Supply chain accounting clearing and settlement system and method based on deep learning

Abstract

The application discloses a supply chain accounting clearing and settlement system and method based on deep learning, S1, collecting accounting data in a supply chain and performing standardization processing; S2, processing the accounting data through a capsule network to generate high-level feature representation of each bill element; S3, modeling complex relationships between bills by using a relationship reasoning module, establishing a relationship graph between bills, and obtaining a final settlement sequence and an amount distribution scheme of the bills; S4, using historical accounting data as a training set to train the capsule network and the relationship reasoning module; S5, performing real-time automatic clearing and settlement processing on newly generated bill data, and updating the settlement state in the supply chain; S6, generating a detailed settlement report and returning the settlement report to a supply chain node. The application can provide an efficient and scientific optimization scheme in supply chain accounting clearing and settlement, and brings significant technical value and economic benefits for practical application.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of account reconciliation, and particularly relates to a supply chain account reconciliation system and method based on deep learning. BACKGROUND

[0002] With the globalization of trade and the complexity of supply chain management, supply chain account reconciliation has become a crucial part of enterprise operations. In traditional account reconciliation, manual operations or rule-based automated systems are often used to check each bill one by one to determine the payment amount and payment sequence of each bill. Existing technologies mainly rely on simple rule matching and data comparison. Although these methods can meet the basic account settlement needs, they are inefficient and prone to errors when dealing with large-scale supply chain data. In addition, these methods are difficult to handle complex relationships between bills, such as multi-level payment sequences, amount allocation, and dynamic factors such as transaction interactions between suppliers, resulting in a cumbersome and error-prone account settlement process.

[0003] Currently, account data analysis methods based on traditional neural networks have been proposed and applied to some account reconciliation systems. However, traditional convolutional neural networks mainly focus on extracting local features when processing account data, especially for image data feature extraction. They fail to effectively capture the global dependency between bills and the complex dynamic interactions between bills. For example, although convolutional neural networks can handle visual and numerical features of bills, they have limitations in modeling complex relationships such as payment sequence and amount allocation between bills. In addition, existing account reconciliation systems are difficult to automatically derive long-distance dependencies and spatial relationships between bills, resulting in constraints on the accuracy and efficiency of settlement.

[0004] In this context, the introduction of capsule networks and relationship reasoning provides a new solution for account reconciliation, which has stronger ability than traditional convolutional neural networks, can effectively capture the spatial and pose relationships in account data, and solve the shortcomings of traditional neural networks in long-distance dependence and spatial feature expression. Through the capsule unit, the system can better represent the complex relationships and structured dependencies between bill elements, providing a theoretical basis for more accurate account reconciliation. However, although the capsule network has advantages in processing account data, the existing account reconciliation system still faces two major challenges: on the one hand, the complex relationships between bills such as payment order and amount allocation are still difficult to model effectively through traditional methods; on the other hand, the existing technology lacks a reasoning mechanism that can effectively utilize the dynamic relationships between bills, resulting in poor adaptability of the settlement model in real-time data and complex transaction scenarios. In addition, the introduction of the relationship reasoning module, although it can effectively capture the complex relationships between bills, in practical applications, how to reasonably design and optimize the modeling method of these complex relationships to improve the accuracy and efficiency of the settlement process is still a problem to be solved.

[0005] Based on the above background, this paper proposes a supply chain account reconciliation method based on capsule networks and relationship reasoning, aiming to optimize the account reconciliation process using deep learning and dynamic reasoning mechanisms, accurately model the complex relationships between bills, and improve the intelligent level of supply chain settlement through multi-level reasoning, thereby solving the shortcomings of existing technologies. SUMMARY

[0006] One object of the present application is to propose a supply chain account reconciliation system and method based on deep learning, which can provide an efficient and scientific optimization scheme for supply chain account reconciliation, bringing significant technical value and economic benefits to practical applications.

