Ship subassembly manufacturing iterative optimization scheduling method and system based on artificial neural network

Abstract

This invention discloses an iterative optimization scheduling method and system for ship section manufacturing based on artificial neural networks. The method involves establishing a ship section manufacturing workshop model based on its state characteristic values ​​and constraints; acquiring historical production plans from the workshop; enumerating production plans using different scheduling rules; establishing a scheduling experience feedback pool; constructing and training a backpropagation (BP) neural network for iterative optimization scheduling of ship section manufacturing; calculating the scheduling rule priority weight matrix using the BP neural network based on the workshop's state characteristic values ​​and optimization target weights; executing the scheduling rules according to the priority weight matrix and recording the scheduling results in the scheduling experience feedback pool; and randomly selecting small batches of experience samples to update the parameters of the BP neural network. This invention achieves good scheduling results and can improve production efficiency and resource utilization in the ship section manufacturing process.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of ship block manufacturing, and particularly relates to a ship block manufacturing iterative optimization scheduling method and system based on an artificial neural network. BACKGROUND

[0002] Shipbuilding industry is an important strategic industry related to national defense, national strength and people's livelihood. Shipbuilding industry takes supply-side structural reform as the main line and constantly promotes the transformation of the industry to high-quality development. China has become a major shipbuilding country in terms of industrial scale, and ranks first in the world in terms of shipbuilding completion, hand-held ship orders, and new ship orders. However, there is still a big gap in key indicators such as shipbuilding efficiency, cycle, and quality compared with international advanced shipbuilding countries and regions. Due to the existence of practical factors such as multiple workstations, scattered layout, large number of intermediate products, high site uncertainty, and long production planning cycle in ship block manufacturing, the production efficiency and resource utilization rate in the ship block manufacturing process are low. The scheduling of ship block manufacturing is an NP-hard problem, and the current scheduling of most shipbuilding enterprises relies on the experience and knowledge of management personnel. The production efficiency and resource utilization rate of manual scheduling cannot be guaranteed, so it is urgent to improve the current situation of high dependence on manual scheduling in ship block manufacturing.

[0003] With the rapid development of information technology, a large number of scheduling rules have been accumulated in shipbuilding industry, but there is no scheduling rule that can be applied to all situations. Different scheduling rules should be used according to actual production needs. An optimization scheduling method that can dynamically adapt to actual production needs is particularly important. The rapid development of artificial intelligence technology brings new possibilities to the scheduling of ship block manufacturing. Artificial neural networks can simulate biological neural systems and can learn from training samples, automatically adjust the weights and biases of their networks, and have strong learning and adaptation capabilities.

[0004] Therefore, the design of a ship block manufacturing iterative optimization scheduling method and system based on an artificial neural network not only reduces the dependence of scheduling in ship block manufacturing on manual work, but also realizes adaptive scheduling, which has important practical significance for ship block manufacturing. SUMMARY

[0005] To solve the problems in the prior art, the present application aims to provide a ship block manufacturing iterative optimization scheduling method and system based on an artificial neural network.

[0006] To achieve the above purpose, the technical scheme adopted by the present application is as follows:

[0007] A ship block manufacturing iterative optimization scheduling method based on an artificial neural network, comprising the steps of:

[0008] Step one, according to the state characteristic value and the constraint condition of the ship section manufacturing workshop, a ship section manufacturing workshop model is established;

[0009] Step two, the historical production plan of the ship section manufacturing workshop is obtained, different scheduling rules are enumerated, the scheduling performance index is recorded, and a scheduling experience review pool is established;

[0010] Step three, a ship section manufacturing iterative optimization scheduling BP neural network model is constructed;

[0011] Step four, the BP neural network is trained, the Adam optimizer is used to update the weights and biases of the BP neural network through the gradient descent algorithm until the training round reaches the training preset requirement;

[0012] Step five, according to the state characteristic value and the optimization target weight of the ship section manufacturing workshop, the BP neural network is used to calculate the scheduling rule priority weight matrix;

[0013] Step six, according to the scheduling rule priority weight matrix, the scheduling rule is executed to perform automatic scheduling, and the optimization target output scheduling result is met;

[0014] Step seven, the scheduling result is recorded in the scheduling experience review pool, thereby helping the subsequent BP neural network update and work record query;

[0015] Step eight, after a certain number of uses, a small batch of experience samples are randomly extracted from the scheduling experience review pool, and the parameters of the BP neural network are updated.

