Non-replacement flow shop scheduling method and device based on neural network

Abstract

The present invention relates to intelligent decision-making technology, and discloses a non-permutation flow shop scheduling method based on a neural network, including: obtaining historical flow shop scheduling problem data, and converting the historical flow shop problem data into an initial disjunctive graph according to a preset disjunctive graph conversion relationship; recording each node in the initial disjunctive graph, and performing decision modeling based on each node using a preset model framework to obtain an initial decision model; using a preset graph neural network model to perform feature extraction and feature embedding on each node in the initial disjunctive graph, and using a preset reinforcement learning algorithm to train the preset graph neural network model to obtain a standard graph neural network model; using the standard graph neural network model to optimize the initial decision model to obtain a target decision model. The present invention also proposes a non-permutation flow shop scheduling device based on a neural network. The present invention can reduce the difficulty of implementing non-permutation flow shop scheduling problems and improve practicality.

Description

Technical Field

[0001] The present invention relates to the field of intelligent decision-making technology, and in particular to a non-replacement flow shop scheduling method and device based on a neural network. Background Art

[0002] Production scheduling is an important part of intelligent manufacturing. Among scheduling problems, non-permutation flowshop scheduling (NPFS) is a generalization of traditional permutation flowshop scheduling (PFS), which allows the order of workpieces to be changed on different machines. The scale of feasible solutions to the non-permutation flowshop scheduling problem is much larger than that of the permutation flowshop scheduling problem, and the challenge of obtaining high-quality scheduling solutions is greater. Therefore, the study of this problem has greater theoretical significance and application value. Due to its wide application in modern manufacturing, the NPFS (non-permutation flowshop scheduling) problem has been proven to be a non-deterministic polynomial (NP) problem of polynomial complexity.

[0003] At present, exact algorithms, heuristic algorithms and meta-heuristic algorithms are classic algorithms for solving flow shop problems. Exact algorithms include mathematical modeling and branch-and-bound methods, which can obtain the optimal solution for small-scale problems. However, in actual workshop scheduling problems, if the production scale is large, it is difficult to guarantee the optimal solution using exact solutions. Exact algorithms refer to a class of methods that solve small-scale scheduling problems by enumerating all feasible solutions and obtain exact solutions. Some researchers have proposed a mixed integer programming model to minimize total advance and delay. However, when we need to solve larger-scale NPFS problems, exact methods cannot always guarantee the optimal solution, and have a high dependence on the scale and parameters of the problem, and have poor versatility. Therefore, approximate methods have been developed to approximately solve such problems. Approximate methods include heuristic algorithms and meta-heuristic algorithms.

[0004] Heuristic algorithms are search algorithms based on experience and intuition. They can be used to solve complex optimization problems, including NP-hard problems such as NPFS. They usually guide the search through some simple rules to make the search more efficient. These rules may be based on the characteristics of the problem or previous experience, such as domain knowledge or previous search results. This type of algorithm has high solution efficiency, but the solution quality is average.

[0005] Metaheuristic algorithms refer to a class of higher-level heuristic algorithms that can automatically design more effective heuristic algorithms through methods such as adaptation, learning, and evolution. Metaheuristic algorithms usually contain multiple sub-algorithms, and cross and mutate in different problem domains to generate more adaptable heuristic algorithms. Although metaheuristic algorithms are algorithms that can solve NP-difficult problems within an acceptable time, the process of solving NPFS problems with this type of method is very complicated. For large-scale practical scheduling problems, although metaheuristic algorithms can obtain approximate optimal solutions in a shorter time, there are problems such as excessive computational complexity.

[0006] The above-mentioned traditional exact algorithms and meta-heuristic algorithms have their own limitations in solving large-scale problems, which makes the solution process more complicated, and thus leads to poor solution quality and poor versatility. Summary of the invention

[0007] The present invention provides a neural network-based non-permutation flow shop scheduling method and device, which can reduce the difficulty of implementing the neural network-based non-permutation flow shop scheduling problem and improve practicality.

[0008] To achieve the above object, the present invention provides a non-permutation flow shop scheduling method based on a neural network, comprising:

[0009] Acquire historical flow shop scheduling problem data, and transform the historical flow shop problem data into an initial disjunctive graph according to a preset disjunctive graph transformation relationship;

[0010] Recording each node in the initial disjunctive graph, and performing decision modeling based on each node using a preset model framework to obtain an initial decision model;

[0011] Using a preset graph neural network model to perform feature extraction and feature embedding on each node in the initial disjunctive graph, and using a preset reinforcement learning algorithm to train the preset graph neural network model, and after the training is completed, a standard graph neural network model is obtained;

[0012] Optimizing the initial decision model using the standard graph neural network to obtain a target decision model;

[0013] The target decision model is used to perform scheduling analysis on the flow shop scheduling problem data to be solved to obtain a scheduling result.

