Double-branch order information prediction method, computer device and storage medium

Abstract

The invention provides a double-branch order information prediction method, a computer device and a storage medium, and the method comprises the steps: creating an order, and inputting the front information of the order; performing prediction according to the preposed information by using a preset time prediction model to obtain initial time consumption parameter information related to an execution link of the order; generating a preset number of order execution alternative schemes according to the initial time consumption parameter information; and performing period violation risk prediction on all the order execution alternative schemes by using a preset period violation risk assessment model, and outputting period violation risk information corresponding to each order execution alternative scheme. By applying the double-branch order information prediction method, the accuracy of order prediction can be improved, order risk assessment is optimized, and the method adapts to dynamically changing market environments and production scheduling requirements.

Description

Technical Field

[0001] The present invention relates to the technical field of order management, and in particular, to a dual-branch order information prediction method, a computer device applying the dual-branch order information prediction method, and a computer-readable storage medium applying the dual-branch order information prediction method. Background Art

[0002] Order management and forecasting play an important role in enterprise operations and optimization. At present, the market mainly uses traditional order management tools or simple time forecasting methods, but these technologies have the following shortcomings in practical applications: Most of the existing order forecasting systems rely on linear regression or simple time series models. Although these models can predict order trends to a certain extent, they are highly dependent on data and often cannot effectively capture the complex interactive relationships between order features. For example, traditional methods usually improve the accuracy of predictions by introducing more dimensions or levels of data, but such methods often only rely on the increase in data dimensions to improve model performance, ignoring the interactions and synergies between different features. When faced with high-dimensional data, these methods usually fall into overfitting or ignore the complex dependencies between data, which affects the final prediction results.

[0003] Moreover, current order risk management methods mainly focus on simple estimation of default probability, usually only providing a static default prediction result, but lacking in-depth analysis of the causes of default and corresponding optimization strategies. Especially in a dynamic environment, traditional systems are often unable to combine real-time order data, predict multiple possible optimization solutions, and provide scientific risk assessment suggestions, which makes it difficult for existing prediction systems to meet the needs of enterprises for dynamic adjustment of order processes. For example, most existing risk assessment methods are based on regression models or neural network models to generate a probability value to estimate default risk. However, these models are often limited to static data analysis and do not deeply consider the complex dynamic factors in the order and their profound impact on risk prediction. For example, the time required to complete an order is not only related to the basic attributes of the order itself, but also closely related to the continuous changes in multiple core production links. The time consumption of each core link is dynamically changing, especially in a complex production environment, where delays or changes in different links may have a decisive impact on the overall completion time of the order. Existing models usually only predict the probability of default based on historical data or fixed assumptions, ignoring the profound impact of these dynamic changes on the overall order process. Therefore, even if neural networks or more complex time series models are used, these methods still mainly focus on estimating the probability of default, but lack sufficient coping capabilities in processing dynamically changing real-time data.

[0004] Therefore, a more optimized order information prediction method needs to be considered. Summary of the invention

[0005] The first object of the present invention is to provide a dual-branch order information prediction method that can improve the accuracy of order prediction, optimize order risk assessment, and adapt to dynamically changing market environments and production scheduling requirements.

[0006] The second object of the present invention is to provide a computer device that can improve the accuracy of order forecasting, optimize order risk assessment, and adapt to dynamically changing market environments and production scheduling requirements.

[0007] The third object of the present invention is to provide a computer-readable storage medium that can improve the accuracy of order forecasting, optimize order risk assessment, and adapt to dynamically changing market environments and production scheduling requirements.

[0008] In order to achieve the above-mentioned first purpose, the dual-branch order information prediction method provided by the present invention includes: creating an order and inputting the prerequisite information of the order; using a preset time prediction model to predict based on the prerequisite information to obtain initial time-consuming parameter information related to the execution link of the order; generating a preset number of order execution alternative plans based on the initial time-consuming parameter information; using a preset delinquency risk assessment model to predict the delinquency risk of all order execution alternative plans, and outputting the delinquency risk information corresponding to each order execution alternative plan; outputting the optimal order execution plan among a preset number of order execution alternative plans based on all delinquency risk information.