[0007] The supply chain account reconciliation method based on deep learning according to an embodiment of the present application comprises the following steps:

[0008] S1, collecting account data in the supply chain and performing standardization processing;

[0009] S2, processing the account data through a capsule network, distributing and updating the features based on a dynamic routing mechanism, capturing the spatial and pose relationships between account data elements, and generating high-level feature representations of each bill element;

[0010] S3, based on the high-level feature representations of the bill elements, modeling the complex relationships between bills using a relationship reasoning module, establishing a relationship graph between bills, and deriving the priority of each bill through a reasoning mechanism to obtain the final settlement order and amount allocation scheme of the bills;

[0011] S4, adopt historical accounting data as a training set, train the capsule network and the relationship reasoning module through back propagation and gradient descent method;

[0012] S5, apply the trained capsule network and the relationship reasoning module to real-time automatic clearing and settlement processing of newly generated billing data, match the amounts between bills, determine the payment order according to the reasoning result, and automatically update the settlement state in the supply chain;

[0013] S6, according to the result of automatic clearing and settlement, generate a detailed settlement report, and return the settlement report to the supply chain node for payment processing.

[0014] Optionally, the S1 includes the following steps:

[0015] S11, collect accounting data in the supply chain, including invoices, bills, payment records, and supplier information data;

[0016] S12, pre-process the collected accounting data, including data cleaning, removing invalid and duplicate records, filling in missing data, and standardizing different formats of data;

[0017] S13, normalize the accounting data.

[0018] Optionally, the S2 includes the following steps:

[0019] S21, use YOLOv10 to extract preliminary features from the collected accounting data, and input the preliminary feature vectors extracted by YOLOv10 into the capsule unit in the capsule network for further processing;

[0020] S22, in the capsule unit, the features are allocated and updated through a dynamic routing mechanism to capture the spatial and pose relationships between the accounting data elements;

[0021] Let the output vector of the ith capsule be Through linear transformation and nonlinear activation function processing, the transformed output vector is obtained

[0022] s ij =R ij ·(v i +b i )+b ij ;

[0023] Where, R ij is the rotation matrix from capsule i to capsule j, b i is the bias term of capsule i, b ij is the bias term of capsule i to capsule j, through the rotation matrix and affine transformation, the model can capture the spatial transformation relationship between the bills;

[0024] S23, establish the relationship between capsules by using the routing mechanism, and calculate the routing coefficient c between capsules ij to obtain the influence of each capsule:

[0025]

[0026] wherein, is the weight vector of capsule j, s ij is the output vector of capsule i to capsule j, is the set of adjacent nodes of capsule i, exp(.) is the natural exponential function, c ij is the normalized routing coefficient, according to the influence of the output of each capsule on the adjacent capsules, dynamically adjust the weight transmission between capsules, so that the network can dynamically adjust the information transmission according to the output intensity of each capsule;

[0027] S24, update the output vector of each capsule by weighted sum of the transformation vector s ij of each capsule and the routing coefficient c ij

[0028]

[0029] wherein, ||.|| represents the vector modulus operation, ∈ is a small constant to prevent division by zero error, ||.|| represents the Euclidean norm; 2

[0030] S25, generate the advanced feature representation of each billing element according to the output result v′ j of the capsule network.

[0031] Optionally, the S3 comprises the following steps:

[0032] S31, according to the advanced feature representation of each billing element obtained from the capsule network, use the relationship reasoning module to model the complex relationship between bills, analyze the correlation, payment order and amount allocation information between bills;

[0033] S32, construct the relationship graph between bills by calculating the similarity S ij between bill i and bill j, each node represents a billing element, and the edge represents the relationship between billing elements:

[0034]

[0035] wherein, sim(.) represents the similarity function, v i is the feature vector of bill i, v j is the feature vector of bill j, v ik and v​​jk are the kth element in the feature vector respectively, d is the node degree;

[0036] S33, the feature update formula of each billing node i in the graph neural network is:

[0037]

[0038] wherein, represents the feature vector of billing i after the (l+1)th layer graph convolution, A ij is the relationship matrix element between billing i and billing j, is the degree of node i, represents the set of neighbor nodes adjacent to node i, and σ is the sigmoid activation function;

[0039] S34, on the basis of graph convolution, further non-linear modeling of the relationship between billing nodes is performed through a multi-layer perception, by inputting the features and of each billing into the multi-layer perception to calculate the payment order and amount allocation relationship between the billings:

[0040]

[0041] wherein, and are the output feature vectors of billing i and billing j at the Lth layer in the graph convolution network respectively, and MLP(·) represents a multi-layer perception, which learns the complex relationship between the billings through multi-layer nonlinear mapping, and the model is used to predict the settlement rules and priorities between the billings;

[0042] S35, based on the similarity between the billings and the output of the graph neural network, a settlement priority matrix P is constructed:

[0043]

[0044] wherein, w is a weight vector, and are the outputs of billing i and billing j at the Lth layer in the graph convolution network respectively, and P ij represents the settlement priority between billing i and billing j, exp(.) is the natural exponential function, and α is the adjustment factor;

[0045] S36, the derived settlement priority matrix is combined with the amount allocation strategy to obtain the final settlement order and amount allocation scheme of the billings.

