Supply chain financial demand analysis system and method based on big data analysis

Abstract

The invention relates to the technical field of supply chain financial investment funds, and discloses a supply chain financial demand analysis system and method based on big data analysis, and the system comprises a data collection module, a data storage module, a data processing module, a data analysis module, a data dependence relation generation module, a risk early warning module, and a fusion module. The data dependency relationship generation module is used for generating a time dependency relationship of the predicted credit based on the time characteristics and a long-short-term memory network; the risk early warning module is used for generating risk information of the target enterprise based on an adversarial network and the noise reduction data; the fusion module is used for carrying out feature fusion on the time dependency relationship and the risk information to obtain enterprise demand features of the target enterprise; and the application module is used for judging whether the target enterprise needs to be popularized based on the enterprise demand characteristics. The supply chain financial demand analysis system can solve the problem that an existing supply chain financial demand analysis system is low in working efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of supply chain finance investment funds, and particularly to a supply chain finance demand analysis system and method based on big data analysis. Background Art

[0002] Supply chain finance demand analysis is a typical paradigm of technology-driven financial innovation. Its core lies in reshaping the risk, efficiency, and cost structure of supply chain finance through data assetization, decision-making intelligence, and service ecologicalization.

[0003] Existing technologies mostly rely on static data such as historical transactions and finance to train models, lacking the ability to perceive real-time dynamic data. The models cannot predict the risk of future cash flow breaks in a timely manner, resulting in a lag in financing decisions. At the same time, they overly rely on structured financial data, which is not conducive to the effective allocation of funds and risk prevention and control. Therefore, how to improve the work efficiency of supply chain finance demand analysis has become a problem to be solved. Summary of the Invention

[0004] The present invention provides a supply chain finance demand analysis system and method based on big data analysis, and its main purpose is to solve the problem of low work efficiency of the existing supply chain finance demand analysis system.

[0005] To achieve the above object, the supply chain finance demand analysis system based on big data analysis provided by the present invention includes:

[0006] A data acquisition module: used to acquire the panoramic data of the supply chain business of the target enterprise;

[0007] A data storage module: used to store the panoramic data of the supply chain business into a big data storage system, and clean and aggregate the panoramic data of the supply chain business to obtain integrated supply chain data;

[0008] A data processing module: used to perform auto-encoding processing on the integrated supply chain data to obtain denoised data;

[0009] A data analysis module; used to perform forward layer-by-layer mapping on the denoised data based on a preset credit scoring algorithm to obtain predicted credit;

[0010] A data dependency generation module: used to extract the time characteristics of the predicted credit by using a pre-constructed credit scoring time model, and generate the time dependency of the predicted credit based on the time characteristics and a long short-term memory network;

[0011] A risk warning module: used to generate risk information of the target enterprise based on an adversarial network and the denoised data;

[0012] Fusion module: used to perform feature fusion on the time-dependent relationship and the risk information to obtain the enterprise demand characteristics of the target enterprise;

[0013] Application module: used to determine whether to promote the target enterprise based on the enterprise demand characteristics.

[0014] Optionally, the collection of the panoramic data of the supply chain business of the target enterprise includes: basic enterprise information, early warning information, financial information, and transaction information.

[0015] Optionally, the big data storage system includes:

[0016] Establish a data lake;

[0017] Store the panoramic data of the supply chain business in the data lake in its original format, and use the data lake to clean the panoramic data of the supply chain business;

[0018] Aggregate the cleaned panoramic data of the supply chain business to obtain the supply chain integration data of the panoramic data of the supply chain business.

[0019] Optionally, the auto-encoding process of the supply chain integration data to obtain the noise-reduced data includes:

[0020] Perform auto-encoding on the supply chain integration data based on a preset auto-encoding algorithm to obtain the noise-reduced data, where the preset auto-encoding algorithm is:

[0021]

[0022] In the formula: x is the supply chain integration data, is the feature vector, is the activation function, is the weight matrix of the encoding layer, is the encoding bias vector, is the weight matrix of the input layer, is the output bias vector, is the number of features of the input layer, is the dimension of the low-dimensional space after encoding, is the noise-reduced data, is the decoding function, represents d-dimensional real number space, represents m-dimensional real number space, represents the set composed of all real matrices with m rows and n columns, represents the set composed of all real matrices with n rows and m columns.

