Supply chain artificial intelligence financial risk decision-making system

Abstract

The invention discloses a supply chain artificial intelligence financial risk decision-making system, and belongs to the technical field of data processing. A learning data customization module and a learning data label module are arranged, so that a user can customize a risk analysis task, personalized data learning can be realized, and the user experience is improved. And meanwhile, the intelligent library module, the data learning module and the model operation module are arranged, so that the data relationship specified by the user can be accurately learned, the user is assisted to make a financial risk decision, and the processing efficiency of the user can be effectively improved.

Description

Technical Field

[0001] This invention belongs to the field of data processing technology, specifically relating to a supply chain artificial intelligence financial risk decision-making system. Background Technology

[0002] The continuous increase in cross-border trade has led to a more complex product supply chain, with the number of stakeholders increasing at an accelerating pace. As more stakeholders join the supply chain, the resulting bottlenecks and obstacles become more diverse. Due to limited administrative resources, existing supply chain management systems often rely on manual analysis and experience-based judgment when dealing with financial risks, resulting in inefficiency and insufficient accuracy. To improve the stability and resilience of the supply chain, it is necessary to develop a system capable of automatically identifying and assessing financial risks. Summary of the Invention

[0003] This invention provides a supply chain AI financial risk decision-making system to address the technical problems of low efficiency and insufficient accuracy in existing technologies.

[0004] A supply chain AI-powered financial risk decision-making system includes: a learning data customization module, a learning data tagging module, an AI library module, a data learning module, and a model execution module; The learning data customization module is used to respond to the historical supply chain data input by the user through human-computer interaction and provide the historical supply chain data to the data learning module; The learning data tagging module is used to respond to the financial risk tagging data corresponding to the historical supply chain data input by the user through human-computer interaction, and to provide the financial risk tagging data to the data learning module after associating the historical supply chain data with the data. The artificial intelligence library module is used to provide a variety of different artificial intelligence models, and in response to the model determination command input by the user through human-computer interaction, to use the artificial intelligence model corresponding to the model determination command as the target artificial intelligence model; The data learning module is used to train the target artificial intelligence model based on historical supply chain data and corresponding financial risk label data, using a multi-information hybrid learning algorithm, so that the target artificial intelligence model learns the correlation between historical supply chain data and corresponding financial risk label data, and obtains a financial risk decision model. The model operation module is used to connect to the supply chain system to collect real-time supply chain data corresponding to historical supply chain data, and to use a financial risk decision-making model to identify the real-time supply chain data to obtain financial risk auxiliary decision-making results.

[0005] Furthermore, it also includes: a financial risk report generation module; The financial risk report generation module is used to generate a financial risk report based on the real-time supply chain data and the financial risk auxiliary decision-making results, according to a preset template, and transmit the financial risk report to the user-specified device, account, and / or system.

[0006] Furthermore, it also includes a historical project verification module; The historical project verification module is used to perform clustering processing on real-time supply chain data and historical supply chain data using the K-means clustering algorithm to obtain target historical supply chain data belonging to the same category as the real-time supply chain data, and to put the N target historical supply chain data that are closest to the real-time supply chain data in terms of Euclidean distance into the financial risk report.

[0007] Furthermore, the artificial intelligence library module includes CNN, CNN-LSTM, LSTM, BP neural network models, and ResNet.

