Contract multi-return prediction method, device and equipment and medium

Abstract

The invention relates to the technical field of artificial intelligence, and discloses a contract multi-return prediction method and device, equipment and a medium, and the method comprises the steps: obtaining a contract A party type, a contract amount of a to-be-predicted contract, an appointed return number, each appointed return amount and each appointed return time; acquiring historical refunding data of the historical contract; training a contract return classification prediction model according to the historical return data; and inputting the contract A party type, the contract amount, the agreed money return number, each agreed money return amount and each agreed money return time corresponding to the to-be-predicted contract into the contract money return classification prediction model to obtain the money return risk category and the money return default probability of the contract A party customer, thereby realizing money return risk management and default prediction of different customer types. Fine management risk prediction can be conveniently carried out on the return money, corresponding coping strategies can be conveniently adopted for different return money risks, and the occurrence probability of bad debts and dull debts is reduced.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method, device, equipment and medium for predicting multiple payment returns under a contract. Background Art

[0002] Collection problem definition Collection problem refers to the situation where, after the company provides products or services, the customer fails to pay for the goods on the agreed time, resulting in difficulties in the company's cash flow and may even lead to a financial crisis. Current situation of collection problem In the context of fierce market competition and complex and changing economic environment, collection problem has become a difficult problem faced by many companies. Due to the lack of effective collection mechanisms and measures, some companies have continuously increased accounts receivable and increased bad debt risks, which has seriously affected the normal operation and financial status of the company. In the existing technology, the management of collection by enterprises is relatively chaotic. It is impossible to manage customer collection in a refined manner and adjust the collection strategy in a targeted manner, which makes the collection work difficult to implement and inefficient. Summary of the invention

[0003] In view of this, the present invention provides a method for predicting multiple payment collections for a contract, which solves the technical problems in the prior art that enterprises have chaotic management of payment collections, are unable to manage customer payment collections in a refined manner, and cannot adjust payment collection strategies in a targeted manner, resulting in difficulties in implementing payment collection work and low efficiency.

[0004] According to a first aspect of the present invention, a method for predicting multiple payment returns of a contract is provided, comprising:

[0005] Obtain the type of Party A of the contract, the contract amount of the contract to be predicted, the number of agreed repayments, the agreed repayment amount for each repayment, and the agreed repayment time for each repayment. The type of Party A of the contract includes the institution attributes, the number of historical cooperation times, and the industry field. The institution attributes include state-owned enterprises, private enterprises, government units, private non-enterprises, and public institutions.

[0006] Obtain historical payment data for historical contracts, where payment data includes contract amount, number of payment collections agreed in the contract, agreed payment time for each payment, agreed payment amount for each payment, number of actual payment collections for each payment, actual payment time for each payment, and actual payment amount for each payment;

[0007] Based on historical payment collection data, train the contract payment collection classification prediction model;

[0008] The contract party A type, contract amount, agreed number of repayments, each agreed repayment amount and each agreed repayment time corresponding to the contract to be predicted are input into the contract repayment classification prediction model to obtain the repayment risk category and repayment default probability of the contract party A customers, where the repayment risk categories include high credit rating customers, medium credit rating customers, low credit rating customers and bad debt customers.

[0009] According to a second aspect of the present invention, a method and device for predicting multiple payment collections of a contract is provided, comprising:

[0010] The first acquisition module is used to obtain the type of Party A of the contract, the contract amount of the contract to be predicted, the number of agreed repayments, the agreed repayment amount of each repayment, and the agreed repayment time of each repayment. The type of Party A of the contract includes the institution attributes, the number of historical cooperation times, and the industry field. The institution attributes include state-owned enterprises, private enterprises, government units, private non-enterprises, and public institutions.

[0011] The second acquisition module is used to obtain historical payment data of historical contracts, where the payment data includes the contract amount, the number of payment collections agreed in the contract, the agreed payment time for each payment, the agreed payment amount for each payment, the number of actual payment collections for each payment, the actual payment time for each payment, and the actual payment amount for each payment;

[0012] Model training module, used to train contract payment classification prediction model based on historical payment data;

[0013] The model prediction module is used to input the contract party A type, contract amount, agreed number of repayments, each agreed repayment amount and each agreed repayment time corresponding to the contract to be predicted into the contract repayment classification prediction model to obtain the repayment risk category and repayment default probability of the contract party A customer, where the repayment risk categories include high credit rating customers, medium credit rating customers, low credit rating customers and bad debt customers.