[0014] Optionally, the converting the historical flow shop problem data into a disjunctive graph according to a preset disjunctive graph conversion relationship includes:

[0015] Extracting the scheduling elements of the historical flow shop problem data, and constructing disjunctive graph nodes using the workpiece elements in the scheduling elements, and assigning different colors to the disjunctive graph nodes corresponding to different workpieces;

[0016] Connecting the nodes of the disjunctive graph using directed connecting lines of a first color to represent the sequential processing order of the fixed workpieces in the scheduling element on each machine, thereby obtaining a first connecting graph;

[0017] Using directed connecting lines of a second color to connect workpieces of the same color processed by the same machine, a second connecting line diagram is obtained;

[0018] The first connection graph and the second connection graph are integrated to obtain the disjunction graph.

[0019] Optionally, performing decision modeling according to each node using a preset model framework to obtain an initial decision model includes:

[0020] Taking the state of the next arbitrary node of the initial disjunctive graph starting from the starting node as the initial state;

[0021] Taking the next decision action of the workpiece in the initial state as the current selectable action;

[0022] Calculate the difference between the maximum estimated completion time generated by the initial state and the maximum estimated completion time generated by the next state, and use the difference as a reward;

[0023] Utilizing the reward to optimize the currently selectable action to obtain an optimal scheduling state;

[0024] Taking the next schedule generated by the optimal scheduling state as an update action, and using the update action to update the initial disjunctive graph;

[0025] The above steps are repeated, and the current state of the updated initial disjunctive graph is used as the updated state, and the updated initial disjunctive graph is integrated to obtain a scheduling decision model.

[0026] Optionally, the using a preset graph neural network model to perform feature extraction and feature embedding on each node in the initial decision model includes:

[0027] Convert each node in the initial decision model into a three-dimensional original feature;

[0028] Using a graph isomorphism network model to convert the three-dimensional original features into feature vectors;

[0029] Map the feature vector to the graph neural network model.

[0030] Optionally, the converting the three-dimensional original features into feature vectors by using a graph isomorphic network model includes:

[0031] Iteratively calculating feature embedding of the three-dimensional original features, and performing batch normalization to obtain normalized three-dimensional features;

[0032] The normalized three-dimensional features are pooled using a preset pooling function to obtain the feature vector.

[0033] Optionally, the iterative calculation of the feature embedding of the three-dimensional original feature includes:

[0034] The feature embedding of the three-dimensional original feature is iteratively calculated using the following formula:

[0035]

[0036] Where k is the number of iterations, h v (k) is the feature embedding of node v iterated k times, is a multilayer perceptron with the parameter θ for the number of iterations k k , ε (k) is an arbitrary number that can be learned, N (v) is the neighborhood of v.

[0037] Optionally, the using a preset reinforcement learning algorithm to train the preset graph neural network model includes:

[0038] Obtain the scheduling information of the graph neural network model, input the scheduling information into the new Actor network of the preset reinforcement learning algorithm, and construct a normal distribution of each execution action under the current state data;

[0039] Filtering an execution action from the normal distribution, and then inputting it into the graph neural network model to obtain a reward scalar value and next state data under the current state data;

[0040] Input the next state data into the Actor new network, and cyclically obtain the reward scalar value and the next state data until a preset number of scheduling information, reward scalar values, and the next execution action are obtained;

[0041] Inputting the next state data obtained in the last step of the loop into the Critic network of the preset reinforcement learning algorithm to obtain the state value function of the current state data;

[0042] Calculate the reward scalar value discount function according to the state value function until the entire scheduling process is completed, and update the Critic network according to the first loss function in the preset reinforcement learning algorithm by back propagation;

[0043] Inputting the obtained preset amount of scheduling information into the old Actor network and the new Actor network respectively to obtain a first normal distribution and a second normal distribution;

[0044] Inputting the obtained preset number of next execution actions into the first normal distribution and the second normal distribution respectively to obtain a first probability value and a second probability value;

[0045] Calculating a relative rate of change according to the first probability value and the second probability value;

[0046] According to the relative rate of change, updating the new Actor network using a second loss function;

[0047] The updating process of the Critic network and the Actor new network is iteratively trained k times to obtain the standard graph neural network model.

[0048] In order to solve the above problems, the present invention also provides a non-permutation flow shop scheduling device based on a neural network, the device comprising:

[0049] A disjunctive graph conversion module, used to obtain historical flow shop scheduling problem data, and convert the historical flow shop problem data into an initial disjunctive graph according to a preset disjunctive graph conversion relationship;

[0050] The target decision model establishment module is used to record each node in the initial disjunctive graph, and use a preset model framework to perform decision modeling based on each node to obtain an initial decision model; use a preset graph neural network model to extract and embed features for each node in the initial disjunctive graph, and use a preset reinforcement learning algorithm to train the preset graph neural network model to obtain a standard graph neural network model; use the standard graph neural network model to optimize the initial decision model to obtain a target decision model.

[0051] In order to solve the above problem, the present invention further provides an electronic device, the electronic device comprising:

[0052] at least one processor; and,

[0053] a memory communicatively connected to the at least one processor; wherein,

[0054] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the above-mentioned neural network-based non-replacement flow shop scheduling method.

[0055] In order to solve the above problems, the present invention also provides a computer-readable storage medium, in which at least one computer program is stored. The at least one computer program is executed by a processor in an electronic device to implement the above-mentioned neural network-based non-replacement flow shop scheduling method.

