Workshop robust scheduling method and system considering global disturbance, medium and equipment

Abstract

The invention provides a workshop robust scheduling method and system considering global disturbance, a medium and equipment, and the method comprises the steps: constructing a robustness evaluation model under the global disturbance, and carrying out the robust performance evaluation of a scheduling scheme under the influence of the global disturbance; establishing a global scheduling model considering a robustness index, determining constraint conditions in a scheduling problem, and constructing optimization targets of the robustness index and the maximum completion time efficiency index of the process start deviation time and the order delivery deviation time; and designing a master-apprentice evolutionary algorithm based on an agent model, and generating a global scheduling scheme. Aiming at typical discrete manufacturing characteristics of multiple varieties and variable / small batches of aerospace products, global disturbances such as emergency order insertion, process change and process deviation are considered, the robustness of a scheduling scheme is evaluated by establishing a robust evaluation model, and multi-target comprehensive optimization of the global disturbances is realized based on a master-apprentice evolutionary algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of workshop scheduling, and in particular to a workshop robust scheduling method, system, medium and equipment considering global disturbance. Background Art

[0002] The demand for a new generation of aerospace equipment, such as launch vehicles and hypersonic vehicles, has surged. This increase in demand is accompanied by the characteristics of multiple varieties, variable batches, and short cycles, which puts higher requirements on the development and production of aerospace products. Faced with the new characteristics of multi-model co-production and short-cycle high-intensity delivery, the task arrangement of aerospace manufacturing workshops has become more complicated, and in the production process of the workshop, it is inevitable to encounter the influence of global disturbance factors such as emergency orders, process changes, and process deviations. How to comprehensively consider various uncertainties in the production process to formulate a reasonable scheduling plan for the workshop, reduce the impact of disturbance factors, and improve the stability of workshop scheduling is a key issue of research.

[0003] For the production scheduling problem of aerospace manufacturing workshops with complex constraints and multi-type uncertainty influences, meta-heuristic intelligent optimization algorithms can better meet the needs of workshops in terms of uncertainty processing and solution efficiency. However, the fitness function of current intelligent optimization algorithms usually has a clear analytical expression. In view of the needs of robust scheduling optimization, how to establish a proxy evaluation mechanism of the fitness function from the perspective of data analysis and integrate it with the traditional intelligent optimization algorithm is the core and key of the research.

[0004] Patent application document CN115249121A discloses a robust scheduling optimization method for discrete manufacturing workshops based on deep reinforcement learning, including: taking historical processing data, using neural networks to fit the function of the process processing time on equipment, operators and the initial time of production; establishing a processing environment model of the factory workshop, the processing environment model includes the number of operable equipment, on-the-job operators and the inventory of intermediate products; obtaining the number of products that need to be processed on the day; constructing a discrete manufacturing workshop robust scheduling problem based on the processing environment model and the number of products that need to be processed on the day, the objective function of the discrete manufacturing workshop robust scheduling problem is to minimize the maximum completion time and minimize the difference between the completion time and the delivery time; solving the discrete manufacturing workshop robust scheduling problem. However, this patent cannot completely solve the current technical problems, nor can it meet the needs of the present invention. Summary of the invention

[0005] In view of the defects in the prior art, the object of the present invention is to provide a workshop robust scheduling method, system, medium and equipment taking into account global disturbances.

[0006] The robust workshop scheduling method considering global disturbance provided by the present invention includes:

[0007] Step 1: Evaluate the robust performance of the scheduling scheme under the influence of global disturbances;

[0008] Step 2: Establish a global scheduling model that takes robustness into account, construct constraints in the scheduling problem, and determine the optimization objectives for the robustness indicators of process start deviation time, order delivery deviation time, and maximum completion time efficiency indicator;

[0009] Step 3: Generate a global scheduling solution based on the master-apprentice evolutionary algorithm of the agent model.

[0010] Preferably, the step 1 comprises:

[0011] Construct a disjunctive graph model to describe the scheduling plan from the dimensions of process route and equipment competition relationship;

[0012] The encoder based on the graph structure describes the relationship between the process sequence and equipment competition in the scheduling plan;

[0013] Based on the decoder of the timing transfer relationship, the effects of disturbances along the two dimensions of process execution sequence and equipment load accumulation are analyzed, and two robustness indicators, process start time deviation and product delivery time deviation, are calculated.

[0014] Preferably, step 2 comprises:

[0015] Establish a global scheduling model that comprehensively considers robustness indicators and efficiency indicators, describe the process tasks, process sequence, process time constraints under the influence of disturbances, and the allocation of process tasks to the selected equipment and the production sequence scheduling of process tasks on the equipment;

[0016] The global scheduling model includes the following assumptions and objective functions:

[0017] Part manufacturing follows the predecessor-successor process relationship, and each operation has multiple predecessor operations and at most one successor operation; each operation is produced on only one machine in the optional equipment set; each machine performs at most one operation at a time; after the machine starts to execute a process operation, it will not be interrupted before the execution is completed, and other operations cannot be temporarily inserted; at the beginning of the scheduling cycle, all parts and machines are available;

[0018] The objective function is: Minimize Obj = λ1makespan + λ2SD + λ3DD, where λ1, λ2, λ3 are weight coefficients, makespan is the maximum completion time describing the efficiency index, SD and DD are the process start time deviation and delivery time deviation describing the robustness index.

