Supply chain management method

Abstract

The invention discloses a supply chain management method, and belongs to the technical field of supply chain management, and the method comprises the steps: collecting a pre-stored supply chain management recognition task or a supply chain management recognition task inputted by a worker through human-computer interaction, enabling the worker to achieve a customized supply chain management strategy, thereby assisting the worker to improve the application flexibility, and improving the management efficiency. Then training a target deep learning model by adopting data specified in the supply chain management identification task, identifying the supply chain real-time data through the trained target deep learning model, and determining a supply chain management tag corresponding to the supply chain real-time data; finally, the automatic management of the supply chain can be realized according to the supply chain management label corresponding to the supply chain real-time data, and workers can be effectively assisted to improve the management efficiency and the management accuracy of the supply chain.

Description

Technical Field

[0001] The present invention belongs to the technical field of supply chain management, and particularly relates to a supply chain management method. Background Art

[0002] With the rapid development of the global economy, supply chain management plays an increasingly important role in various industries. Supply chain management is a systematic management method aimed at effectively integrating the resources and activities of each link in the supply chain to achieve the efficient and low-cost flow of products from raw materials to the final consumer. It includes multiple links such as demand forecasting, procurement, production, inventory, transportation, and distribution. By sharing information, collaborating, and making optimized decisions, the overall cost is reduced and customer satisfaction is improved. With technological progress, digitalization and intelligentization have become important trends in supply chain management, such as using big data, cloud computing, blockchain, etc. to enhance transparency and traceability. Effective supply chain management is the key for modern enterprises to enhance competitiveness and respond to market changes.

[0003] Traditional risk identification methods mainly rely on manual experience and statistical analysis, and have problems such as low identification efficiency, poor accuracy, and difficulty in dealing with sudden risks. Therefore, there is an urgent need for an efficient and accurate supply chain risk identification method. Summary of the Invention

[0004] The present invention provides a supply chain management method to solve the problems of low identification efficiency and poor accuracy caused by relying on manual experience and statistical analysis in the prior art.

[0005] A supply chain management method includes: Collecting a supply chain management identification task stored in advance or input by a staff member through human-computer interaction; wherein, the supply chain management identification task includes supply chain historical data and a supply chain management label corresponding to the supply chain historical data; Obtaining a deep learning model scheduling instruction input by a staff member through human-computer interaction, and in response to the deep learning model scheduling instruction, selecting a target deep learning model from a model library; Training the target deep learning model according to the supply chain historical data and the supply chain management label corresponding to the supply chain historical data in the supply chain management identification task by using a multi-mechanism parameter optimization algorithm, and obtaining the trained target deep learning model; Collecting real-time supply chain data to be identified, and scheduling the trained target deep learning model to identify the real-time supply chain data, and determining a supply chain management label corresponding to the real-time supply chain data; Based on the supply chain management label corresponding to the real-time supply chain data, performing supply chain management by using a preset supply chain management rule.

[0006] In a possible implementation, the supply chain historical data is image data or text data.

[0007] In a possible implementation, the model library includes one or more of LeNet model, AlexNet model, VGGNet model, ResNet model, LSTM model, GRU model, and BP Net model.

[0008] In a possible implementation, according to the supply chain historical data in the supply chain management recognition task and the supply chain management label corresponding to the supply chain historical data, a multi-mechanism parameter optimization algorithm is used to train the target deep learning model to obtain the trained target deep learning model, including: Based on the hyperparameters of the target deep learning model, use the Logistic chaotic mapping sequence to initialize the population to obtain multiple individuals; According to the supply chain historical data in the supply chain management recognition task and the supply chain management label corresponding to the supply chain historical data, obtain the fitness of each individual, and determine the current optimal individual according to the fitness of the individual; For any individual, based on the current optimal individual, use the spiral position selection mechanism to select the exploration area for the individual to obtain the individual after the exploration area selection; For any individual after the exploration area selection, use the spiral search mechanism with search boundary limitation to perform neighborhood exploration on the individual to obtain the individual after the neighborhood exploration; For any individual after the neighborhood exploration, based on the current optimal individual, use the correlation adaptive search mechanism to perform balanced search on the individual to obtain the individual after the balanced search; Judge whether the training end condition is met. If so, determine the target optimal individual according to the individual after the balanced search, and obtain the trained target deep learning model according to the target current optimal individual. Otherwise, return to the step of determining the current optimal individual.