[0014] According to a third aspect of the present invention, there is provided a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned method for predicting multiple payment collections under a contract when executing the computer program.

[0015] According to a fourth aspect of the present invention, there is provided a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-mentioned method for predicting multiple payment collections under a contract.

[0016] Through the above technical scheme, the present invention provides a method, device, equipment and medium for predicting multiple collections of contracts, by obtaining the type of Party A of the contract, the contract amount of the contract to be predicted, the number of agreed collections, the amount of each agreed collection and the time of each agreed collection; obtaining the historical collection data of historical contracts; training a contract collection classification prediction model based on the historical collection data; inputting the type of Party A of the contract, the contract amount, the number of agreed collections, the amount of each agreed collection and the time of each agreed collection corresponding to the contract to be predicted into the contract collection classification prediction model to obtain the collection risk category and the collection default probability of the Party A customer of the contract, thereby realizing collection risk management and default prediction for different customer types, facilitating refined management risk prediction of collections, facilitating the adoption of corresponding response strategies for different collection risks, and reducing the probability of bad debts and non-performing debts.

[0017] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention and implement it according to the contents of the specification, and to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0019] Figure 1 A schematic diagram of an application scenario of a method for predicting multiple payment returns for a contract provided in an embodiment of the present invention is shown;

[0020] Figure 2 A flowchart of a method for predicting multiple payment returns for a contract provided in the first embodiment of the present invention is shown;

[0021] Figure 3 The network structure diagram of the double-layer LSTM model provided in the second embodiment of the present invention is shown;

[0022] Figure 4 A flowchart of obtaining the optimal hyperparameters of the LSTM model based on the firefly optimization algorithm provided in the third embodiment of the present invention is shown;

[0023] Figure 5 A schematic diagram of the structure of a method device for predicting multiple payment returns for a contract provided in an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0024] The specific implementation of the present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments. It should be noted that the embodiments and features in the embodiments of the present invention can be combined with each other without conflict.

[0025] The electric vehicle throttle sound effect linkage generation method provided by the embodiment of the present invention can be applied to Figure 1 In the system, the database server connects the contract management platform and the financial management platform to obtain the type of Party A of the contract, the contract amount of the contract to be predicted, the number of agreed repayments, the amount of each agreed repayment and the agreed repayment time of each repayment, among which the type of Party A of the contract includes the institution attributes, the number of historical cooperation times and the industry field, and the institution attributes include state-owned enterprises, private enterprises, government units, private non-enterprises and public institutions; obtain the historical repayment data of historical contracts, among which the repayment data includes the contract amount, the number of agreed repayments of the contract, the agreed repayment time of each repayment, the amount of each agreed repayment, The actual number of collections, the actual collection time, and the actual collection amount of each corresponding to each transaction; the data analysis server trains the contract collection classification prediction model based on the historical collection data obtained, and inputs the contract party A type, contract amount, agreed collection number, agreed collection amount, and agreed collection time of each contract corresponding to the predicted contract into the contract collection classification prediction model to obtain the collection risk category and collection default probability of the contract party A customer, among which the collection risk category includes high credit rating customers, medium credit rating customers, low credit rating customers, and bad debt customers. The system realizes collection risk management and default prediction for different customer types, facilitates the refined management risk prediction of collection, and facilitates the adoption of corresponding response strategies for different collection risks to reduce the probability of bad debts.

[0026] This system analyzes the reasons for the differences in contract payment collection by segmenting them. Common reasons include macro factors, Party A's payment ability, sales staff performance, etc. The classification prediction model is used to evaluate the impact of each reason on the sales payment difference and determine the main reason.