[0008] Furthermore, based on historical supply chain data and corresponding financial risk label data, a multi-information hybrid learning algorithm is used to train the target artificial intelligence model. This enables the target artificial intelligence model to learn the correlation between historical supply chain data and corresponding financial risk label data, resulting in a financial risk decision-making model, including: The model parameters of the target artificial intelligence model are initialized multiple times to obtain multiple model parameter codes; Using historical supply chain data as input and financial risk label data corresponding to the historical supply chain data as the expected output, the error value corresponding to each model parameter code is obtained, and the optimal model parameter code is obtained based on the error value corresponding to each model parameter code. Regarding the first t In the initial model parameter encoding during training, the position of the optimal model parameter encoding is used as a reference, and the model parameter encoding is updated using a population information position adaptive learning strategy to obtain the model parameter encoding after one information mixing. For the model parameter encoding after the first information mixing, a multi-encoding information interaction strategy is adopted to update the model parameter encoding to obtain the model parameter encoding after the second information mixing. For the model parameter encoding after secondary information mixing, a dual encoding collaboration strategy is adopted to update the model parameter encoding, resulting in the model parameter encoding after tertiary information mixing. For the model parameter encoding after three information mixing, the Levy flight information variation mixing strategy is used to update the model parameter encoding, resulting in the model parameter encoding after four information mixing. Determine whether the iteration termination condition is met. If so, re-obtain the optimal model parameter encoding based on the model parameter encoding after four information mixing steps, and use the re-obtained optimal model parameter encoding and decoding as the final parameters to obtain the financial risk decision model. Otherwise, return to the first information mixing step.

[0009] Furthermore, using the position of the optimal model parameter encoding as a reference, a population information position adaptive learning strategy is employed to update the model parameter encoding, resulting in the model parameter encoding after one information mixing step, including: Based on the error value corresponding to each model parameter encoding, the quality of the corresponding position of the model parameter encoding is determined as follows:

[0010] in, Indicates the first i The quality of the location corresponding to the encoding of each model parameter This represents the error value corresponding to the encoding of the i-th model parameter. This represents the first constant term, which is less than 0.001. i =1,2,…,M, where M represents the total number of model parameter codes; Based on the quality of the position corresponding to the model parameter encoding, determine the first i The degree of information mixing between the individual model parameter encodings and the optimal model parameter encodings is:

[0011] in, Indicates the first i The degree of information mixing between the individual model parameter encodings and the optimal model parameter encodings. This indicates the quality of the position corresponding to the optimal model parameter encoding; According to the i The degree of information mixing between the encoding of the first model parameter and the encoding of the optimal model parameter, for the first... i The model parameter encoding is updated as follows:

[0012] in, Indicates the first t During the training process, the first i Each model parameter is encoded. This represents the encoding of model parameters after a single information mixing. , Indicates the first t Encoding of optimal model parameters during each training process. This indicates an adaptive update step size, and , Let T represent a random number between (0,1), and let T represent the maximum number of training iterations. c This represents the second constant term between (0,2).

[0013] Furthermore, for the model parameter encoding after the first information mixing, a multi-encoding information interaction strategy is used to update the model parameter encoding, resulting in the model parameter encoding after the second information mixing, including: Based on the current number of training iterations, the adaptive update coefficients are obtained as follows:

[0014] in, Indicates adaptively updated coefficients. Represents a random number between (0,1). Represents pi; For any model parameter encoding after a single information mixing, a first paired model parameter encoding for information interaction is randomly matched to the model parameter encoding. For the model parameter codes after one information mixing, the error value corresponding to each model parameter code is re-acquired and arranged in ascending order of error value to obtain the arranged model parameter codes; For the rearranged model parameter codes, the first model parameter code is updated based on the first paired model parameter code as follows:

[0015] in, Indicates the number after arrangement m The model parameter encoding, Represents a random number between (0,1). Indicates the number after arrangement m The first paired model parameter code corresponding to each model parameter code. This represents the encoding of model parameters after the secondary information is mixed. ; For the sorted model parameter codes, the other model parameter codes are updated based on the first paired model parameter codes as follows:

[0016] in, Indicates the number after arrangement m -1 model parameter encoding.

[0017] Furthermore, for the model parameter encoding after secondary information mixing, a dual-encoding collaboration strategy is adopted to update the model parameter encoding, resulting in the model parameter encoding after tertiary information mixing as follows:

[0018] in, This represents the second information after the second information is mixed. n Each model parameter is encoded. n =1,2,…,M, where M represents the total number of model parameter codes. This represents the encoding of model parameters after mixing the three sets of information. , Represents a random number between [0, 2π]. This represents a third random number between [0, π]. , , Indicates the first intermediate coefficient. Indicates the second intermediate coefficient. Denotes the constant term, and .