[0056] The embodiment of the present invention can simplify the difficulty of flow shop problem data by converting historical flow shop problem data into disjunctive graphs and converting text features into image features, and can convert each node of the initial disjunctive graph into feature embedding of a graph neural network. The initial decision model can be optimized through the graph neural network, and then the decision optimization of the disjunctive graph can be made to achieve the scheduling decision of the flow shop scheduling problem. The optimized scheduling model can be tested with test data to test the performance of the optimized scheduling model and ensure the generalization ability of the model. The present invention can reduce the difficulty of realizing the non-permutation flow shop scheduling problem based on a neural network and improve practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 A flowchart of a non-permutation flow shop scheduling method based on a neural network provided by an embodiment of the present invention;

[0058] Figure 2 A flowchart of a PPO algorithm of a non-permutation flow shop scheduling method based on a neural network provided in one embodiment of the present invention;

[0059] Figure 3 An overall algorithm flow chart of a non-permutation flow shop scheduling method based on a neural network provided by an embodiment of the present invention;

[0060] Figure 4 A functional module diagram of a non-permutation flow shop scheduling device based on a neural network provided by an embodiment of the present invention;

[0061] Figure 5 A schematic diagram of the structure of an electronic device for implementing the neural network-based non-replacement flow shop scheduling method provided in one embodiment of the present invention.

[0062] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0063] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0064] The embodiment of the present application provides a non-permutation flow shop scheduling method based on a neural network. The execution subject of the non-permutation flow shop scheduling method based on a neural network includes but is not limited to at least one of the electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided by the embodiment of the present application. In other words, the non-permutation flow shop scheduling method based on a neural network can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc. The server can be an independent server, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN), and big data and artificial intelligence platforms.

[0065] Reference Figure 1 FIG. 1 is a flow chart of a non-permutation flow shop scheduling method based on a neural network provided by an embodiment of the present invention. In this embodiment, the non-permutation flow shop scheduling method based on a neural network includes:

[0066] S1. Obtain historical flow shop scheduling problem data, and transform the historical flow shop problem data into an initial disjunctive graph according to a preset disjunctive graph transformation relationship.

[0067] In the embodiment of the present invention, the historical flow shop problem data refers to scheduling data related to arranging multiple workpieces on a production line to be processed on multiple machines, including processing workpiece data and processing machine data.

[0068] The embodiment of the present invention can obtain historical flow shop scheduling problem data by using relevant crawler tools to capture large-scale examples of flow shop problem data on web pages.

[0069] In the embodiment of the present invention, the preset disjunctive graph conversion relationship refers to the corresponding relationship between the flow shop problem data and the disjunctive graph, for example, converting the workpieces into nodes in the disjunctive graph and converting the processing sequence into connecting lines in the disjunctive graph.

[0070] As an embodiment of the present invention, according to a preset disjunctive graph transformation relationship, the historical flow shop problem data is transformed into an initial disjunctive graph, including:

[0071] Extract the scheduling elements of historical flow shop problem data, and use the workpiece elements in the scheduling elements to construct disjunctive graph nodes, and assign different colors to the disjunctive graph nodes corresponding to different workpieces;

[0072] The nodes of the disjunctive graph are connected by using directed connecting lines of the first color to represent the processing order of the fixed workpieces in the scheduling element on each machine, thereby obtaining a first connecting graph;

[0073] Using directed connecting lines of a second color to connect workpieces of the same color processed by the same machine, a second connecting line diagram is obtained;

[0074] The first connection graph and the second connection graph are integrated to obtain the initial disjunction graph.

[0075] The embodiment of the present invention converts historical flow shop problem data into a disjunctive graph according to a preset disjunctive graph conversion relationship, can realize the conversion of text features into image features, and can simplify the difficulty of flow shop problem data.

[0076] S2. Record each node in the initial disjunctive graph, and perform decision modeling based on each node using a preset model framework to obtain an initial decision model.

[0077] In an embodiment of the present invention, the preset model framework refers to a model framework that integrates the connection lines between each node in the disjunctive graph into a decision-making process. The Markov decision process model can be used. The Markov decision process model provides a framework for modeling and making decisions for random processes, which can help us solve complex scheduling problems. The Markov property means that the next state depends only on the current state and action, and is independent of the previous historical state.

[0078] As an embodiment of the present invention, the decision modeling is performed according to each node using a preset model framework to obtain an initial decision model, including:

[0079] Taking the state of the next arbitrary node of the initial disjunctive graph starting from the starting node as the initial state;

[0080] Taking the next decision action of the workpiece in the initial state as the current selectable action;

[0081] Calculate the difference between the maximum estimated completion time generated by the initial state and the maximum estimated completion time generated by the next state, and use the difference as a reward;

[0082] Utilizing the reward to optimize the currently selectable action to obtain an optimal scheduling state;

[0083] Taking the next schedule generated by the optimal scheduling state as an update action, and using the update action to update the initial disjunctive graph;

[0084] The above steps are repeated, and the current state of the updated initial disjunctive graph is used as the updated state, and the updated initial disjunctive graph is integrated to obtain an initial decision model.