[0019] Preferably, the step 3 comprises:

[0020] The master-apprentice evolutionary algorithm based on the agent mechanism updates the scheduling scheme based on the randomly generated two scheduling schemes by using the taboo search TS operator for improving the solution performance and the path reconnection PR operator for updating the offspring;

[0021] For the two randomly generated scheduling solutions, two child solutions are generated respectively by executing the path reconnection operator, and taboo search is performed on the two child solutions to generate two new scheduling solutions. Then, the maximum completion time makespan, process start time deviation SD and delivery time deviation DD corresponding to the new scheduling solutions are calculated respectively by using the robustness evaluation method.

[0022] The robust workshop scheduling system considering global disturbance provided by the present invention includes:

[0023] Module M1: Robust performance evaluation of scheduling schemes under the influence of global disturbances;

[0024] Module M2: Establish a global scheduling model that takes robustness indicators into consideration, construct constraints in the scheduling problem, and determine the optimization objectives for the robustness indicators of process start deviation time, order delivery deviation time, and maximum completion time efficiency indicators;

[0025] Module M3: A master-apprentice evolutionary algorithm based on an agent model to generate a global scheduling solution.

[0026] Preferably, the module M1 comprises:

[0027] Construct a disjunctive graph model to describe the scheduling plan from the dimensions of process route and equipment competition relationship;

[0028] The encoder based on the graph structure describes the relationship between the process sequence and equipment competition in the scheduling plan;

[0029] Based on the decoder of the timing transfer relationship, the effects of disturbances along the two dimensions of process execution sequence and equipment load accumulation are analyzed, and two robustness indicators, process start time deviation and product delivery time deviation, are calculated.

[0030] Preferably, the module M2 comprises:

[0031] Establish a global scheduling model that comprehensively considers robustness indicators and efficiency indicators, describe the process tasks, process sequence, process time constraints under the influence of disturbances, and the allocation of process tasks to the selected equipment and the production sequence scheduling of process tasks on the equipment;

[0032] The global scheduling model includes the following assumptions and objective functions:

[0033] Part manufacturing follows the predecessor-successor process relationship, and each operation has multiple predecessor operations and at most one successor operation; each operation is produced on only one machine in the optional equipment set; each machine performs at most one operation at a time; after the machine starts to execute a process operation, it will not be interrupted before the execution is completed, and other operations cannot be temporarily inserted; at the beginning of the scheduling cycle, all parts and machines are available;

[0034] The objective function is: Minimize Obj = λ1makespan + λ2SD + λ3DD, where λ1, λ2, λ3 are weight coefficients, makespan is the maximum completion time describing the efficiency index, SD and DD are the process start time deviation and delivery time deviation describing the robustness index.

[0035] Preferably, the module M3 comprises:

[0036] The master-apprentice evolutionary algorithm based on the agent mechanism updates the scheduling scheme based on the randomly generated two scheduling schemes by using the taboo search TS operator for improving the solution performance and the path reconnection PR operator for updating the offspring;

[0037] For the two randomly generated scheduling solutions, two child solutions are generated respectively by executing the path reconnection operator, and taboo search is performed on the two child solutions to generate two new scheduling solutions. Then, the maximum completion time makespan, process start time deviation SD and delivery time deviation DD corresponding to the new scheduling solutions are calculated respectively by using the robustness evaluation method.

[0038] According to the computer-readable storage medium storing a computer program provided by the present invention, when the computer program is executed by a processor, the steps of the method for robust workshop scheduling considering global disturbances are implemented.

[0039] The electronic device provided according to the present invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the steps of the method for robust workshop scheduling considering global disturbances are implemented when the computer program is executed by the processor.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] (1) This paper aims at the global scheduling problem of aerospace manufacturing workshops, considers global disturbance events such as emergency order insertion, process change and process deviation, and proposes an optimization goal that considers completion time and robustness, including: minimizing the maximum completion time, process start time deviation, and delivery time deviation. It transforms the traditional single-objective optimization into multi-objective optimization, which is closer to actual production and has practical significance for solving similar problems in enterprises.