[0019] Furthermore, for the model parameter encoding after three information mixtures, the Lévy flight information mutation mixing strategy is used to update the model parameter encoding, resulting in the model parameter encoding after four information mixtures as follows:

[0020] in, Indicates the first t The third set of information mixtures during the training process u Each model parameter is encoded. u =1,2,…,M, where M represents the total number of model parameter codes. This represents the encoding of model parameters after four levels of information mixing. , This indicates Levi's flight stride.

[0021] Further, determine whether the iteration termination condition is met, including: Determine if the current number of training iterations is greater than the maximum number of training iterations. If so, the iteration termination condition is met; otherwise, the iteration termination condition is not met.

[0022] This invention provides a supply chain AI-powered financial risk decision-making system. By setting up a learning data customization module and a learning data tagging module, users can customize risk analysis tasks and achieve personalized data learning. At the same time, by setting up an intelligent library module, a data learning module, and a model running module, it can accurately learn the data relationships specified by the user, thereby assisting users in making financial risk decisions and effectively improving the user's processing efficiency. Attached Figure Description

[0023] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0024] Figure 1 This is a schematic diagram of the structure of a supply chain artificial intelligence financial risk decision-making system provided in an embodiment of the present invention.

[0025] Figure 2 This is a flowchart for obtaining a financial risk decision-making model provided in an embodiment of the present invention.

[0026] The accompanying drawings have illustrated specific embodiments of the invention, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the invention in any way, but rather to illustrate the concept of the invention to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0027] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims.

[0028] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0029] like Figure 1 As shown, this embodiment of the invention provides a supply chain AI financial risk decision-making system, including: a learning data customization module, a learning data tagging module, an AI library module, a data learning module, and a model running module; The learning data customization module is used to respond to the historical supply chain data input by the user through human-computer interaction and provide the historical supply chain data to the data learning module.

[0030] The data provided by users is not limited to any particular type; it can include various operational data as well as other user data. By identifying historical supply chain data by the user, users can customize data learning and data recognition tasks, which greatly improves data processing efficiency and helps users make more decisions.

[0031] Optionally, to ensure that the data can be accurately identified, the collected supply chain historical data should be in numerical form, and data that is not in numerical form should be converted.

[0032] The learning data tagging module is used to respond to the financial risk tagging data corresponding to the historical supply chain data input by the user through human-computer interaction, and to provide the financial risk tagging data to the data learning module after associating the financial risk tagging data with the historical supply chain data.

[0033] Learning data tags refers to various decision-making types of users. By learning the relationship between historical supply chain data and financial risk tag data specified by the user, the data tag can perform tasks on behalf of the user, helping the user improve the efficiency of data and risk supervision.

[0034] The artificial intelligence library module is used to provide a variety of different artificial intelligence models, and in response to the model determination command input by the user through human-computer interaction, to use the artificial intelligence model corresponding to the model determination command as the target artificial intelligence model; Since different data may be better identified using different artificial intelligence models, a variety of artificial intelligence models can be provided so that users can choose the one that suits them best.

[0035] The data learning module is used to train the target artificial intelligence model based on historical supply chain data and corresponding financial risk label data, using a multi-information hybrid learning algorithm, so that the target artificial intelligence model learns the correlation between historical supply chain data and corresponding financial risk label data, and obtains a financial risk decision model. Existing artificial intelligence optimization algorithms often suffer from slow convergence speed and poor convergence performance (e.g., prone to getting trapped in local optima), making it difficult to effectively learn data relationships and ultimately failing to accurately complete the user-specified data analysis tasks. Therefore, this invention provides a multi-information hybrid learning algorithm to address these technical problems, improve data relationship learning performance, and ultimately assist users in more accurately predicting financial risks, thereby increasing user work efficiency.

[0036] The model operation module is used to connect to the supply chain system to collect real-time supply chain data corresponding to historical supply chain data, and to use a financial risk decision-making model to identify the real-time supply chain data to obtain financial risk auxiliary decision-making results.