[0085] In the embodiment of the present invention, the initial state refers to the first state of converting the flow shop problem data into a disjunctive graph, which is the initial state, and the machine has not yet started to make scheduling decisions for the workpieces.

[0086] In the embodiment of the present invention, the reward refers to the expectation of hitting other states at the next moment in a certain state.

[0087] The embodiment of the present invention models the disjunctive graph as a scheduling decision model according to a preset model framework, so that the disjunctive graph can be selected according to the decision rules of the scheduling decision model, thereby optimizing the disjunctive process of the disjunctive graph.

[0088] S3. Use a preset graph neural network model to perform feature extraction and feature embedding on each node in the initial disjunctive graph, and use a preset reinforcement learning algorithm to train the preset graph neural network model and optimize the initial decision model to obtain a target decision model.

[0089] In the embodiment of the present invention, the preset graph neural network model refers to a model that uses a neural network to process graph structure data. For example, a graph isomorphism network model can be used to embed the target disjunctive graph into the graph neural network model, wherein the graph isomorphism network model is a variant of the graph neural network model, and its main advantage is that it has strong fitting and generalization capabilities.

[0090] As an embodiment of the present invention, the use of a preset graph neural network model to extract and embed features for each node in the initial decision model includes:

[0091] Convert each node in the initial decision model into a three-dimensional original feature;

[0092] Using a graph isomorphism network model to convert the three-dimensional original features into feature vectors;

[0093] Map the feature vector to the graph neural network model.

[0094] Furthermore, the graph isomorphism network model is used to transform the three-dimensional original features into feature vectors, including:

[0095] Iteratively calculate the feature embedding of the original three-dimensional features and perform batch normalization to obtain normalized three-dimensional features;

[0096] The normalized three-dimensional features are pooled using a preset pooling function to obtain a feature vector.

[0097] In the embodiment of the present invention, the pooling function refers to an operation used to reduce the dimension of the feature graph, thereby obtaining a global representation of the entire graph. The pooling function can take the embedding of all nodes as input, which can reduce the spatial size of the feature graph, reduce the number of parameters in the network, and avoid the occurrence of overfitting. For example, the average pooling function can be used to perform pooling processing on the normalized three-dimensional features using a preset pooling function.

[0098] Furthermore, the feature embedding of the three-dimensional original features is iteratively calculated, including:

[0099] The feature embedding of the three-dimensional original feature is iteratively calculated using the following formula:

[0100]

[0101] Where k is the number of iterations, h v (k) is the feature embedding of node v iterated k times, is a multilayer perceptron with the parameter θ for the number of iterations k k , ε (k) is an arbitrary number that can be learned, N (v) is the neighborhood of v.

[0102] Reference Figure 2 As shown, the method of training the preset graph neural network model using a preset reinforcement learning algorithm includes:

[0103] Obtain the scheduling information of the graph neural network model, input the scheduling information into the new Actor network of the preset reinforcement learning algorithm, and construct a normal distribution of each execution action under the current state data;

[0104] Filtering an execution action from the normal distribution, and then inputting it into the graph neural network model to obtain a reward scalar value and next state data under the current state data;

[0105] Input the next state data into the Actor new network, and cyclically obtain the reward scalar value and the next state data until a preset number of scheduling information, reward scalar values, and the next execution action are obtained;

[0106] Inputting the next state data obtained in the last step of the loop into the Critic network of the preset reinforcement learning algorithm to obtain the state value function of the current state data;

[0107] Calculate the reward scalar value discount function according to the state value function until the entire scheduling process is completed, and update the Critic network according to the first loss function in the preset reinforcement learning algorithm by back propagation;

[0108] Inputting the obtained preset amount of scheduling information into the old Actor network and the new Actor network respectively to obtain a first normal distribution and a second normal distribution;

[0109] Inputting the obtained preset number of next execution actions into the first normal distribution and the second normal distribution respectively to obtain a first probability value and a second probability value;

[0110] Calculating a relative rate of change according to the first probability value and the second probability value;

[0111] According to the relative rate of change, updating the new Actor network using a second loss function;

[0112] The updating process of the Critic network and the Actor new network is iteratively trained k times to obtain the standard graph neural network model.

[0113] In the embodiment of the present invention, the Actor network and the Critic network refer to the Actor-Critic architecture used in the reinforcement learning algorithm, that is, the Actor network is used to output the strategy, and the Critic network is used to output the value function.

[0114] The embodiment of the present invention embeds features of the target disjunctive graph into a graph neural network model, and trains the graph neural network model with a preset reinforcement learning algorithm. After the training, an optimized scheduling model is obtained, which can realize feature embedding of the disjunctive graph into a graph neural network model, and can make decision optimization of the disjunctive graph through the graph neural network model, thereby realizing the scheduling decision of the flow shop problem. Among them, by training the graph neural network model with a preset reinforcement learning algorithm, the processing accuracy of the graph neural network model can be improved, and the scheduling accuracy of the flow shop problem can be further optimized.

[0115] S4. Utilize the standard graph neural network model to optimize the initial decision model to obtain a target decision model.