[0023] Optionally, the calculation formula of the credit scoring algorithm is as follows:

[0024]

[0025] In the formula: is the noise-reduced data, is the linear transformation function, is the output vector of the is the output vector of the is the output vector of the first layer, is the bias vector of the ) is the non-linear activation function, is the weight matrix of the l-th layer, is the formula for judging whether the enterprise defaults, is the predicted credit, is the output layer weight matrix, is the bias vector of the output layer, is the output vector of the is the total number of layers of the neural network, and l is a certain layer in the network;

[0026] Before analyzing the noise-reduced data based on the preset credit scoring algorithm, it further includes:

[0027] Construct the to-be-trained credit scoring time model of the target enterprise;

[0028] Generate the data set of the to-be-trained credit scoring time model;

[0029] Use the preset loss function and the data set to train the to-be-trained credit scoring time model, where the preset loss function is:

[0030]

[0031] In the formula: is the value of the cross-entropy loss function, is the number of training samples, is the true label of the i-th sample, is the predicted credit of the module for the i-th sample.

[0032] Optionally, the extracting the time features of the predicted credit by using the pre-constructed credit scoring time model includes:

[0033] Extract the time features of the predicted credit according to the time convolutional layer in the pre-constructed credit scoring time model, where the convolution algorithm of the time convolutional layer is:

[0034]

[0035] Among them, is the time feature of the predicted credit, is the noise-reduced data, is a one-dimensional convolutional kernel, is the bias term, is the time step, is the current time step.

[0036] Optionally, generating the time dependence of the predicted credit based on the time feature and the long short-term memory network includes:

[0037] Inputting the time feature into the long short-term memory network, and performing layer-by-layer processing on the time feature according to the input gate, forget gate, output gate, cell state update layer, and hidden state update layer of the long short-term memory network to obtain the time dependence of the predicted credit.

[0038] Optionally, generating the risk information of the target enterprise based on the adversarial network and the noise-reduced data includes:

[0039] The algorithm formula of the adversarial network is:

[0040]

[0041] Among them, is the noise-reduced data, is the warning information corresponding to the noise-reduced data, is to give the warning information of the risk information, is the generator in the adversarial network, is the discriminator in the adversarial network, means that the goal of the generator is to minimize its loss function, means that the goal of the discriminator is to maximize its loss function, is the expected value, means the noise-reduced data and the warning information corresponding to the noise-reduced data of the joint distribution, is the discriminator's judgment value of the noise-reduced data and the warning information corresponding to the noise-reduced data of the logarithm, is the random noise, is the random noise of the probability distribution, is the logarithm of calculating that the risk information is misclassified as false information by the discriminator, It is the discriminator's evaluation of risk information;

[0042] Based on the noise-reduced data and the early warning information, use the adversarial network to generate the risk information of the target enterprise.

[0043] Optionally, the feature fusion of the time dependence and the risk information to obtain the enterprise demand features of the target enterprise includes:

[0044] Based on the attention mechanism, perform feature fusion on the time dependence and the risk information to obtain the enterprise demand features of the target enterprise.

[0045] Optionally, a method for generating reference information based on artificial intelligence and smart home, characterized in that the method includes:

[0046] Used to collect the panoramic data of the supply chain business of the target enterprise;

[0047] Used to store the panoramic data of the supply chain business in the big data storage system, and clean and aggregate the panoramic data of the supply chain business to obtain the supply chain integration data;

[0048] Used to perform auto-encoding processing on the supply chain integration data to obtain noise-reduced data;

[0049] Used to perform a forward layer-by-layer mapping on the noise-reduced data based on a preset credit scoring algorithm to obtain a predicted credit;

[0050] Used to extract the time features of the predicted credit using a pre-constructed credit scoring time model, and generate the time dependence of the predicted credit based on the time features and the long short-term memory network;

[0051] Used to generate the risk information of the target enterprise based on the adversarial network and the noise-reduced data;

[0052] Used to perform feature fusion on the time dependence and the risk information to obtain the enterprise demand features of the target enterprise;

[0053] Used to determine whether to promote the target enterprise based on the enterprise demand features.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] The present invention integrates data, denoises data, and predicts credit through the supply chain, and constructs a credit scoring time model to dynamically associate time features with predicted credit. Through the fusion module, it solves the problems of the prior art relying on historical information, lacking the ability to perceive real-time dynamic data, and being unable to predict the risk of future cash flow breakage in a timely manner, and improves the efficiency of the financial demand analysis system. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a schematic flowchart of a supply chain financial demand analysis system based on big data analysis provided by an embodiment of the present invention;

[0057] Figure 2 It is a schematic flowchart of a supply chain financial demand analysis method based on big data analysis provided by an embodiment of the present invention; DETAILED DESCRIPTION OF THE EMBODIMENTS

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0059] The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The singular forms "the" and "said" used in the embodiments of the present invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. "Plurality" generally includes at least two.