[0037] This invention provides a supply chain AI-powered financial risk decision-making system. By setting up a learning data customization module and a learning data tagging module, users can customize risk analysis tasks and achieve personalized data learning. At the same time, by setting up an intelligent library module, a data learning module, and a model running module, it can accurately learn the data relationships specified by the user, thereby assisting users in making financial risk decisions and effectively improving the user's processing efficiency.

[0038] In this embodiment of the invention, it further includes: a financial risk report generation module; The financial risk report generation module is used to generate a financial risk report based on the real-time supply chain data and the financial risk auxiliary decision-making results, according to a preset template, and transmit the financial risk report to the user-specified device, account, and / or system.

[0039] By generating financial risk reports, users can gain a more intuitive understanding of the data and its risks, enabling them to perform verification processes. At the same time, financial risk reports are also easy to store and trace.

[0040] In this embodiment of the invention, a historical project verification module is also included; The historical project verification module is used to perform clustering processing on real-time supply chain data and historical supply chain data using the K-means clustering algorithm to obtain target historical supply chain data belonging to the same category as the real-time supply chain data, and to put the N target historical supply chain data that are closest to the real-time supply chain data in terms of Euclidean distance into the financial risk report.

[0041] By providing users with historical data from N similar target supply chains, users can more easily find and handle problems, thereby improving their work efficiency.

[0042] In this embodiment of the invention, the artificial intelligence library module includes CNN (Convolutional Neural Networks), CNN-LSTM, LSTM (Long Short-Term Memory), BP (Back Propagation) neural network models, and ResNet (Residual Network).

[0043] It is worth noting that, in addition to the aforementioned artificial intelligence models, other artificial intelligence models can also be used to learn and classify data.

[0044] like Figure 2 As shown, based on historical supply chain data and corresponding financial risk label data, a multi-information hybrid learning algorithm is used to train the target artificial intelligence model. This enables the target artificial intelligence model to learn the correlation between historical supply chain data and corresponding financial risk label data, resulting in a financial risk decision-making model, including: The model parameters of the target artificial intelligence model are initialized multiple times to obtain multiple model parameter codes; For example, model parameters (such as weight parameters) can be randomly initialized between the upper and lower limits to obtain model parameter codes. This process can be repeated multiple times to obtain multiple different model parameter codes.

[0045] Using historical supply chain data as input and financial risk label data corresponding to the historical supply chain data as the expected output, the error value corresponding to each model parameter code is obtained, and the optimal model parameter code is obtained based on the error value corresponding to each model parameter code. Optionally, embodiments of the present invention may use the cross-entropy error function or the mean square error function to obtain the error value corresponding to the encoding of each model parameter.

[0046] Regarding the first t In the initial model parameter encoding during training, the position of the optimal model parameter encoding is used as a reference, and the model parameter encoding is updated using a population information position adaptive learning strategy to obtain the model parameter encoding after one information mixing. For the model parameter encoding after the first information mixing, a multi-encoding information interaction strategy is adopted to update the model parameter encoding to obtain the model parameter encoding after the second information mixing. For the model parameter encoding after secondary information mixing, a dual encoding collaboration strategy is adopted to update the model parameter encoding, resulting in the model parameter encoding after tertiary information mixing. For the model parameter encoding after three information mixing, the Levy flight information variation mixing strategy is used to update the model parameter encoding, resulting in the model parameter encoding after four information mixing. Determine whether the iteration termination condition is met. If so, re-obtain the optimal model parameter encoding based on the model parameter encoding after four information mixing steps, and use the re-obtained optimal model parameter encoding and decoding as the final parameters to obtain the financial risk decision model. Otherwise, return to the first information mixing step.

[0047] The multi-information hybrid learning algorithm provided in this invention can achieve better learning of data relationships, thereby providing more accurate data analysis functions, improving the prediction effect of financial risks, and ultimately analyzing data features more accurately, thus improving users' work efficiency.

