steps:
[0006] Step S1, the server obtains a large amount of historical textile production data, and constructs a data set after preprocessing each of the historical textile production data;
[0007] Step S2: The server creates a textile production scheduling model for outputting a textile production scheduling plan based on a convolutional neural network, a recurrent neural network, a long short-term memory network, and a feature fusion network, and sets a loss function of the textile production scheduling model;
[0008] Step S3: the server trains the textile production scheduling model using the data set and the loss function, and deploys the trained textile production scheduling model;
[0009] Step S4: The server periodically sends a data acquisition instruction to the PLC located in the textile production line based on the modbus protocol. The PLC signs and encrypts the real-time textile production data based on the received data acquisition instruction to obtain an encrypted data packet and uploads it to the server;
[0010] Step S5: the server decrypts and verifies the received encrypted data packet to obtain real-time textile production data, and inputs the real-time textile production data into the deployed textile production scheduling model to obtain a textile production scheduling plan;
[0011] Step S6: The server optimizes the textile production schedule by using a genetic algorithm and input production constraints to obtain a textile production schedule decision;
[0012] Step S7: The server optimizes the textile production scheduling model based on the textile production scheduling decision, the real-time textile production data and the execution result of the textile production scheduling decision.
[0013] Furthermore, the step S1 is specifically as follows:
[0014] The server obtains a large amount of historical textile production data including at least order information, raw material inventory, machine status, production progress and worker skill level, performs preprocessing on each of the historical textile production data including at least data cleaning, data integration and data normalization, and constructs a data set based on each of the preprocessed historical textile production data.
[0015] Furthermore, in step S2, the convolutional neural network is used to extract local features of different time periods from the textile production data through different convolution kernels; the recurrent neural network is used to extract short-distance dependency features in adjacent textile production data through a hidden layer; the long short-term memory network is used to extract long-distance dependency features in textile production data with a long time span; the feature fusion network is used to fuse local features, short-distance dependency features and long-distance dependency features to output a predicted textile production schedule; the loss function adopts a cross entropy loss function or a mean square error loss function.
[0016] Furthermore, the step S3 is specifically as follows:
[0017] The server divides the data set into a training set, a validation set, and a test set based on a preset ratio;
[0018] The textile production scheduling model is trained by the training set. During the training process, the gradient information of the loss function is calculated based on the stochastic gradient descent algorithm, and the model parameters of the textile production scheduling model are updated based on the gradient information, so that the loss value of the loss function converges in the direction of minimization until the loss value is less than a preset loss threshold;
[0019] Verifying the trained textile production scheduling model through the verification set, and adjusting the hyperparameters of the textile production scheduling model based on the verification result;
[0020] The verified textile production scheduling model is tested by using the test set, and the textile production scheduling model that passes the test is deployed.
[0021] Furthermore, the step S4 is specifically as follows:
[0022] Based on the modbus protocol, the server periodically sends data acquisition instructions to the PLC installed in the textile production line. Based on the received data acquisition instructions, the PLC signs the new real-time textile production data generated within a preset period with a preset private key to obtain a signature value, encrypts the real-time textile production data and the signature value into an encrypted data packet through the DES algorithm, and uploads the encrypted data packet to the server in real time.
[0023] In a second aspect, the present invention provides a textile production scheduling system based on deep learning, comprising the following modules:
[0024] A data set construction module is used for the server to obtain a large amount of historical textile production data, and to construct a data set after preprocessing each of the historical textile production data;
[0025] A textile production scheduling model creation module, used for the server to create a textile production scheduling model for outputting a textile production scheduling plan based on a convolutional neural network, a recurrent neural network, a long short-term memory network, and a feature fusion network, and to set a loss function of the textile production scheduling model;
[0026] A textile production scheduling model training module, used for the server to train the textile production scheduling model through the data set and the loss function, and deploy the trained textile production scheduling model;
[0027] A data acquisition instruction sending module is used for the server to periodically send data acquisition instructions to the PLC located in the textile production line based on the modbus protocol. The PLC signs and encrypts the real-time textile production data based on the received data acquisition instructions to obtain an encrypted data packet and uploads it to the server;
[0028] A textile production scheduling generation module, which is used for the server to decrypt and verify the received encrypted data packet to obtain real-time textile production data, and input the real-time textile production data into the deployed textile production scheduling model to obtain a textile production scheduling plan;
[0029] A textile production scheduling decision generation module is used for the server to optimize the textile production scheduling plan through a genetic algorithm and input production constraints to obtain a textile production scheduling decision;
[0030] The textile production scheduling model optimization module is used for the server to optimize the textile production scheduling model based on the textile production scheduling decision, real-time textile production data and the execution result of the textile production scheduling decision.