[0009] It can be seen from the above scheme that the dual-branch order information prediction method of the present invention introduces a dual-branch structure of a preset time prediction model and a preset default risk assessment model, which are responsible for order time prediction and default risk assessment respectively. Each branch independently carries out the prediction task, which can maximize the processing efficiency and accuracy of each task. Specifically, the time prediction branch is responsible for predicting the initial time-consuming parameter information related to the execution link of the order, and dynamically integrates the initial time-consuming parameter information to obtain multiple order execution alternatives. The risk assessment branch focuses on the potential risks in the order execution process, based on the time prediction results and the existing order information data, combined with the execution difficulty and uncertainty of different links, to evaluate the default risk, and can select the best solution from multiple alternatives, so that it has strong adaptability when processing diversified orders, and adapts to the dynamically changing market environment and production scheduling needs.

[0010] In a further scheme, the step of generating a preset number of order execution alternative plans based on the initial time-consuming parameter information includes: adjusting the initial time-consuming parameter information according to preset rules to obtain a preset number of different time-consuming parameter information; combining each time-consuming parameter information with the previous information to obtain the corresponding order execution alternative plan.

[0011] It can be seen that by adjusting the initial time-consuming parameter information through preset rules, various uncertain factors in the actual order execution process can be taken into account. For example, the production process may encounter delays in the supply of raw materials, equipment failures, etc., and the transportation process may be affected by factors such as traffic conditions and weather. Generating multiple different time-consuming parameter information can cover these possible changes, allowing companies to prepare response plans in advance. This method can flexibly generate alternative order execution plans suitable for different scenarios.

[0012] In a further solution, the step of adjusting the initial time-consuming parameter information according to a preset rule includes: adjusting the time consumption of each time-consuming parameter in the initial time-consuming parameter information to increase or decrease the time consumption.

[0013] It can be seen that by increasing or decreasing each time-consuming parameter, a variety of different time-consuming combinations can be generated relatively quickly. This adjustment method can cover the most basic time changes in the order execution process. In actual business, the time consumption of each link of an order may become longer or shorter due to various factors. For example, the production link may be shortened due to equipment upgrades, or it may be extended due to workers taking leave. By adjusting each time-consuming parameter, these actual possible situations can be simulated to a certain extent.

[0014] In a further solution, the default risk information includes the default probability and the default duration.

[0015] It can be seen that by complementing each other, the probability of default and the duration of default provide a more comprehensive assessment of the risk of order default, which helps companies formulate more refined response strategies.

[0016] In a further solution, the preset time prediction model is a time prediction model that adopts a feature-driven attention mechanism; and / or the preset delinquency risk assessment model is a delinquency risk assessment model that adopts a feature-driven attention mechanism.

[0017] It can be seen that the time prediction model and default risk assessment model based on the feature-driven attention mechanism can dynamically capture the complex interactive relationship between order information, for example, how unstructured information such as order difficulty, category, and additional notes affect the time requirements of each production link. Through the feature-driven attention mechanism, the model can automatically discover potential key factors and their interactions in multi-dimensional features, thereby improving the accuracy of prediction. This deep feature learning not only enhances the model's predictive ability, but also effectively improves the reliability and adaptability of the system when processing complex orders.

[0018] In a further solution, before the step of using a preset time prediction model to predict according to the pre-information, it also includes: using a preset encoder to perform data processing on the pre-information to obtain data that conforms to the preset time prediction model.

[0019] It can be seen that the pre-information may exist in various raw formats, such as a mixture of text and numerical values, or the dimensions do not match the input format expected by the preset time prediction model. The preset encoder can convert these raw data into a format that the model can effectively process. After being processed by the preset encoder, the data is more in line with the requirements of the preset time prediction model, and the model can better learn the patterns and rules in the data, thereby improving the accuracy of predicting the order execution time. The existence of the preset encoder enables the system to process various formats and types of pre-information, enhancing the versatility of the system. At the same time, because the data has been standardized, the model training process is more stable and less susceptible to interference from data outliers or different scales, which improves the stability and reliability of the entire order prediction system.