[0116] As an embodiment of the present invention, optimizing the initial decision model by using the standard graph neural network model includes: As an embodiment of the present invention, optimizing the initial decision model by using the standard graph neural network model includes:

[0117] Using the standard graph neural network model to predict the action value function of each action and the state value function of each state in the initial decision model;

[0118] The initial decision model is optimized using each action-value function and each state-value function.

[0119] The embodiment of the present invention can simplify the difficulty of flow shop problem data by converting historical flow shop problem data into disjunctive graphs and converting text features into image features, and can convert each node of the initial disjunctive graph into feature embedding of a graph neural network. The initial decision model can be optimized through the graph neural network, and then the decision optimization of the disjunctive graph can be made to realize the scheduling decision of the flow shop scheduling problem. The present invention can reduce the difficulty of realizing the non-permutation flow shop scheduling problem based on a neural network and improve practicality.

[0120] Reference Figure 3 , which is an overall algorithm flow chart of a neural network-based non-replacement flow shop scheduling method provided in one embodiment of the present invention.

[0121] like Figure 4 , which is a functional module diagram of a neural network-based non-replacement flow shop scheduling device provided by one embodiment of the present invention.

[0122] The non-permutation flow shop scheduling device 100 based on a neural network of the present invention can be installed in an electronic device. According to the functions implemented, the non-permutation flow shop scheduling device 100 based on a neural network can include a disjunctive graph conversion module 101 and a target decision model establishment module 102.

[0123] The module described in the present invention may also be referred to as a unit, which refers to a series of computer program segments that can be executed by a processor of an electronic device and can complete fixed functions, and is stored in a memory of the electronic device.

[0124] In this embodiment, the functions of each module / unit are as follows:

[0125] The disjunctive graph conversion module 101 is used to obtain historical flow shop scheduling problem data and convert the historical flow shop problem data into an initial disjunctive graph according to a preset disjunctive graph conversion relationship.

[0126] In the embodiment of the present invention, the historical flow shop problem data refers to scheduling data related to arranging multiple workpieces on a production line to be processed on multiple machines, including processing workpiece data and processing machine data.

[0127] The embodiment of the present invention can obtain historical flow shop scheduling problem data by using relevant crawler tools to capture large-scale examples of flow shop problem data on web pages.

[0128] In the embodiment of the present invention, the preset disjunctive graph conversion relationship refers to the corresponding relationship between the flow shop problem data and the disjunctive graph, for example, converting the workpieces into nodes in the disjunctive graph and converting the processing sequence into connecting lines in the disjunctive graph.

[0129] As an embodiment of the present invention, according to a preset disjunctive graph transformation relationship, the historical flow shop problem data is transformed into an initial disjunctive graph, including:

[0130] Extract the scheduling elements of historical flow shop problem data, and use the workpiece elements in the scheduling elements to construct disjunctive graph nodes, and assign different colors to the disjunctive graph nodes corresponding to different workpieces;

[0131] The nodes of the disjunctive graph are connected by using directed connecting lines of the first color to represent the processing order of the fixed workpieces in the scheduling element on each machine, thereby obtaining a first connecting graph;

[0132] Using directed connecting lines of a second color to connect workpieces of the same color processed by the same machine, a second connecting line diagram is obtained;

[0133] The first connection graph and the second connection graph are integrated to obtain a disjunction graph.

[0134] The embodiment of the present invention converts historical flow shop problem data into a disjunctive graph according to a preset disjunctive graph conversion relationship, can realize the conversion of text features into image features, and can simplify the difficulty of flow shop problem data.

[0135] The target decision model establishment module 102 is used to record each node in the initial disjunctive graph, and perform decision modeling based on each node using a preset model framework to obtain an initial decision model; use a preset graph neural network model to perform feature extraction and feature embedding on each node in the initial disjunctive graph, and use a preset reinforcement learning algorithm to train the preset graph neural network model to obtain a standard graph neural network model; use the standard graph neural network model to optimize the initial decision model to obtain a target decision model.

[0136] In an embodiment of the present invention, the preset model framework refers to a model framework that integrates the connection lines between each node in the disjunctive graph into a decision-making process. The Markov decision process model can be used. The Markov decision process model provides a framework for modeling and making decisions for random processes, which can help us solve complex scheduling problems. The Markov property means that the next state depends only on the current state and action, and is independent of the previous historical state.

[0137] As an embodiment of the present invention, the decision modeling is performed according to each node using a preset model framework to obtain an initial decision model, including:

[0138] Taking the state of the next arbitrary node of the initial disjunctive graph starting from the starting node as the initial state;

[0139] Taking the next decision action of the workpiece in the initial state as the current selectable action;

[0140] Calculate the difference between the maximum estimated completion time generated by the initial state and the maximum estimated completion time generated by the next state, and use the difference as a reward;

[0141] Utilizing the reward to optimize the currently selectable action to obtain an optimal scheduling state;

[0142] Taking the next schedule generated by the optimal scheduling state as an update action, and using the update action to update the initial disjunctive graph;

[0143] The above steps are repeated, and the current state of the updated initial disjunctive graph is used as the updated state, and the updated initial disjunctive graph is integrated to obtain an initial decision model.