[0031] Furthermore, the data set construction module is specifically used for:
[0032] The server obtains a large amount of historical textile production data including at least order information, raw material inventory, machine status, production progress and worker skill level, performs preprocessing on each of the historical textile production data including at least data cleaning, data integration and data normalization, and constructs a data set based on each of the preprocessed historical textile production data.
[0033] Furthermore, in the textile production scheduling model creation module, the convolutional neural network is used to extract local features of different time periods from the textile production data through different convolution kernels; the recurrent neural network is used to extract short-distance dependency features in adjacent textile production data through a hidden layer; the long short-term memory network is used to extract long-distance dependency features in textile production data with a long time span; the feature fusion network is used to fuse local features, short-distance dependency features and long-distance dependency features to output a predicted textile production scheduling plan; the loss function adopts a cross entropy loss function or a mean square error loss function.
[0034] Furthermore, the textile production scheduling model training module is specifically used for:
[0035] The server divides the data set into a training set, a validation set, and a test set based on a preset ratio;
[0036] The textile production scheduling model is trained by the training set. During the training process, the gradient information of the loss function is calculated based on the stochastic gradient descent algorithm, and the model parameters of the textile production scheduling model are updated based on the gradient information, so that the loss value of the loss function converges in the direction of minimization until the loss value is less than a preset loss threshold;
[0037] Verifying the trained textile production scheduling model through the verification set, and adjusting the hyperparameters of the textile production scheduling model based on the verification result;
[0038] The verified textile production scheduling model is tested by using the test set, and the textile production scheduling model that passes the test is deployed.
[0039] Furthermore, the data acquisition instruction sending module is specifically used for:
[0040] Based on the modbus protocol, the server periodically sends data acquisition instructions to the PLC installed in the textile production line. Based on the received data acquisition instructions, the PLC signs the new real-time textile production data generated within a preset period with a preset private key to obtain a signature value, encrypts the real-time textile production data and the signature value into an encrypted data packet through the DES algorithm, and uploads the encrypted data packet to the server in real time.
[0041] The advantages of the present invention are:
[0042] A large amount of historical textile production data is obtained through the server and a data set is constructed after preprocessing; then a textile production scheduling model for outputting a textile production schedule is created based on a convolutional neural network, a recurrent neural network, a long short-term memory network, and a feature fusion network, and a loss function of the textile production scheduling model is set; then the textile production scheduling model is trained through the data set and the loss function, and the trained textile production scheduling model is deployed; then the server periodically sends data acquisition instructions to the PLC located in the textile production line based on the modbus protocol, and the PLC signs and encrypts the real-time textile production data based on the data acquisition instructions to obtain an encrypted data packet and uploads it to the server; then the server decrypts and verifies the encrypted data packet to obtain real-time textile production data and inputs it into the deployed textile production scheduling model to obtain a textile production schedule, and optimizes the textile production schedule through a genetic algorithm and the input production constraints to obtain a textile production scheduling decision; finally, the server optimizes the textile production scheduling model based on the textile production scheduling decision, the real-time textile production data, and the execution results of the textile production scheduling decision; that is, the textile production scheduling model is optimized through the big data of historical textile production data The convolutional neural network of the textile production scheduling model is used to extract local features of different time periods from the textile production data through different convolution kernels, the recurrent neural network is used to extract short-distance dependency features in adjacent textile production data through hidden layers, and the long short-term memory network is used to extract long-distance dependency features in textile production data with a long time span. The feature fusion network is used to fuse local features, short-distance dependency features and long-distance dependency features, so that the textile production scheduling model can effectively extract the characteristics and laws of the textile production scheduling process from the textile production data, ensure the reliability of the generation of the textile production scheduling plan, and optimize the textile production scheduling plan in combination with genetic algorithms and production constraints, so that the final textile production scheduling decision can be more in line with the actual production needs, and the textile production scheduling model is continuously optimized based on the textile production scheduling decision, real-time textile production data and the execution results of the textile production scheduling decision, and the quality of the textile production scheduling plan is continuously improved. Subsequently, the textile production scheduling decision can be quickly output through the acquired real-time textile production data, without manual operation, and can quickly adapt to production changes, which ultimately greatly improves the quality and efficiency of textile production scheduling. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] The present invention will be further described below in conjunction with embodiments with reference to the accompanying drawings.
[0044] Figure 1 It is a flow chart of a textile production scheduling method based on deep learning of the present invention.
