A manufacturing industry chain optimization method based on a double carbon target

Abstract

This invention discloses a manufacturing supply chain optimization method based on dual carbon objectives, belonging to the field of manufacturing supply chain optimization technology. It collects the economic benefits and carbon emissions of enterprises at each node of the supply chain, as well as the parameters of each node in a complex network, as samples. An initial population of the supply chain is collected. A dual-carbon supply chain collaborative optimization model is established. The samples and the initial population are input into the dual-carbon supply chain collaborative optimization model. An improved artificial bee colony algorithm is used to optimize the BP neural network. Based on the feedback from the termination node, the initial population of the supply chain is evaluated and input into the optimization system. The clonal selection algorithm is then used to optimize the supply chain. This invention provides a manufacturing supply chain optimization method based on dual carbon objectives that can optimize both the economic benefits and carbon emissions of the supply chain, solving the problems of low efficiency and high cost throughout the entire supply chain.

Description

Technical Field

[0001] This invention relates to the field of manufacturing supply chain optimization technology, and more specifically to a manufacturing supply chain optimization method based on dual carbon objectives. Background Technology

[0002] Currently, chip manufacturing is one of the most complex and globalized industries in the world's industrial and supply chains. In the chip production process, wafer fabrication is one of the most economically efficient stages in the entire semiconductor value chain. However, it is also the manufacturing stage with the greatest environmental impact, as it involves the use of large amounts of water, energy, and raw materials, generating corresponding waste and emissions. The rapid growth in chip demand and production at present exacerbates the environmental impact of different stages of its life cycle. A review of existing literature reveals a scarcity of scientific case studies on environmental impact assessments in the semiconductor industry; most assessments focus on the perspectives of companies or technologies. The manufacturing of modern semiconductor devices involves a series of complex energy-intensive and resource-intensive processes, generating substantial amounts of waste. Higgs et al., using Intel as an example, studied the impact of CO2 emissions on various aspects of semiconductor companies, including manufacturing operations, product use, supply chain, logistics, and employee travel. Liu et al. assessed the potential environmental impact of five production processes for double data rate synchronous dynamic random access memory (DRAM) based on relevant ecological indicators. Boyd et al. studied the life cycle assessment of digital logic chips at the 7th generation technology level and analyzed the impact of relevant stages under various environmental indicators. Huang et al. developed a parameter-based tool to help identify key parameters in complex manufacturing processes in the semiconductor industry. Through process and statistical analysis of 7114 wafer products, they established six CFP regression models. Guo et al. studied the impact categories involved in the production of embedded non-volatile memory (eNVM) chips, addressing issues such as climate change, water use, particulate matter, and resource utilization of minerals and metals. They provided suggestions for low-carbon development of semiconductor companies from a life-cycle perspective. Zhang Fan et al.'s research has demonstrated the decisive role of technological change and efficiency improvement in the impact of carbon emission intensity; however, not all companies can bear the high costs brought about by technological change. Therefore, how to better coordinate the efficiency and energy consumption status of various companies in the industrial chain, transcend the limitations of individual companies' carbon emission reduction, systematically improve the efficiency of the entire industrial chain, reduce corporate energy consumption, and ensure the optimal efficiency of the entire industrial chain is a question worth exploring.

[0003] The aforementioned studies only analyzed a single link in chip manufacturing or a single company, addressing the impact assessment of a single manufacturing process or production stage. To date, no research has focused on the overall impact on the industry or its supply chain. This is primarily due to: the confidentiality of technologies and patents, the reliability of product lifecycle data, the inherent complexity of the industrial sector, the complexity and variability of manufacturing technologies, the global nature of the chip supply chain, and the diversity of product manufacturing technologies. For the same reasons, it is difficult to compare the environmental impact of different chip manufacturers and different manufacturing technologies, and it is also difficult to explore overall economic benefits and carbon emissions from a supply chain perspective.

[0004] Therefore, how to break down the barriers between upstream and downstream of the industrial chain from the perspective of the whole industrial chain, and propose efficient optimization methods based on existing algorithms and models to optimize the economic benefits and carbon emissions of the industrial chain from a holistic perspective, is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a manufacturing supply chain optimization method based on dual carbon objectives, which can optimize the economic benefits and carbon emissions of the supply chain based on existing algorithms and models, and solve the problems of low efficiency and high cost of the entire supply chain.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A manufacturing supply chain optimization method based on dual carbon objectives includes the following steps:

[0008] Step 1: Collect the economic benefits and carbon emissions of enterprises at each node of the industrial chain, as well as the parameters of each node in the complex network, as samples; collect the initial population of the industrial chain.

[0009] Step 2: Establish a dual-carbon industrial chain collaborative optimization model; bring the samples and the initial population of the industrial chain into the dual-carbon industrial chain collaborative optimization model, wherein the dual-carbon industrial chain collaborative optimization model uses an improved artificial bee colony algorithm to optimize the initial weights and thresholds of the BP neural network, and based on the feedback of the termination node, selectively feeds back the optimal weights, thresholds and related errors of the neural network to the next step to complete the evaluation of the initial population of the industrial chain.