[0060] Depending on the context, the words "if" or "when" as used herein may be interpreted as "when" or "while" or "in response to determining" or "in response to detecting". Similarly, depending on the context, the phrase "if determined" or "if detecting (stated condition or event)" may be interpreted as "when determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)".

[0061] The embodiments of the present application provide a supply chain finance demand analysis system and method based on big data analysis. The execution subjects of the supply chain finance demand analysis system and method based on big data analysis include, but are not limited to, at least one of electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided by the embodiments of the present application. In other words, the supply chain finance demand analysis system and method based on big data analysis can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to: a single server, a server cluster, a cloud server, or a cloud server cluster, etc. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, Content Delivery Network (CDN), and big data and artificial intelligence platforms.

[0062] As Figure 1 shown, it is a schematic flowchart of the supply chain finance demand analysis system based on big data analysis of the present invention. The supply chain finance demand analysis system 100 based on big data analysis of the present invention can be installed in an electronic device. According to the functions achieved, the supply chain finance demand analysis system 100 based on big data analysis can include a data collection module 101, a data storage module 102, a data processing module 103, a data analysis module 104, a data dependency generation module 105, a risk warning module 106, a fusion module 107, and an application module 108. The modules of the present invention can also be referred to as units, which refer to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, and are stored in the memory of the electronic device.

[0063] In this embodiment, the functions of each module / unit are as follows:

[0064] The data collection module 101 is used to collect the panoramic data of the supply chain business of the target enterprise. The collection of the panoramic data of the supply chain business of the target enterprise includes: enterprise basic information, warning information, financial information, and transaction information.

[0065] Specifically, the enterprise basic information includes the enterprise registered capital, the establishment years, the industry category, etc. The financial information includes the key indicators of the balance sheet and the profit and loss statement. The transaction information includes multi-dimensional data such as the transaction frequency, the transaction amount, and the payment timeliness.

[0066] Specifically, the registered capital of an enterprise is the total capital registered with the registration and management agency, which is the total amount of capital subscribed by the enterprise's shareholders or the total amount of share capital subscribed. It reflects the initial capital scale of the enterprise and the limit of its liability assumption. A higher registered capital usually means that the enterprise has relatively strong financial strength at the time of establishment and may have certain advantages in market competition, business expansion, and risk assumption. For example, when participating in the bidding for large projects, a higher registered capital may be an important reference indicator for the enterprise to possess corresponding strength.

[0067] Specifically, the establishment years reflect the enterprise's survival time in the market. Enterprises with a longer establishment years usually have richer market experience, more stable customer groups, and more mature operation and management models.

[0068] Specifically, the industry category reflects the economic field and business scope in which the enterprise is located. Different industries have different market environments, competition levels, development trends, and risk characteristics. For example, the high-tech industry usually features high investment, high risk, and high return, with a fast pace of technological innovation and fierce market competition; while traditional manufacturing industries pay more attention to production scale, cost control, and product quality. Understanding the industry category of an enterprise can effectively analyze the market opportunities and challenges faced by the enterprise and evaluate its competitiveness and development prospects in the industry. Industrial and commercial registration information: When an enterprise conducts industrial and commercial registration, it will clearly fill in the industry category to which it belongs, and the industry code and corresponding industry name registered by the enterprise can be queried through the National Enterprise Credit Information Publicity System.

[0069] There are corresponding industry associations and competent departments for different industries, and they have relatively detailed records of the enterprise information within their respective industries. The industry category information of an enterprise can be obtained by contacting the relevant industry associations or competent departments.

[0070] Specifically, the enterprise's financial information includes: current assets, non-current assets, current liabilities, non-current liabilities, and owners' equity.

[0071] Current assets include monetary funds, accounts receivable, inventory, etc. Monetary funds reflect the enterprise's cash reserves and payment capabilities; accounts receivable represent the amounts that should be recovered after the enterprise sells goods or provides services, and their scale and aging distribution affect the enterprise's capital recovery speed and bad debt risks; inventory is the goods held by the enterprise for production or sales, and the inventory turnover speed reflects the enterprise's operation efficiency.

[0072] Non-current assets such as fixed assets, intangible assets, etc. Fixed assets are the long-term assets used by the enterprise for production and operation, and their scale and quality affect the enterprise's production capacity and competitiveness; intangible assets include patents, trademarks, etc., which are of great value to some high-tech enterprises and knowledge-intensive enterprises.