[0046] Optionally, the S4 comprises the following steps:

[0047] S41, the historical accounting data is trained as a training set;

[0048] S42, the gradient descent method is used to optimize the loss function, and the gradient of each parameter theta j is calculated The formula is as follows:

[0049]

[0050] Wherein, eta is the learning rate, controlling the step size of each parameter update, Indicates the gradient of the loss function L(theta) to the parameter theta j ;

[0051] S43, the gradient of each layer of the network is calculated by using the back propagation algorithm, and the parameters of the capsule model are updated layer by layer, and the output error delta (k) of each layer is calculated, and the formula is as follows:

[0052] Delta (k) =(W (k+1) ) T Delta (k+1) *f'(z (k) );

[0053] Wherein, delta (k) is the error term of the kth layer, W (k+1) is the weight matrix of the k+1th layer, z (k) is the weighted input of the kth layer, and f'(z (k) ) is the derivative of the activation function.

[0054] Optionally, the S5 comprises the following steps:

[0055] S51, real-time automatic clearing and settlement processing is performed on the newly generated billing data, the new billing data is input into the trained capsule network and the relationship reasoning model, and the clearing and settlement information of each bill is obtained;

[0056] S52, based on the reasoning result, the amount matching between bills is calculated, the payment order of each bill is determined, each bill is allocated a payment time order, and the settlement state in the supply chain is automatically updated.

[0057] The supply chain billing clearing and settlement system based on deep learning according to the embodiment of the application comprises the following modules:

[0058] The data acquisition module is used for acquiring the billing data in the supply chain and performing standardization processing;

[0059] The feature extraction module is used for performing preliminary feature extraction on the collected billing data by using YOLOv10, obtaining the basic visual and numerical information of the billing data, and further extracting the spatial and posture relationship between bills by using the capsule network;

[0060] a relationship modeling module configured to model complex relationships between bills based on the relationship reasoning module, analyze payment order and amount allocation factors between bills, and generate priority and settlement rules between bills;

[0061] a model training module configured to train the capsule network and the relationship reasoning module using historical accounting data as a training set through back propagation and gradient descent optimization methods;

[0062] a real-time clearing and settlement module configured to perform real-time automatic clearing and settlement processing on newly generated bill data, determine amount matching and payment order between bills according to the reasoning result, and automatically update the settlement state in the supply chain;

[0063] a settlement report generation module configured to generate a detailed settlement report according to the automatic clearing and settlement result, and return the settlement report to the supply chain node for payment processing.

[0064] The present application has the following advantages:

[0065] (1) The present application effectively improves the spatial relationship modeling capability in the accounting clearing and settlement process by introducing a capsule network and a relationship reasoning module. Compared with traditional convolutional neural networks, the present application can capture the spatial and pose relationships between elements in bill data and accurately represent the complex dependency relationships between bills. This enables the system to more accurately understand and process long-distance dependencies and global relationships in accounting data, improving the accuracy and reliability of accounting clearing.

[0066] (2) The present application can accurately model dynamic relationships such as payment order and amount allocation between bills by introducing a relationship reasoning module. On this basis, the system can automatically derive the settlement priority and rules of bills, thereby avoiding the limitations of traditional methods relying on static rules and simple matching. The present application significantly improves the efficiency and accuracy of the accounting clearing and settlement process by analyzing and reasoning the complex relationships between bills.