[0010] Step 3: Input the evaluated initial population of the industry chain into the optimization system and combine it with the clonal selection algorithm to optimize the industry chain.

[0011] Preferably, step 1 specifically includes:

[0012] The parameters of each node in a complex network include: average path length and node importance ranking;

[0013] The average path is represented as:

[0014] The sum of the distances of all edges on the shortest path connecting nodes i and j is defined as the distance d between the two nodes. ij The average distance between any two nodes is defined as the average path length l of the network, where N is the number of nodes in the network;

[0015] The ranking of node importance is expressed as follows:

[0016] Among them, PR j (t-1) represents the PR value of node j at time t-1, c represents the random jump probability, and e jk This represents the economic benefit of node k connected to node j. Assume node j points to m nodes. If node j points to node i, then a jt The value is 1 if it is set to 1, and 0 otherwise.

[0017] The initial population of the industrial chain is divided into the first initial population of the industrial chain and the second initial population of the industrial chain.

[0018] Preferably, step 2 specifically includes:

[0019] Step 2.1: Substitute each individual in the initial population of the first industrial chain into the BP network A in the dual-carbon industrial chain collaborative optimization model, initially set the initial weights and thresholds of the BP network A, and obtain the corrected weights and thresholds for each individual through sample training;

[0020] Step 2.2: After all individuals in the initial population of the first industry chain have completed the calculation, determine whether the result meets the termination condition. If it does, proceed to step 2.3. If it does not, update the individuals in the initial population of the first industry chain, use the updated population as the initial population of the first industry chain, and return to step 2.1.

[0021] Step 2.3: Use the initial population of the second industrial chain as the sample of BP network B, and use the optimal weights and thresholds obtained by BP network A as the initial weights and thresholds of BP network B to train the sample of BP network B.

[0022] Preferably, step 2.1 further includes setting progress nodes R, R1, and R2, with an initial value of 1 for each progress node. After receiving a sample, the model is initialized. R1 counts the number of iterations for BP network A, R2 counts the number of iterations for BP network B, and R counts the number of iterations for the clone selection algorithm. The sample is input to the network input layer EnterA.

[0023] Preferably, the sample training in step 2.1 specifically includes training the first individual received by the input layer EnterA of the network with the sample once, calculating the network error e1 at this time when the training is completed, and correcting the weights and thresholds of the BP network A based on the partial derivatives of each neuron in the output layer with respect to the network error e1.

[0024] Preferably, obtaining the corrected weights and thresholds of each individual in step 2.1 specifically includes: determining whether the value of the signal R1 is equal to the number u of the population. When R1 < u, R1 = R1 + 1, and substituting the next individual in the population into the BP network A to calculate the network error e at this time i , and optimizing the weights and thresholds at this time according to the calculation results; when R1 = u, it indicates that the error calculation of each individual in the population has been completed.

[0025] Preferably, step 2.2 specifically includes: determining whether the error calculation result meets the termination condition through the artificial bee colony algorithm. If it meets, execute step 2.3; if not, update the relevant nectar sources of the individuals in the initial population of the first industrial chain through the following formula, and the foraging bees and scout bees perform according to the following formula respectively:

[0026] x′ id =x id +φ id (x id -x kd )

[0027]

[0028] Among them, the foraging bee corresponding to the i-th nectar source searches for a new nectar source x′ id , i = 1, 2,... SN, SN is the total number of foraging bees, φ id is a random number on the interval [-1, 1], x id represents the value of the i-th nectar source, i ≠ k; the scout bee searches for a new solution x id , r is a random number on the interval [0, 1], and are the lower and upper bounds of the d-th dimension;

[0029] Take the updated population as the initial population of the first industrial chain and return to step 2.1.

[0030] Preferably, in step 2.2, returning to step 2.1 specifically includes: R2 detects whether all individuals in the updated initial population of the first industrial volume have been updated for crossover and mutation, and sends a reset signal to R1 after the update is completed. At this time, the value of R1 is 1, and continue with step 2.1, using the updated population to calculate the new round of benefit values.

[0031] Preferably, step 2.3 specifically includes: the model determines that the artificial bee colony algorithm meets the termination condition, the initial population of the second industrial chain is used as the sample of BP network B, the system uses the optimal initial weights and thresholds obtained by BP network A as the initial weights and thresholds of BP network B, and BP network B is trained on the sample of BP network B after receiving the data input.

[0032] Preferably, step 3 includes: sorting the individuals in the population according to their evaluation values ​​and feeding the sorted population data back to the optimization system; the optimization system receiving the population data from the BP network B then performing mutation and crossover operations using a clonal selection algorithm.