[0144] In the embodiment of the present invention, the initial state refers to the first state of converting the flow shop problem data into a disjunctive graph, which is the initial state, and the machine has not yet started to make scheduling decisions for the workpieces.

[0145] In the embodiment of the present invention, the reward refers to the expectation of hitting other states at the next moment in a certain state.

[0146] The embodiment of the present invention can realize the selection of the disjunctive graph according to the decision rules of the scheduling decision model by modeling each node according to a preset model framework, thereby optimizing the decision process of the disjunctive graph.

[0147] In the embodiment of the present invention, the preset graph neural network model refers to a model that uses a neural network to process graph structure data. For example, a graph isomorphism network model can be used to embed the target disjunctive graph into the graph neural network model, wherein the graph isomorphism network model is a variant of the graph neural network model, and its main advantage is that it has strong fitting and generalization capabilities.

[0148] As an embodiment of the present invention, the use of a preset graph neural network model to extract and embed features for each node in the initial decision model includes:

[0149] Convert each node in the initial decision model into a three-dimensional original feature;

[0150] Using a graph isomorphism network model to convert the three-dimensional original features into feature vectors;

[0151] Map the feature vector to the graph neural network model.

[0152] Furthermore, the graph isomorphism network model is used to transform the three-dimensional original features into feature vectors, including:

[0153] Iteratively calculate the feature embedding of the original three-dimensional features and perform batch normalization to obtain normalized three-dimensional features;

[0154] The normalized three-dimensional features are pooled using a preset pooling function to obtain a feature vector.

[0155] In the embodiment of the present invention, the pooling function refers to an operation used to reduce the dimension of the feature graph, thereby obtaining a global representation of the entire graph. The pooling function can take the embedding of all nodes as input, which can reduce the spatial size of the feature graph, reduce the number of parameters in the network, and avoid the occurrence of overfitting. For example, the average pooling function can be used to perform pooling processing on the normalized three-dimensional features using a preset pooling function.

[0156] Furthermore, the feature embedding of the three-dimensional original features is iteratively calculated, including:

[0157] The feature embedding of the three-dimensional original feature is iteratively calculated using the following formula:

[0158]

[0159] Where k is the number of iterations, h v (k) is the feature embedding of node v iterated k times, is a multilayer perceptron with the parameter θ for the number of iterations k k , ε (k) is an arbitrary number that can be learned, N (v) is the neighborhood of v.

[0160] As an embodiment of the present invention, the step of training the preset graph neural network model using a preset reinforcement learning algorithm includes:

[0161] Obtain the scheduling information of the graph neural network model, input the scheduling information into the new Actor network of the preset reinforcement learning algorithm, and construct a normal distribution of each execution action under the current state data;

[0162] Filtering an execution action from the normal distribution, and then inputting it into the graph neural network model to obtain a reward scalar value and next state data under the current state data;

[0163] Input the next state data into the Actor new network, and cyclically obtain the reward scalar value and the next state data until a preset number of scheduling information, reward scalar values, and the next execution action are obtained;

[0164] Inputting the next state data obtained in the last step of the loop into the Critic network of the preset reinforcement learning algorithm to obtain the state value function of the current state data;

[0165] Calculate the reward scalar value discount function according to the state value function until the entire scheduling process is completed, and update the Critic network according to the first loss function in the preset reinforcement learning algorithm by back propagation;

[0166] Inputting the obtained preset amount of scheduling information into the old Actor network and the new Actor network respectively to obtain a first normal distribution and a second normal distribution;

[0167] Inputting the obtained preset number of next execution actions into the first normal distribution and the second normal distribution respectively to obtain a first probability value and a second probability value;

[0168] Calculating a relative rate of change according to the first probability value and the second probability value;

[0169] According to the relative rate of change, updating the new Actor network using a second loss function;

[0170] The updating process of the Critic network and the Actor new network is iteratively trained k times to obtain the standard graph neural network model.

[0171] In the embodiment of the present invention, the Actor network and the Critic network refer to the Actor-Critic architecture used in the reinforcement learning algorithm, that is, the Actor network is used to output the strategy, and the Critic network is used to output the value function.

[0172] The embodiment of the present invention embeds features of the target disjunctive graph into a graph neural network model, and trains the graph neural network model with a preset reinforcement learning algorithm. After the training, an optimized scheduling model is obtained, which can realize feature embedding of the disjunctive graph into a graph neural network model, and can make decision optimization of the disjunctive graph through the graph neural network model, thereby realizing the scheduling decision of the flow shop problem. Among them, by training the graph neural network model with a preset reinforcement learning algorithm, the processing accuracy of the graph neural network model can be improved, and the scheduling accuracy of the flow shop problem can be further optimized.

[0173] As an embodiment of the present invention, the optimizing the initial decision model by using the standard graph neural network model includes:

[0174] Using the standard graph neural network model to predict the action value function of each action and the state value function of each state in the initial decision model;

[0175] The initial decision model is optimized using each action-value function and each state-value function.

[0176] The electronic device may include a processor 10, a memory 11, a communication bus 12 and a communication interface 13, and may also include a computer program stored in the memory 11 and executable on the processor 10, such as a non-replacement flow shop scheduling method program based on a neural network.