[0009] In a possible implementation, based on the hyperparameters of the target deep learning model, use the Logistic chaotic mapping sequence to initialize the population to obtain multiple individuals, including: Perform random initialization between the upper and lower limits corresponding to the hyperparameters of the target deep learning model, and form a vector with the hyperparameters after random initialization to obtain the initial individual; Based on the initial individual, obtain multiple individuals as:

[0010] Wherein, represents the i th individual, andi When = 1, represents the initial individual, represents the i +1th individual, represents the chaos coefficient set between [0, 4].

[0011] In a possible implementation manner, according to the supply chain historical data in the supply chain management recognition task and the supply chain management labels corresponding to the supply chain historical data, the fitness of each individual is obtained, and the current optimal individual is determined according to the fitness of all individuals, including: For any individual, apply the hyperparameters included in the individual to the target deep learning model, and use the supply chain historical data in the supply chain management recognition task as the input and the supply chain management label corresponding to the supply chain historical data as the expected output to obtain the cross-entropy loss function value corresponding to the individual; After adding the cross-entropy loss function value corresponding to the individual to the preset constant term, obtain the intermediate parameter value; wherein, the preset constant term is less than or equal to 0.001; Take the reciprocal of the intermediate parameter value to obtain the fitness corresponding to the individual; According to the fitness of all individuals, determine the individual with the maximum fitness as the optimal individual.

[0012] In a possible implementation manner, for any individual, based on the current optimal individual, use the spiral position selection mechanism to select the exploration area for the individual to obtain the individual after the exploration area selection, including:

[0013] Among them, represents the optimal individual, represents the kth individual in the tth training process, k = 1, 2,..., K, and K represents the total number of individuals, represents the individual after the kth exploration area selection, represents the natural constant, represents a random number between (0, 1), represents the pi, represents the cosine function.

[0014] In a possible implementation manner, for any individual after the exploration area selection, use the spiral search mechanism with search boundary limitation to perform neighborhood exploration on the individual to obtain the individual after the neighborhood exploration, including: For any individual after the exploration area selection, obtain the neighborhood exploration boundary as:

[0015]

[0016] in, Indicates t During the training process m The d-th dimension hyperparameter of the individual after the exploration area selection, d = 1, 2, ..., D, D represents the total dimension of the hyperparameters contained in the individual, Indicates d The lower bound of the dimensional hyperparameter, represents the upper limit of the hyperparameter of the dth dimension, represents the neighborhood exploration boundary control factor, and , Indicates the preset maximum number of training times. Represents an individual The lower limit of the neighborhood exploration boundary corresponding to the d-th dimension hyperparameter, Represents an individual The upper limit of the neighborhood exploration boundary corresponding to the d-th dimension hyperparameter; According to the neighborhood exploration boundary, the individual is subjected to neighborhood exploration, and the individual after the neighborhood exploration is obtained as follows:

[0017] in, Indicates m After exploring the neighborhood, the individual d dimensional hyperparameters, represents a natural constant, Represents a random number between (0,1), represents pi, Represents the cosine function.