[0073] Current liabilities include short-term borrowings, accounts payable, etc. Short-term borrowings reflect the debts that an enterprise needs to repay in the short term, and accounts payable reflect the amount owed by the enterprise to its suppliers. Excessive current liabilities may lead to greater short-term debt repayment pressure on the enterprise.

[0074] Non-current liabilities include long-term borrowings, bonds payable, etc. The scale and maturity structure of long-term liabilities affect the long-term financial stability and financing costs of an enterprise.

[0075] Owner's equity is the residual interest enjoyed by the owners after deducting liabilities from the assets of the enterprise, including paid-in capital, capital reserve, surplus reserve, and undistributed profits, etc. Owner's equity reflects the net asset scale of the enterprise and the equity status of the shareholders.

[0076] Specifically, for listed companies, they are required to disclose financial statements, including the balance sheet, regularly in accordance with relevant laws and regulations. The annual reports, semi-annual reports, and quarterly reports of listed companies can be obtained through platforms such as the websites of stock exchanges, and the balance sheet information can be extracted from them. For non-listed companies, their balance sheets can be directly obtained from the enterprise by establishing a cooperative relationship with the enterprise and signing a confidentiality agreement, etc. In the scenario of supply chain finance, core enterprises may require upstream and downstream enterprises to provide financial statements to evaluate their credit status. At the same time, enterprises need to submit financial statements to the tax authorities, and the tax authorities may have information on the balance sheets of enterprises. Under the condition of compliance with laws, regulations, and relevant procedures, relevant data can be obtained through the tax authorities.

[0077] Specifically, the key indicators of the income statement are obtained in a similar way to those of the balance sheet. The income statement of listed companies can be obtained from platforms such as the websites of stock exchanges; for non-listed companies, it can be obtained internally within the enterprise or from the tax authorities under the premise of being legal and compliant.

[0078] Furthermore, operating revenue represents the total revenue realized by an enterprise in its daily business activities and is an important manifestation of the enterprise's operating results. The growth of operating revenue reflects the enterprise's market expansion ability and business development trend.

[0079] Cost of sales represents the costs incurred by an enterprise in producing products or providing services, including direct materials, direct labor, and manufacturing expenses, etc. The control level of the cost of sales affects the profitability of the enterprise.

[0080] Operating profit represents the balance after deducting the cost of sales, taxes and surcharges, selling expenses, administrative expenses, R & D expenses, and financial expenses, etc. from the operating revenue. Operating profit reflects the profitability of the enterprise's core business.

[0081] Total profit represents the amount obtained by adding non-operating income to the operating profit and then subtracting non-operating expenses. Total profit reflects the overall profitability of the enterprise during a certain accounting period.

[0082] Net profit represents the net amount obtained by subtracting income tax expenses from the total profit, which is the ultimate profit result of an enterprise. The level of net profit directly affects the return to shareholders of the enterprise and its own development capabilities.

[0083] The data storage module 102 is used to store the panoramic data of the supply chain business in the big data storage system, and clean and aggregate the panoramic data of the supply chain business to obtain integrated supply chain data. The big data storage system includes:

[0084] Build a data lake, store the panoramic data of the supply chain business in the data lake in its original format, use the data lake to clean the panoramic data of the supply chain business, and aggregate the cleaned panoramic data of the supply chain business to obtain the integrated supply chain data of the panoramic data of the supply chain business.

[0085] Specifically, the integrated supply chain data refers to a structured and themed dataset formed after processing the original panoramic data of the supply chain business, such as cleaning, standardizing, and aggregating. Its core objectives are to eliminate data silos: integrate fragmented data scattered in different systems (such as ERP, TMS, e-commerce platforms), enhance data value: transform the original data into high-value information that can be directly used for analysis through business logic abstraction, and support decision-making: provide underlying data support for core functions such as credit assessment, risk warning, and demand forecasting in supply chain finance.

[0086] Specifically, use streaming data collection tools such as Kafka and Flume to receive data from the core enterprise ERP system, logistics TMS, and e-commerce platform in real time, store the original data through the HDFS distributed file system, and store the original data "as it is" to form the basis of the data lake.

[0087] Specifically, the data cleaning and aggregation process can significantly improve data quality, ensure the accuracy of subsequent analysis, and the high-quality integrated data is more easily and efficiently utilized by models to support accurate decision-making.