[0027] During the application of this system, for each actual payment, the system will automatically display all agreed payment times and amounts contained in the corresponding contract through the front-end interface. The actual amount of each agreed payment in the corresponding contract before the actual payment is received is calculated and displayed. The corresponding calculation method and processing method are as follows:

[0028] (1) Add up all the actual payments received under the contract, and then subtract the current payment from the total amount of actual payments received before the current payment. Split the total amount of actual payments according to the agreed payment amount and the time sequence of each agreed payment using a recursive loop method, i.e., the actual payment amount is split to the earliest agreed payment by default, and the agreed payments that are not split are displayed as 0 yuan;

[0029] (2) Subtract the corresponding amount of repayment before this repayment from the agreed repayment amount to obtain the remaining repayment amount of each agreed repayment before this repayment; according to the remaining repayment amount of each agreed repayment, the current repayment amount is split according to the time sequence of each agreed repayment, using a recursive loop method to obtain the repayment split details of each agreed repayment amount corresponding to this repayment amount;

[0030] (3) The enterprise calculates the rewards for the agreed payment amounts of each contract based on the payment amount of this payment after comprehensively considering factors such as the number of days of early payment, and then adds them up to get the final payment progress reward for this payment amount. The system allows users to modify and review the agreed payment time, agreed payment amount, and the split details corresponding to each actual payment amount of each contract to deal with special circumstances.

[0031] (4) To help enterprises collect payments as quickly as possible, penalties are calculated once for each fixed time period, rather than a one-time penalty on the day of overdue payment (for example, the last day of each month is the settlement date); the agreed payment time for all contracts and their corresponding amount details are listed; each agreed payment time is subtracted from the settlement date, and then filtered and added to obtain the agreed total payment amount for each contract before the threshold number of days; the agreed total payment amount for each contract before the threshold number of days is subtracted from the actual total payment amount on the settlement date. If the result is greater than 0, the value is the amount of arrears that needs to be punished; for sales payment progress rewards, the problem of difficulty in determining which (several) agreed payments are returned and how much each payment accounts for, caused by the many-to-many relationship between the agreed payment and the actual payment, is solved. In the embodiment of the present invention, the time interval between each agreed payment time and the actual payment is different, and the corresponding reward coefficients may also be different, so they must be calculated separately for accurate calculation; for sales payment penalties, it is triggered once in each specified time period, rather than once for each overdue payment, which helps to improve the progress of payment completion. After exceeding the threshold, no matter how many days the payment is overdue, the relevant penalty coefficient is the same, so there is no need to calculate separately according to the agreed payment ratios in the contract. By directly calculating the total amount of the agreed payment before the threshold days and the total amount of money collected on the day, the outstanding balance that needs to be punished for all contracts is obtained in batches at one time, and then directly multiplied by the specified ratio to get the corresponding penalty amount. The processing method provided by the embodiment of the present invention solves the problem that it is difficult to determine the specific number of days in arrears and the overdue amount due to the existence of multiple agreed payment times and amounts.

[0032] The present invention is described in detail below through specific embodiments.

[0033] Embodiment 1

[0034] like Figure 2 As shown, a method for predicting multiple payment returns for a contract provided in an embodiment of the present invention includes:

[0035] Step 201: Obtain the type of Party A of the contract, the contract amount of the contract to be predicted, the number of agreed repayments, the agreed repayment amount for each repayment, and the agreed repayment time for each repayment;

[0036] Among them, the type of Party A of the contract includes institutional attributes, historical cooperation times, and industry fields. The institutional attributes include state-owned enterprises, private enterprises, government units, private non-enterprises, and public institutions;

[0037] Step 202: Obtain historical payment data of historical contracts;

[0038] The payment collection data includes the contract amount, the number of payment collections agreed in the contract, the agreed payment collection time for each payment, the agreed payment collection amount for each payment, the actual number of payment collections for each payment, the actual payment collection time for each payment, and the actual payment collection amount for each payment;

[0039] Step 203: training a contract payment classification prediction model based on historical payment collection data;

[0040] Wherein, step 203 specifically includes:

[0041] Step 203-1: cluster the historical payment collection data based on the local outlier factor LOF to obtain payment collection risk categories of different Party A customers;

[0042] Step 203-2: Sort the historical payment collection data by time, set the sliding window length T, and perform sliding window sampling on the sample sets of payment collection risk categories of different Party A customers to obtain a sample set based on the sliding window;

[0043] Among them, the sliding window length T is less than the number of samples in the smallest category;

[0044] Step 203-3, constructing a long short-term memory LSTM model based on the attention mechanism;

[0045] Step 203-4: input the sample set based on the sliding window into the LSTM model based on the attention mechanism, perform back propagation and gradient calculation on the model parameters based on the binary cross entropy loss function, and use the Adam optimizer to iteratively update the model parameters of the LSTM model based on the attention mechanism according to the gradient calculation results;

[0046] Step 203-5: When the change in the loss function is less than the preset threshold or the number of iterations is equal to the maximum number of iterations, the LSTM model based on the attention mechanism is output as the contract multiple payment prediction model.