[0067] (3) The present application achieves higher accuracy and flexibility in settlement priority calculation by combining multi-dimensional information such as bill similarity, payment order and amount allocation. Compared with the prior art, the present application can automatically adjust the settlement order and amount allocation of bills in real time according to the dynamic changes in accounting data, improving the adaptability and global optimization capability of settlement decisions. BRIEF DESCRIPTION OF DRAWINGS

[0068] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, illustrate embodiments of the present application, and are used to explain the present application, and do not constitute a limitation on the present application. In the drawings:

[0069] Figure 1The flow chart of the supply chain accounting clearing and settlement method based on deep learning proposed in the application;

[0070] Figure 2 The structural diagram of the supply chain accounting clearing and settlement system based on deep learning proposed in the application. DETAILED DESCRIPTION

[0071] The application will be further described in detail in conjunction with the accompanying drawings. These drawings are simplified schematic diagrams, and only schematically show the basic structure of the application, and therefore only show the components related to the application.

[0072] REFERENCE Figure 1 The supply chain accounting clearing and settlement method based on deep learning comprises the following steps:

[0073] S1, collecting accounting data in the supply chain and performing standardization processing;

[0074] S2, processing the accounting data through a capsule network, distributing and updating the features based on a dynamic routing mechanism, capturing the spatial and pose relationships between the accounting data elements, and generating high-level feature representations of each billing element;

[0075] S3, based on the high-level feature representations of the billing elements, using a relationship reasoning module to model the complex relationships between the bills, establishing a relationship graph between the bills, and deriving the priority of each bill through an inference mechanism to obtain the final settlement order and amount allocation scheme of the bills;

[0076] S4, using historical accounting data as a training set, training the capsule network and the relationship reasoning module through back propagation and gradient descent method;

[0077] S5, applying the trained capsule network and relationship reasoning module to real-time automatic clearing and settlement processing of newly generated billing data, performing amount matching and payment order determination between the bills according to the inference results, and automatically updating the settlement status in the supply chain;

[0078] S6, generating a detailed settlement report according to the results of automatic clearing and settlement, and returning the settlement report to the supply chain nodes for payment processing.

[0079] The application collects accounting data in the supply chain, including invoices, bills, payment records, etc., and performs standardization processing on the data, ensuring data consistency and integrity in the subsequent analysis process. This step enables the system to collect accounting information from multiple sources and preprocess it, thereby improving data quality and providing a reliable foundation for subsequent feature extraction and reasoning modeling. Through standardized processing, errors caused by inconsistent data formats are avoided, ensuring the efficiency and accuracy of the clearing and settlement process.

[0080] In this embodiment, S1 comprises the following steps:

[0081] S11, collecting account data in the supply chain, including invoices, bills, payment records, and supplier information data;

[0082] S12, preprocessing the collected account data, including data cleaning, removing invalid and duplicate records, filling in missing data, and standardizing different formats of data;

[0083] S13, normalizing the account data.

[0084] The present application further processes the account data through capsules, which can capture the spatial and pose relationships between bill elements. This step enhances the representation of long-range dependencies in account data through capsule units, solving the problem that traditional neural network methods cannot effectively handle complex spatial relationships. The introduction of capsule networks can effectively improve the system's ability to model global relationships between bills, enhancing the accuracy and flexibility of the settlement process, especially in the processing of complex bill data, which can accurately identify and retain hierarchical features between bills.

[0085] In this embodiment, S2 comprises the following steps:

[0086] S21, using YOLOv10 to extract preliminary features from the collected account data, and inputting the preliminary feature vectors extracted by YOLOv10 into the capsule unit in the capsule network for further processing;

[0087] S22, in the capsule unit, features are assigned and updated through a dynamic routing mechanism to capture the spatial and pose relationships between account data elements;

[0088] Let the output vector of the i-th capsule be Through linear transformation and nonlinear activation function processing, the transformed output vector is obtained

[0089] s ij =R ij ·(v i +b i )+b ij ;

[0090] where R ij is the rotation matrix from capsule i to capsule j, b i is the bias term of capsule i, and b ij is the bias term of capsule i to capsule j. Through rotation matrix and affine transformation, the model can capture the spatial transformation relationship between bills.