[0042] (2) The present invention designs an extended disjunctive graph representation method that describes the structural characteristics of the scheduling plan, and constructs a conditional graph convolutional neural network encoding unit that considers multiple types of correlations between processes based on the process knowledge in the extended disjunctive graph, and a recurrent neural network decoding unit that considers the influence of the timing of the previous and next processes. The transmission relationship of uncertain disturbances along the two dimensions of process and equipment is accurately expressed, and the robustness evaluation of the workshop scheduling plan driven by the graph network data model is realized. This method can effectively evaluate the impact of global disturbance events such as emergency insertion, process change and process deviation on the scheduling plan;

[0043] (3) The present invention proposes an improved master-apprentice evolutionary algorithm that combines a recombination operator with a taboo search operator. By introducing the recombination operator and the taboo search operator, the excellent individual characteristics can be better retained during the search process and the local optimal solution can be effectively avoided, thereby improving the search efficiency and the quality of the solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0045] Figure 1 A flowchart of a workshop robust scheduling method considering global disturbances according to the present invention;

[0046] Figure 2 It is a flow chart of the scheduling scheme robustness evaluation method of the present invention;

[0047] Figure 3 It is a bar chart of comparative experimental results of the global scheduling method in the embodiment of the present invention. DETAILED DESCRIPTION

[0048] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0049] Example 1

[0050] The technical problem to be solved by the present invention is: to provide a workshop robust scheduling method taking global disturbance into consideration, on the one hand, to realize the robustness evaluation of the scheduling scheme under global disturbance; on the other hand, to realize the robust scheduling optimization of the workshop production system under uncertain environment based on the master-apprentice evolutionary algorithm, and to generate a global scheduling scheme with multi-objective comprehensive optimization.

[0051] The technical solution adopted by the method of the present invention is: a robust workshop scheduling method considering global disturbances, such as Figure 1 , including the following steps:

[0052] Step 1: Design a scheduling scheme robustness evaluation method under global disturbances, and evaluate the robustness performance of the scheduling scheme under the influence of global disturbances;

[0053] The scheduling scheme robustness evaluation method under the global disturbance specifically includes the following sub-steps:

[0054] Step 1.1: Construct a disjunctive graph model, which consists of a node set and a disjunctive arc set, wherein the node set represents all product processes involved in production scheduling in an aerospace manufacturing workshop; the process route is represented by a solid line connecting arc between process nodes, and the equipment competition relationship between processes is represented by two reverse dotted disjunctive arcs;

[0055] Step 1.2: Based on the disjunctive graph model, a graph encoder based on the Conditional Graph Convolutional Neural Network (R-GCN) is constructed to describe the relationships between process sequences and equipment competition in the scheduling scheme. First, the adjacency matrix and degree matrix corresponding to the disjunctive graph are constructed to reflect the correlation between process nodes and the number of connecting edges of each process node. On this basis, the three adjacent information of forward process relationship, backward process relationship and equipment sharing relationship in the disjunctive graph are aggregated through subgraph sampling operation and adjacency aggregation operation. The node embedding function is trained using data samples, and finally the feature vector corresponding to each process node is extracted;

[0056] Step 1.3: Based on the use of conditional graph convolutional neural networks to describe the process and equipment constraints between process nodes, a recurrent neural network (RNN) based on a recursive structure is used to describe the layer-by-layer transmission relationship of the process node states in the encoder, so as to realize the information association, expression and transmission between different processes under the influence of the two-dimensional relationship of process / equipment, analyze the effects of disturbances along the two dimensions of process execution sequence and equipment load accumulation, and calculate the two robustness indicators of process start time deviation and product delivery time deviation.

[0057] Step 2: Establish a global scheduling model that takes into account robustness indicators, including building constraints and objective functions. The parameters and decision variables in the model are defined as shown in the table:

[0058] Table 1 Parameters and decision variables of the robust scheduling process

[0059]

[0060]

[0061] According to the parameter definition, the global scheduling mathematical model is specifically expressed as follows:

[0062] a. Objective function

[0063] MinimizeObj=λ1makespan+λ2SD+λ3DD (1)

[0064] b. Constraints

[0065]

[0066] In formula (1), λ1, λ2, λ3 are weight coefficients, makespan is the maximum completion time describing the efficiency index, PD and CD are the process start time deviation and delivery time deviation describing the robustness index; formula (2) is the equipment uniqueness constraint, which means that each process can and can only be assigned to one device in the optional equipment set; formula (3) is the equipment occupancy constraint, which means that any machine can only execute one process at any time; formula (4) is the event point usage constraint, which means that on each device, the next event point can be used only after the previous event point is allocated; formulas (5)-(7) are production sequence constraints, formula (5) means that a new process can only be started after the previous process on the same machine is completed, and formula (6) means that once a process starts, it cannot be interrupted. The number of processes is affected by disturbances such as urgent insertion. Formula (7) indicates that between processes with assembly constraints, the subsequent process can only start after the previous process is completed, and the order of processes is affected by disturbances such as process changes. Formulas (8)-(9) are used to establish the corresponding relationship between the completion time of each process and the event point of each equipment, and the completion time of the process is affected by disturbances such as process deviation. Formula (10) is the calculation method of the maximum completion time, that is, the maximum completion time is equal to the maximum value of the completion time of the last task on each device. Formula (11) is the calculation method of the process start time deviation and the delivery time deviation. Based on the recurrent neural network decoder in step 1.3, two robustness indicators, namely the process start time deviation SD and the delivery time deviation DD, are output.