[0047] Step 204: input the contract party A type, contract amount, agreed number of repayments, each agreed repayment amount and each agreed repayment time corresponding to the contract to be predicted into the contract repayment classification prediction model to obtain the repayment risk category and repayment default probability of the contract party A customer.

[0048] Among them, the collection risk categories include high credit rating customers, medium credit rating customers, low credit rating customers and bad debt and non-performing debt customers.

[0049] An embodiment of the present invention provides a method for predicting multiple collections of a contract, by obtaining the type of Party A of the contract, the contract amount of the contract to be predicted, the number of agreed collections, the amount of each agreed collection, and the time of each agreed collection of the contract; obtaining historical collection data of historical contracts; training a contract collection classification prediction model based on the historical collection data; inputting the type of Party A of the contract, the contract amount, the number of agreed collections, the amount of each agreed collection, and the time of each agreed collection corresponding to the contract to be predicted into the contract collection classification prediction model to obtain the collection risk category and the probability of collection default of the Party A customer of the contract, thereby realizing collection risk management and default prediction for different customer types, facilitating refined management risk prediction of collections, facilitating the adoption of corresponding response strategies for different collection risks, and reducing the probability of bad debts and non-performing debts.

[0050] After step 204, different repayment strategies are adopted for different risk repayment customers and default risk probabilities, including

[0051] Set corresponding payment adjustment strategies based on the payment risk category and payment default probability of Party A’s customers in the contract;

[0052] Set different collection strategies for customers of different risk categories. For example:

[0053] Low-risk customers: Normal payment collection cycle, no special adjustments required.

[0054] Medium-risk customers: To shorten the payment collection cycle, you may need to remind them in advance or take some preventive measures.

[0055] High-risk customers: Take immediate measures, such as early collection, adjustment of agreed repayment amount, adjustment of agreed repayment number, etc., to reduce the pressure of single repayment for high-risk customers and increase the success rate of repayment.

[0056] Set different performance reward incentive strategies based on different probability of default in repayments.

[0057] In order to motivate the efficiency of payment collection, different performance reward incentive strategies are set for sales staff, such as:

[0058] Low probability of default: Set lower reward standards to encourage the team to maintain efficient collection of payments.

[0059] Medium probability of default: Set a medium reward standard to motivate the team to increase collection efforts.

[0060] High probability of default: Setting high reward standards encourages teams to take more proactive measures to reduce the risk of default.

[0061] Through the method of the first embodiment of the present invention, it is possible to cluster and classify the types of Party A of different contracts and predict the probability of default in contract payment collection, and adopt corresponding payment collection strategies in a targeted manner, thereby improving the flexibility and targeting of contract payment collection and improving the efficiency of payment collection work.

[0062] Embodiment 2

[0063] LSTM can only process fixed-length sequence information and lacks certain adaptability to different sliding window lengths T. LSTM is not suitable for the temporal characteristics of different features in each sample. Based on the above two defects, the second embodiment of the present invention introduces the Attention mechanism, which allocates attention to different parameters for each feature of the sample from the level of feature temporal characteristics, so that the output of the model can adapt to the temporal distribution at the feature level rather than just the sample level. Figure 3 As shown, the long short-term memory LSTM model based on the attention mechanism includes a first LSTM layer, an attention layer and a second LSTM layer, wherein the first LSTM layer is used to receive a sample set input based on a sliding window, the attention layer is used to calculate the attention weight output by the first LSTM layer, and the second LSTM layer is used to receive the attention weighted value output by the first LSTM layer.

[0064] The traditional LSTM model only focuses on the temporal nature of the sample level, and does not actually allocate too many parameters for the temporal nature of the feature level. Adding the attention mechanism as the middle layer of the feature collector can attach the state of the memory unit to the feature state, so that each feature can be fitted to different degrees of temporal nature, making the model have a certain degree of interpretability similar to that of the tree model, and having certain advantages in parameter adjustment and structural optimization.