[0048] In this embodiment of the invention, taking the position of the optimal model parameter encoding as a reference, the model parameter encoding is updated using a population information position adaptive learning strategy to obtain the model parameter encoding after one information mixing, including: Based on the error value corresponding to each model parameter encoding, the quality of the corresponding position of the model parameter encoding is determined as follows:

[0049] in, Indicates the first i The quality of the location corresponding to the encoding of each model parameter This represents the error value corresponding to the encoding of the i-th model parameter. This represents the first constant term, which is less than 0.001. i =1,2,…,M, where M represents the total number of model parameter codes; Based on the quality of the position corresponding to the model parameter encoding, determine the first i The degree of information mixing between the individual model parameter encodings and the optimal model parameter encodings is:

[0050] in, Indicates the first i The degree of information mixing between the individual model parameter encodings and the optimal model parameter encodings. This indicates the quality of the position corresponding to the optimal model parameter encoding; According to the i The degree of information mixing between the encoding of the first model parameter and the encoding of the optimal model parameter, for the first... i The model parameter encoding is updated as follows:

[0051] in, Indicates the first t During the training process, the first i Each model parameter is encoded. This represents the encoding of model parameters after a single information mixing. , Indicates the first t Encoding of optimal model parameters during each training process. This indicates an adaptive update step size, and , Let T represent a random number between (0,1), and let T represent the maximum number of training iterations. c This represents the second constant term between (0,2).

[0052] The population information location adaptive learning strategy provided in this embodiment of the invention can enable better codes to learn more optimal codes, while worse codes tend to search in their own vicinity. This can improve the convergence speed of the algorithm, maintain code diversity, and enhance global search capabilities.

[0053] In this embodiment of the invention, for the model parameter encoding after a first information mixing, a multi-encoding information interaction strategy is used to update the model parameter encoding to obtain the model parameter encoding after a second information mixing, including: Based on the current number of training iterations, the adaptive update coefficients are obtained as follows:

[0054] in, Indicates adaptively updated coefficients. Represents a random number between (0,1). Represents pi; For any model parameter encoding after a single information mixing, a first paired model parameter encoding for information interaction is randomly matched to the model parameter encoding. For the model parameter codes after one information mixing, the error value corresponding to each model parameter code is re-acquired and arranged in ascending order of error value to obtain the arranged model parameter codes; For the rearranged model parameter codes, the first model parameter code is updated based on the first paired model parameter code as follows:

[0055] in, Indicates the number after arrangement m The model parameter encoding, Represents a random number between (0,1). Indicates the number after arrangement m The first paired model parameter code corresponding to each model parameter code. This represents the encoding of model parameters after the secondary information is mixed. ; For the sorted model parameter codes, the other model parameter codes are updated based on the first paired model parameter codes as follows:

[0056] in, Indicates the number after arrangement m -1 model parameter encoding.

[0057] The multi-encoding information interaction provided in this embodiment of the invention can enable each encoder to learn random encoding and encoding information that is better than itself, which can improve the ability to explore unknown areas and also enhance the diversity of algorithms.

[0058] In this embodiment of the invention, for the model parameter encoding after secondary information mixing, a dual-encoding cooperation strategy is adopted to update the model parameter encoding, resulting in the model parameter encoding after tertiary information mixing as follows:

[0059] in, This represents the second information after the second information is mixed. n Each model parameter is encoded. n =1,2,…,M, where M represents the total number of model parameter codes. This represents the encoding of model parameters after mixing the three sets of information. , Represents a random number between [0, 2π]. This represents a third random number between [0, π]. , , Indicates the first intermediate coefficient. Indicates the second intermediate coefficient. Denotes the constant term, and .

[0060] The dual-encoding cooperation strategy provided in this invention allows the encoding to explore between its own position and the optimal position in an irregular manner, which helps to improve local search capabilities and final search efficiency.

[0061] In this embodiment of the invention, for the model parameter encoding after three information mixing steps, the Levy flight information mutation mixing strategy is used to update the model parameter encoding, resulting in the model parameter encoding after four information mixing steps:

[0062] in, Indicates the first t The third set of information mixtures during the training process u Each model parameter is encoded. u =1,2,…,M, where M represents the total number of model parameter codes. This represents the encoding of model parameters after four levels of information mixing. , This indicates Levi's flight stride.

[0063] While the three search methods above can provide local search and high search accuracy, their global search capabilities are generally limited. Therefore, we will perform a jump search to provide powerful global search capabilities.