[0067] Step 3: Generate a global scheduling plan based on the master-apprentice evolutionary algorithm of the agent mechanism. On the basis of constructing the global scheduling model, a master-apprentice evolutionary algorithm as shown in Table 2 is proposed to generate a scheduling plan that comprehensively minimizes the maximum completion time, process start time deviation, and delivery time deviation.

[0068] Table 2 The process of the master-apprentice evolutionary algorithm for the global scheduling model

[0069]

[0070]

[0071] As shown in Table 2, the MAE algorithm updates the scheduling scheme based on the random generation of two scheduling schemes, using the Tabu Search (TS) operator to improve the solution performance and the Path Relinking (PR) operator to update the offspring, thereby improving the population diversity in the evolutionary process and achieving effective search of the scheduling solution space. Specifically, for the two randomly generated scheduling solutions Sol1 and Sol2, two offspring solutions Sol′1 and Sol′2 are generated respectively by executing the path relinking operator, and two new scheduling solutions Sol1 and Sol2 are generated by performing taboo search on the two offspring solutions respectively. The maximum completion time makespan, process start time deviation SD and delivery time deviation DD corresponding to the new scheduling solution are calculated using the robustness evaluation method, and Sol1, Sol2 and the optimal solution of the current cycle are obtained according to the step 2 formula (1). The corresponding objective function value is selected, and the solution that minimizes the objective function value is selected as the new optimal solution for the current cycle And if The corresponding objective function value is less than the global optimal solution Sol * , then further Update to Sol * , thus completing the cnth learning iteration of the gth generation master-apprentice combination (i.e., the two scheduling solutions Sol1 and Sol2). In addition, in order to maintain the diversity of the two solutions Sol1 and Sol2, the distance metric function d(Sol1, Sol2) is introduced to judge the similarity between the two scheduling solutions. When the distance does not exceed a given constant MinD, the scheduling solution Sol2 will be reinitialized. In the MAE algorithm, the maximum number of generations of the master-apprentice cycle and the maximum number of iterations of the master-apprentice learning are constrained by given constants Maxgen and MaxcycleN, forming a total of g≤Maxgen generations of master-apprentice combination cycles. Each generation of master-apprentice combinations performs cn≤MaxcycleN mutual learning iterations. According to the optimal solution Sol generated in the above process, * , obtain the scheduling scheme that minimizes the objective function. The scheduling decoding method, reorganization operator and taboo search algorithm involved in the algorithm are introduced in detail below:

[0072] (1) Encoding method

[0073] The scheduling plan for aerospace manufacturing workshops needs to complete the allocation of process tasks to the selected equipment and the sorting of process tasks on the equipment. According to the variable definition of Table 1 in step 2, the design consists of two strings of the same length TD = {td i ∈K i |i=1,2,...,n} and TA={ta i∈I|i=1,2,...,n}, where each position of the TD field represents a process task, and the value of each position td i Represents the selected equipment number, which means that the i-th process task is assigned to the td-th process task. i Each position in the TA field represents the priority level of the process task on the same device. The value of each position is ta. i Represents the process task number, which means that if production is carried out on the same equipment, the (i-1) The first process task is in ta i Production is carried out between the processes to obtain the production sequence of the processes on each device.

[0074] According to the encoding method, a TD is randomly generated at each position i in the TD field. i ∈K i , and randomly assigning elements in {1,2,...,n} to each position of the TA field, a random scheduling solution can be obtained, that is, the Initial() operation in Table 2 is completed. The distance metric between the two scheduling solutions Sol1 and Sol2 is calculated using the following formula:

[0075]

[0076] Among them: the scheduling solution Sol1 is encoded and Describe, schedule the solution Sol2 to encode and Give a description.

[0077] (2) Reorganization Operator

[0078] The reorganization operator NS1(Sol1,Sol2) uses the reconstruction and update of the solution to make the original scheduling solution Sol1 learn from the scheduling solution Sol2, thereby generating a new scheduling solution Sol′1. To achieve the above goals, the encoding string TD of the scheduling solution Sol1 will be rewritten in the reorganization operator. 1 and TA 1 Perform three operations:

[0079] Equipment reselection: Randomly select two positions on the TD field, and reselect machines from their optional equipment sets for the process tasks at these two positions;

[0080] Task exchange: Randomly select two positions on the TA field and exchange the process tasks at these two positions;

[0081] Task insertion: Randomly select a position on the TA field and insert the process task at that position into any position after the next process task of the same equipment.

[0082] A new temporary scheduling solution is created after each operation is completed. According to formula (12), Distance to learning object Sol2 If satisfied Then replace Sol1 with Otherwise, try the next operation. After trying all three operations, obtain the new scheduling solution Sol′1 based on the last scheduling solution Sol1, that is, Sol′1←NS1(Sol1,Sol2).