[0091] S23, a routing mechanism is used to establish the relationship between capsules, and the routing coefficient c between capsules is calculated ij to obtain the influence of each capsule:

[0092]

[0093] wherein, is the weight vector of capsule j, s ij is the output vector of capsule i to capsule j, is the set of adjacent nodes of capsule i, exp(.) is the natural exponential function, c ij is the normalized routing coefficient, according to the influence of the output of each capsule on the adjacent capsule, the weight transmission between capsules is dynamically adjusted, so that the network can dynamically adjust the information transmission according to the output intensity of each capsule;

[0094] S24, update the output vector of each capsule by weighted sum of the transformation vector s ij of each capsule and the routing coefficient c ij

[0095]

[0096] wherein, ||.|| represents the vector modulus operation, ∈ is a small constant to prevent division by zero error, and ||.|| represents the Euclidean norm; 2

[0097] S25, according to the output result v′ j of the capsule network, generate the advanced feature representation of each billing element.

[0098] The present application can dynamically analyze the payment order, amount allocation and other factors between bills by modeling the complex relationship between bills through the relationship reasoning module, which effectively constructs the relationship graph between bills and deduces the settlement rules and priority of each bill through the reasoning mechanism. By considering the complex relationship between bills, the present application avoids the simplified processing of traditional methods for account settlement, and ensures accurate matching and reasonable allocation in the settlement process, greatly improving the decision-making ability of the system.

[0099] In the embodiment, S3 includes the following steps:

[0100] S31, according to the advanced feature representation of each billing element obtained from the capsule network, the relationship reasoning module is used to model the complex relationship between bills, and the correlation, payment order and amount allocation information between bills are analyzed;

[0101] S32, the similarity S ij ​​To construct a graph between bills, each node represents a bill element, and the edge represents the relationship between bill elements:

[0102]

[0103] where sim(.) represents the similarity function, v i is the feature vector of bill i, v j is the feature vector of bill j, v ik and v jk are the kth elements in the feature vector, and d is the node degree.

[0104] S33, the feature update formula of each bill node i in the graph neural network is:

[0105]

[0106] where, represents the feature vector of bill i after the l+1 layer graph convolution, A ij is the relationship matrix element between bill i and bill j, is the degree of node i, represents the set of neighbor nodes adjacent to node i, and sigma is the sigmoid activation function.

[0107] S34, on the basis of graph convolution, further nonlinear modeling of the relationship between bill nodes is carried out through multilayer perception, by inputting the features and of each bill into the multilayer perception to calculate the payment order and amount allocation relationship between bills:

[0108]

[0109] where, and are the output feature vectors of bill i and bill j at the Lth layer in the graph convolution network, respectively, and MLP(·) represents the multilayer perception, which learns the complex relationship between bills through multilayer nonlinear mapping. This model is used to predict the settlement rules and priorities between bills.

[0110] S35, based on the similarity between bills and the output of the graph neural network, a settlement priority matrix P is constructed:

[0111]

[0112] where w is the weight vector, and are the outputs of bill i and bill j at the Lth layer in the graph convolution network, respectively, and P ijrepresents the settlement priority between bill i and bill j, exp(.) is the natural exponential function, and a is an adjustment factor;

[0113] S36, combine the derived settlement priority matrix with the amount allocation strategy to obtain the final settlement order and amount allocation scheme of the bills.

[0114] The present application trains the model through historical accounting data, and adjusts the parameters of the model by using optimization methods such as back propagation and gradient descent, so as to improve the accuracy and efficiency in the process of accounting clearing and settlement. Through training, the present application can continuously optimize the adaptability of the model under different accounting data environments, and improve the performance of the system in actual application. This step provides the system with the ability of self-learning, so that it can continuously improve the settlement accuracy with the accumulation of data, and avoids the limitations of manual intervention and rule hard coding.

[0115] In the embodiment, S4 includes the following steps:

[0116] S41, training the historical accounting data as a training set;

[0117] S42, using the gradient descent method to optimize the loss function, and calculating the gradient of each parameter j . The formula is as follows:

[0118]

[0119] Wherein, η is the learning rate, controlling the step size of each parameter update, represents the gradient of the loss function L(θ) with respect to the parameter j .

[0120] S43, using the back propagation algorithm to calculate the gradient of each layer of the network, and updating the parameters of the capsule model layer by layer, by calculating the output error (k) of each layer, the formula is as follows:

[0121] (k) (k+1) T (k+1) (k)

[0122] Wherein, δ (k) is the error term of the kth layer, W (k+1) is the weight matrix of the k+1th layer, z (k) is the weighted input of the kth layer, and f′(z (k) ) is the derivative of the activation function.