[0033] As can be seen from the above technical solution, compared with the prior art, this invention discloses a manufacturing supply chain optimization method based on dual-carbon objectives. It analyzes the economic benefits and carbon emissions of the supply chain from a holistic perspective, and the proposed BP-CS model for overall evaluation and optimization of the supply chain can effectively evaluate and optimize the "dual-carbon" supply chain. Under stable supply chain structure, optimizing the combination of enterprises at supply chain nodes can effectively achieve overall supply chain optimization. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0035] Figure 1 The attached figure is a diagram of the overall method architecture provided by the present invention;

[0036] Figure 2 The attached figure is a schematic diagram of the chip industry chain nodes provided by the present invention;

[0037] Figure 3 The attached figure is a diagram of the complex network topology of the chip industry chain provided by this invention;

[0038] Figure 4 The attached figure is an analytical diagram of the BP-CS model provided by this invention;

[0039] Figure 5 The attached figure is a comparison chart of the optimized data provided by the present invention;

[0040] Figure 6 The attached figure is a radar chart showing the various indicators and total benefits provided by this invention;

[0041] Figure 7-1 The attached figure is a box plot of the Economic benefits index provided by this invention.

[0042] Figure 7-2 The attached figure is a box plot of the Carbonemissions index provided by this invention.

[0043] Figure 7-3 The attached figure is a box plot of the Averageimportance metric provided by this invention.

[0044] Figure 7-4 The attached figure is a box plot of the pathlength metric provided by this invention.

[0045] Figure 8 The attached figure is a flowchart of the BP-CS model provided by the present invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] like Figure 1 , 8 As shown, this invention discloses a manufacturing supply chain optimization method based on dual-carbon objectives. It introduces complex network theory into the study of supply chain optimization problems. The BP neural network can be used to evaluate or predict new data through extensive training on existing data. A complex network-based supply chain collaborative optimization model is constructed by combining the BP neural network with the clonal selection algorithm. This model comprehensively examines the functions and benefits of each enterprise in the supply chain nodes. Finally, based on the model solution, a supply chain collaborative optimization scheme is presented. This is the model's approach, including the following steps:

[0048] Step 1: Collect the economic benefits and carbon emissions of enterprises at each node of the industrial chain, as well as the parameters of each node in the complex network, as samples; collect the initial population of the industrial chain.

[0049] Step 2: Establish a dual-carbon industrial chain collaborative optimization model; bring the samples and the initial population of the industrial chain into the dual-carbon industrial chain collaborative optimization model. The dual-carbon industrial chain collaborative optimization model uses an improved artificial bee colony algorithm to optimize the initial weights and thresholds of the BP neural network, and based on the feedback from the termination node, selectively feeds back the optimal weights, thresholds and related errors of the neural network to the next step to complete the evaluation of the initial population of the industrial chain.

[0050] Step 3: Input the evaluated initial population of the industry chain into the optimization system and combine it with the clonal selection algorithm to optimize the industry chain.

[0051] In one specific embodiment, step 1 specifically includes:

[0052] The parameters of each node in a complex network include: average path length and node importance ranking;

[0053] The average path is represented as:

[0054] The sum of the distances of all edges on the shortest path connecting nodes i and j is defined as the distance d between the two nodes. ij The average distance between any two nodes is defined as the average path length l of the network, where N is the number of nodes in the network;

[0055] Furthermore, the average path length reflects the smoothness of the system, which has strong practical significance in supply chain decision-making, especially in information flow decisions. This parameter reflects the looseness of the complex network structure; the smaller the parameter, the smaller the network and the more compact the structure, and vice versa. In this method, the average path length directly depends on the distance between nodes in different combinations.

[0056] In one specific embodiment, in a directed weighted network, there may be dangling nodes, i.e., nodes that only have edges pointing to them, but no edges pointing from them to other nodes. In such cases, the random walk process cannot continue. To solve this problem, a random jump probability *c* is introduced. During the iteration process, each node's PageRank (PR) value is distributed equally among all nodes in the network with a probability of (1-c), and distributed to the nodes it points to with a probability of *c* according to their economic benefits. Different nodes have different influences on the same node. The PageRank algorithm stipulates that nodes will distribute their values ​​proportionally to the nodes they point to according to the economic benefits of each node. The PR value of a node is used to characterize the importance of each node in the industry chain. The node importance ranking is expressed as:

[0057]

[0058] Among them, PR j (t-1) represents the PR value of node j at time t-1, c represents the random jump probability, and e jk This represents the economic benefit of node k connected to node j. Assume node j points to m nodes. If node j points to node i, then a jt The value is 1 if it is set to 1, and 0 otherwise.

[0059] The initial population of the industrial chain is divided into the initial population of the first industrial chain and the initial population of the second industrial chain.

[0060] In one specific embodiment, step 2 specifically includes:

[0061] Step 2.1: Substitute each individual in the initial population of the first industrial chain into BP network A in the collaborative optimization model of the dual-carbon industrial chain. Initially set the initial weights and thresholds of BP network A, and obtain the corrected weights and thresholds of each individual through sample training.

[0062] Step 2.2: After all individuals in the initial population of the first industrial chain have completed the calculation, determine whether the result meets the termination condition. If it meets, execute Step 2.3; if not, update the individuals in the initial population of the first industrial chain, use the updated population as the initial population of the first industrial chain, and return to Step 2.1.