[0045] Figure 2 It is a structural schematic diagram of a textile production scheduling system based on deep learning in the present invention. DETAILED DESCRIPTION
[0046] The technical solution in the embodiment of the present application has the following overall idea: the textile production scheduling model is trained through the big data of historical textile production data, and the convolutional neural network of the textile production scheduling model is used to extract local features of different time periods from the textile production data through different convolution kernels, the recurrent neural network is used to extract short-distance dependency features in adjacent textile production data through hidden layers, the long short-term memory network is used to extract long-distance dependency features in each textile production data with a long time span, and the feature fusion network is used to fuse local features, short-distance dependency features and long-distance dependency features, so that the textile production scheduling model can effectively extract the characteristics and laws in the textile production scheduling process from the textile production data, ensure the reliability of the generation of the textile production scheduling plan, and also optimize the textile production scheduling plan in combination with genetic algorithms and production constraints, so that the final textile production scheduling decision can be more in line with actual production needs, and the textile production scheduling model is continuously optimized based on the textile production scheduling decision, real-time textile production data and the execution results of the textile production scheduling decision, and the quality of the textile production scheduling plan is continuously improved. Subsequently, the textile production scheduling decision can be quickly output through the acquired real-time textile production data, without manual operation, and can quickly adapt to production changes to improve the quality and efficiency of textile scheduling.
[0047] Please refer to Figure 1 to Figure 2 As shown, a preferred embodiment of a textile production scheduling method based on deep learning of the present invention comprises the following steps:
[0048] Step S1, the server obtains a large amount of historical textile production data, and constructs a data set after preprocessing each of the historical textile production data;
[0049] Step S2, the server creates a textile production scheduling model for outputting a textile production scheduling plan based on a convolutional neural network (CNN), a recurrent neural network (RNN), a long short-term memory network (LSTM), and a feature fusion network, and sets a loss function of the textile production scheduling model;
[0050] Step S3: the server trains the textile production scheduling model using the data set and the loss function, and deploys the trained textile production scheduling model;
[0051] Step S4: The server periodically sends a data acquisition instruction to the PLC located in the textile production line based on the modbus protocol. The PLC signs and encrypts the real-time textile production data based on the received data acquisition instruction to obtain an encrypted data packet and uploads it to the server;
[0052] Step S5: the server decrypts and verifies the received encrypted data packet to obtain real-time textile production data, and inputs the real-time textile production data into the deployed textile production scheduling model to obtain a textile production scheduling plan;
[0053] Step S6: The server optimizes the textile production schedule by using a genetic algorithm and input production constraints to obtain a textile production schedule decision;
[0054] Step S7: The server optimizes the textile production scheduling model based on the textile production scheduling decision, the real-time textile production data and the execution result of the textile production scheduling decision, that is, realizes the dynamic optimization of the textile production scheduling decision.
[0055] The step S1 is specifically as follows:
[0056] The server obtains a large amount of historical textile production data including at least order information, raw material inventory, machine status, production progress and worker skill level, performs preprocessing on each of the historical textile production data including at least data cleaning, data integration and data normalization, and constructs a data set based on each of the preprocessed historical textile production data.
[0057] In specific implementation, the historical textile production data can be mapped through the object model to perform simple calculations on related data, data field mapping and other operations. The object model usually adopts the following methods: 1. JSON Schema: define the properties, commands and events of the device in JSON format to facilitate machine parsing and processing; 2. XML Schema: use XML format to describe the object model, which has good scalability and readability; 3. Custom protocol: use custom protocols and formats to implement the object model to meet specific business needs.
[0058] In step S2, the convolutional neural network is used to extract local features of different time periods from the textile production data through different convolution kernels, that is, different convolution kernels are responsible for capturing local features of different dimensions, and multiple convolution layers containing convolution kernels are often set. As the number of convolution layers increases, the local features are gradually abstracted and advanced.
[0059] The recurrent neural network is used to extract short-distance dependency features in adjacent textile production data through hidden layers; the uniqueness of the hidden layer is that each hidden layer has a feedback loop, which can process sequence data and use the output of the previous moment as one of the inputs of the current moment, which has advantages in processing time series data of textile production (such as data on production progress changing over time, data on machine status at different time points, etc.), and can capture the sequence dependency relationship in the data; the input layer of the recurrent neural network is used to receive pre-processed time series data, and input it into the network in sequence according to time steps, so that the network can learn and extract features along the time dimension; the output layer outputs the corresponding prediction value according to the processing result of the hidden layer on the entire time series data, such as predicting the production progress of the next stage or whether the machine is likely to fail, etc.;
[0060] The long short-term memory network is used to extract long-distance dependency features in various textile production data with a long time span; that is, the long short-term memory network is used to better remember the key features and dependency relationships in the time series data with a longer time span in textile production, such as information such as the long-term production efficiency change trend;
[0061] The feature fusion network is used to fuse local features, short-distance dependent features and long-distance dependent features to output a predicted textile production schedule;
[0062] The loss function adopts a cross entropy loss function or a mean square error loss function; in specific implementation, the cross entropy loss function can be selected for classification tasks, and the mean square error loss function can be selected for regression tasks.