[0123] ​​​​​​The application performs real-time clearing and settlement processing on newly generated billing data through the trained model. This step reduces the need for manual intervention and improves the real-time nature and efficiency of account clearing and settlement by automatically processing new billing data. The inference results of the model not only enable amount matching and payment sequence determination between bills, but also automatically update the settlement status in the supply chain, thereby improving the response capability and accuracy of the system in a rapidly changing supply chain environment.

[0124] In this embodiment, S5 includes the following steps:

[0125] S51, real-time automatic clearing and settlement processing of newly generated billing data, inputting the new billing data into the trained capsule network and relationship reasoning model to obtain clearing and settlement information of each bill;

[0126] S52, based on the inference result, calculating the amount matching between bills, determining the payment sequence of each bill, assigning a payment time sequence to each bill, and automatically updating the settlement status in the supply chain.

[0127] Reference Figure 2 The deep learning-based supply chain account clearing and settlement system includes the following modules:

[0128] A data acquisition module for acquiring account data in the supply chain and performing standardization processing;

[0129] A feature extraction module for performing preliminary feature extraction on the collected account data using YOLOv10 to obtain basic visual and numerical information of the account data, and further extracting spatial and pose relationships between bills through a capsule network;

[0130] A relationship modeling module for modeling the complex relationships between bills based on the relationship reasoning module, analyzing the payment sequence and amount allocation factors between bills, and generating priority and settlement rules between bills;

[0131] A model training module for training the capsule network and relationship reasoning module using historical account data as a training set through backpropagation and gradient descent optimization methods;

[0132] A real-time clearing and settlement module for real-time automatic clearing and settlement processing of newly generated billing data, performing amount matching and payment sequence determination between bills based on the inference results, and automatically updating the settlement status in the supply chain;

[0133] A settlement report generation module for generating detailed settlement reports based on the results of automatic clearing and settlement, and returning the settlement reports to the supply chain nodes for payment processing.

[0134] Embodiment:

[0135] The embodiment is applied in a large electronic product manufacturing enterprise which cooperates with multiple component suppliers around the world. The accounting data involved is updated in large quantities every day. The bills provided by each supplier contain different information fields, including invoice amount, payment terms, discounts, taxes, and payment methods. Since multiple suppliers provide bills at the same time and there are complex payment sequences and amount allocation relationships among the bills, the traditional rule-based accounting clearing system cannot handle these complex situations, and the common problems in traditional systems are high manual involvement, slow processing speed, and easy errors.

[0136] In this context, the enterprise introduces a deep learning method based on capsule networks and relationship reasoning to realize the automation and intelligentization of accounting clearing settlement, optimize the supply chain settlement process, and the system needs to automatically identify the payment sequence and amount allocation relationship among bills and give the optimal clearing settlement scheme.

[0137] After the enterprise applies the accounting clearing settlement method of the present application, the data collection module first collects bill information from different suppliers, and standardizes these bill data. The system converts the information in the bills (such as amount, invoice number, supplier information, etc.) into numerical feature vectors for subsequent deep learning model processing.

[0138] Then, YOLOv10 performs preliminary feature extraction on these data, extracting local features in the accounting data, such as bill amount, supplier information, and payment conditions. Next, the capsule network module models the spatial relationship among the bills, enabling the system to capture complex dependency relationships and long-distance dependencies among the bills, such as payment sequence and amount allocation among multiple bills from the same supplier.

[0139] The relationship reasoning module then reasons the complex relationships among the bills, analyzes the payment sequence, amount allocation, and other factors among the bills, and generates the priority and settlement rules among the bills. In this way, the system not only identifies the direct relationships among the bills, but also handles the implicit relationships among the bills, greatly improving the accuracy and flexibility of the settlement.

[0140] Through the training of historical accounting data, the system can automatically clear and settle new bills in real-time processing, reducing manual intervention and human error. After each settlement, the system also automatically generates a detailed settlement report, including payment details, amount allocation, settlement status, and other information of the bills, for further auditing and payment by relevant supply chain nodes.