[0018] In a possible implementation, for any individual after neighborhood exploration, based on the current optimal individual, a correlation adaptive search mechanism is used to perform a balanced search on the individual to obtain the individual after the balanced search, including: For any individual after neighborhood exploration, a target individual is randomly matched for the individual after neighborhood search, and the similarity between the individual and the target individual is obtained as follows:

[0019] in, Indicates n The similarity between the individuals after neighborhood exploration and their corresponding target individuals, Indicates the tth training process n After exploring the neighborhood, the individual d dimensional hyperparameters, Indicates the tth training process n The target individual corresponding to the individual after the neighborhood explorationd The dimensional hyperparameter, where D represents the total dimension of the hyperparameters; When the similarity is less than the preset threshold, the balanced search for the individual after the n -th neighborhood exploration is:

[0020] where, represents the lower limit of the d -th dimensional hyperparameter, represents the upper limit of the d-th dimensional hyperparameter, represents a random number between (0, 1), represents the n -th dimensional hyperparameter of the individual after the d -th balanced search; When the similarity is greater than or equal to the preset threshold, the balanced search for the individual after the n -th neighborhood exploration is:

[0021]

[0022] where, represents the individual after the n -th neighborhood exploration during the t-th training process, represents the individual after the n -th balanced search, represents the adaptive interaction factor, represents the target individual corresponding to the individual after the n -th neighborhood exploration, represents a random number between (0, 1), represents the optimal individual.

[0023] In a possible implementation manner, based on the supply chain management label corresponding to the supply chain real-time data, supply chain management is performed by using preset supply chain management rules, including: In the preset supply chain management rules, query the automatic management operation corresponding to the supply chain management label of the pre-stored supply chain real-time data, and execute the automatic management operation to implement supply chain management.

[0024] A supply chain management method provided by the present invention first collects supply chain management recognition tasks pre-stored or input by staff through human-computer interaction, enabling staff to implement customized supply chain management strategies, thereby assisting staff in improving application flexibility. Then, the target deep learning model is trained using the specified data in the supply chain management recognition task, and the trained target deep learning model is used to identify the supply chain real-time data to determine the supply chain management label corresponding to the supply chain real-time data. Finally, the automated management of the supply chain can be realized according to the supply chain management label corresponding to the supply chain real-time data, which can effectively assist staff in improving supply chain management efficiency and management accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings herein are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present invention and used together with the specification to explain the principles of the present invention.

[0026] Figure 1 It is a schematic flowchart of a supply chain management method provided by an embodiment of the present invention.

[0027] Through the above accompanying drawings, specific embodiments of the present invention have been shown, and there will be more detailed descriptions hereinafter. These drawings and textual descriptions are not intended to limit the scope of the inventive concept in any way, but to illustrate the concept of the present invention to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0028] Exemplary embodiments will be described in detail herein, and examples thereof are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0029] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0030] As Figure 1 shown, an embodiment of the present invention provides a supply chain management method, including: S1. Collect supply chain management recognition tasks pre-stored or input by staff through human-computer interaction; wherein, the supply chain management recognition tasks include supply chain historical data and supply chain management labels corresponding to the supply chain historical data; The supply chain management method provided by the embodiments of the present invention supports staff to customize supply chain management recognition tasks, that is, input supply chain historical data and the corresponding supply chain management tags of the supply chain historical data. After training the deep learning model with the supply chain historical data and the corresponding supply chain management tags of the supply chain historical data, label custom recognition can be achieved, thereby assisting staff to recognize supply chain information and improving supply chain management efficiency.

[0031] For example, the supply chain historical data can be the image information corresponding to the supply chain products, and the supply chain management tags corresponding to the supply chain historical data can be the quality tags corresponding to the supply chain products (such as normal tags, surface scratch tags, crack tags, depression tags, etc.). Then, by setting this supply chain management recognition task, the product quality of the supply chain can be automatically detected, thereby assisting staff in the quality management of the supply chain.

[0032] However, it should be noted that the above supply chain historical data and the corresponding supply chain management tags of the supply chain historical data are the distances of the embodiments of the present invention, and other supply chain management recognition tasks can also be set. For example, various data of the supply chain are collected to identify supply chain risks.