[0063] The step S3 is specifically as follows:
[0064] The server divides the data set into a training set, a validation set, and a test set based on a preset ratio;
[0065] The textile production scheduling model is trained by the training set. During the training process, the gradient information of the loss function is calculated based on the stochastic gradient descent algorithm (SGD), and the model parameters of the textile production scheduling model are updated based on the gradient information, so that the loss value of the loss function converges in the direction of minimization, and the accuracy and effectiveness of extracting key features and rules of the production process are continuously improved until the loss value is less than the preset loss threshold; the model parameters at least include the weights of the convolution kernels in CNN, and the weights of each gate in RNN and LSTM; in specific implementation, variant algorithms of the stochastic gradient descent algorithm, such as Adagrad, Adade l ta, Adam, etc., can also be selected;
[0066] Verifying the trained textile production scheduling model through the verification set, and adjusting the hyperparameters of the textile production scheduling model based on the verification result to prevent overfitting and the like;
[0067] The verified textile production scheduling model is tested by using the test set to evaluate the generalization ability of the textile production scheduling model, and the textile production scheduling model that passes the test is deployed.
[0068] The step S4 is specifically as follows:
[0069] Based on the modbus protocol, the server periodically sends data acquisition instructions to the PLC installed in the textile production line. Based on the received data acquisition instructions, the PLC signs the new real-time textile production data generated within a preset period with a preset private key to obtain a signature value, encrypts the real-time textile production data and the signature value into an encrypted data packet through the DES algorithm, and uploads the encrypted data packet to the server in real time.
[0070] The real-time textile production data is signed with a private key to obtain a signature value, and then the real-time textile production data and the signature value are encrypted into an encrypted data packet through the DES algorithm to prevent the real-time textile production data from being stolen in plain text during transmission. The signature value can be used for subsequent verification to determine the data integrity, thereby greatly improving the security of uploading real-time textile production data.
[0071] The step S5 is specifically as follows:
[0072] The server decrypts the received encrypted data packet through the DES algorithm to obtain real-time textile production data and a signature value. After verifying the signature value through a preset public key matching the private key, the real-time textile production data is input into the deployed textile production scheduling model to obtain a textile production schedule.
[0073] In step S6, the production constraints are exemplified by machine capacity, worker scheduling, and delivery date; in specific implementation, the textile production scheduling decision includes an optimistic decision and a pessimistic decision;
[0074] Optimistic decision-making is reflected in textile production scheduling as follows: assuming that the production scheduling process is relatively smooth, a better scheduling plan can be found quickly; for example, the optimistic decision obtained based on the heuristic optimistic scheduling algorithm will give priority to textile tasks with high production efficiency and high order urgency, and construct the initial production schedule in a more radical way, such as assuming ideal conditions such as no equipment failure and timely supply of raw materials.
[0075] Taking the optimistic application of genetic algorithms as an example, when generating the initial population (scheduling plan), it tends to choose a plan that seems to be able to complete the order quickly. In the crossover and mutation operations, new combinations will be boldly tried in the hope of finding the global optimal or near-global optimal scheduling plan more quickly. It is relatively loose in dealing with various constraints in the scheduling process (such as equipment maintenance time, worker scheduling, etc.), and believes that these constraints can be met through rapid adjustments. Application scenario: When the demand for textile orders is relatively simple, the equipment status is stable, and the order delivery period is relatively loose, the optimistic scheduling algorithm can play its advantages. For example, in the production of conventional textiles, when the equipment has just been maintained and is unlikely to fail in the near future, the optimistic scheduling algorithm can quickly formulate a scheduling plan.
[0076] The manifestation of pessimistic decision in textile production scheduling: Based on the pessimistic decision obtained by the pessimistic scheduling algorithm, various possible adverse factors will be carefully considered in textile production scheduling, and factors such as equipment failure rate, uncertainty in raw material supply, and worker leave will be analyzed in detail; when constructing a scheduling plan, priority will be given to meeting the most basic requirements of the order, such as ensuring that the order can be delivered on time, even in the worst case. For example, in the scheduling algorithm based on constraint planning, the pessimistic scheduling algorithm will strictly incorporate constraints such as equipment maintenance time and raw material arrival time into the core considerations of scheduling, and arrange production tasks in a conservative way, such as reserving enough time for equipment maintenance, even if the equipment is currently in good condition. When arranging the production sequence of orders, priority will be given to those orders with the tightest delivery period and the least risk factors. Application scenarios and advantages: When the textile production environment is complex and changeable, such as aging equipment, unstable raw material supply, strict order requirements, etc., the pessimistic scheduling algorithm is more suitable. For example, in the production of high-end customized textiles, the equipment needs to adjust the process parameters frequently, and the raw materials are purchased from multiple suppliers with uncertain supply time. The pessimistic scheduling algorithm can effectively avoid order delays caused by various uncertain factors. Its advantages lie in stability and reliability. The scheduling plan generated by the pessimistic scheduling algorithm can still guarantee the delivery of orders on time with a high probability when various unfavorable factors occur. It can deal with emergencies well. For example, when equipment fails, due to the reservation of a certain buffer time or backup plan, production can be transitioned relatively smoothly, reducing the impact on the entire production plan.