[0020] In a further embodiment, the preset encoder includes a BERT model for processing text data and a multi-layer perceptron for processing numerical data.

[0021] It can be seen that in the order prediction scenario, both text data and numerical data contain important prediction clues. By processing text data with the BERT model and numerical data with the multi-layer perceptron, the value of both types of data can be fully explored and the information in the previous information can be fully utilized. The combination of the BERT model and the multi-layer perceptron can provide richer and more representative features for subsequent time prediction models. The BERT model and the multi-layer perceptron effectively process the data types they are good at respectively, and the generated features complement each other, which helps the time prediction model better learn the complex relationship between data and order execution time, thereby improving the accuracy and reliability of the prediction.

[0022] In a further solution, when outputting the best order execution solution among a preset number of order execution alternative solutions based on all default risk information, a solution suggestion is also output.

[0023] It can be seen that by outputting the optimal order execution plan and solution suggestions, it provides clear guidance for the company's order management decision-making. Companies do not need to blindly choose from many alternatives, but can quickly make high-quality decisions based on scientific evaluation methods and targeted suggestions, reasonably arrange the order execution process, and improve resource utilization efficiency.

[0024] In order to achieve the second purpose of the present invention, the present invention provides a computer device including a processor and a memory, the memory stores a computer program, and when the computer program is executed by the processor, the steps of the above-mentioned dual-branch order information prediction method are implemented.

[0025] In order to achieve the third purpose of the present invention, the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a controller, the steps of the above-mentioned dual-branch order information prediction method are implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flow chart of an embodiment of the dual-branch order information prediction method of the present invention.

[0027] Figure 2 It is a schematic diagram of the calculation of the attention mechanism of the preset time prediction model and the preset default risk assessment model in the embodiment of the dual-branch order information prediction method of the present invention.

[0028] The present invention is further described below in conjunction with the accompanying drawings and embodiments. DETAILED DESCRIPTION

[0029] The dual-branch order information prediction method of the present invention is a computer program applied to a computer device for performing order information prediction.

[0030] Dual-branch order information prediction method embodiment: See also Figure 1 In this embodiment, when the dual-branch order information prediction method is working, step S1 is first executed to create an order and input the pre-information of the order. When creating a new order, the person in charge of the order needs to enter the pre-information, which is a necessary input for the order to start flowing, ensuring that the execution link of the order has complete initial conditions. The pre-information can be set according to the application scenario of the order. For example, in one scenario, the pre-information includes the order creation time, order type, order difficulty, expected completion time, and remarks.

[0031] After inputting the pre-information of the order, step S2 is executed to use the preset time prediction model to predict based on the pre-information and obtain the initial time-consuming parameter information related to the execution link of the order. Once the pre-information is entered, the preset time prediction model starts calculating and outputs the initial time-consuming parameter information related to the execution link of the order based on the input order features.

[0032] The preset time prediction model is usually constructed based on a large amount of historical order data. These historical data contain the pre-information of various orders (such as order type, product quantity, customer requirements, etc.) and the corresponding actual execution time records (such as production time, transportation time, delivery time, etc.). The model uses machine learning algorithms or statistical methods to mine the potential relationship between the pre-information in the data and the time consumption of each execution link. In this embodiment, the preset time prediction model is a time prediction model that uses a feature-driven attention mechanism. Preferably, the preset time prediction model is a time prediction model that uses a feature-driven multi-head attention mechanism. The deep learning architecture based on the feature-driven multi-head attention mechanism can accurately capture the complex interactions between different features, thereby generating highly reliable prediction results.

[0033] In this embodiment, before the step of using a preset time prediction model to make a prediction based on the pre-information, it also includes: using a preset encoder to process the pre-information to obtain data that conforms to the preset time prediction model. The pre-information may exist in various original formats, such as a mixture of text and numerical values, or the dimension does not match the input format expected by the preset time prediction model. The preset encoder can convert these raw data into a format that the model can effectively process. After being processed by the preset encoder, the data is more in line with the requirements of the preset time prediction model, and the model can better learn the patterns and laws in the data, thereby improving the accuracy of predicting the execution time of the order. The existence of the preset encoder enables the system to process pre-information of various formats and types, enhancing the versatility of the system. At the same time, since the data has been standardized, the model training process is more stable and is not easily affected by data outliers or interference of different scales, thereby improving the stability and reliability of the entire order prediction system.