[0016] Further, in step one, the state characteristic value includes a to-be-built section, the number of work tasks, the time required to complete the work task, the work team available in the section workshop, and the equipment available for use in the section workshop.

[0017] Further, in step two, the scheduling rules include the "shortest completion time" rule, the "earliest completion deadline" rule, the "first come first served" rule, the "longest remaining time" rule, and the "maximum time ratio" rule.

[0018] Further, in step two, the scheduling performance index includes total completion time, average process time, equipment utilization rate variance, and production energy consumption; the above scheduling performance index is the optimization target of the ship section manufacturing workshop.

[0019] Further, in step three, the BP neural network has a 5-layer structure, including 1 input layer, 1 output layer, and 3 hidden layers; wherein, the input layer nodes correspond to the sum of the number of state characteristic values of the ship section manufacturing workshop and the number of optimization targets, and the output layer nodes correspond to the number of scheduling rules;

[0020] A full connection BP neural network is used to build the BP neural network with a Relu activation function.

[0021] Further, in step four, the scheduling performance index data set of the ship block manufacturing workshop in step two is further normalized.

[0022] Further, in step five, the plurality of optimization objectives are comprehensively considered, different weights are assigned to different optimization objectives based on the importance of the optimization objectives, and the optimal production plan is obtained.

[0023] Further, in step six, each element of the scheduling rule priority weight matrix corresponds to a scheduling rule, and the value range is [0, 1]. The closer the priority weight of the scheduling rule is to 1, the higher the priority of the scheduling rule in the scheduling.

[0024] The priority of a single work task is calculated by the scheduling rule priority weight, and the calculation formula is as follows:

[0025]

[0026] In the formula, K i represents the priority of the task in the scheduling rule i, and alpha i represents the priority weight of the scheduling rule.

[0027] Further, in step seven, the scheduling result includes the state characteristic value of the ship block manufacturing workshop, the optimization objective weight matrix, and the scheduling rule priority weight matrix.

[0028] A ship block manufacturing iterative optimization scheduling system based on an artificial neural network, which is used to realize the ship block manufacturing iterative optimization scheduling method based on the artificial neural network.

[0029] The beneficial effects of the present application are that, compared with the prior art, the present application takes the actual demand of ship block manufacturing as the starting point, combines BP neural network and scheduling rules, and can improve the production efficiency and resource utilization rate in the ship block manufacturing process. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 is a whole schematic diagram of the ship block manufacturing iterative optimization scheduling method based on the artificial neural network described in the present application;

[0031] Figure 2 is a historical production plan processing schematic diagram of a ship block manufacturing workshop;

[0032] Figure 3 is a BP neural network schematic diagram based on ship block manufacturing optimization scheduling;

[0033] Figure 4This is a schematic diagram illustrating the execution of scheduling rules based on the priority weight matrix of the scheduling rules. Detailed Implementation

[0034] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of this application.

[0035] like Figure 1 As shown, the iterative optimization scheduling method for ship section manufacturing based on artificial neural networks described in this invention includes the following steps:

[0036] Step 1: Establish a model of the ship section manufacturing workshop based on the state characteristic values ​​and constraints of the workshop;

[0037] The state characteristic values ​​of a ship section manufacturing workshop are a direct reflection of the actual production situation in the workshop. The ship section manufacturing workshop is a multi-factor, multi-constraint, and nonlinear manufacturing system.

[0038] The state characteristics of a ship section manufacturing workshop include: the section to be built M, the number of work tasks N, the time required to complete the work tasks t, the work groups P that can work in the section workshop, and the equipment K that can be used in the section workshop.