[0077] A preferred embodiment of a textile production scheduling system based on deep learning of the present invention includes the following modules:
[0078] A data set construction module is used for the server to obtain a large amount of historical textile production data, and to construct a data set after preprocessing each of the historical textile production data;
[0079] A textile production scheduling model creation module is used for the server to create a textile production scheduling model for outputting a textile production scheduling plan based on a convolutional neural network (CNN), a recurrent neural network (RNN), a long short-term memory network (LSTM) and a feature fusion network, and to set a loss function of the textile production scheduling model;
[0080] A textile production scheduling model training module, used for the server to train the textile production scheduling model through the data set and the loss function, and deploy the trained textile production scheduling model;
[0081] A data acquisition instruction sending module is used for the server to periodically send data acquisition instructions to the PLC located in the textile production line based on the modbus protocol. The PLC signs and encrypts the real-time textile production data based on the received data acquisition instructions to obtain an encrypted data packet and uploads it to the server;
[0082] A textile production scheduling generation module, which is used for the server to decrypt and verify the received encrypted data packet to obtain real-time textile production data, and input the real-time textile production data into the deployed textile production scheduling model to obtain a textile production scheduling plan;
[0083] A textile production scheduling decision generation module is used for the server to optimize the textile production scheduling plan through a genetic algorithm and input production constraints to obtain a textile production scheduling decision;
[0084] The textile production scheduling model optimization module is used for the server to optimize the textile production scheduling model based on the textile production scheduling decision, real-time textile production data and the execution result of the textile production scheduling decision, that is, to achieve dynamic optimization of the textile production scheduling decision.
[0085] The data set construction module is specifically used for:
[0086] The server obtains a large amount of historical textile production data including at least order information, raw material inventory, machine status, production progress and worker skill level, performs preprocessing on each of the historical textile production data including at least data cleaning, data integration and data normalization, and constructs a data set based on each of the preprocessed historical textile production data.
[0087] In specific implementation, the historical textile production data can be mapped through the object model to perform simple calculations on related data, data field mapping and other operations. The object model usually adopts the following methods: 1. JSON Schema: define the properties, commands and events of the device in JSON format to facilitate machine parsing and processing; 2. XML Schema: use XML format to describe the object model, which has good scalability and readability; 3. Custom protocol: use custom protocols and formats to implement the object model to meet specific business needs.
[0088] In the textile production scheduling model creation module, the convolutional neural network is used to extract local features of different time periods from textile production data through different convolution kernels, that is, different convolution kernels are responsible for capturing local features of different dimensions, and multiple convolution layers containing convolution kernels are often set. As the number of convolution layers increases, local features are gradually abstracted and advanced.
[0089] The recurrent neural network is used to extract short-distance dependency features in adjacent textile production data through hidden layers; the uniqueness of the hidden layer is that each hidden layer has a feedback loop, which can process sequence data and use the output of the previous moment as one of the inputs of the current moment, which has advantages in processing time series data of textile production (such as data on production progress changing over time, data on machine status at different time points, etc.), and can capture the sequence dependency relationship in the data; the input layer of the recurrent neural network is used to receive pre-processed time series data, and input it into the network in sequence according to time steps, so that the network can learn and extract features along the time dimension; the output layer outputs the corresponding prediction value according to the processing result of the hidden layer on the entire time series data, such as predicting the production progress of the next stage or whether the machine is likely to fail, etc.;
[0090] The long short-term memory network is used to extract long-distance dependency features in various textile production data with a long time span; that is, the long short-term memory network is used to better remember the key features and dependency relationships in the time series data with a longer time span in textile production, such as information such as the long-term production efficiency change trend;
[0091] The feature fusion network is used to fuse local features, short-distance dependent features and long-distance dependent features to output a predicted textile production schedule;
[0092] The loss function adopts a cross entropy loss function or a mean square error loss function; in specific implementation, the cross entropy loss function can be selected for classification tasks, and the mean square error loss function can be selected for regression tasks.