[0033] S2. Obtain the deep learning model scheduling instruction input by the staff through human-computer interaction, and in response to the deep learning model scheduling instruction, select a target deep learning model from the model library; The model library is a library containing various deep learning models, which supports staff to select the required deep learning models by themselves, so as to achieve the recognition of various types of data. At the same time, deep learning models with different complexities can be applied to devices with different performances.

[0034] S3. According to the supply chain historical data and the corresponding supply chain management tags in the supply chain management recognition task, use the multi-mechanism parameter optimization algorithm to train the target deep learning model, and obtain the trained target deep learning model; In the prior art, during the training process of the deep learning model, it is easy to fall into local optimality, and the training effect of the algorithm is poor, which will cause the trained deep learning model to be unable to accurately recognize the supply chain management tags. Therefore, the embodiments of the present invention use the multi-mechanism parameter optimization algorithm to train the target deep learning model, improve the training effect and global training ability of the deep learning model, thereby improving the recognition accuracy of the supply chain management tags.

[0035] S4. Collect the real-time supply chain data to be recognized, and schedule the trained target deep learning model to recognize the real-time supply chain data, and determine the supply chain management tags corresponding to the real-time supply chain data; Optionally, the trained target deep learning model can be deployed in the cloud to utilize the powerful data processing capabilities of the cloud to achieve fast and large-scale data processing.

[0036] S5. Based on the supply chain management tags corresponding to the real-time supply chain data, perform supply chain management using preset supply chain management rules.

[0037] Optionally, corresponding automatic management operations can be set in advance for each supply chain management tag. After the supply chain management tag is recognized, the corresponding automatic management operation can be executed to achieve supply chain management.

[0038] In a possible implementation manner, the supply chain historical data is image data or text data. It should be noted that these two types of data are only examples of the embodiments of the present invention, and other data can also be used for supply chain management.

[0039] In a possible implementation manner, the model library includes one or more of the LeNet model (a classic convolutional neural network), AlexNet model (a classic convolutional neural network), VGGNet model (a deep convolutional neural network model), ResNet model (a residual neural network), LSTM model (a long short-term memory network), GRU model (an improved recurrent neural network), and BP Net (BP neural network) model.

[0040] It should be noted that the above deep learning model is only a preferred implementation manner of the embodiments of the present invention, and other deep learning models can also be used to construct the model library.

[0041] In a possible implementation manner, according to the supply chain historical data and the supply chain management tags corresponding to the supply chain historical data in the supply chain management recognition task, a multi-mechanism parameter optimization algorithm is used to train the target deep learning model to obtain the trained target deep learning model, including: Based on the hyperparameters of the target deep learning model, initialize the population using the Logistic chaotic mapping sequence to obtain multiple individuals; According to the supply chain historical data and the supply chain management tags corresponding to the supply chain historical data in the supply chain management recognition task, obtain the fitness of each individual, and determine the current optimal individual according to the fitness of the individual; For any individual, based on the current optimal individual, use the spiral position selection mechanism to select the exploration area for the individual to obtain the individual after the exploration area is selected; For any individual after exploration area selection, a spiral search mechanism with search boundary limitation is adopted to explore the neighborhood of the individual to obtain the individual after neighborhood exploration; For any individual after neighborhood exploration, based on the current optimal individual, a correlation adaptive search mechanism is adopted to perform balanced search on the individual to obtain the individual after balanced search; Determine whether the training end condition is satisfied. If so, based on the individual after balanced search, determine the target optimal individual, and obtain the target deep learning model after training according to the target current optimal individual. Otherwise, return to the step of determining the current optimal individual.

[0042] Optionally, after the spiral position selection mechanism, the spiral search mechanism with search boundary limitation, and the correlation adaptive search mechanism are executed, an out-of-limit processing method can also be adopted to process the out-of-limit hyperparameters. For example, when the hyperparameters are out of limit, they can be randomly generated between the upper limit and the lower limit.