[0039] The model of the ship section manufacturing workshop needs to meet the following constraints:

[0040] (1) Each segment has its fixed process route. Subsequent tasks can only be carried out after the preceding tasks are completed. Before the preceding tasks are completed, the work team carrying out the subsequent tasks enters a waiting state after the previous task is completed. The relevant constraints can be expressed as:

[0041] T mn -T m(n-1) >t mn n≠1

[0042] Among them, T mn t represents the completion time of task n in segment m. mn This represents the time required to complete task n in segment m.

[0043] (2) At any given time, a work group can only carry out one work task and cannot carry out multiple work tasks simultaneously. All work groups can work normally at t=0. The relevant constraints can be expressed as:

[0044]

[0045] Where P represents the total number of work groups, p represents a work group, and X pmnTo control the variable, when the work task n in the segment m is completed by the p team, it is 1, if the work task is not completed, it is 0.

[0046] (3) The completion time of each work task in the segment is positive, and the completion time of the segment is greater than the sum of the completion times of all work tasks in the segment, and the related constraint can be expressed as:

[0047] t pmn >0, T max >∑t pmn

[0048] (4) All devices in the segment are available at t=0;

[0049] (5) All devices can be scheduled for maintenance, and no unexpected downtime will occur before a workpiece is processed.

[0050] Step two: obtain the historical production plan of the ship segment manufacturing workshop, enumerate the production plan using different scheduling rules, record the scheduling performance index, and establish a scheduling experience review pool; as shown in Figure 2

[0051] The scheduling rules include: "shortest completion time" rule, "earliest completion deadline" rule, "first come first served" rule, "longest remaining time" rule, and "maximum time ratio" rule.

[0052] (1) "Shortest completion time" rule, that is, the priority of the work task with shorter required time is higher, and the scheduling rule can be expressed as:

[0053]

[0054] In the formula, K n represents the priority of work task n, and the value range is [0, 1], the closer the priority is to 1, the earlier the work task is in the scheduling, T n represents the required time of work task n, and T represents the longest required time of work task.

[0055] (2) "Earliest completion deadline" rule, that is, the priority of the work task with earlier completion deadline is higher, and the scheduling rule can be expressed as:

[0056]

[0057] In the formula, T nend represents the completion deadline of work task n.

[0058] (3) "First come first served" rule, that is, the priority of the work task with earlier release time is higher, and the scheduling rule can be expressed as:

[0059] ​

[0060] where t nassign represents the release time of the work task n.

[0061] (4) The "longest remaining time" rule, i.e. the work task with the longest time required for completion has the highest priority, which can be expressed as:

[0062]

[0063] where t nremain represents the remaining time required for completion of the work task n.

[0064] (5) The "maximum time ratio" rule, i.e. the work task with the larger difference between the waiting time and the required time has the higher priority, which can be expressed as:

[0065]

[0066] The scheduling performance indicators include: total completion time, average flow time, equipment utilization variance, and production energy consumption. The above scheduling performance indicators are the optimization objectives of the ship section manufacturing workshop.

[0067] (1) Total completion time T max , whose calculation formula is as follows:

[0068] T max = max{T m}

[0069] where T m is the completion time of the section m.

[0070] (2) Average flow time T flow , whose calculation formula is as follows:

[0071]

[0072] where N is the number of work tasks of the ship section manufacturing workshop, t pmi is the time consumed by the team p of the m section to complete the work task i.

[0073] (3) Equipment utilization variance σ, whose calculation formula is as follows:

[0074]

[0075] (4) Production energy consumption E, whose calculation formula is as follows:

[0076]

[0077] where Wik The energy consumption of the equipment representing the ship block manufacturing workshop when processing the i-th work task at the k-th process, e k The energy consumption of the equipment k when on standby, t k Representing the total standby time of equipment k in the production plan.