[0093] The textile production scheduling model training module is specifically used for:
[0094] The server divides the data set into a training set, a validation set, and a test set based on a preset ratio;
[0095] The textile production scheduling model is trained by the training set. During the training process, the gradient information of the loss function is calculated based on the stochastic gradient descent algorithm (SGD), and the model parameters of the textile production scheduling model are updated based on the gradient information, so that the loss value of the loss function converges in the direction of minimization, and the accuracy and effectiveness of extracting key features and rules of the production process are continuously improved until the loss value is less than the preset loss threshold; the model parameters at least include the weights of the convolution kernels in CNN, and the weights of each gate in RNN and LSTM; in specific implementation, variant algorithms of the stochastic gradient descent algorithm, such as Adagrad, Adade l ta, Adam, etc., can also be selected;
[0096] Verifying the trained textile production scheduling model through the verification set, and adjusting the hyperparameters of the textile production scheduling model based on the verification result to prevent overfitting and the like;
[0097] The verified textile production scheduling model is tested by using the test set to evaluate the generalization ability of the textile production scheduling model, and the textile production scheduling model that passes the test is deployed.
[0098] The data acquisition instruction sending module is specifically used for:
[0099] Based on the modbus protocol, the server periodically sends data acquisition instructions to the PLC installed in the textile production line. Based on the received data acquisition instructions, the PLC signs the new real-time textile production data generated within a preset period with a preset private key to obtain a signature value, encrypts the real-time textile production data and the signature value into an encrypted data packet through the DES algorithm, and uploads the encrypted data packet to the server in real time.
[0100] The real-time textile production data is signed with a private key to obtain a signature value, and then the real-time textile production data and the signature value are encrypted into an encrypted data packet through the DES algorithm to prevent the real-time textile production data from being stolen in plain text during transmission. The signature value can be used for subsequent verification to determine the data integrity, thereby greatly improving the security of uploading real-time textile production data.
[0101] The textile production scheduling plan generation module is specifically used for:
[0102] The server decrypts the received encrypted data packet through the DES algorithm to obtain real-time textile production data and a signature value. After verifying the signature value through a preset public key matching the private key, the real-time textile production data is input into the deployed textile production scheduling model to obtain a textile production schedule.
[0103] In the textile production scheduling decision generation module, the production constraint conditions are exemplified by machine capacity, worker scheduling, and delivery date; in specific implementation, the textile production scheduling decision includes optimistic decision and pessimistic decision;
[0104] Optimistic decision-making is reflected in textile production scheduling as follows: assuming that the production scheduling process is relatively smooth, a better scheduling plan can be found quickly; for example, the optimistic decision obtained based on the heuristic optimistic scheduling algorithm will give priority to textile tasks with high production efficiency and high order urgency, and construct the initial production schedule in a more radical way, such as assuming ideal conditions such as no equipment failure and timely supply of raw materials.
[0105] Taking the optimistic application of genetic algorithms as an example, when generating the initial population (scheduling plan), it tends to choose a plan that seems to be able to complete the order quickly. In the crossover and mutation operations, new combinations will be boldly tried in the hope of finding the global optimal or near-global optimal scheduling plan more quickly. It is relatively loose in dealing with various constraints in the scheduling process (such as equipment maintenance time, worker scheduling, etc.), and believes that these constraints can be met through rapid adjustments. Application scenario: When the demand for textile orders is relatively simple, the equipment status is stable, and the order delivery period is relatively loose, the optimistic scheduling algorithm can play its advantages. For example, in the production of conventional textiles, when the equipment has just been maintained and is unlikely to fail in the near future, the optimistic scheduling algorithm can quickly formulate a scheduling plan.
[0106] The manifestation of pessimistic decision in textile production scheduling: Based on the pessimistic decision obtained by the pessimistic scheduling algorithm, various possible adverse factors will be carefully considered in textile production scheduling, and factors such as equipment failure rate, uncertainty in raw material supply, and worker leave will be analyzed in detail; when constructing a scheduling plan, priority will be given to meeting the most basic requirements of the order, such as ensuring that the order can be delivered on time, even in the worst case. For example, in the scheduling algorithm based on constraint planning, the pessimistic scheduling algorithm will strictly incorporate constraints such as equipment maintenance time and raw material arrival time into the core considerations of scheduling, and arrange production tasks in a conservative way, such as reserving enough time for equipment maintenance, even if the equipment is currently in good condition. When arranging the production sequence of orders, priority will be given to those orders with the tightest delivery period and the least risk factors. Application scenarios and advantages: When the textile production environment is complex and changeable, such as aging equipment, unstable raw material supply, strict order requirements, etc., the pessimistic scheduling algorithm is more suitable. For example, in the production of high-end customized textiles, the equipment needs to adjust the process parameters frequently, and the raw materials are purchased from multiple suppliers with uncertain supply time. The pessimistic scheduling algorithm can effectively avoid order delays caused by various uncertain factors. Its advantages lie in stability and reliability. The scheduling plan generated by the pessimistic scheduling algorithm can still guarantee the delivery of orders on time with a high probability when various unfavorable factors occur. It can deal with emergencies well. For example, when equipment fails, due to the reservation of a certain buffer time or backup plan, production can be transitioned relatively smoothly, reducing the impact on the entire production plan.