[0088] The data processing module 103 is used to perform auto-encoding processing on the integrated supply chain data to obtain denoised data; the performing auto-encoding processing on the integrated supply chain data to obtain denoised data includes: performing auto-encoding processing on the integrated supply chain data based on a preset auto-encoding algorithm to obtain denoised data, where the preset auto-encoding algorithm is:

[0089]

[0090] In the formula: x is the integrated supply chain data, is the feature vector, is the activation function, is the weight matrix of the encoding layer, is the encoding bias vector, is the output bias vector, is the weight matrix of the input layer, is the number of features in the input layer, is the dimension of the encoded low-dimensional space, is the denoised data, is the decoding function, represents d-dimensional real space, represents m-dimensional real space, represents the set of all real matrices with m rows and n columns, represents the set of all real matrices with n rows and m columns.

[0091] Specifically, the feature vector is the structured data input into the neural network, which is a high-dimensional vector formed by preprocessing the panoramic data of the target enterprise's supply chain business (such as financial indicators, transaction records, logistics data, etc.).

[0092] Specifically, the activation function is the key component that introduces non-linearity in the neural network, determining whether a neuron is activated, that is, the intensity of the output signal.

[0093] Specifically, the weight matrix is the connection parameter between adjacent layers in the neural network, representing the connection strength between neurons. The dimension of the matrix is , and the weights are adjusted through training to quantify the contribution of each input feature to the output.

[0094] Specifically, the encoding bias vector is the trainable parameter of each neuron in the neural network, used to adjust the threshold of the activation function. In an autoencoder (AE) or variational autoencoder (VAE), the encoding bias vector specifically refers to the bias of the encoder part.

[0095] Specifically, the decoding function is a component in an autoencoder or generative adversarial network (GAN), which restores the low-dimensional encoded vector to an approximation of the original input. In a credit assessment model, the decoding function is used for reconstruction verification after feature dimensionality reduction.

[0096] Specifically, the number of features in the input layer, similar to the dimension of the input data in an autoencoder, is the total number of features used for credit assessment, such as the number of features like the registered capital of the enterprise, the number of years since establishment, and past transaction records.

[0097] Specifically, the dimension of the encoded low-dimensional space is the dimension of the encoded low-dimensional space. One of the purposes of an autoencoder is to compress high-dimensional data into a low-dimensional space, and the dimension of the encoded low-dimensional space is the number of features represented after compression, and .

[0098] Specifically, the auto - encoding algorithm learns the latent feature representation of data through unsupervised learning, and can retain the key information of the data while reducing noise. This makes the processed data more robust to noise and outliers, especially suitable for the complex scenarios of multi - source heterogeneous data in the supply chain.

[0099] Specifically, the auto - encoding algorithm can learn the internal structure of data without labeled data, which is particularly important for the massive unlabeled data in the supply chain. Through auto - encoding processing, the hidden patterns and associations in the data can be mined, providing better feature inputs for supply chain intelligent applications.

[0100] The data analysis module 104 is used to perform a forward - layer - by - layer mapping on the denoised data based on a preset credit scoring algorithm to obtain a predicted credit. The calculation formula of the credit scoring algorithm is:

[0101]

[0102] In the formula: is the denoised data, is the linear transformation function, is the output vector of the is the output vector of the is the output vector of the first layer, is the bias vector of the ) is the non - linear activation function, is the weight matrix of the l - th layer, is the formula for judging whether an enterprise defaults, is the predicted credit, is the output - layer weight matrix, is the output - layer bias vector, is the output vector of the is the total number of layers of the neural network, and l is a certain layer in the network.

[0103] Specifically, the total number of layers of the neural network includes an input layer, hidden layers, and an output layer. For example, in a simple three - layer neural network (input layer, one hidden layer, output layer), then .

[0104] Specifically, l is usually used to represent a certain layer in the network. It is a variable, and its value range is from 1 to L. Each l corresponds to a specific layer in the network.

[0105] Before analyzing the denoised data based on the preset credit scoring algorithm, it also includes:

[0106] Build the to-be-trained credit scoring time model of the target enterprise;

[0107] Generate a data set for the to-be-trained credit scoring time model;

[0108] Use a preset loss function and the data set to train the to-be-trained credit scoring time model, where the preset loss function is:

[0109]

[0110] In the formula: is the value of the cross-entropy loss function, is the number of training samples, is the true label of the i-th sample, is the predicted credit of the module for the i-th sample.

[0111] Specifically, the number of samples is the number of samples included in the data set used to train the autoencoder. For example, if the relevant data of 1000 enterprises is used for training, then N = 1000.

[0112] Data dependency generation module 105: used to extract the time features of the predicted credit by using a pre-constructed credit scoring time model, and generate the time dependency of the predicted credit based on the time features and a long short-term memory network. The extracting the time features of the predicted credit by using a pre-constructed credit scoring time model includes: extracting the time features of the predicted credit according to the time convolutional layer in the pre-constructed credit scoring time model, where the convolutional algorithm of the time convolutional layer is:

[0113]

[0114] Among them, is the time feature of the predicted credit, is the noise-reduced data, is a one-dimensional convolutional kernel, is a bias term, is the time step size, is the current time step.