[0083] (3) Taboo Search Operator

[0084] The taboo search algorithm further improves the search capability of the MAE algorithm through local search on the basis of the scheduling solution reconstruction update by the reorganization operator, that is, the encoding string TD′ of the new scheduling solution Sol′1 produced 1 and TA′ 1 , find a better solution by taking the following actions:

[0085] Equipment load reduction: For the process tasks that determine the makespan, analyze the equipment where they are located, randomly select one of the process tasks assigned to the equipment, put the current equipment on the taboo list, and reallocate other equipment to be selected;

[0086] Task filling: For all process tasks on the same equipment, for any two process tasks in a sequential relationship, if the difference between the start times of the two is greater than the production time of a process task after the two, then the process task will be inserted between the two process tasks without violating the process route constraints, and the adjacency relationship between the two processes will be added to the taboo list;

[0087] A new temporary scheduling solution is created after each operation is completed. According to formula (1), The corresponding objective function value, if satisfied Then replace Sol′1 with Otherwise, try the next operation. After trying both operations, obtain a new scheduling solution Sol1 based on the last scheduling solution Sol′1, that is, Sol1←NS2(Sol′1).

[0088] Based on the encoding method and iterative operator, the MAE algorithm uses the algorithm flow shown in Table 2 to continuously reconstruct and update the scheduling solution, search the solution space of the global scheduling problem with a diversified population, and finally obtain a global scheduling solution that comprehensively considers robustness indicators and efficiency indicators.

[0089] Example 2

[0090] Aiming at the global disturbances of urgent order insertion, process change and process deviation in the production process of aerospace discrete manufacturing workshop, the present invention constructs a global scheduling model with the robustness index of process start deviation time, order delivery deviation time and maximum completion time efficiency index as optimization targets, designs a master-apprentice evolutionary algorithm based on agent mechanism, and proposes a workshop robust scheduling method considering global disturbance, which includes the following steps:

[0091] Step 1: Design a scheduling scheme robustness evaluation method under global disturbances, and evaluate the robustness performance of the scheduling scheme under the influence of global disturbances;

[0092] First, construct a disjunctive graph model, such as Figure 2 As shown, the disjunctive graph G = (P, RA, RB) of this embodiment is composed of a node set P and disjunctive arc sets RA and RB. The node set P = {p1, p2, p3, ..., p n} represents all n product processes involved in production scheduling in aerospace manufacturing workshop; RA = {ra 01 ,ra 03 ,ra 05 ,ra 12 ,…} The process route is represented by solid arcs connecting process nodes, such as ra 12 The corresponding p1→p2 indicates that process p1 is the immediate predecessor of process p2, and 12 The value on represents the production time of the preceding process p1; RB = {rb 13 ,rb 25 ,rb 28 ,…} Two reverse dashed lines are used to extract arcs to represent the equipment competition relationship between processes, such as rb 13 Indicates that processes p1 and p3 need to be produced on the same equipment, rb 13 =0 means that process p1 is produced before process p3 on the equipment, rb 13=1 means that process p1 is produced later than process p3 on the equipment; U and V represent the virtual start process and end process respectively, and the corresponding connection arc value is 0. In this embodiment, the scheduling plan needs to start from U. First, the node set P and the solid connection arc set RA in the disjunctive graph are generated according to the process route, and the elements in RA are assigned according to the production time. Then, according to the allocation of process tasks to the selected equipment, a bidirectional dotted connection arc set RB is generated, and then each element in RB is assigned according to the sorting of process tasks on each piece of equipment; finally, the nodes are connected through the transfer relationship based on the connection arc to realize the scheduling execution process to the end process V, forming a description of the scheduling plan.

[0093] Secondly, a graph-based encoder is proposed to describe the relationships between process order, equipment competition, etc. in the scheduling scheme;

[0094] This embodiment is based on the disjunctive graph model and designs a graph encoder based on the Conditional Graph Convolutional Neural Network (R-GCN), and the output is a low-dimensional vector representation that describes the scheduling scheme. Figure 2 As shown, firstly, the adjacency matrix and degree matrix corresponding to the graph G are constructed to reflect the correlation between the process nodes and the number of connecting edges of each process node, and are reflected in the conditional graph convolutional neural network in the form of the following formula:

[0095]

[0096] Where Matrix R It represents the adjacency relationship matrix between process nodes in the disjunctive graph G. The matrix dimension is affected by the number of elements in the process node set P. The matrix value is affected by the disjunctive arc sets RA and RB. 1 is the number of layers of the conditional graph convolutional neural network. Represents the process node p in the (l+1)th layer i The feature vector of , δ is a nonlinear activation function, Represents process node p i The adjacent process node, W (l) Represents the parameters of the neural network to be trained. On this basis, through subgraph sampling operations and adjacency aggregation operations, the three adjacent information of forward process relationship, backward process relationship and equipment sharing relationship in the disjunction graph are aggregated, and the node embedding function of the following formula is trained using data samples:

[0097]

[0098] in Represents the process node p i The set of first-order adjacent process nodes with relation r, represents the weight of the relationship r, They represent the three adjacency relationships of predecessor, successor and shared equipment respectively. When l=0, input multi-dimensional information such as process type, start time, production time, preparation time, completion time, equipment number, etc. According to the above formula, the node embedding function of process route and equipment conflict in the disjunctive graph is reflected. The relationship aggregation of process nodes is realized by layer-by-layer calculation, and finally each process node p is extracted. i The corresponding eigenvector Z i ,Right now