[0107] In summary, the advantages of the present invention are:
[0108] A large amount of historical textile production data is obtained through the server and a data set is constructed after preprocessing; then a textile production scheduling model for outputting a textile production schedule is created based on a convolutional neural network, a recurrent neural network, a long short-term memory network, and a feature fusion network, and a loss function of the textile production scheduling model is set; then the textile production scheduling model is trained through the data set and the loss function, and the trained textile production scheduling model is deployed; then the server periodically sends data acquisition instructions to the PLC located in the textile production line based on the modbus protocol, and the PLC signs and encrypts the real-time textile production data based on the data acquisition instructions to obtain an encrypted data packet and uploads it to the server; then the server decrypts and verifies the encrypted data packet to obtain real-time textile production data and inputs it into the deployed textile production scheduling model to obtain a textile production schedule, and optimizes the textile production schedule through a genetic algorithm and the input production constraints to obtain a textile production scheduling decision; finally, the server optimizes the textile production scheduling model based on the textile production scheduling decision, the real-time textile production data, and the execution results of the textile production scheduling decision; that is, the textile production scheduling model is optimized through the big data of historical textile production data The convolutional neural network of the textile production scheduling model is used to extract local features of different time periods from the textile production data through different convolution kernels, the recurrent neural network is used to extract short-distance dependency features in adjacent textile production data through hidden layers, and the long short-term memory network is used to extract long-distance dependency features in textile production data with a long time span. The feature fusion network is used to fuse local features, short-distance dependency features and long-distance dependency features, so that the textile production scheduling model can effectively extract the characteristics and laws of the textile production scheduling process from the textile production data, ensure the reliability of the generation of the textile production scheduling plan, and optimize the textile production scheduling plan in combination with genetic algorithms and production constraints, so that the final textile production scheduling decision can be more in line with the actual production needs, and the textile production scheduling model is continuously optimized based on the textile production scheduling decision, real-time textile production data and the execution results of the textile production scheduling decision, and the quality of the textile production scheduling plan is continuously improved. Subsequently, the textile production scheduling decision can be quickly output through the acquired real-time textile production data, without manual operation, and can quickly adapt to production changes, which ultimately greatly improves the quality and efficiency of textile production scheduling.
[0109] Although the specific implementation modes of the present invention are described above, those skilled in the art should understand that the specific implementation modes described are only illustrative and are not intended to limit the scope of the present invention. Equivalent modifications and changes made by those skilled in the art in accordance with the spirit of the present invention should be included in the scope of protection of the claims of the present invention.
Claims
1. A textile production scheduling method based on deep learning, characterized in that: The steps include: Step S1, the server obtains a large amount of historical textile production data, and constructs a data set after preprocessing each of the historical textile production data; Step S2: The server creates a textile production scheduling model for outputting a textile production scheduling plan based on a convolutional neural network, a recurrent neural network, a long short-term memory network, and a feature fusion network, and sets a loss function of the textile production scheduling model; Step S3: the server trains the textile production scheduling model using the data set and the loss function, and deploys the trained textile production scheduling model; Step S4: The server periodically sends a data acquisition instruction to the PLC located in the textile production line based on the modbus protocol. The PLC signs and encrypts the real-time textile production data based on the received data acquisition instruction to obtain an encrypted data packet and uploads it to the server; Step S5: the server decrypts and verifies the received encrypted data packet to obtain real-time textile production data, and inputs the real-time textile production data into the deployed textile production scheduling model to obtain a textile production scheduling plan; Step S6: The server optimizes the textile production schedule by using a genetic algorithm and input production constraints to obtain a textile production schedule decision; Step S7: The server optimizes the textile production scheduling model based on the textile production scheduling decision, the real-time textile production data and the execution result of the textile production scheduling decision.
2. A textile production scheduling method based on deep learning as claimed in claim 1, characterized in that: The step S1 is specifically as follows: The server obtains a large amount of historical textile production data including at least order information, raw material inventory, machine status, production progress and worker skill level, performs preprocessing on each of the historical textile production data including at least data cleaning, data integration and data normalization, and constructs a data set based on each of the preprocessed historical textile production data.
3. A textile production scheduling method based on deep learning as claimed in claim 1, characterized in that: In step S2, the convolutional neural network is used to extract local features of different time periods from the textile production data through different convolution kernels; the recurrent neural network is used to extract short-distance dependency features in adjacent textile production data through a hidden layer; and the long short-term memory network is used to extract long-distance dependency features in textile production data with a long time span; The feature fusion network is used to fuse local features, short-distance dependency features and long-distance dependency features to output a predicted textile production schedule; the loss function adopts a cross entropy loss function or a mean square error loss function.