[0063] Step 2.3: Use the initial population of the second industrial chain as the samples of BP network B, use the optimal weights and thresholds obtained by BP network A as the initial weights and thresholds of BP network B, and train the samples of BP network B.

[0064] In a specific embodiment, Step 2.1 further includes setting progress nodes R, R1, and R2, with the initial values of all progress nodes being 1. After receiving the samples, the model is initialized. R1 counts for BP network A, R2 counts for BP network B, and R counts for the number of loops of the clonal selection algorithm. The samples are input into the network input layer EnterA.

[0065] In a specific embodiment, the sample training in Step 2.1 specifically includes training the first individual received by the network input layer EnterA through the samples once. When the training is completed, calculate the network error e1 at this time, and correct the weights and thresholds of BP network A based on the partial derivatives of each neuron in the output layer with respect to the network error e1.

[0066] In a specific embodiment, obtaining the corrected weights and thresholds of each individual in Step 2.1 specifically includes: determining whether the value of signal R1 is equal to the population size u. When R1 < u, R = R1 + 1, and substitute the next individual in the population into BP network A to calculate the network error e [[ID= sixteen]] i and optimize the weights and thresholds at this time according to the calculation results; when R1 = u, it indicates that the error calculation of each individual in the population has been completed.

[0067] In a specific embodiment, Step 2.2 specifically includes: using the artificial bee colony algorithm to determine whether the error calculation result meets the termination condition. If it meets, execute Step 2.3; if not, update the relevant nectar sources of the individuals in the initial population of the first industrial chain through the following formula. The forager bees and scout bees are updated according to the following formula respectively:

[0068] x′ id =x id +φ id (x id -xkd )

[0069]

[0070] Among them, the foraging bees corresponding to the i-th nectar source search for a new nectar source x′ according to the update formula. id i = 1, 2, ..., SN, where SN is the total number of foraging bees, φ id Let x be a random number in the interval [-1, 1]. id Let x represent the value of the i-th nectar source, i ≠ k; the scout bee searches for a new solution x. id r is a random number in the interval [0,1]. and Let be the lower and upper bounds of the d-th dimension;

[0071] Use the updated population as the initial population for the first industry chain and return to step 2.1.

[0072] In one specific embodiment, step 2.2, returning to step 2.1 specifically includes: R2 detecting whether all individuals in the updated initial population of the first industry quantity have completed crossover and mutation, and after the update is completed, sending a reset signal to R1. At this time, the value of R1 is 1, and continuing to step 2.1, using the updated population to calculate a new round of benefit values.

[0073] In one specific embodiment, step 2.3 specifically includes: the model determines that the artificial bee colony algorithm meets the termination condition, the initial population of the second industry chain is used as the sample of BP network B, the system uses the optimal initial weights and thresholds obtained by BP network A as the initial weights and thresholds of BP network B, and BP network B is trained on the sample of BP network B after receiving the data input.

[0074] In one specific embodiment, step 3 includes: sorting the individuals in the population according to their evaluation values ​​and feeding the sorted population back to the optimization system; the optimization system, after receiving the population from the BP network B, performs mutation and crossover operations using a clonal selection algorithm.

[0075] In one specific embodiment, the initial population of the industry chain is collected, such as... Figure 2 As shown, the structure of the chip industry chain is simulated, in which R4 chip design, R5 chip lithography, R13 wafer processing, R21 packaging and testing, and R24 market distribution are the main components of the chain. Figure 2 Each directional line in the diagram represents the production relationship between groups and the transfer of production materials; that is, the production of a product under a given supply chain relationship requires related production nodes as prerequisites. Each node in the diagram has multiple enterprise options, the number of which is represented by the number of innermost circles. The node connections under the upstream, midstream, and downstream relationships of the supply chain are considered as a complex network. Each node enterprise in the supply chain is independent of the others. The complex network topology diagram is as follows: Figure 3 As shown, the main nodes in the industrial chain are marked. The size of the circles in the figure reflects the importance of each node, and the number of the innermost circles reflects the number of substitutable enterprises.

[0076] In one specific embodiment, the TOPSIS or DEA method in supply chain or industrial chain management is one of the methods for comprehensive evaluation. When evaluating the benefits of the industrial chain, the TOPSIS model based on the entropy weight method is used to achieve a comprehensive evaluation of the dual-carbon industrial chain, and this is used as the output index of the BP neural network.

[0077] The manufacturing supply chain under dual-carbon architecture aims to maximize the economic benefits of the entire chain while minimizing carbon emission costs. Based on previous research experience and combined with important metrics in complex networks, such as average path length l and clustering coefficient Ci, the overall benefit function T of the manufacturing supply chain under dual-carbon architecture is given as follows:

[0078] T = F(p) i ,b i ,l,C,n) (1)

[0079] In the formula, p i Let b represent the economic benefit value of the i-th unit in the industrial chain. i Let p represent the carbon emission cost of the i-th unit in the industrial chain, l represent the average path length of the industrial chain, C represent the average node importance of the entire industrial chain, reflecting the stability of the industrial chain system, and n represent the number of node enterprises in the industrial chain. Therefore, the industrial chain benefit value can be regarded as p i ,b i Functions of l, C, n.