[0065] The newly added memory unit not only serves as a representation of feature parameters, but also as a carrier for cell state transmission. While inheriting the LSTM's historical dependence on states at different stages, it ensures that the model's focus can always be maintained on a single or a group of samples that have a greater impact on the final output state.

[0066] The LSTM model is a deep learning model used to process time series data. It can effectively capture temporal dependencies. However, using the LSTM model alone may not achieve the desired effect. In this case, you can consider adding a Dense (fully connected) layer after the LSTM model to further improve the model performance. Figure 3 As shown, in the long short-term memory LSTM model based on the attention mechanism in the second embodiment of the present invention, the second LSTM layer is sequentially connected to a fully connected Dense layer and a discarded Dropout layer.

[0067] The main purpose of adding a Dense layer is to increase nonlinear transformations. Since LSTM itself already has nonlinear characteristics, the Dense layer is usually used to further enhance the nonlinear capabilities of the LSTM model. Specifically, the Dense layer usually uses activation functions such as ReLU to introduce nonlinearity, so that the model can better adapt to complex time series data. Adding a Dense layer can also help the LSTM model better solve some problems. For example, when the input sequence length is long, the LSTM model may have overfitting problems. At this time, Dropout or L2 regularization can be added to prevent overfitting. However, these methods all need to be implemented by increasing the network structure, while the Dense layer can directly perform corresponding regularization to reduce the occurrence of overfitting.

[0068] Adding a Dense layer can help the LSTM model better handle multi-classification problems. In the LSTM model, Softmax is usually used as the activation function of the last layer for classification, but this method is only applicable to the case where the number of output categories is small. If you want to process a large number of different categories, you need to use a more flexible classifier. In the second embodiment of the present invention, the output of the LSTM model is sent to a Dense layer, and then classified using activation functions such as Softmax, which can effectively improve the performance of the classifier.

[0069] Embodiment 3

[0070] The hyperparameter setting of LSTM has a significant impact on the training and performance of the model. In practical applications, it is usually necessary to find the parameter setting that best suits the current task through experiments, and more experience-based adjustments are required. In the third embodiment of the present invention, the hyperparameter optimization is achieved by introducing the firefly optimization algorithm. Before step 203-4, the step of obtaining the optimal hyperparameters of the LSTM model is included, such as Figure 4 As shown, specifically including:

[0071] Step 401, create a firefly population F;

[0072] Each firefly f i A set of hyperparameters θi express;

[0073] Step 402: Based on historical payment data D train , define the fitness function J(θ i ,D train );

[0074] Among them, the loss function is used as the fitness function: In machine learning model training, the fitness function often directly uses the loss function, such as mean square error (MSE), cross entropy loss, etc. In this case, the lower the value of the fitness function, the better the performance of the model.

[0075] Step 403, calculate the brightness of each firefly and sort them to obtain the position of the firefly with the highest brightness;

[0076] Step 404: Determine whether the number of iterations reaches the preset number of iterations or the fitness function J(θ i ,D train ) When the transformation amount is less than the preset threshold, the position and brightness of the firefly with the largest brightness are output, otherwise the execution continues;

[0077] Among them, the position of the firefly with the largest output brightness and its brightness correspond to the optimal hyperparameters for LSTM model training.

[0078] Step 405: Update the formula based on the position Update the firefly position, perform random perturbations on the fireflies at the best position, increase the number of iterations by 1, and jump to step 403 to continue execution.

[0079] Among them, x i and x j is the spatial position of fireflies i and j, α is the step size factor, rand is a random number in [0,1], β0 is the maximum attraction, γ is the light absorption coefficient, r i,j is the spatial distance between fireflies i and j.