[0115] The generating the time dependency of the predicted credit based on the time features and a long short-term memory network includes: inputting the time features into the long short-term memory network, and performing layer-by-layer processing on the time features according to the input gate, forget gate, output gate, cell state update layer and hidden state update layer of the long short-term memory network to obtain the time dependency of the predicted credit.

[0116] Specifically, extract time-related features from the panoramic data of supply chain operations, such as a company's historical credit scores, financial metrics at different time points (such as monthly debt-to-asset ratios, quarterly net profits, etc.), time series of transaction records (such as weekly transaction amounts, annual transaction frequencies), and so on.

[0117] Furthermore, the long short-term memory network can process multiple time features simultaneously, comprehensively analyze information from multiple aspects such as a company's financial data, transaction records, and credit history, so as to more comprehensively evaluate the company's credit risk. Compared with traditional credit assessment methods, the long short-term memory network can uncover more potential information and improve the accuracy of predictions.

[0118] Specifically, adapt to complex patterns: In reality, credit data often has complex patterns and non-linear relationships. Through its non-linear activation functions and multi-layer structure, the long short-term memory network can learn these complex patterns, thus better fitting the data and improving the accuracy of predictions.

[0119] Specifically, the long short-term memory network has a certain ability to resist interference from noise in time series data. Through the mechanism of the forget gate, it can filter out some unimportant information and only retain information that has an important impact on predictions, thereby improving the robustness of the model.

[0120] Specifically, in practical applications, there may be missing values in time series data. The long short-term memory network can, to a certain extent, compensate for the impact of missing data through its memory function and still make relatively accurate predictions.

[0121] The risk warning module 106 is used to generate risk information for the target company based on the adversarial network and the noise-reduced data. Generating the risk information for the target company based on the adversarial network and the noise-reduced data includes:

[0122] The algorithm formula of the adversarial network is:

[0123]

[0124] Where, is the noise-reduced data, is the warning information corresponding to the noise-reduced data, is the given warning information of the risk information, is the generator in the adversarial network, is the discriminator in the adversarial network, indicates that the goal of the generator is to minimize its loss function, indicates that the goal of the discriminator is to maximize its loss function, is the expected value, representing the noise reduction data and the warning information corresponding to the noise reduction data of the joint distribution is the discriminator for the noise reduction data and the warning information corresponding to the noise reduction data the logarithm of the decision value is random noise is random noise of the probability distribution is to calculate the logarithm of the risk information being misclassified as false information by the discriminator is the discriminator's assessment of the risk information;

[0125] Based on the noise reduction data and the warning information, use the adversarial network to generate the risk information of the target enterprise.

[0126] The fusion module 107 is used to perform feature fusion on the time dependence relationship and the risk information to obtain the enterprise demand characteristics of the target enterprise;

[0127] The performing feature fusion on the time dependence relationship and the risk information to obtain the enterprise demand characteristics of the target enterprise includes:

[0128] Based on the attention mechanism, perform feature fusion on the time dependence relationship and the risk information to obtain the enterprise demand characteristics of the target enterprise.

[0129] Specifically, the attention mechanism can automatically assign different weights to different features in the time dependence relationship and the risk information. In the time dependence relationship, the data or trends at certain time points may have a more critical impact on the enterprise demand characteristics. For example, in the risk information, certain specific risk factors (such as major legal lawsuits, market fluctuations of core businesses, etc.) may have a decisive role in the enterprise's demand. The attention mechanism can focus on these important features and ignore the irrelevant information, thereby more accurately depicting the enterprise's demand characteristics.

[0130] Specifically, as the data changes, the attention mechanism will dynamically adjust the weights of each feature. For example, when the enterprise faces different development stages or market environments, the importance of certain stage features in the time dependence relationship and certain risk factors in the risk information will change. The attention mechanism can capture this change in real time, adjust the weight allocation, and ensure that it always focuses on the most critical information.

[0131] Specifically, by integrating supply chain data, denoising data, predicting credit, and constructing a credit scoring time model, time features are dynamically associated with predicted credit. Through the fusion module, the problems of the existing technology relying on historical information, lacking the ability to perceive real-time dynamic data, and being unable to predict the risk of future cash flow breaks in a timely manner are solved, improving the efficiency of the financial demand analysis system.