[0080] In one specific embodiment, such as Figure 4 As shown, the dual-carbon industrial chain collaborative optimization model (BP-CS model) is a network model based on the fusion of BP neural network, improved artificial bee colony algorithm, complex network, and clonal selection algorithm. This model overcomes the interference of initial weight threshold on the training of traditional BP neural network and the problem that it can only achieve local optima rather than global optima. It also overcomes the problems of low training accuracy and poor mapping effect of traditional BP neural network. The combined algorithm automatically processes and filters the features in the dataset, solves the problems of high correlation and low influence in the dataset, and finally achieves overall evaluation and optimization of the industrial chain.

[0081] Among them, the model takes the set of economic benefits and carbon emissions of enterprises at each node of a given industrial chain and the parameters of each node in a complex network as sample inputs, establishes a relevant BP neural network for this sample, uses an improved artificial bee colony algorithm to optimize the initial weights and thresholds of the BP neural network, calculates the minimum network error through simulation, and based on the feedback of the termination node C', selectively feeds back the optimal weights, thresholds and related errors of the neural network to subsequent steps.

[0082] The steps of the BP-CS model are as follows:

[0083] Step 1: System initialization and start receiving samples. After the sample input layer receives the samples passed in from the outside world, it is initialized. The initial values of the progress nodes R, R1, and R2 are all 1. R1 counts for the BP network A, R2 counts for the BP network B, R counts for the number of loops of the clonal selection algorithm, and the sample set is input to the network input layer EnterA;

[0084] Step 2: After the system receives the samples passed in from the network input layer, the first individual in the population is substituted into the BP network A, and the weights and thresholds of the hidden layer and output layer of the network A are initially set. Then, start training the first individual received by the network input layer EnterA once. When the training is completed, calculate the network error e1 at this time, and correct the weights and thresholds of the neural network A based on the partial derivatives of each neuron in the output layer with respect to the network error e1;

[0085] Step 3: Judge whether the value of the signal R1 is equal to the population size u. When R1 < u, R2 transmits a working signal to R1, R1 = R1 + 1, and substitute the next individual in the population into the network A to calculate the network error e1 at this time, and optimize the weights and thresholds at this time according to the calculation results; when R1 = u, it indicates that the error calculation of each individual in the population has been completed, and at this time, execute Step 4;

[0086] Step 4: The system judges whether the result of the improved artificial bee colony algorithm meets the termination condition of 200 iterations. If the termination condition is not met, execute Step 5; otherwise, execute Step 7;

[0087] Step 5: Update the relevant nectar sources for all individuals in the original population. The foraging bees and scout bees are carried out according to the following formulas respectively:

[0088] x′ id = x id + φ id (x id - x kd )

[0089]

[0090] Artificial bee colony algorithms employ three types of bees: foraging bees, follower bees, and scout bees. Each nectar source represents a possible solution to the problem, and the amount of nectar from a source corresponds to the fitness of that solution. Each foraging bee is associated with a specific nectar source. Foraging and follower bees always seek new nectar sources near known feasible solutions, while scout bees randomly roam the entire space to explore for new nectar sources.

[0091] The standard artificial bee colony algorithm treats the solution process of the optimization problem as a search in a D-dimensional search space. In the formula, the foraging bees corresponding to the i-th nectar source search for a new nectar source x′ using an update formula. id Where i = 1, 2, ..., SN, SN is the total number of foraging bees, φ id It is a random number in the interval [-1, 1], x id Let x represent the value of the i-th nectar source, i ≠ k; the scout bee searches for new possible solutions. id Where r is a random number in the interval [0,1]. and It is the lower and upper bound of the d-th dimension.

[0092] Step 6: R2 checks whether all individuals in the population have completed crossover and mutation updates, and sends a reset signal to R1 after the update is complete. At this time, the value of R1 is 1, and Step 2 is continued to use the updated population to calculate a new round of benefit values.

[0093] Step 7: The model determines that the artificial bee colony algorithm meets the termination condition. The system uses the optimal initial weights and thresholds obtained from neural network A as the initial weights and thresholds of BP network B. After receiving the data input, network B first performs 600 training iterations on the sample set of neural network B. After training is complete, each individual in the population is substituted into the BP neural network for evaluation. Finally, the population individuals are sorted according to the evaluation values ​​and fed back to the optimization subsystem.

[0094] Step 8: After the optimization system receives the population transmitted by neural network B, it performs mutation and crossover operations, and substitutes the improved individuals into neural network B for evaluation, and determines whether the evaluation value after substitution has achieved optimization.

[0095] In the above steps, since BP network A participates in the improved artificial bee colony algorithm, and its main role is to train network B with suitable initial weights and thresholds, network A does not need to perform a large amount of repetitive calculations. To ensure the efficiency of the algorithm, BP network A only performs 30 iterations during training. After the improved artificial bee colony algorithm optimization is completed, the network error needs to be accurately calculated. Therefore, BP network B is used for 600 training iterations to obtain the final weights and thresholds, and the optimization results are obtained.