[0177] In some embodiments, the processor 10 may be composed of an integrated circuit, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips. The processor 10 is the control core (ControlUnit) of the electronic device, and uses various interfaces and lines to connect various components of the entire electronic device, and executes or executes programs or modules stored in the memory 11 (for example, executing a non-replacement flow shop scheduling method program based on a neural network, etc.), and calls data stored in the memory 11 to execute various functions of the electronic device and process data.

[0178] The memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (for example: SD or DX memory, etc.), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of an electronic device, such as a mobile hard disk of the electronic device. In other embodiments, the memory 11 can also be an external storage device of an electronic device, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device. Further, the memory 11 can also include both an internal storage unit of the electronic device and an external storage device. The memory 11 can not only be used to store application software and various types of data installed in the electronic device, such as a code of a non-replacement flow shop scheduling method program based on a neural network, but can also be used to temporarily store data that has been output or is to be output.

[0179] The communication bus 12 may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The bus is configured to realize connection and communication between the memory 11 and at least one processor 10, etc.

[0180] The communication interface 13 is used for communication between the above-mentioned electronic device and other devices, including a network interface and a user interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device and other electronic devices. The user interface may be a display (Display), an input unit (such as a keyboard (Keyboard)), and optionally, the user interface may also be a standard wired interface, a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, and an OLED (Organic Light-Emitting Diode, organic light-emitting diode) touch device, etc. Among them, the display may also be appropriately referred to as a display screen or a display unit, which is used to display information processed in the electronic device and to display a visual user interface.

[0181] Figure 5 Only an electronic device with components is shown, and those skilled in the art will understand that Figure 5 The structure shown does not constitute a limitation on the electronic device, and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0182] For example, although not shown, the electronic device may also include a power source (such as a battery) for supplying power to each component. Preferably, the power source may be logically connected to the at least one processor 10 through a power management device, so that the power management device can realize functions such as charging management, discharging management, and power consumption management. The power source may also include one or more DC or AC power sources, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

[0183] It should be understood that the embodiment is for illustration only and the scope of the patent application is not limited to this structure.

[0184] The program of the non-replacement flow shop scheduling method based on neural network stored in the memory 11 of the electronic device is a combination of multiple instructions. When running in the processor 10, it can achieve:

[0185] Acquire historical flow shop scheduling problem data, and transform the historical flow shop problem data into an initial disjunctive graph according to a preset disjunctive graph transformation relationship;

[0186] Recording each node in the initial disjunctive graph, and performing decision modeling based on each node using a preset model framework to obtain an initial decision model;

[0187] Using a preset graph neural network model to perform feature extraction and feature embedding on each node in the initial disjunctive graph, and using a preset reinforcement learning algorithm to train the preset graph neural network model to obtain a standard graph neural network model;

[0188] The standard graph neural network model is used to optimize the initial decision model to obtain a target decision model.

[0189] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to the description of the relevant steps in the corresponding embodiment of the accompanying drawings, which will not be repeated here.

[0190] Furthermore, if the module / unit integrated in the electronic device 1 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM).

[0191] The present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor of an electronic device, the computer program can implement:

[0192] Acquire historical flow shop scheduling problem data, and transform the historical flow shop problem data into an initial disjunctive graph according to a preset disjunctive graph transformation relationship;

[0193] Recording each node in the initial disjunctive graph, and performing decision modeling based on each node using a preset model framework to obtain an initial decision model;

[0194] Using a preset graph neural network model to perform feature extraction and feature embedding on each node in the initial disjunctive graph, and using a preset reinforcement learning algorithm to train the preset graph neural network model to obtain a standard graph neural network model;

[0195] The standard graph neural network model is used to optimize the initial decision model to obtain a target decision model.

[0196] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0197] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0198] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0199] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0200] Therefore, no matter from which point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the appended claims rather than the above description, so it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any attached figure mark in the claims should not be regarded as limiting the claims involved.

[0201] The blockchain referred to in this invention is a new application mode of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0202] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0203] In addition, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the system claim can also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any particular order.

[0204] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A non-replacement flow shop scheduling method based on neural network, characterized in that: The method comprises: Acquire historical flow shop scheduling problem data, and transform the historical flow shop problem data into an initial disjunctive graph according to a preset disjunctive graph transformation relationship; Recording each node in the initial disjunctive graph, and performing decision modeling based on each node using a preset model framework to obtain an initial decision model; A preset graph neural network model is used to perform feature extraction and feature embedding on each node in the initial disjunctive graph, and a preset reinforcement learning algorithm is used to train the preset graph neural network model to obtain a standard graph neural network model, wherein the method of training the preset graph neural network model using a preset reinforcement learning algorithm includes: obtaining scheduling information of the graph neural network model, and inputting the scheduling information into the Actor new network of the preset reinforcement learning algorithm, and constructing a normal distribution of each execution action under the current state data; selecting an execution action from the normal distribution, and then inputting it into the graph neural network model to obtain a reward scalar value and next state data under the current state data; inputting the next state data into the Actor new network, and cyclically obtaining the reward scalar value and the next state data until a preset number of scheduling information, reward scalar values ​​and the next execution action are obtained; the next state number obtained in the last step of the loop is obtained. According to the input into the Critic network of the preset reinforcement learning algorithm, the state value function of the current state data is obtained; the reward scalar value discount function is calculated according to the state value function until the entire scheduling process is completed, and the Critic network is updated by back propagation according to the first loss function in the preset reinforcement learning algorithm; the obtained preset number of scheduling information is respectively input into the Actor old network and the Actor new network to obtain the first normal distribution and the second normal distribution; the obtained preset number of next execution actions are respectively input into the first normal distribution and the second normal distribution to obtain the first probability value and the second probability value; the relative change rate is calculated according to the first probability value and the second probability value; according to the relative change rate, the Actor new network is updated using the second loss function; the update process of the Critic network and the Actor new network is iteratively trained k times to obtain the standard graph neural network model; The standard graph neural network model is used to optimize the initial decision model to obtain a target decision model, wherein the method of optimizing the initial decision model using the standard graph neural network model includes: using the standard graph neural network model to predict the action value function of each action and the state value function of each state in the initial decision model; and optimizing the initial decision model using each action value function and each state value function.