4. A textile production scheduling method based on deep learning as claimed in claim 1, characterized in that: The step S3 is specifically as follows: The server divides the data set into a training set, a validation set, and a test set based on a preset ratio; The textile production scheduling model is trained by the training set. During the training process, the gradient information of the loss function is calculated based on the stochastic gradient descent algorithm, and the model parameters of the textile production scheduling model are updated based on the gradient information, so that the loss value of the loss function converges in the direction of minimization until the loss value is less than a preset loss threshold; Verifying the trained textile production scheduling model through the verification set, and adjusting the hyperparameters of the textile production scheduling model based on the verification result; The verified textile production scheduling model is tested by using the test set, and the textile production scheduling model that passes the test is deployed.
5. A textile production scheduling method based on deep learning as claimed in claim 1, characterized in that: The step S4 is specifically as follows: Based on the modbus protocol, the server periodically sends data acquisition instructions to the PLC installed in the textile production line. Based on the received data acquisition instructions, the PLC signs the new real-time textile production data generated within a preset period with a preset private key to obtain a signature value, encrypts the real-time textile production data and the signature value into an encrypted data packet through the DES algorithm, and uploads the encrypted data packet to the server in real time.
6. A textile production scheduling system based on deep learning, characterized by: Includes the following modules: A data set construction module is used for the server to obtain a large amount of historical textile production data, and to construct a data set after preprocessing each of the historical textile production data; A textile production scheduling model creation module, used for the server to create a textile production scheduling model for outputting a textile production scheduling plan based on a convolutional neural network, a recurrent neural network, a long short-term memory network, and a feature fusion network, and to set a loss function of the textile production scheduling model; A textile production scheduling model training module, used for the server to train the textile production scheduling model through the data set and the loss function, and deploy the trained textile production scheduling model; A data acquisition instruction sending module is used for the server to periodically send data acquisition instructions to the PLC located in the textile production line based on the modbus protocol. The PLC signs and encrypts the real-time textile production data based on the received data acquisition instructions to obtain an encrypted data packet and uploads it to the server; A textile production scheduling generation module, which is used for the server to decrypt and verify the received encrypted data packet to obtain real-time textile production data, and input the real-time textile production data into the deployed textile production scheduling model to obtain a textile production scheduling plan; A textile production scheduling decision generation module is used for the server to optimize the textile production scheduling plan through a genetic algorithm and input production constraints to obtain a textile production scheduling decision; The textile production scheduling model optimization module is used for the server to optimize the textile production scheduling model based on the textile production scheduling decision, real-time textile production data and the execution result of the textile production scheduling decision.
7. A textile production scheduling system based on deep learning as claimed in claim 6, characterized in that: The data set construction module is specifically used for: The server obtains a large amount of historical textile production data including at least order information, raw material inventory, machine status, production progress and worker skill level, performs preprocessing on each of the historical textile production data including at least data cleaning, data integration and data normalization, and constructs a data set based on each of the preprocessed historical textile production data.
8. A textile production scheduling system based on deep learning as claimed in claim 6, characterized in that: In the textile production scheduling model creation module, the convolutional neural network is used to extract local features of different time periods from the textile production data through different convolution kernels; the recurrent neural network is used to extract short-distance dependency features in adjacent textile production data through hidden layers; the long short-term memory network is used to extract long-distance dependency features in textile production data with a long time span; The feature fusion network is used to fuse local features, short-distance dependency features and long-distance dependency features to output a predicted textile production schedule; the loss function adopts a cross entropy loss function or a mean square error loss function.
9. A textile production scheduling system based on deep learning as claimed in claim 6, characterized in that: The textile production scheduling model training module is specifically used for: The server divides the data set into a training set, a validation set, and a test set based on a preset ratio; The textile production scheduling model is trained by the training set. During the training process, the gradient information of the loss function is calculated based on the stochastic gradient descent algorithm, and the model parameters of the textile production scheduling model are updated based on the gradient information, so that the loss value of the loss function converges in the direction of minimization until the loss value is less than a preset loss threshold; Verifying the trained textile production scheduling model through the verification set, and adjusting the hyperparameters of the textile production scheduling model based on the verification result; The verified textile production scheduling model is tested by using the test set, and the textile production scheduling model that passes the test is deployed.
10. A textile production scheduling system based on deep learning as claimed in claim 6, characterized in that: The data acquisition instruction sending module is specifically used for: Based on the modbus protocol, the server periodically sends data acquisition instructions to the PLC installed in the textile production line. Based on the received data acquisition instructions, the PLC signs the new real-time textile production data generated within a preset period with a preset private key to obtain a signature value, encrypts the real-time textile production data and the signature value into an encrypted data packet through the DES algorithm, and uploads the encrypted data packet to the server in real time.