[0096] This system uses a backpropagation (BP) neural network with three important types of neurons: an input layer (EnterA) with n nodes, a hidden layer (V) with s nodes, and an output layer (W) with h nodes. The input, hidden, and output layers constitute a three-layer BP neural network. The model contains two BP networks: BP network A and BP network B.

[0097] Each node in the EnterA receiver layer corresponds to a neuron in the input layer, and its total input set is EnterA.

[0098] EnterA = {X i}i=1,2,···n (2)

[0099] The network error E during the training of a BP neural network is expressed as:

[0100]

[0101] In the above formula, p represents the total number of samples, r represents the number of neurons in the output layer, and represents the expected output value corresponding to the pi-th sample in the j-th component dimension. Let f represent the output value of the j-th neuron in the corresponding output layer after the pi-th sample is input and trained. v Let y represent the hidden layer transfer function. t The output of hidden layer neuron t is represented by f. w Let c represent the transfer function of the output layer neuron. k Let k represent the k-th input to the input layer. Then, expanding equation (3) to the hidden layer according to the principle of the BP neural network, we can obtain:

[0102]

[0103] Expanding equation (7) to the input layer, we get:

[0104]

[0105] in:

[0106]

[0107] In equation (5), w j,t The value of w varies with the value of t. When t takes a value between 1 and s, w j,t The weights of the t-th neuron in the hidden layer corresponding to the j-th neuron in the output layer of the neural network, when t is 0, w j,t The threshold corresponding to the j-th neuron in the output layer; similarly, v t,k The value of v changes with the value of k. When k takes a number between 1 and m, vt,k This represents the weight of the k-th neuron in the input layer corresponding to the t-th neuron in the hidden layer. When k is 0, v t,k This represents the threshold of the first neuron in the hidden layer.

[0108] From equations (4), (5), and (6), it can be seen that the network error E of the neural network is related to the input sample set EnterA, the expected output set, the weights of each layer in the network, and the threshold w. j,t v t,k All of these are related. The BP neural network continuously optimizes and adjusts the weights and thresholds of each layer by using different sample inputs, thereby continuously reducing the network error E. This process is achieved using the gradient descent algorithm. This algorithm is prone to getting trapped in local optima and cannot achieve global optima. Therefore, the network error obtained by this algorithm is different for different initial weights and thresholds.

[0109] Since this model is designed to train on an existing sample input set to predict subsequent data, it calculates the network error using a BP neural network built from sample genes. This network error is then used to improve the threshold and weights of the neural network. Therefore, the desired outcome is that the final trained network error E is only affected by the sample data. However, as the above analysis clearly shows, the initial weights and thresholds in each layer of the BP neural network all influence the network error. Therefore, to minimize the impact of initial weights and thresholds on the network error, a two-layer nested BP neural network is used. Neural network A is combined with an improved artificial bee colony algorithm to optimize the initial weights and thresholds of BP neural network B. Neural network B, based on the optimization by neural network A and the improved artificial bee colony algorithm, uses the optimal weights and thresholds obtained from training neural network A as its initial values ​​and undergoes another round of training, thereby achieving accurate prediction of the sample set.

[0110] The BP neural network B transmits the population after multiple training and sorting to the optimization system. The industrial chain set P(G) is selected and optimized from the perspectives of economic benefits and carbon emissions after multiple generations of screening and optimization by the clonal selection algorithm.

[0111] The population of the clonal selection algorithm in the subsystem is (p1,...p r ...,p u There are u particles in total, of which particle p r for Represents the t-th node {U} in the dual-carbon industrial chain. t C t The set of}, U t C represents the industrial benefits of node t. t This represents the carbon emissions at node t.

[0112] Furthermore, the model of the cloning selection algorithm is described as follows:

[0113] Step 1: Let g be an algebra, g = 0. Initialize the population P(G).

[0114] Step 2: (1) Calculate the evaluation value of each antibody in P(G) using Eq. (2) Sort the antibodies in ascending order according to their evaluation values.

[0115] Step 3: The cloned population P(G) is cloned to obtain P(G1). P(G) is randomly crossed with P(G1), and random genes of random antibodies in the population are exchanged to obtain Pc(G).

[0116] Step 4: Each clone in Pc(G) undergoes mutation. The mutation rate of each antibody is inversely proportional to its evaluation value; that is, the higher the evaluation value of each antibody, the lower the mutation rate. The mutation yields a new population P(G)m.

[0117] Step 5: Select antibodies with higher evaluation values ​​to form a population S(G) from P(G) and Pm(G).

[0118] Step 6: Add antibodies with lower evaluation values ​​from P(G) to S(G) with low probability. Use S(G) as the new antibody population P(G) and return to step 1 until the required number of generations for the population is met.