2. The non-replacement flow shop scheduling method based on neural network according to claim 1, characterized in that: The step of converting the historical flow shop problem data into a disjunctive graph according to a preset disjunctive graph conversion relationship includes: Extracting the scheduling elements of the historical flow shop problem data, and constructing disjunctive graph nodes using the workpiece elements in the scheduling elements, and assigning different colors to the disjunctive graph nodes corresponding to different workpieces; Connecting the nodes of the disjunctive graph using directed connecting lines of a first color to represent the sequential processing order of the fixed workpieces in the scheduling element on each machine, thereby obtaining a first connecting graph; Using directed connecting lines of a second color to connect workpieces of the same color processed by the same machine, a second connecting line diagram is obtained; The first connection graph and the second connection graph are integrated to obtain the initial disjunction graph.

3. The non-replacement flow shop scheduling method based on neural network as claimed in claim 1, characterized in that: The decision modeling is performed according to each node using a preset model framework to obtain an initial decision model, including: Taking the state of the next arbitrary node of the initial disjunctive graph starting from the starting node as the initial state; Taking the next decision action of the workpiece in the initial state as the current selectable action; Calculate the difference between the maximum estimated completion time generated by the initial state and the maximum estimated completion time generated by the next state, and use the difference as a reward; Utilizing the reward to optimize the currently selectable action to obtain an optimal scheduling state; Taking the next schedule generated by the optimal scheduling state as an update action, and using the update action to update the initial disjunctive graph; The above steps are repeated, and the current state of the updated initial disjunctive graph is used as the updated state, and the updated initial disjunctive graph is integrated to obtain an initial decision model.

4. The non-replacement flow shop scheduling method based on neural network as claimed in claim 1, characterized in that: The using of a preset graph neural network model to extract and embed features for each node in the initial decision model includes: Convert each node in the initial decision model into a three-dimensional original feature; Using a graph isomorphism network model to convert the three-dimensional original features into feature vectors; Map the feature vector to the graph neural network model.

5. The non-replacement flow shop scheduling method based on neural network as claimed in claim 4, characterized in that: The converting the three-dimensional original features into feature vectors by using a graph isomorphic network model includes: Iteratively calculating feature embedding of the three-dimensional original features, and performing batch normalization to obtain normalized three-dimensional features; The normalized three-dimensional features are pooled using a preset pooling function to obtain the feature vector.

6. The non-replacement flow shop scheduling method based on neural network as claimed in claim 5, characterized in that: The iterative calculation of the feature embedding of the three-dimensional original feature includes: The feature embedding of the three-dimensional original feature is iteratively calculated using the following formula: Where k is the number of iterations, h v (k) is the feature embedding of node v iterated k times, is a multilayer perceptron with the parameter θ for the number of iterations k k , ε (k) is an arbitrary number that can be learned, N (v) is the neighborhood of v.

7. A non-replacement flow shop scheduling device based on neural network, characterized in that: The device can implement the non-replacement flow shop scheduling method based on neural network as claimed in any one of claims 1 to 6, and the device comprises: A disjunctive graph conversion module, used to obtain historical flow shop scheduling problem data, and convert the historical flow shop problem data into an initial disjunctive graph according to a preset disjunctive graph conversion relationship; A target decision model establishment module is used to record each node in the initial disjunctive graph, and perform decision modeling based on each node using a preset model framework to obtain an initial decision model; use a preset graph neural network model to perform feature extraction and feature embedding on each node in the initial disjunctive graph, and use a preset reinforcement learning algorithm to train the preset graph neural network model to obtain a standard graph neural network model; use the standard graph neural network model to optimize the initial decision model to obtain a target decision model, wherein the use of the standard graph neural network model to optimize the initial decision model includes: using the standard graph neural network model to predict the action value function of each action in the initial decision model and the state value function of each state; and using each action value function and each state value function to optimize the initial decision model.

8. An electronic device, characterized in that: The electronic device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the neural network-based non-replacement flow shop scheduling method as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the neural network-based non-replacement flow shop scheduling method as described in any one of claims 1 to 6 is implemented.

Images