[0141] In a practical application, the system was used by the enterprise to clear and settle the accounting data from January 1, 2025 to January 31, 2025, involving 2200 invoices from 50 suppliers. The traditional manual accounting clearing method required 7 days to complete the settlement, and the settlement results often had large errors. After the system was put into use, the settlement time was shortened to 2 days when processing the same number of invoices, and the settlement error rate was reduced from 15% of the traditional system to 2%. The following is a specific data comparison:

[0142] Table 1 Performance comparison of the present application and the traditional method in accounting data clearing and settlement

[0143] Item Conventional system System of the present application Bill quantity 2200 pieces 2200 pieces Total settlement time 7 days 2 days Total amount About 1.2 billion RMB About 1.2 billion RMB Settlement error rate 15％ 2％ Manual intervention times 150 times 30 times Time for automatically generating settlement report Cannot automatically generate 30 minutes Supplier settlement correction times 50 times 5 times

[0144] In the data comparison, it can be seen that by applying the method of the present application, the settlement time is greatly shortened, the error rate is significantly reduced, and the number of manual interventions is reduced, improving the overall settlement efficiency. At the same time, the system can automatically generate a settlement report, reducing the workload of manually generating reports and avoiding potential errors caused by manual operation.

[0145] This embodiment fully demonstrates the application effect of the supply chain accounting clearing and settlement method based on capsule network and relationship reasoning in practical scenarios. Through intelligent accounting clearing and settlement, the enterprise significantly improves the efficiency and accuracy of settlement, reduces the need for manual intervention, optimizes the supply chain settlement process, and the application of the system not only improves the accuracy of settlement, but also improves the overall efficiency and transparency of supply chain management by automatically generating settlement reports and processing accounting data in real time.

[0146] The above is only the preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can make equivalent replacements or changes to the technical solutions and inventive concepts of the present application within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A deep learning-based supply chain accounting clearing and settlement method, characterized in that, Comprise the following steps: S1, collecting the accounting data in the supply chain, and performing standardization processing; S2, processing the accounting data through the capsule network, distributing and updating the features based on the dynamic routing mechanism, capturing the spatial and posture relationships between the accounting data elements, and generating the advanced feature representation of each billing element; S3, based on the advanced feature representation of the billing element, using the relationship reasoning module to model the complex relationship between the bills, establishing the relationship graph between the bills, and deriving the priority of each bill through the reasoning mechanism to obtain the final settlement order and the amount allocation scheme of the bill; The S3 comprises the following steps: S31, according to the advanced feature representation of each billing element obtained from the capsule network, using the relationship reasoning module to model the complex relationship between the bills, analyzing the correlation, payment order and amount allocation information between the bills; S32, constructing a graph of bills by calculating the similarity between the bills and the bills to build a relationship graph between the bills, each node representing a bill element, and the edges representing the relationship between the bill elements:​ ； wherein, denotes a similarity function, is a bill feature vector, is a bill feature vector, and are the i-th and j-th element of the feature vector, respectively, is the node degree; S33、each bill node The feature update formula in the graph neural network is: ； in, Indicates bill In the Feature vectors after layer graph convolution It's a bill. and bills Relationship matrix elements It is a node The degree, Represents nodes The set of adjacent neighbor nodes, It is the sigmoid activation function; S34. Building upon graph convolution, a multilayer perceptron is further used to nonlinearly model the relationships between bill nodes, by using the features of each bill... and The data is fed into a multilayer perceptron to calculate the payment order and amount allocation strategy among the bills. ； wherein, and respectively are the bills and the bills the output feature vector of the Lth layer in the graph convolution network, denotes a multi-layer perceptron that learns complex relationships between bills through multi-layer nonlinear mapping, which is used to predict the settlement rules and priorities between bills; S35、based on the similarity between the bills and the graph neural network output, construct a settlement priority matrix : ； wherein, is a weight vector, and are the bills and the bill at the output of the -th layer in the graph convolution network, denotes the settlement priority between the bills and the bill , is a natural exponential function, is a tuning factor; S36, combining the derived settlement priority matrix with the amount allocation strategy to obtain the final settlement order and the amount allocation scheme of the bill; S4, using historical accounting data as a training set, training the capsule network and the relationship reasoning module through back propagation and gradient descent method; S5, applying the trained capsule network and relationship reasoning module to real-time automatic clearing and settlement processing of newly generated billing data, matching the amount between the bills, determining the payment order, and automatically updating the settlement status in the supply chain according to the reasoning result; S6, according to the result of automatic clearing and settlement, generating a detailed settlement report, and returning the settlement report to the supply chain node for payment processing.