[0119] In the aforementioned methodological study, a supply chain network of 27 nodes was constructed. Each network node has multiple enterprise options, represented by the number of innermost circles. Each option has different economic benefits (in dollars), carbon emissions (in tons), and distances to neighboring nodes (in kilometers). A supply chain benefit verification analysis was conducted on 215 enterprise nodes in 73 cities. Data was sourced from publicly available data from some enterprises and average data from various industries. The verification experiment was conducted from both the macro-level overall benefit of the supply chain and the micro-level benefit of individual enterprise nodes. This method acquired 2213 sets of sample data from the chip supply chain. Missing values ​​and duplicate samples were removed from these sets, leaving 2124 valid sets. To control the modeling training time and ensure the representativeness of the sample data, stratified sampling was used to collect 1200 sets of supply chain sample data, which were then divided into two groups of 600 samples each. These were used for BP neural network A and BP neural network B, respectively, with 100 sets of data from network B used for verification.

[0120] The BP-CS model is a fusion of the clonal selection algorithm and the BP neural network. Its parameters are as follows: the initial population has 300 individuals and 600 generations; the BP neural network has 1 hidden layer, and after verification, it was determined that the number of hidden layer nodes is 3 and the number of output layer nodes is 1. At the same time, the Sigmoid function and the linear function are used as the transfer functions between the hidden layer and the output layer neurons. No test set is set in BP network A, and the ratio of training set to test set is set to 5:1 in BP network B. In BP-CS, the number of generations selected for training BP network A is 30, and the number of generations selected for training network B is 600.

[0121] The validation results of neural network B in the model using the test sample data show that the actual output value is basically consistent with the test value, with small error and high accuracy. Therefore, it can be concluded that the BP-CS collaborative optimization model can complete the comprehensive evaluation of the collaborative driving level of the industrial chain and supply chain, and has good effect and accuracy.

[0122] The clonal selection algorithm is used to optimize a given chip industry chain. Mutation rate, crossover probability, and the overall industry chain benefit F are negatively correlated. Simultaneously, the improvement probability of a node is also negatively correlated with the importance of nodes in the complex network model, ensuring that nodes playing crucial roles in the industry chain cannot be replaced. In the mutation operation, raw material company A in the chip industry chain increases costs to reduce carbon emissions, assuming that all related inputs are converted into corresponding carbon emissions based on carbon emission trading market prices. In the crossover operation, wafer packaging and testing segments within the entire industry chain are replaced with wafer packaging and testing segments from other chains. After 200 population screenings, the optimized results are compared with the original data. Figure 5 As shown, the overall improvement rate is approximately 2.09%.

[0123] The optimized radar charts for each indicator are as follows: Figure 6 As shown, it is evident that the average economic benefit, average carbon emission benefit, and average node importance have all improved. A shorter average path indicates higher overall efficiency, while the average path intersection rate decreased compared to before the improvement. The following lists the explanations for the changes in the four variables: Figure 7-1 The paper presents a box plot comparing the Economic Benefits index. Compared to the value before the improvement, the index has improved to a certain extent, which reflects the important position of economic benefits in the overall index and is consistent with the dependence of the entire chip industry chain on economic benefits. Figure 7-2 The chart provides a box plot comparing the carbon emission index, clearly showing that the carbon emission index did not improve significantly after optimization, and in fact, some optimized values ​​decreased in actual data comparisons. This demonstrates that from the perspective of the entire industry chain, allowing some companies to exceed carbon emission limits actually optimizes the overall efficiency of the entire chain. Figure 7-3 The box plot of the Average Importance metric is provided, showing that the median remains largely unchanged, while the mean has slightly increased. The average node importance is calculated using the LinkRank algorithm based on the individual importance of each node. The increased node importance after optimization indicates that the optimization model, when optimizing the entire industry chain, aligns the chain towards more reliable nodes, thus helping to improve the overall importance of the industry chain. Figure 7-4 Box plots of the path length metric are presented. The median and mean of the metric have both decreased, which is consistent with the initial setting of the average path length. That is, the average path length reflects the overall feedback efficiency of the industry chain to a certain extent. The shorter the average path, the shorter the reaction time of adjacent nodes in the industry chain. The optimized chip industry chain is consistent with the setting of the BP-CS model to optimize the average path length in the direction of smaller values.

[0124] In summary, this invention's analysis of the economic benefits and carbon emissions of the industrial chain from a holistic perspective is essential. The BP-CS model proposed in this paper, which evaluates and optimizes the entire industrial chain, can effectively assess and optimize "dual-carbon" industrial chains. Under stable industrial chain structures, optimizing the combination of enterprises at each node can effectively achieve overall industrial chain optimization, further confirming the necessity of implementing a holistic strategy.