[0132] The application module 108 is used to determine whether to promote the target enterprise based on the enterprise demand characteristics.

[0133] Specifically, customers are classified into four categories according to enterprise demand characteristics: high-value customers, potential customers, risky customers (such as those on the dishonesty list or with major default records), and long-tail customers: those with a small funding gap, unclear demands but active on the supply chain platform.

[0134] Furthermore, for high-value customers, door-to-door service by customer managers is the main approach, combined with pop-up notifications on the supply chain platform, and a dedicated service hotline is synchronously opened. The promotion content emphasizes "exclusive quota" and "green channel", such as "Your exclusive financing quota has been increased to 10 million yuan".

[0135] Potential customers: Reach them through message push on the supply chain platform and targeted emails, and assist with invitations to industry conferences. The content focuses on "time-limited offers", such as "The first financing interest rate for new customers is reduced by 0.5%".

[0136] Long-tail customers: Cover them with text messages and automated emails, and push concise copywriting, such as "Apply online in 3 minutes, and the fastest loan can be released on the same day". Combine with social media advertising to expand the coverage.

[0137] Refer to Figure 2 As shown, it is a schematic flowchart of the supply chain financial demand analysis method based on big data analysis provided by an embodiment of the present invention. In this embodiment, the supply chain financial demand analysis method based on big data analysis includes:

[0138] S1. Collect the panoramic data of the supply chain business of the target enterprise;

[0139] S2. Store the panoramic data of the supply chain business in the big data storage system, and clean and aggregate the panoramic data of the supply chain business to obtain supply chain integration data;

[0140] S3. Perform auto-encoding processing on the supply chain integration data to obtain denoised data;

[0141] S4. Based on a preset credit scoring algorithm, perform forward layer-by-layer mapping on the denoised data to obtain predicted credit;

[0142] S5. Extract the time features of the predicted credit using a pre-constructed credit scoring time model, and generate the time dependence of the predicted credit based on the time features and a long short-term memory network;

[0143] S6. Generate the risk information of the target enterprise based on an adversarial network and the noise-reduced data;

[0144] S7. Perform feature fusion on the time dependence and the risk information to obtain the enterprise demand features of the target enterprise;

[0145] S8. Determine whether to promote the target enterprise based on the enterprise demand features.

[0146] In several embodiments provided by the present invention, it should be understood that the disclosed method and system can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0147] The modules described as separate components may or may not be physically separated. The components shown as modules may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0148] In addition, in each embodiment of the present invention, the functional modules can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated unit can be implemented in the form of hardware or in the form of a hardware plus software functional module.

[0149] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention.

[0150] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence is the theory, method, technology and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0151] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A supply chain finance demand analysis system based on big data analysis, characterized in that, The system includes: A data acquisition module: used to acquire the panoramic data of the supply chain business of the target enterprise; A data storage module: used to store the panoramic data of the supply chain business in a big data storage system, and clean and aggregate the panoramic data of the supply chain business to obtain integrated supply chain data; A data processing module: used to perform auto-encoding processing on the integrated supply chain data to obtain denoised data; A data analysis module; used to perform forward layer-by-layer mapping on the denoised data based on a preset credit scoring algorithm to obtain predicted credit; A data dependency generation module: used to extract the time characteristics of the predicted credit using a pre-constructed credit scoring time model, and generate the time dependency of the predicted credit based on the time characteristics and a long short-term memory network; A risk warning module: used to generate risk information of the target enterprise based on a confrontation network and the denoised data; A fusion module: used to perform feature fusion on the time dependency and the risk information to obtain the enterprise demand characteristics of the target enterprise; An application module: used to determine whether to promote the target enterprise based on the enterprise demand characteristics.

2. The supply chain finance demand analysis system based on big data analysis according to claim 1, characterized in that, The acquisition of the panoramic data of the supply chain business of the target enterprise includes: basic enterprise information, warning information, financial information, and transaction information.

3. The supply chain finance demand analysis system based on big data analysis according to claim 2, characterized in that The big data storage system includes: Establishing a data lake; Storing the panoramic data of the supply chain business in the data lake in its original format, and using the data lake to clean the panoramic data of the supply chain business; Aggregating the cleaned panoramic data of the supply chain business to obtain the integrated supply chain data of the panoramic data of the supply chain business.