[0078] Step three: build a ship block manufacturing iterative optimization scheduling BP neural network model;

[0079] As Figure 3 shown, the network has a 5-layer structure, including 1 input layer, 1 output layer, and 3 hidden layers. The input layer nodes correspond to the sum of the number of state feature values and the number of optimization objectives of the ship block manufacturing workshop, and the output layer nodes correspond to the number of scheduling rules. To make the BP neural network have stronger intelligent decision-making ability, a fully connected BP neural network is used in combination with a Relu activation function to build the network model. The expression of the Relu function is:

[0080] F(x) = max(0, x)

[0081] The number of neurons in the hidden layer of the BP neural network model is calculated using the Kolmogorov formula:

[0082] N hid = 2N in +1

[0083] In the formula, N id is the number of neurons in the hidden layer, and N in is the number of neurons in the input layer.

[0084] Step four: train the BP neural network, use the Adam optimizer to update the weights and biases of the BP neural network through the gradient descent algorithm until the training round reaches the training preset requirements.

[0085] Because the dimensions of the input may have large differences, some dimensions dominate during training, causing the network to converge slowly.

[0086] To make the BP neural network have stronger robustness, the scheduling performance index dataset of the ship block manufacturing workshop is normalized. Let the original dataset be X, x new be the normalized data, x min represents the minimum value in X, x max represents the maximum value in X, and x i is changed to a value in the interval [0, 1]. The larger the scheduling performance index, the better the performance, and the calculation formula is as follows:

[0087]

[0088] Step five: according to the state characteristic value of the ship section manufacturing workshop and the optimization target weight, the priority weight matrix of the scheduling rule is calculated by using the BP neural network;

[0089] The optimization target includes: total completion time T max , average process time T flow , equipment utilization variance σ, and production energy consumption PE.

[0090] The present application comprehensively considers multiple optimization targets, and different weights are allocated to different optimization targets based on the importance of the optimization targets, so as to obtain an optimal production plan. The value range of the optimization target weight is [0, 1], and the closer the value of the priority weight is to 1, the more important the optimization target is, and vice versa. The sum of the weights of all optimization targets is 1.

[0091] Step six: according to the priority weight matrix of the scheduling rule, the scheduling rule is executed to automatically schedule, so as to meet the optimization target and output the scheduling result; as shown in the following formula: Figure 4

[0092] Each element of the priority weight matrix of the scheduling rule corresponds to a scheduling rule, and the value range is [0, 1]. The closer the priority weight of the scheduling rule is to 1, the higher the priority of the scheduling rule in scheduling is. The priority of a single work task is calculated by the priority weight of the scheduling rule, and the calculation formula is as follows:

[0093]

[0094] In the formula, K i represents the priority of the task in the scheduling rule i, and α i represents the priority weight of the scheduling rule.

[0095] Step seven: the scheduling result is recorded in the scheduling experience feedback pool, thereby providing help for subsequent BP neural network updating and work record query;

[0096] The scheduling result includes: the state characteristic value of the ship section manufacturing workshop, the optimization target weight matrix, and the priority weight matrix of the scheduling rule.

[0097] Step eight: after a certain number of uses, a small batch of experience samples are randomly extracted from the scheduling experience feedback pool, and the parameters of the BP neural network are updated, so that the decision of the BP neural network is more in line with the actual production needs.

[0098] After a certain number of uses, a small batch of experience samples are randomly extracted from the scheduling experience feedback pool, and the weights and biases of the BP neural network are updated by using the Adam optimizer through the gradient descent algorithm until the training round reaches the preset requirement.

[0099] ​The application has the beneficial effect that, compared with the prior art, the application takes the actual demand of ship block manufacturing as the starting point, combines the BP neural network and the scheduling rule, and can improve the production efficiency and resource utilization rate in the ship block manufacturing process.

[0100] The applicant of the present application has made a detailed description and explanation of the embodiments of the present application in combination with the drawings of the specification, but those skilled in the art should understand that the above embodiments are only preferred embodiments of the present application, and the detailed description is only to help the reader better understand the spirit of the present application, and is not a limitation on the protection scope of the present application, on the contrary, any improvement or modification based on the spirit of the present application should fall within the protection scope of the present application.