[0064] In this embodiment of the invention, determining whether the iteration termination condition is met includes: Determine if the current number of training iterations is greater than the maximum number of training iterations. If so, the iteration termination condition is met; otherwise, the iteration termination condition is not met.

[0065] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. It should be understood that the invention is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A supply chain AI-powered financial risk decision-making system, characterized in that, include: The learning data customization module, learning data tagging module, artificial intelligence library module, data learning module, and model running module are included. The learning data customization module is used to respond to the historical supply chain data input by the user through human-computer interaction and provide the historical supply chain data to the data learning module; The learning data tagging module is used to respond to the financial risk tagging data corresponding to the historical supply chain data input by the user through human-computer interaction, and to provide the financial risk tagging data to the data learning module after associating the historical supply chain data with the data. The artificial intelligence library module is used to provide a variety of different artificial intelligence models, and in response to the model determination command input by the user through human-computer interaction, to use the artificial intelligence model corresponding to the model determination command as the target artificial intelligence model; The data learning module is used to train the target artificial intelligence model based on historical supply chain data and corresponding financial risk label data, using a multi-information hybrid learning algorithm, so that the target artificial intelligence model learns the correlation between historical supply chain data and corresponding financial risk label data, and obtains a financial risk decision model. The model operation module is used to connect to the supply chain system to collect real-time supply chain data corresponding to historical supply chain data, and to use a financial risk decision-making model to identify the real-time supply chain data to obtain financial risk auxiliary decision-making results.

2. The supply chain AI financial risk decision-making system according to claim 1, characterized in that, Also includes: Financial risk report generation module; The financial risk report generation module is used to generate a financial risk report based on the real-time supply chain data and the financial risk auxiliary decision-making results, according to a preset template, and transmit the financial risk report to the user-specified device, account, and / or system.

3. The supply chain AI financial risk decision-making system according to claim 2, characterized in that, It also includes a historical project verification module; The historical project verification module is used to perform clustering processing on real-time supply chain data and historical supply chain data using the K-means clustering algorithm to obtain target historical supply chain data belonging to the same category as the real-time supply chain data, and to put the N target historical supply chain data that are closest to the real-time supply chain data in terms of Euclidean distance into the financial risk report.

4. The supply chain AI financial risk decision-making system according to claim 1, characterized in that, The artificial intelligence library module includes CNN, CNN-LSTM, LSTM, BP neural network models, and ResNet.

5. The supply chain AI financial risk decision-making system according to claim 1, characterized in that, Based on historical supply chain data and corresponding financial risk label data, a multi-information hybrid learning algorithm is used to train the target artificial intelligence model. This enables the target artificial intelligence model to learn the correlation between historical supply chain data and corresponding financial risk label data, resulting in a financial risk decision-making model, including: The model parameters of the target artificial intelligence model are initialized multiple times to obtain multiple model parameter codes; Using historical supply chain data as input and financial risk label data corresponding to the historical supply chain data as the expected output, the error value corresponding to each model parameter code is obtained, and the optimal model parameter code is obtained based on the error value corresponding to each model parameter code. Regarding the first t In the initial model parameter encoding during training, the position of the optimal model parameter encoding is used as a reference, and the model parameter encoding is updated using a population information position adaptive learning strategy to obtain the model parameter encoding after one information mixing. For the model parameter encoding after the first information mixing, a multi-encoding information interaction strategy is adopted to update the model parameter encoding to obtain the model parameter encoding after the second information mixing. For the model parameter encoding after secondary information mixing, a dual encoding collaboration strategy is adopted to update the model parameter encoding, resulting in the model parameter encoding after tertiary information mixing. For the model parameter encoding after three information mixing, the Levy flight information variation mixing strategy is used to update the model parameter encoding, resulting in the model parameter encoding after four information mixing. Determine whether the iteration termination condition is met. If so, re-obtain the optimal model parameter encoding based on the model parameter encoding after four information mixing steps, and use the re-obtained optimal model parameter encoding and decoding as the final parameters to obtain the financial risk decision model. Otherwise, return to the first information mixing step.