[0099] Then, based on the decoder of the timing transfer relationship, the effects of disturbance along the two dimensions of process execution order and equipment load accumulation are analyzed, and two robustness indicators, process start time deviation and product delivery time deviation, are calculated;

[0100] Based on the description of process and equipment constraints between process nodes by conditional graph convolutional neural network, this embodiment further analyzes the cumulative transmission effect of process time along the process adjacency relationship under the influence of disturbance to obtain the final robustness index of the scheduling scheme. Based on the recurrent neural network (RNN) containing a recursive structure, the process node status in the encoder is described. The layer-by-layer transmission relationship realizes the information association, expression and transmission between different processes under the influence of the two-dimensional relationship of process / equipment. Figure 2 As shown in the figure, S0 represents the feature vector under the theoretical state, that is, the theoretical start time of each process node, to reflect the overall information of the whole picture without interference. Each hidden state S (i+1) The hidden state S affected by the previous process disturbance i It is determined together with the characteristic vectors Z1, Z2, Z3, ... that reflect the adjacency relationship between process nodes, as shown in the following formula:

[0101]

[0102] where δ(·) is the hyperbolic tangent tanh activation function, Represents the process node p i The set of first-order adjacent process nodes with relation r, Represents three types of adjacency relationships: preceding, succeeding, and shared equipment. ij and b i is the network parameter of the recurrent neural network, c i is the bias constant. The following formula is used to calculate the process start time deviation:

[0103] y i =δ(S i +d) (16)

[0104] Where δ(·) is the sigmoid activation function, d represents the bias constant, and according to y1,y1,…y n By calculating the mean and maximum value of , we can obtain robustness indicators such as process start time deviation and delivery time deviation in the execution of the scheduling plan. Based on the above decoding architecture, on the basis of network parameter initialization, the mean square error (MSE) between the process start time deviation value and the actual sample deviation value is calculated by the decoder as the training loss function of the recurrent neural network to achieve the network parameter a ij , b i 、c i Etc., and finally achieve the robustness evaluation index of the scheduling scheme.

[0105] Step 2: Establish a global scheduling model that takes into account robustness indicators, including constructing constraints and objective functions;

[0106] The parameters and decision variables in this embodiment are defined as shown in the following table.

[0107] Table 3 Parameters and decision variables of the robust scheduling process

[0108]

[0109]

[0110] According to the parameter definition, the global scheduling mathematical model is specifically expressed as follows:

[0111] a. Objective function

[0112] Minimize Obj=λ1makespan+λ2SD+λ3DD (17)

[0113] b. Constraints

[0114]

[0115] In formula (17), λ1,λ2,λ3 are weight coefficients, makespan is the maximum completion time describing the efficiency index, PD and CD are the process start time deviation and delivery time deviation describing the robustness index; formula (18) is the equipment uniqueness constraint, which means that each process can and can only be assigned to one device in the optional equipment set; formula (19) is the equipment occupancy constraint, which means that any machine can only execute one process at any time; formula (20) is the event point usage constraint, which means that on each device, the next event point can be used only after the previous event point is allocated; formulas (21)-(23) are production sequence constraints. Formula (21) means that a new process can only be started after the previous process on the same machine is completed, and formula (22) means that once a process starts, It cannot be interrupted, and the number of processes is affected by disturbances such as emergency insertion. Formula (23) indicates that between processes with assembly constraints, the subsequent process can only start after the previous process is completed, and the order of processes is affected by disturbances such as process changes; Formulas (24)-(25) are used to establish the corresponding relationship between the completion time of each process and the event point of each equipment, and the process completion time is affected by disturbances such as process deviation; Formula (26) is the calculation method of the maximum completion time, that is, the maximum completion time is equal to the maximum value of the completion time of the last task on each device; Formula (27) is the calculation method of the process start time deviation and the delivery time deviation. Based on the recurrent neural network decoder in step 1, two robustness indicators, the process start time deviation SD and the delivery time deviation DD, are output.

[0116] Step 3: Generate a global scheduling solution based on the master-apprentice evolutionary algorithm based on the agent mechanism;

[0117] This embodiment uses the master-apprentice evolutionary algorithm shown in the following table to generate a scheduling plan that comprehensively minimizes the maximum completion time, process start time deviation, and delivery time deviation.