[0043] In a possible implementation manner, based on the hyperparameters of the target deep learning model, a population is initialized using a Logistic chaotic mapping sequence to obtain multiple individuals, including: Randomly initialize between the upper limit and the lower limit corresponding to the hyperparameters of the target deep learning model, and form a vector with the hyperparameters after random initialization to obtain an initial individual; Based on the initial individual, the multiple individuals obtained are:

[0044] Among them, represents the i th individual, and when i = 1, represents the initial individual, represents the i + 1th individual, represents the chaotic coefficient set between [0, 4].

[0045] The embodiment of the present invention initializes the population using a Logistic chaotic mapping sequence, which can improve the distribution uniformity of the population at the initial moment, thereby enhancing the ability to find the global optimal solution and the training speed of the optimization algorithm.

[0046] In a possible implementation manner, according to the supply chain historical data in the supply chain management recognition task and the supply chain management labels corresponding to the supply chain historical data, the fitness of each individual is obtained, and the current optimal individual is determined according to the fitness of all individuals, including: For any individual, apply the hyperparameters included in the individual to the target deep learning model, and use the supply chain historical data in the supply chain management recognition task as the input, and the supply chain management label corresponding to the supply chain historical data as the expected output to obtain the cross-entropy loss function value corresponding to the individual; After adding the cross-entropy loss function value corresponding to the individual to a preset constant term, obtain an intermediate parameter value; wherein, the preset constant term is less than or equal to 0.001; Take the reciprocal of the intermediate parameter value to obtain the fitness corresponding to the individual; According to the fitness of all individuals, determine the individual with the largest fitness as the optimal individual.

[0047] In the embodiment of the present invention, the larger the fitness, the better the position of the individual in the solution space. Therefore, the individual with the largest fitness is the optimal individual.

[0048] In a possible implementation manner, for any individual, based on the current optimal individual, use a spiral position selection mechanism to select an exploration area for the individual, and obtain the individual after the exploration area selection, including:

[0049] Wherein, represents the optimal individual, represents the kth individual in the tth training process, k = 1, 2,..., K, and K represents the total number of individuals, represents the individual after the kth exploration area selection, represents the natural constant, represents a random number between (0, 1), represents the pi, represents the cosine function.

[0050] The spiral position selection mechanism provided by the embodiment of the present invention can enable the individual to select its respective search area in a divergent manner based on the position of the optimal individual, and at the same time applies a spiral search mechanism, which can improve the ability of the algorithm to jump out of the local optimum, and is provided with an exponential function that decreases non-linearly with the number of training times, which can gradually narrow the overall search range in the later stage of the algorithm and improve the search accuracy.

[0051] In a possible implementation manner, for any individual after the exploration area selection, use a spiral search mechanism with search boundary limitation to perform neighborhood exploration on the individual, and obtain the individual after the neighborhood exploration, including: For any individual after the exploration area selection, obtain the neighborhood exploration boundary as:

[0052]

[0053] Among them, represents the d-th dimensional hyperparameter of the individual after the selection of the t -th exploration area in the m -th training process, where d = 1, 2, …, D, and D represents the total dimension of the hyperparameters included in the individual. represents the lower limit of the d -th dimensional hyperparameter. represents the upper limit of the d-th dimensional hyperparameter. represents the neighborhood exploration boundary control factor, and , represents the preset maximum number of training times. represents the lower limit of the neighborhood exploration boundary corresponding to the d-th dimensional hyperparameter of the individual . represents the upper limit of the neighborhood exploration boundary corresponding to the d-th dimensional hyperparameter of the individual ; According to the neighborhood exploration boundary, perform neighborhood exploration on the individual to obtain the individual after neighborhood exploration as:

[0054] Among them, represents the m -th dimensional hyperparameter of the individual after the d -th neighborhood exploration. represents the natural constant. represents a random number between (0, 1). represents pi. represents the cosine function.