[0080] In the third embodiment of the present invention, a balance is achieved between global exploration and local development to maximize the optimization benefit. In the traditional firefly optimization algorithm, the constant step size hinders effective search and needs to be dynamically adjusted to achieve the best exploration-convergence balance. In the early stage, a larger initial step size is required to achieve balanced identification and development, and as the number of iterations increases, it is gradually reduced to maintain the balance. The selection of a larger or smaller step size depends on the definition space of the optimization objective. In order to maintain a balance between identification and development capabilities, the initial step size (α) should be relatively large and gradually decrease as the number of iterations increases. The selection of a larger or smaller search step size depends on the definition space of the optimization objective; high-dimensional space requires a larger search step size, while low-dimensional space benefits from a small step size, and the ability of the optimization algorithm to solve diverse optimization challenges, as described in formula (2). Here, the existing maximum number of iterations = number of iterations. Therefore, the step size factor α is a variable step size factor α(t) = α(0) / (1+exp(δ×(t-maxiteration) / 3)), where is the number of iterations, the maximum number of iterations, and δ is the step size transformation coefficient.

[0081] In the third embodiment of the present invention, the variable step size firefly optimization algorithm is integrated with the two-layer LSTM to optimize the architecture and hyperparameters for this purpose. The two-layer LSTM captures time dependencies, while the variable step size firefly optimization algorithm explores the hyperparameter space for key configurations. The potential of this hybrid technology can be further realized through careful parameter tuning and rigorous model validation.

[0082] The variable-step firefly optimization algorithm systematically evaluates the ideal number of hidden neurons in each hidden layer. The initial weights of the network are assigned random values, and the Adam optimizer based on the gradient descent principle is used, such as: maximum epoch = 100, batch size = 512, and initial learning rate = 0.001, to iteratively refine these network weights. Subsequently, the model was thoroughly tested to measure the performance of the variable-step firefly optimization algorithm after adjustment. This approach not only highlights the key role of weight optimization in fine-tuning the prediction ability of the two-layer LSTM, but also highlights the importance of determining the optimal number of hidden neurons to improve the overall model efficiency. The variable-step firefly optimization algorithm is integrated with the two-layer LSTM to achieve the best prediction accuracy in the context of contract receivables bad debt data classification. Let w denote the weight of the hidden layer of the two-layer LSTM, and the model is denoted as two-layer LSTM(w). The parameter θ obtained from the variable-step firefly optimization algorithm is used to fine-tune the weight of the two-layer LSTM model. The variable-step firefly optimization algorithm model is denoted as (θ, w*), where w* is the adjusted weight. The optimization process involves iteratively updating the parameter θ using the VSSFF algorithm.

[0083] Further, as Figures 2 to 4 The specific implementation of the method, the embodiment of the present invention provides a contract multiple payment forecasting device, such as Figure 5 As shown, the device comprises:

[0084] The first acquisition module 510 is used to acquire the type of Party A of the contract, the contract amount of the contract to be predicted, the number of agreed repayments, the agreed repayment amount of each repayment, and the agreed repayment time of each repayment, wherein the type of Party A of the contract includes the institution attribute, the number of historical cooperation times, and the industry field, and the institution attribute includes state-owned enterprises, private enterprises, government units, private non-enterprises, and public institutions;

[0085] The second acquisition module 520 is used to obtain historical payment collection data of historical contracts, wherein the payment collection data includes the contract amount, the number of payment collections agreed in the contract, the agreed payment collection time for each payment, the agreed payment collection amount for each payment, the number of actual payment collections corresponding to each payment, the actual payment collection time for each payment, and the actual payment collection amount for each payment;

[0086] Model training module 530, used to train a contract payment classification prediction model based on historical payment data;

[0087] The model prediction module 540 is used to input the contract party A type, contract amount, agreed number of repayments, each agreed repayment amount and each agreed repayment time corresponding to the contract to be predicted into the contract repayment classification prediction model to obtain the repayment risk category and repayment default probability of the contract party A customer, wherein the repayment risk category includes high credit rating customers, medium credit rating customers, low credit rating customers and bad debt customers.

[0088] In an embodiment of the present invention, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, steps of a method for predicting multiple payment collections under a contract are implemented, including:

[0089] Obtain the type of Party A of the contract, the contract amount of the contract to be predicted, the number of agreed repayments, the agreed repayment amount for each repayment, and the agreed repayment time for each repayment. The type of Party A of the contract includes the institution attributes, the number of historical cooperation times, and the industry field. The institution attributes include state-owned enterprises, private enterprises, government units, private non-enterprises, and public institutions.