6. The supply chain AI financial risk decision-making system according to claim 5, characterized in that, Using the position of the optimal model parameter encoding as a reference, a population information position adaptive learning strategy is employed to update the model parameter encoding, resulting in the model parameter encoding after one information mixing step, including: Based on the error value corresponding to each model parameter encoding, the quality of the corresponding position of the model parameter encoding is determined as follows: in, Indicates the first i The quality of the location corresponding to the encoding of each model parameter This represents the error value corresponding to the encoding of the i-th model parameter. This represents the first constant term, which is less than 0.

001. i =1,2,…,M, where M represents the total number of model parameter codes; Based on the quality of the position corresponding to the model parameter encoding, determine the first i The degree of information mixing between the individual model parameter encodings and the optimal model parameter encodings is: in, Indicates the first i The degree of information mixing between the individual model parameter encodings and the optimal model parameter encodings. This indicates the quality of the position corresponding to the optimal model parameter encoding; According to the i The degree of information mixing between the encoding of the first model parameter and the encoding of the optimal model parameter, for the first... i The model parameter encoding is updated as follows: in, Indicates the first t During the training process, the first i Each model parameter is encoded. This represents the encoding of model parameters after a single information mixing. , Indicates the first t Encoding of optimal model parameters during each training process. This indicates an adaptive update step size, and , Let T represent a random number between (0,1), and let T represent the maximum number of training iterations. c This represents the second constant term between (0,2).

7. The supply chain AI financial risk decision-making system according to claim 6, characterized in that, For the model parameter encoding after primary information mixing, a multi-encoding information interaction strategy is used to update the model parameter encoding, resulting in the model parameter encoding after secondary information mixing, including: Based on the current number of training iterations, the adaptive update coefficients are obtained as follows: in, Indicates adaptively updated coefficients. Represents a random number between (0,1). Represents pi; For any model parameter encoding after a single information mixing, a first paired model parameter encoding for information interaction is randomly matched to the model parameter encoding. For the model parameter codes after one information mixing, the error value corresponding to each model parameter code is re-acquired and arranged in ascending order of error value to obtain the arranged model parameter codes; For the rearranged model parameter codes, the first model parameter code is updated based on the first paired model parameter code as follows: in, Indicates the number after arrangement m The model parameter encoding, Represents a random number between (0,1). Indicates the number after arrangement m The first paired model parameter code corresponding to each model parameter code. This represents the encoding of model parameters after the secondary information is mixed. ; For the sorted model parameter codes, the other model parameter codes are updated based on the first paired model parameter codes as follows: in, Indicates the number after arrangement m -1 model parameter encoding.

8. The supply chain AI financial risk decision-making system according to claim 6, characterized in that, For the model parameter encoding after secondary information mixing, a dual-encoding collaboration strategy is adopted to update the model parameter encoding, resulting in the model parameter encoding after tertiary information mixing: in, This represents the second information after the second information is mixed. n Each model parameter is encoded. n =1,2,…,M, where M represents the total number of model parameter codes. This represents the encoding of model parameters after mixing the three sets of information. , Represents a random number between [0, 2π]. This represents a third random number between [0, π]. , , Indicates the first intermediate coefficient. Indicates the second intermediate coefficient. Denotes the constant term, and .

9. The supply chain AI financial risk decision-making system according to claim 6, characterized in that, For the model parameter encoding after three information mixtures, the Lévy flight information mutation mixing strategy is used to update the model parameter encoding, resulting in the model parameter encoding after four information mixtures: in, Indicates the first t The third set of information mixtures during the training process u Each model parameter is encoded. u =1,2,…,M, where M represents the total number of model parameter codes. This represents the encoding of model parameters after four levels of information mixing. , This indicates Levi's flight stride.

10. The supply chain AI financial risk decision-making system according to claim 9, characterized in that, Determine whether the iteration termination condition is met, including: Determine if the current number of training iterations is greater than the maximum number of training iterations. If so, the iteration termination condition is met; otherwise, the iteration termination condition is not met.

Images