[0055] The spiral search mechanism with search boundary limitation provided by the embodiments of the present invention can enable each individual to perform neighborhood search around its own position, and at the same time, the search path is an irregular curve, which can effectively implement neighborhood search and improve the ability to find the global optimal solution.

[0056] In a possible implementation manner, for any individual after neighborhood exploration, based on the current optimal individual, a correlation adaptive search mechanism is used to perform balance search on the individual to obtain the individual after balance search, including: For any individual after neighborhood exploration, randomly match a target individual for the individual after neighborhood search, and obtain the similarity between the individual and the target individual as:

[0057] Among them, represents the similarity between the n -th individual after neighborhood exploration and its corresponding target individual. Denote the n d-dimensional hyperparameter of the individual after the d t-th neighborhood exploration during the t-th training process, Denote the n d-dimensional hyperparameter of the target individual corresponding to the individual after the d t-th neighborhood exploration during the t-th training process, where D represents the total dimension of the hyperparameters; When the similarity is less than the preset threshold, then the balance search for the individual after the n t-th neighborhood exploration is:

[0058] wherein, denotes the lower limit of the d d-th dimensional hyperparameter, denotes the upper limit of the d-th dimensional hyperparameter, denotes a random number between (0, 1), denotes the n d-dimensional hyperparameter of the individual after the d t-th balance search; When the similarity is greater than or equal to the preset threshold, then the balance search for the individual after the n t-th neighborhood exploration is:

[0059]

[0060] wherein, denotes the individual after the n t-th neighborhood exploration during the t-th training process, denotes the individual after the n t-th balance search, denotes the adaptive interaction factor, denotes the n target individual corresponding to the individual after the t-th neighborhood exploration, denotes a random number between (0, 1),

[0061] The correlation adaptive search mechanism provided by the embodiments of the present invention can separate and search for relatively similar individuals, while performing collaborative search for dissimilar individuals, effectively balancing global search and local search, enhancing the ability of the algorithm to jump out of the global optimum, and at the same time ensuring that more blank areas are searched, greatly improving the training effect of the algorithm.

[0062] Optionally, an annealing simulation algorithm can also be used to control the balance search process to improve the training speed of the algorithm.

[0063] Through the mutual cooperation and influence of the above several mechanisms, the embodiments of the present invention effectively solve the technical problems of the prior art being prone to falling into local optimality and having poor training accuracy, and finally improve the recognition accuracy of supply chain management labels.

[0064] In a possible implementation manner, based on the supply chain management label corresponding to the supply chain real-time data, supply chain management is performed by using a preset supply chain management rule, including: In the preset supply chain management rule, query the automatic management operation corresponding to the supply chain management label corresponding to the pre-stored supply chain real-time data, and execute the automatic management operation to implement supply chain management.

[0065] For example, when controlling the quality of supply chain products, it can be to detect the surface defects of supply chain products, and the supply chain management label corresponding to the supply chain real-time data should be the specific defect type. For different defect types, different automatic management operations can be set (such as data warehousing, giving warnings, calculating the product qualification rate, etc.). For example, in the process of quality management of supply chain products, the qualification rate of supply chain products can be counted or product quality warnings can be realized by identifying supply chain management labels.

[0066] A supply chain management method provided by the present invention first collects a supply chain management recognition task pre-stored or input by a staff through human-computer interaction, which can enable the staff to implement a customized supply chain management strategy, thereby assisting the staff to improve application flexibility. Then, the target deep learning model is trained with the data specified in the supply chain management recognition task, and the supply chain real-time data is recognized by the trained target deep learning model to determine the supply chain management label corresponding to the supply chain real-time data. Finally, the automatic management of the supply chain can be realized according to the supply chain management label corresponding to the supply chain real-time data, which can effectively assist the staff to improve the supply chain management efficiency and management accuracy.