2. The deep learning based supply chain accounting clearing and settlement method according to claim 1, characterized in that, The S1 comprises the following steps: S11, collecting the accounting data in the supply chain, including invoices, bills, payment records, and supplier information data; S12, preprocessing the collected accounting data, including data cleaning, removing invalid and duplicate records, filling missing data, and standardizing different formats of data; S13, normalizing the accounting data. 3.The deep learning based supply chain accounting clearing and settlement method according to claim 1, characterized in that, The S2 comprises the following steps: S21, using YOLOv10 to extract the initial features of the collected accounting data, and inputting the initial feature vector extracted by YOLOv10 into the capsule unit in the capsule network for further processing; S22, in the capsule unit, distributing and updating the features through the dynamic routing mechanism to capture the spatial and posture relationships between the accounting data elements; Let the output vector of the th capsule be , which is processed by a linear transformation and a nonlinear activation function to obtain the transformed output vector : ； wherein, is a rotation matrix from capsule to capsule , is a bias term for capsule , is a bias term for capsule to capsule , through rotation matrix and affine transformation, the model can capture the spatial transformation relationship between the bills; S23, adopt the route mechanism to establish the relationship between the capsules, calculate the route coefficient between the capsules to obtain the influence of each capsule: ； in, It is a capsule The weight vector, It is a capsule To capsule The output vector, For capsules The set of adjacent nodes, It is a natural exponential function. These are the normalized routing coefficients; S24, update the output vector of each capsule by weighted sum of the transformed vectors of each capsule with the routing coefficients update the output vector of each capsule by weighted sum of the transformed vectors of each capsule ； wherein denotes a vector modulo operation, e is a small constant to prevent division by zero errors, denotes the Euclidean norm; S25, generating a high-level feature representation for each bill element based on the output of the capsule network , generating a high-level feature representation for each bill element based on the output of the capsule network 4. The deep learning based supply chain accounting clearing and settlement method according to claim 1, characterized in that, The S4 comprises the following steps: S41, training the historical accounting data as a training set; S42, adopt gradient descent method to optimize the loss function, calculate the gradient of each parameter of the formula as follows: ​ ； wherein, is the learning rate, controlling the step size of each parameter update, denotes the loss function for the parameters of the gradient; S43, the gradient of each layer of the network is calculated by using the back propagation algorithm, and the parameters of the capsule model are updated layer by layer, and the output error of each layer is calculated , the formula is as follows: ； in, For the first Layer error term, For the first The weight matrix of the layer, For the first Weighted input of the layer, It is the derivative of the activation function.

5. The deep learning based supply chain accounting clearing and settlement method according to claim 1, characterized in that, The S5 comprises the following steps: S51, real-time automatic clearing and settlement processing of newly generated billing data, inputting the new billing data into the trained capsule network and relationship reasoning model to obtain the clearing and settlement information of each bill; S52, based on the reasoning result, calculating the amount matching between the bills, determining the payment order of each bill, assigning a payment time order to each bill, and automatically updating the settlement status in the supply chain.

6. A deep learning-based supply chain accounting clearing and settlement system, which performs the deep learning-based supply chain accounting clearing and settlement method according to any one of claims 1 to 5, characterized by, Comprise the following modules: A data collection module for collecting accounting data in the supply chain and performing standardization processing; The feature extraction module is configured to perform preliminary feature extraction on the collected accounting data by using YOLOv10, to obtain basic visual and numerical information of the accounting data, and to further extract spatial and posture relationships between bills by using a capsule network. The relationship modeling module is configured to model complex relationships between bills based on the relationship reasoning module, to analyze payment order and amount allocation factors between bills, and to generate priority and settlement rules between bills. The model training module is configured to use historical accounting data as a training set, to train the capsule network and the relationship reasoning module by using a back propagation and gradient descent optimization method. The real-time clearing and settlement module is configured to perform real-time automatic clearing and settlement processing on newly generated bill data, to perform amount matching and payment order determination between bills according to the reasoning result, and to automatically update the settlement state in the supply chain. The settlement report generation module is configured to generate a detailed settlement report according to the result of the automatic clearing and settlement, and to return the settlement report to the supply chain node for payment processing.

Images