[0090] Obtain historical payment data for historical contracts, where payment data includes contract amount, number of payment collections agreed in the contract, agreed payment time for each payment, agreed payment amount for each payment, number of actual payment collections for each payment, actual payment time for each payment, and actual payment amount for each payment;

[0091] Based on historical payment collection data, train the contract payment collection classification prediction model;

[0092] The contract party A type, contract amount, agreed number of repayments, each agreed repayment amount and each agreed repayment time corresponding to the contract to be predicted are input into the contract repayment classification prediction model to obtain the repayment risk category and repayment default probability of the contract party A customers, where the repayment risk categories include high credit rating customers, medium credit rating customers, low credit rating customers and bad debt customers.

[0093] In an embodiment of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of implementing a method for predicting multiple payment collections of a contract include:

[0094] Obtain the type of Party A of the contract, the contract amount of the contract to be predicted, the number of agreed repayments, the agreed repayment amount for each repayment, and the agreed repayment time for each repayment. The type of Party A of the contract includes the institution attributes, the number of historical cooperation times, and the industry field. The institution attributes include state-owned enterprises, private enterprises, government units, private non-enterprises, and public institutions.

[0095] Obtain historical payment data for historical contracts, where payment data includes contract amount, number of payment collections agreed in the contract, agreed payment time for each payment, agreed payment amount for each payment, number of actual payment collections for each payment, actual payment time for each payment, and actual payment amount for each payment;

[0096] Based on historical payment collection data, train the contract payment collection classification prediction model;

[0097] The contract party A type, contract amount, agreed number of repayments, each agreed repayment amount and each agreed repayment time corresponding to the contract to be predicted are input into the contract repayment classification prediction model to obtain the repayment risk category and repayment default probability of the contract party A customers, where the repayment risk categories include high credit rating customers, medium credit rating customers, low credit rating customers and bad debt customers.

[0098] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0099] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0100] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for predicting multiple payment returns from a contract, characterized in that: include: Obtain the type of Party A of the contract, the contract amount of the contract to be predicted, the number of agreed repayments, the agreed repayment amount for each repayment, and the agreed repayment time for each repayment. The type of Party A of the contract includes the institution attributes, the number of historical cooperation times, and the industry field. The institution attributes include state-owned enterprises, private enterprises, government units, private non-enterprises, and public institutions. Obtain historical payment data of historical contracts, wherein the payment data includes the contract amount, the number of payment collections agreed in the contract, the agreed payment time for each payment, the agreed payment amount for each payment, the number of actual payment collections for each payment, the actual payment time for each payment, and the actual payment amount for each payment; Based on the historical payment collection data, a contract payment collection classification prediction model is trained; The contract party A type, contract amount, agreed number of repayments, each agreed repayment amount and each agreed repayment time corresponding to the contract to be predicted are input into the contract repayment classification prediction model to obtain the repayment risk category and repayment default probability of the contract party A customer, wherein the repayment risk category includes high credit rating customers, medium credit rating customers, low credit rating customers and bad debt customers.

2. The method for predicting multiple payment returns under a contract according to claim 1, characterized in that: The step of training a contract multiple payment collection prediction model based on the historical payment collection data includes: Clustering the historical payment collection data based on the local outlier factor LOF to obtain payment collection risk categories for different Party A customers; The historical payment collection data is sorted by time, a sliding window length T is set, and sliding window sampling is performed on sample sets of payment collection risk categories of different Party A customers to obtain a sliding window-based sample set, wherein the sliding window length T is less than the number of samples of the smallest category; Construct a long short-term memory LSTM model based on the attention mechanism; Inputting the sample set based on the sliding window into the LSTM model based on the attention mechanism, performing back propagation and gradient calculation on the model parameters based on the binary cross entropy loss function, and iteratively updating the model parameters of the LSTM model based on the attention mechanism according to the gradient calculation results using the Adam optimizer; When the change in the loss function is less than the preset threshold or the number of iterations is equal to the maximum number of iterations, the LSTM model based on the attention mechanism is output as the contract multiple payment prediction model.