Claims

1. An iterative optimization scheduling method for ship section manufacturing based on artificial neural networks, characterized in that, Including the following steps: Step 1: Establish a model of the ship section manufacturing workshop based on the state characteristic values ​​and constraints of the workshop. Step 2: Obtain the historical production plan of the ship section manufacturing workshop, enumerate the production plan using different scheduling rules, record the scheduling performance indicators, and establish a scheduling experience review pool. Step 3: Construct a BP neural network model for iterative optimization scheduling of ship section manufacturing; Step 4: Train the BP neural network. Use the Adam optimizer to update the weights and biases of the BP neural network through gradient descent algorithm until the training rounds reach the preset training requirements. Step 5: Based on the state characteristic values ​​and optimization target weights of the ship section manufacturing workshop, use a BP neural network to calculate the priority weight matrix of scheduling rules; Step 6: Based on the priority weight matrix of the scheduling rules, execute the scheduling rules to perform automatic scheduling and output the scheduling results to meet the optimization objectives; Step 7: Record the scheduling results in the scheduling experience feedback pool to help with subsequent BP neural network updates and work record queries; Step 8: After a certain number of uses, randomly extract a small batch of experience samples from the scheduling experience revisit pool to update the parameters of the BP neural network.

2. The iterative optimization scheduling method for ship section manufacturing based on artificial neural networks according to claim 1, characterized in that, In step one, the status characteristics include the segment to be built, the number of work tasks, the time required to complete the work tasks, the work teams that can work in the segment workshop, and the equipment available for use in the segment workshop.

3. The iterative optimization scheduling method for ship section manufacturing based on artificial neural networks according to claim 1, characterized in that, In step two, the scheduling rules include the "shortest completion time" rule, the "earliest completion deadline" rule, the "first-come, first-served" rule, the "longest remaining time" rule, and the "maximum time ratio" rule.

4. The iterative optimization scheduling method for ship section manufacturing based on artificial neural networks according to claim 1, characterized in that, In step two, the scheduling performance indicators include total completion time, average process time, equipment utilization variance, and production energy consumption; the above scheduling performance indicators are the optimization targets of the ship section manufacturing workshop.

5. The iterative optimization scheduling method for ship section manufacturing based on artificial neural networks according to claim 1, characterized in that, In step three, the BP neural network has a 5-layer structure, including 1 input layer, 1 output layer, and 3 hidden layers; among them, the input layer nodes correspond to the sum of the number of state feature values ​​and the number of optimization objectives in the ship section manufacturing workshop, and the output layer nodes correspond to the number of scheduling rules; A BP neural network is constructed using a fully connected BP neural network and a ReLU activation function.

6. The iterative optimization scheduling method for ship section manufacturing based on artificial neural networks according to claim 1, characterized in that, Step four also includes normalizing the scheduling performance index dataset of the ship section manufacturing workshop from step two.

7. The iterative optimization scheduling method for ship section manufacturing based on artificial neural networks according to claim 1, characterized in that, In step five, multiple optimization objectives are comprehensively considered, and different weights are assigned to different optimization objectives based on their importance to obtain the optimal production plan.

8. The iterative optimization scheduling method for ship section manufacturing based on artificial neural networks according to claim 1, characterized in that, In step six, each element of the scheduling rule priority weight matrix corresponds to a scheduling rule, with a value range of [0,1]. The closer the priority weight of a scheduling rule is to 1, the higher its priority in the scheduling process. The priority of a single task is calculated using the priority weight of the scheduling rules, as shown in the following formula: In the formula, K i α represents the priority of the task in scheduling rule i. i This represents the priority weight of the scheduling rules.

9. The iterative optimization scheduling method for ship section manufacturing based on artificial neural networks according to claim 1, characterized in that, In step seven, the scheduling results include the state characteristic values ​​of the ship section manufacturing workshop, the optimization target weight matrix, and the scheduling rule priority weight matrix.

10. A ship section manufacturing iterative optimization scheduling system based on artificial neural networks, characterized in that, The system is used to implement the iterative optimization scheduling method for ship section manufacturing based on artificial neural networks as described in any one of claims 1-9.

Images