[0118] Table 4 The process of the master-apprentice evolutionary algorithm for the global scheduling model

[0119]

[0120]

[0121] As shown in Table 4, the MAE algorithm of this embodiment uses the Tabu Search (TS) operator to improve the solution performance and the Path Relinking (PR) operator to update the offspring to update the scheduling scheme based on the random generation of two scheduling schemes, thereby improving the population diversity in the evolutionary process and achieving effective search of the scheduling solution space. Specifically, for the two randomly generated scheduling solutions Sol1 and Sol2, two offspring solutions Sol′1 and Sol′2 are generated respectively by executing the path relinking operator, and two new scheduling solutions Sol1 and Sol2 are generated by performing taboo search on the two offspring solutions respectively. The maximum completion time makespan, process start time deviation SD and delivery time deviation DD corresponding to the new scheduling solution are calculated using the robust evaluation method, and Sol1, Sol2 and the optimal solution of the current cycle are obtained according to the step 2 formula (1). The corresponding objective function value is selected, and the solution that minimizes the objective function value is selected as the new optimal solution for the current cycle And if The corresponding objective function value is less than the global optimal solution Sol * , then further Update to Sol * , thus completing the cnth learning iteration of the gth generation master-apprentice combination (i.e., the two scheduling solutions Sol1 and Sol2). In addition, in order to maintain the diversity of the two solutions Sol1 and Sol2, the distance metric function d(Sol1, Sol2) is introduced to judge the similarity between the two scheduling solutions. When the distance does not exceed a given constant MinD, the scheduling solution Sol2 will be reinitialized. In the MAE algorithm, by constraining the maximum number of generations of the master-apprentice cycle and the maximum number of iterations of the master-apprentice learning by given constants Maxgen and MaxcycleN, a total of g≤Maxgen generations of master-apprentice combination cycles are formed. Each generation of master-apprentice combinations performs cn≤MaxcycleN mutual learning iterations. According to the optimal solution Sol generated in the above process * , and obtain the scheduling solution that minimizes the objective function in equation (1).

[0122] This embodiment compares the performance of the master-apprentice evolutionary algorithm (TP-MAE) that integrates the taboo search (TS) operator and the path reconnection (PR) operator that updates the offspring with the original master-apprentice evolutionary algorithm, the classical particle swarm optimization algorithm for the workshop scheduling problem, and the assignment rule.

[0123] This embodiment uses benchmark examples from workshop scheduling problems, including LA01, LA08, LA13, LA17, LA26, and LA36, with a total number of processes ranging from 50 to 400 and a number of machines ranging from 5 to 20, to reflect the impact of disturbances such as emergency orders and process changes. Considering that the uncertainty of production time caused by process deviation disturbances follows a normal distribution, the working hours p0 in the example are set to the mean μ, and the standard deviations σ1 = 0.05*p0, σ2 = 0.2*p0, and σ3 = 0.35*p0 are set to generate three uncertain production time scenarios. Each uncertain scenario of each example is solved 5 times to retain the mean, as shown in Table 5. The maximum completion time and the robustness index combination are used as the optimization goal, where Best represents the optimal value of the algorithm test, AVE represents the average value in the three scenarios of each example, and RPE represents the percentage deviation of the optimal objective function value of a single algorithm relative to the optimal objective function value of all algorithms.

[0124] Table 5 Comparison of algorithm performance for the combined optimization objective of maximum completion time and robustness index

[0125]

[0126] According to the experimental results in Table 5, and Figure 3 , the performance of the master-apprentice evolutionary algorithm that combines the taboo search operator and the path reconnection operator for updating offspring is better than that of the particle swarm optimization algorithm and the original master-apprentice evolutionary algorithm. For Case I-III, due to the small problem scale, the three algorithms can stably obtain approximate optimal solutions. In particular, in Case I and Case III, the master-apprentice evolutionary algorithm that combines the taboo search operator and the path reconnection operator for updating offspring and the original master-apprentice evolutionary algorithm both obtain the same optimal solution. As the problem scale of Case IV-VI increases, the master-apprentice evolutionary algorithm that combines the taboo search operator and the path reconnection operator for updating offspring has better search and optimization capabilities than the master-apprentice evolutionary algorithm and the particle swarm optimization algorithm due to the use of the taboo search operator and the path reconnection operator. The best results are obtained in terms of optimal solution effect and algorithm stability. The optimal values ​​are improved by 4.6% to 6.5% and 2.4% to 3.7% respectively compared with the particle swarm optimization algorithm and the master-apprentice evolutionary algorithm.

[0127] Those skilled in the art know that, in addition to implementing the system, device and its various modules provided by the present invention in a purely computer-readable program code, it is entirely possible to implement the same program in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers and embedded microcontrollers by logically programming the method steps. Therefore, the system, device and its various modules provided by the present invention can be considered as a hardware component, and the modules included therein for implementing various programs can also be considered as structures within the hardware component; the modules for implementing various functions can also be considered as both software programs for implementing the method and structures within the hardware component.

[0128] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. A robust workshop scheduling method considering global disturbances, characterized in that: include: Step 1: Evaluate the robust performance of the scheduling scheme under the influence of global disturbances; Step 2: Establish a global scheduling model that takes robustness into account, construct constraints in the scheduling problem, and determine the optimization objectives for the robustness indicators of process start deviation time, order delivery deviation time, and maximum completion time efficiency indicator; Step 3: Generate a global scheduling solution based on the master-apprentice evolutionary algorithm of the agent model.