3. The method for predicting multiple payment collections under a contract according to claim 2, characterized in that: The long short-term memory LSTM model based on the attention mechanism includes a first LSTM layer, an attention layer and a second LSTM layer, wherein the first LSTM layer is used to receive a sample set input based on a sliding window, the attention layer is used to calculate the attention weight output by the first LSTM layer, and the second LSTM layer is used to receive the attention weighted value output by the first LSTM layer.

4. The method for predicting multiple payment collections from contracts according to any one of claims 2 or 3, characterized in that: Before the step of inputting the sample set based on the sliding window into the LSTM model based on the attention mechanism, performing back propagation and gradient calculation on the model parameters based on the binary cross entropy loss function, and iteratively updating the model parameters of the LSTM model based on the attention mechanism according to the gradient calculation results using the Adam optimizer, the step of obtaining the optimal hyperparameters of the LSTM model is included, which specifically includes: Create a firefly population F, where each firefly f i A set of hyperparameters θ i express; According to the historical payment data D train , define the fitness function J(θ i ,D train ); Calculate the brightness of each firefly and sort them to get the position of the firefly with the highest brightness; Determine whether the number of iterations reaches the preset number of iterations or the fitness function J(θ i ,D train ) When the transformation amount is less than the preset threshold, the position and brightness of the firefly with the largest brightness are output, otherwise the execution continues; Update the formula based on the position Update the firefly position, randomly perturb the firefly in the best position, increase the number of iterations by 1, and jump to the step of calculating the brightness of each firefly and sorting to get the position of the firefly with the highest brightness, where x i and x j is the spatial position of fireflies i and j, α is the step size factor, rand is a random number in [0,1], β0 is the maximum attraction, γ is the light absorption coefficient, r i,j is the spatial distance between fireflies i and j; Obtain the optimal hyperparameters based on the position and brightness of the firefly with the largest output brightness; Train the two-layer LSTM model according to the optimal hyperparameters to obtain the final weight value W opt .

5. The method for predicting multiple payment returns under a contract according to claim 4, characterized in that: The step size factor α is a variable step size factor α(t)=α(0) / (1+exp(δ×(t-maxiteration) / 3)), where is the number of iterations, is the maximum number of iterations, and δ is the step size transformation coefficient.

6. The method for predicting multiple payment collections under a contract according to claim 3, characterized in that: The second LSTM layer in the long short-term memory LSTM model based on the attention mechanism is sequentially connected to a fully connected Dense layer and a discarded Dropout layer.

7. The method for predicting multiple payment collections under a contract according to claim 1, characterized in that: After the step of inputting the contract party A type, contract amount, agreed number of repayments, each agreed repayment amount and each agreed repayment time corresponding to the contract to be predicted into the contract repayment classification prediction model to obtain the repayment risk category and repayment default probability of the contract party A, the following steps are included: Set corresponding payment adjustment strategies based on the payment risk category and payment default probability of Party A’s customers in the contract; Set different performance reward incentive strategies based on different probability of default in repayments.

8. A method and device for predicting multiple payment returns under a contract, characterized in that: include: The first acquisition module is used to obtain the type of Party A of the contract, the contract amount of the contract to be predicted, the number of agreed repayments, the agreed repayment amount of each repayment, and the agreed repayment time of each repayment, wherein the type of Party A of the contract includes the institution attributes, the number of historical cooperation times, and the industry field, and the institution attributes include state-owned enterprises, private enterprises, government units, private non-enterprises, and public institutions; The second acquisition module is used to obtain historical payment collection data of historical contracts, wherein the payment collection data includes the contract amount, the number of payment collections agreed in the contract, the agreed payment collection time for each payment, the agreed payment collection amount for each payment, the number of actual payment collections corresponding to each payment, the actual payment collection time for each payment, and the actual payment collection amount for each payment; A model training module, used to train a contract payment collection classification prediction model based on the historical payment collection data; The model prediction module is used to input the contract party A type, contract amount, agreed number of repayments, each agreed repayment amount and each agreed repayment time corresponding to the contract to be predicted into the contract repayment classification prediction model to obtain the repayment risk category and repayment default probability of the contract party A customer, wherein the repayment risk category includes high credit rating customers, medium credit rating customers, low credit rating customers and bad debt customers.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method for predicting multiple payment repayments for a contract as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method for predicting multiple payment repayments for a contract as described in any one of claims 1 to 7 are implemented.

Images