[0125] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0126] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for optimizing a manufacturing industry chain based on a double carbon target, characterized in that, include: Step 1: Collect data sets on the economic benefits and carbon emissions of enterprises at each node of the industrial chain, as well as the parameters of each node in the complex network, as samples; Collect the initial population of the industrial chain; Step 2: Establish a dual-carbon industrial chain collaborative optimization model; bring the samples and the initial population of the industrial chain into the dual-carbon industrial chain collaborative optimization model, wherein the dual-carbon industrial chain collaborative optimization model uses an improved artificial bee colony algorithm to optimize the initial weights and thresholds of the BP neural network, and based on the feedback of the termination node, selectively feeds back the optimal weights, thresholds and related errors of the neural network to the next step to complete the evaluation of the initial population of the industrial chain. Step 2 specifically includes: Step 2.1: Introduce each individual in the initial population of the first industrial chain into the BP network A in the dual-carbon industrial chain collaborative optimization model, set the initial weights and thresholds of the BP network A, and obtain the corrected weights and thresholds of each individual through sample training; Step 2.2: After all individuals in the initial population of the first industry chain have completed the calculation, determine whether the result meets the termination condition. If it does, proceed to step 2.

3. If it does not, update the individuals in the initial population of the first industry chain, use the updated population as the initial population of the first industry chain, and return to step 2.

1. Step 2.2 specifically includes: using an artificial bee colony algorithm to determine whether the error calculation result meets the termination condition. If it does, proceed to step 2.3; otherwise, update the relevant nectar sources for individuals in the initial population of the first industry chain using the following formula. Foraging bees and scout bees, update the relevant nectar sources using the following formula respectively: Where the honeybee corresponding to the ith honey source searches for a new honey source according to the update formula is a random number on the interval [-1, 1], represents the value of the ith honey source, i≠k; the scout bee searches for a new solution , r is a random number on the interval [0, 1], and are the lower and upper bounds of the dth dimension; Use the updated population as the initial population for the first industry chain, and return to step 2.1; Step 2.3: Use the initial population of the second industrial chain as the sample of BP network B, and use the optimal weights and thresholds obtained by BP network A as the initial weights and thresholds of BP network B to train the sample of BP network B. Step 3: Input the evaluated initial population of the industry chain into the optimization system and combine it with the clonal selection algorithm to optimize the industry chain.

2. The manufacturing industry chain optimization method based on the double carbon target according to claim 1, wherein Step 1 specifically includes: The parameters of each node in a complex network include: average path length and node importance ranking; The average path is represented as: ; where the distance between two nodes is defined as the sum of the distances of the edges on the shortest path between the two nodes and where the distance between two nodes is defined as the sum of the distances of the edges on the shortest path between the two nodes and the average distance between any two nodes is defined as the average path length l of the network, where is the number of nodes in the network. The ranking of node importance is expressed as follows: ; in, Represents a node exist Moment The value c represents the probability of a random jump. Indicates and Connected nodes Economic benefits, assuming Nodes point to the same Nodes, if node Pointing to node but The value is 1 if it is set to 1, and 0 otherwise. The initial population of the industrial chain is divided into the first initial population of the industrial chain and the second initial population of the industrial chain.

3. The manufacturing supply chain optimization method based on dual carbon objectives according to claim 1, characterized in that, Step 2.1 also includes setting progress nodes. , , The initial value of each progress node is 1. The model is initialized after receiving a sample. Counting is performed on BP network A. Counting is performed on BP network B. The number of iterations for the clone selection algorithm is counted, and the sample is input into the network input layer EnterA.

4. The manufacturing supply chain optimization method based on dual carbon objectives according to claim 1, characterized in that, Step 2.1, specifically the sample training, involves training the network once using the first individual received by the network input layer EnterA with samples, and calculating the network error at this point after training is complete. and with network error The weights and thresholds of the BP network A are adjusted based on the partial derivatives of each neuron in the output layer.

5. The manufacturing supply chain optimization method based on dual carbon targets according to claim 1, characterized in that, Step 2.1, which derives the corrected weights and thresholds for each individual, specifically includes: judging the signal. Is the value equal to the population size? ,when hour, Then, the next individual in the group is substituted into the BP network A to calculate the network error at this time. And optimize the weights and thresholds at this point based on the calculation results; when This indicates that the error calculation for each individual in the group has been completed.

6. The manufacturing supply chain optimization method based on dual carbon targets according to claim 3, characterized in that, In step 2.2, returning to step 2.1 specifically includes: The system checks whether all individuals in the initial population of the first industrial chain have completed crossover and mutation updates, and then sends data to [the relevant authorities] after the update is complete. Send a reset signal at this time The value is 1, and we continue to step 2.1 to calculate a new round of benefit values ​​using the updated population.

7. The manufacturing supply chain optimization method based on dual carbon objectives according to claim 1, characterized in that, Step 2.3 specifically includes: the model determines that the artificial bee colony algorithm meets the termination condition, the initial population of the second industry chain is used as the sample of BP network B, the system uses the optimal initial weights and thresholds obtained by BP network A as the initial weights and thresholds of BP network B, and BP network B is trained on the sample of BP network B after receiving the data input.

8. The manufacturing supply chain optimization method based on dual carbon targets according to claim 7, characterized in that, Step 3 includes: sorting the individuals in the population according to their evaluation values ​​and feeding the sorted population back to the optimization system; the optimization system then receives the population from the BP network B and performs mutation and crossover operations using the clonal selection algorithm.

Images