2. The robust workshop scheduling method considering global disturbance according to claim 1, characterized in that: The step 1 comprises: Construct a disjunctive graph model to describe the scheduling plan from the dimensions of process route and equipment competition relationship; The encoder based on the graph structure describes the relationship between the process sequence and equipment competition in the scheduling plan; Based on the decoder of the timing transfer relationship, the effects of disturbances along the two dimensions of process execution sequence and equipment load accumulation are analyzed, and two robustness indicators, process start time deviation and product delivery time deviation, are calculated.

3. The robust workshop scheduling method considering global disturbance according to claim 1, characterized in that: The step 2 comprises: Establish a global scheduling model that comprehensively considers robustness indicators and efficiency indicators, describe the process tasks, process sequence, process time constraints under the influence of disturbances, and the allocation of process tasks to the selected equipment and the production sequence scheduling of process tasks on the equipment; The global scheduling model includes the following assumptions and objective functions: Part manufacturing follows the predecessor-successor process relationship, and each operation has multiple predecessor operations and at most one successor operation; each operation is produced on only one machine in the optional equipment set; each machine performs at most one operation at a time; after the machine starts to execute a process operation, it will not be interrupted before the execution is completed, and other operations cannot be temporarily inserted; at the beginning of the scheduling cycle, all parts and machines are available; The objective function is: Among them, λ1, λ2, λ3 are weight coefficients, makespan is the maximum completion time describing the efficiency index, SD and DD are the process start time deviation and delivery time deviation describing the robustness index.

4. The robust workshop scheduling method considering global disturbance according to claim 3 is characterized in that: The step 3 comprises: The master-apprentice evolutionary algorithm based on the agent mechanism updates the scheduling scheme based on the randomly generated two scheduling schemes by using the taboo search TS operator for improving the solution performance and the path reconnection PR operator for updating the offspring; For the two randomly generated scheduling solutions, two child solutions are generated respectively by executing the path reconnection operator, and taboo search is performed on the two child solutions to generate two new scheduling solutions. Then, the maximum completion time makespan, process start time deviation SD and delivery time deviation DD corresponding to the new scheduling solutions are calculated respectively by using the robustness evaluation method.

5. A robust workshop scheduling system considering global disturbances, characterized in that: include: Module M1: Robust performance evaluation of scheduling schemes under the influence of global disturbances; Module M2: Establish a global scheduling model that takes robustness indicators into consideration, construct constraints in the scheduling problem, and determine the optimization objectives for the robustness indicators of process start deviation time, order delivery deviation time, and maximum completion time efficiency indicators; Module M3: A master-apprentice evolutionary algorithm based on an agent model to generate a global scheduling solution.

6. The robust workshop scheduling system considering global disturbance according to claim 5, characterized in that: The module M1 comprises: Construct a disjunctive graph model to describe the scheduling plan from the dimensions of process route and equipment competition relationship; The encoder based on the graph structure describes the relationship between the process sequence and equipment competition in the scheduling plan; Based on the decoder of the timing transfer relationship, the effects of disturbances along the two dimensions of process execution sequence and equipment load accumulation are analyzed, and two robustness indicators, process start time deviation and product delivery time deviation, are calculated.

7. The robust workshop scheduling system considering global disturbance according to claim 5, characterized in that: The module M2 comprises: Establish a global scheduling model that comprehensively considers robustness indicators and efficiency indicators, describe the process tasks, process sequence, process time constraints under the influence of disturbances, and the allocation of process tasks to the selected equipment and the production sequence scheduling of process tasks on the equipment; The global scheduling model includes the following assumptions and objective functions: Part manufacturing follows the predecessor-successor process relationship, and each operation has multiple predecessor operations and at most one successor operation; each operation is produced on only one machine in the optional equipment set; each machine performs at most one operation at a time; after the machine starts to execute a process operation, it will not be interrupted before the execution is completed, and other operations cannot be temporarily inserted; at the beginning of the scheduling cycle, all parts and machines are available; The objective function is: Minimize Obj = λ1makespan + λ2SD + λ3DD, where λ1, λ2, λ3 are weight coefficients, makespan is the maximum completion time describing the efficiency index, SD and DD are the process start time deviation and delivery time deviation describing the robustness index.

8. The robust workshop scheduling system considering global disturbance according to claim 7, characterized in that: The module M3 comprises: The master-apprentice evolutionary algorithm based on the agent mechanism updates the scheduling scheme based on the randomly generated two scheduling schemes by using the taboo search TS operator for improving the solution performance and the path reconnection PR operator for updating the offspring; For the two randomly generated scheduling solutions, two child solutions are generated respectively by executing the path reconnection operator, and taboo search is performed on the two child solutions to generate two new scheduling solutions. Then, the maximum completion time makespan, process start time deviation SD and delivery time deviation DD corresponding to the new scheduling solutions are calculated respectively by using the robustness evaluation method.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the robust workshop scheduling method considering global disturbances described in any one of claims 1 to 4 are implemented.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the computer program is executed by a processor, the steps of the robust workshop scheduling method considering global disturbances described in any one of claims 1 to 4 are implemented.

Images