[0067] After considering the specification and practicing the invention disclosed herein, those skilled in the art will readily think of other embodiments of the present invention. The present invention is intended to cover any variations, uses, or adaptations of the present invention, which follow the general principles of the present invention and include common general knowledge or conventional technical means in the technical field not disclosed in the present invention. It should be understood that the present invention is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present invention is only limited by the appended claims.

Claims

1. A supply chain management method, characterized in that, Including: Collecting supply chain management identification tasks that are pre-stored or input by staff through human-computer interaction; wherein, the supply chain management identification tasks include supply chain historical data and supply chain management labels corresponding to the supply chain historical data; Obtaining a deep learning model scheduling instruction input by staff through human-computer interaction, and in response to the deep learning model scheduling instruction, selecting a target deep learning model from a model library; Training the target deep learning model according to the supply chain historical data in the supply chain management identification task and the supply chain management label corresponding to the supply chain historical data by using a multi-mechanism parameter optimization algorithm, and obtaining the trained target deep learning model; Collecting real-time supply chain data to be identified, and scheduling the trained target deep learning model to identify the real-time supply chain data, and determining the supply chain management label corresponding to the real-time supply chain data; Based on the supply chain management label corresponding to the real-time supply chain data, performing supply chain management by using a preset supply chain management rule.

2. The supply chain management method according to claim 1, characterized in that The supply chain historical data is image data or text data.

3. The supply chain management method according to claim 1, wherein The model library includes one or more of a LeNet model, an AlexNet model, a VGGNet model, a ResNet model, an LSTM model, a GRU model, and a BP Net model.

4. The supply chain management method according to claim 1, characterized in that, Training the target deep learning model according to the supply chain historical data in the supply chain management identification task and the supply chain management label corresponding to the supply chain historical data by using a multi-mechanism parameter optimization algorithm, and obtaining the trained target deep learning model, including: Initializing a population by using a Logistic chaotic mapping sequence based on the hyperparameters of the target deep learning model to obtain multiple individuals; Obtaining the fitness of each individual according to the supply chain historical data in the supply chain management identification task and the supply chain management label corresponding to the supply chain historical data, and determining the current optimal individual according to the fitness of the individual; For any individual, based on the current optimal individual, using a spiral position selection mechanism to select an exploration area for the individual, and obtaining the individual after exploration area selection; For any individual after exploration area selection, using a spiral search mechanism with search boundary limitation to perform neighborhood exploration on the individual, and obtaining the individual after neighborhood exploration; For any individual after neighborhood exploration, based on the current optimal individual, using a correlation adaptive search mechanism to perform balance search on the individual, and obtaining the individual after balance search; Judging whether the training end condition is satisfied. If so, determining the target optimal individual according to the individual after balance search, and obtaining the trained target deep learning model according to the target current optimal individual. Otherwise, returning to the step of determining the current optimal individual.

5. The supply chain management method according to claim 4, wherein Initializing a population by using a Logistic chaotic mapping sequence based on the hyperparameters of the target deep learning model to obtain multiple individuals, including: Performing random initialization between the upper limit and the lower limit corresponding to the hyperparameters of the target deep learning model, and forming a vector with the hyperparameters after random initialization to obtain an initial individual; Based on the initial individual, obtaining multiple individuals as follows: Among them, represents the i th individual, and when i = 1, represents the initial individual, represents the i + 1th individual, represents the chaotic coefficient set between [0, 4].