4. The supply chain finance demand analysis system based on big data analysis according to claim 2, wherein The performing auto-encoding processing on the integrated supply chain data to obtain denoised data includes: Performing auto-encoding processing on the integrated supply chain data based on a preset auto-encoding algorithm to obtain denoised data, where the preset auto-encoding algorithm is: Where: x is the supply chain integration data, is the eigenvector, is the activation function, is the weight matrix of the encoding layer, is the output bias vector, is the encoding bias vector, is the weight matrix of the input layer, is the number of input layer features, is the dimension of the encoded low-dimensional space, is the noise-reduced data, is the decoding function, represents d-dimensional real space, represents m-dimensional real space, represents the set consisting of all real matrices of m rows and n columns, represents the set consisting of all real matrices of n rows and m columns.

5. The supply chain finance demand analysis system based on big data analysis according to claim 4, characterized in that The calculation formula of the credit scoring algorithm is: In the formula: is the noise-reduced data, is the linear transformation function, is the output vector of layer is the output vector of layer is the output vector of the first layer, is the bias vector of layer ) is the non-linear activation function, is the weight matrix of layer l, is the formula for judging whether an enterprise defaults, is the predicted credit, is the output layer weight matrix, is the output layer bias vector, is the output vector of layer is the total number of layers of the neural network, and l is a certain layer in the network; Before analyzing the denoised data based on the preset credit scoring algorithm, it further includes: Building a credit scoring time model to be trained for the target enterprise; Generating a data set for the credit scoring time model to be trained; Training the credit scoring time model to be trained using a preset loss function and the data set, where the preset loss function is: In the formula: is the value of the cross-entropy loss function, is the number of training samples, is the true label of the i-th sample, is the predicted credit of the module for the i-th sample.

6. The supply chain finance demand analysis system based on big data analysis according to claim 5, characterized in that The extracting the time characteristics of the predicted credit using the pre-constructed credit scoring time model includes: Extracting the time characteristics of the predicted credit according to the time convolutional layer in the pre-constructed credit scoring time model, where the convolutional algorithm of the time convolutional layer is: Among them, is the time feature of the predicted credit, is the noise-reduced data, is a one-dimensional convolutional kernel, is the bias term, is the time step, is the current time step.

7. The supply chain finance demand analysis system based on big data analysis according to claim 6, characterized in that, The generating the time dependency of the predicted credit based on the time characteristics and a long short-term memory network includes: Inputting the time characteristics into a long short-term memory network, and performing layer-by-layer processing on the time characteristics according to the input gate, forget gate, output gate, cell state update layer, and hidden state update layer of the long short-term memory network to obtain the time dependency of the predicted credit.

8. The supply chain finance demand analysis system based on big data analysis according to claim 2, wherein The generating the risk information of the target enterprise based on a confrontation network and the denoised data includes: The algorithm formula of the confrontation network is: Among them, is the noise reduction data, is the warning information corresponding to the noise reduction data, is the risk information given for the warning information ; is the generator in the adversarial network, is the discriminator in the adversarial network, indicates that the goal of the generator is to minimize its loss function, indicates that the goal of the discriminator is to maximize its loss function, is the expected value, indicates the joint distribution of the noise reduction data and the warning information corresponding to the noise reduction data ; is the logarithm of the decision value of the discriminator for the noise reduction data and the warning information corresponding to the noise reduction data ; is the random noise, is the random noise 's probability distribution, is the logarithm of the value calculated for the risk information being misclassified as false information by the discriminator, is the discriminator's evaluation of the risk information; Based on the noise-reduced data and the warning information, use the adversarial network to generate the risk information of the target enterprise.

9. The supply chain finance demand analysis system based on big data analysis according to claim 1, wherein The feature fusion of the time dependence relationship and the risk information to obtain the enterprise demand characteristics of the target enterprise includes: Based on the attention mechanism, perform feature fusion on the time dependence relationship and the risk information to obtain the enterprise demand characteristics of the target enterprise.

10. A reference information generation method based on artificial intelligence and smart home, characterized in that, The method includes: Collect the panoramic data of the supply chain business of the target enterprise; Store the panoramic data of the supply chain business in the big data storage system, and clean and aggregate the panoramic data of the supply chain business to obtain the supply chain integration data; Perform auto-encoding processing on the supply chain integration data to obtain noise-reduced data; Perform a forward layer-by-layer mapping on the noise-reduced data based on a preset credit scoring algorithm to obtain a predicted credit; Use the pre-constructed credit scoring time model to extract the time characteristics of the predicted credit, and generate the time dependence relationship of the predicted credit based on the time characteristics and the long short-term memory network; Based on the adversarial network and the noise-reduced data, generate the risk information of the target enterprise; Perform feature fusion on the time dependence relationship and the risk information to obtain the enterprise demand characteristics of the target enterprise; Based on the enterprise demand characteristics, determine whether it is necessary to promote the target enterprise.

Images