6. The supply chain management method according to claim 4, wherein According to the supply chain historical data in the supply chain management identification task and the supply chain management labels corresponding to the supply chain historical data, obtaining the fitness of each individual, and determining the current optimal individual according to the fitness of all individuals, including: For any individual, applying the hyperparameters included in the individual to the target deep learning model, using the supply chain historical data in the supply chain management identification task as the input, and using the supply chain management label corresponding to the supply chain historical data as the expected output, to obtain the cross-entropy loss function value corresponding to the individual; After adding the cross-entropy loss function value corresponding to the individual to a preset constant term, obtaining an intermediate parameter value; wherein, the preset constant term is less than or equal to 0.001; Taking the reciprocal of the intermediate parameter value to obtain the fitness corresponding to the individual; According to the fitness of all individuals, determining the individual with the maximum fitness as the optimal individual.

7. The supply chain management method according to claim 4, characterized in that For any individual, based on the current optimal individual, using a spiral position selection mechanism to select an exploration area for the individual, obtaining the individual after exploration area selection, including: Among them, represents the optimal individual, represents the k-th individual in the t-th training process, where k = 1, 2, …, K, and K represents the total number of individuals, represents the individual after the selection of the k-th exploration area, represents the natural constant, represents a random number between (0, 1), represents the pi, represents the cosine function.

8. The supply chain management method according to claim 7, characterized in that For any individual after exploration area selection, using a spiral search mechanism with search boundary limitation to perform neighborhood exploration on the individual, obtaining the individual after neighborhood exploration, including: For any individual after exploration area selection, obtaining the neighborhood exploration boundary as: Among them, represents the d-th dimensional hyperparameter of the individual after the t -th exploration area selection during the m -th training process, where d = 1, 2, …, D, and D represents the total dimension of hyperparameters included in the individual. represents the lower limit of the d -th dimensional hyperparameter. represents the upper limit of the d-th dimensional hyperparameter. represents the neighborhood exploration boundary control factor, and . represents the preset maximum number of training times. represents the lower limit of the neighborhood exploration boundary corresponding to the d-th dimensional hyperparameter of the individual . represents the upper limit of the neighborhood exploration boundary corresponding to the d-th dimensional hyperparameter of the individual . According to the neighborhood exploration boundary, performing neighborhood exploration on the individual, and obtaining the individual after neighborhood exploration as: Among them, represents the m -th hyperparameter of the individual after the d -th neighborhood exploration, represents the natural constant, represents a random number between (0, 1), represents the pi, represents the cosine function.

9. The supply chain management method according to claim 8, wherein For any individual after neighborhood exploration, based on the current optimal individual, using a correlation adaptive search mechanism to perform balance search on the individual, obtaining the individual after balance search, including: For any individual after neighborhood exploration, randomly matching a target individual for the individual after neighborhood search, and obtaining the similarity between the individual and the target individual as: Among them, represents the similarity between the individual after the n -th neighborhood exploration and its corresponding target individual, represents the n -th dimension hyperparameter of the individual after the d -th neighborhood exploration during the t-th training process, represents the n -th dimension hyperparameter of the target individual corresponding to the individual after the d -th neighborhood exploration during the t-th training process, and D represents the total dimension of the hyperparameters; When the similarity is less than the preset threshold, the balance search for the individual after the n th neighborhood exploration is as follows: Among them, represents the lower limit of the d d-th dimensional hyperparameter, represents the upper limit of the d-th dimensional hyperparameter, represents a random number between (0, 1), represents the n d-th dimensional hyperparameter of the individual after d the d-th balanced search; When the similarity is greater than or equal to a preset threshold, the balanced search for the individual after the n -th neighborhood exploration is as follows: in, Indicates the tth training process n After exploring the neighborhood, Indicates n After the equilibrium search, represents the adaptive interaction factor, Indicates n The target individual corresponding to the individual after the neighborhood exploration, Represents a random number between (0,1), represents the optimal individual.

10. The supply chain management method according to claim 1, wherein Based on the supply chain management label corresponding to the supply chain real-time data, performing supply chain management using a preset supply chain management rule, including: In the preset supply chain management rule, querying the automatic management operation corresponding to the supply chain management label corresponding to the pre-stored supply chain real-time data, and executing the automatic management operation to implement supply chain management.

Images