[0034] In this embodiment, the preset encoder includes a BERT model for processing text data and a multi-layer perceptron for processing numerical data. In the order prediction scenario, both text data and numerical data contain important prediction clues. By processing text data with the BERT model and numerical data with the multi-layer perceptron, the value of both types of data can be fully explored and the information in the pre-information can be fully utilized. The combination of the BERT model and the multi-layer perceptron can provide richer and more representative features for subsequent time prediction models. The BERT model and the multi-layer perceptron effectively process the data types they are good at respectively, and the generated features complement each other, which helps the time prediction model to better learn the complex relationship between data and order execution time, thereby improving the accuracy and reliability of the prediction.

[0035] Specifically, the BERT model is responsible for concatenating the text data of the order into a short sentence and breaking the sentence into individual characters using delimiters. Subsequently, the BERT model uses word embedding technology to convert these characters into a 768-dimensional vector representation. In this way, the text information of the order is converted into a high-dimensional vector that can be processed by the preset time prediction model.

[0036] For numerical data, the time fields (such as order creation time, expected completion time, and actual completion time) are first preprocessed. These time data are uniformly converted into a long integer form, for example, "2024 / 6 / 114:15" is converted to "20246011415". Then, these long integers are normalized to map them to a numerical range within 300. For other time data, since they are already regular numerical types, there is no need for excessive preprocessing. After preprocessing, the numerical data is encoded by a multi-layer perceptron. In order to ensure that the expression of numerical data and text data in high-dimensional space is consistent, the multi-layer perceptron is designed as a five-layer neural network architecture, including: the first layer contains 32 neurons, the second layer contains 128 neurons, the third layer contains 512 neurons, the fourth layer contains 768 neurons, and the fifth layer is a nonlinear activation function. The preprocessed data are input into the multi-layer perceptron one by one, and the multi-layer perceptron changes layer by layer to unify the numerical data into a 768-dimensional vector.

[0037] After encoding textual data and numerical data, each order data will generate two vectors of length 768, representing text information and numerical information respectively. Finally, these two 768-dimensional vectors will be directly added together to obtain a high-dimensional vector that combines text and numerical information. This vector will be used as the input of the preset time prediction model to support the model to make accurate predictions and evaluations in practical applications.

[0038] When using the preset time prediction model to make predictions based on the prior information, the specific principles are as follows: like Figure 2 As shown in the figure, the 768-dimensional vector output by the preset encoder is first copied into three copies, representing the query (Query, Q), key (Key, K), and value (Value, V). Q, K, and V are calculated through the conventional attention mechanism to obtain the initial attention weights, and the calculation formula is as follows: ,in, represents the result of conventional attention calculation, is the dimension length of the vector (768 in this embodiment), and the softmax function is used to normalize the result to the probability distribution range. After obtaining the conventional attention weight, the feature-driven attention weight is further calculated. The feature-driven attention calculation process includes the following two formulas: ; ,in, is the feature-driven weight, through the learnable parameters and bias parameters The linear combination of is then calculated using the sigmoid activation function. Initialized with Tensors of uniform dimensions, is the encoded 768-dimensional vector, and ⊙ represents the element-by-element multiplication operation. The final attention result It is the element-wise multiplication of the feature-driven weight and the regular attention weight.

[0039] The above process describes the single-head attention calculation process. In the multi-head attention mechanism, the encoded 768-dimensional vector will be projected into 8 (8-head) independent vector spaces through a projection layer, and each vector space will perform single-head attention calculation independently. The results of the multi-head attention mechanism are synthesized through the following steps: (1) After the calculation results of each head are completed in the independent space, the results of the 8 heads are mapped back to the original 768-dimensional space through a projection layer; (2) The combined results are weighted added to form the final multi-head attention output.

[0040] exist Figure 2 In the process, the result of feature-driven multi-head attention calculation is fused with the encoded original 768-dimensional vector through residual connection and regularization. Then, the result is forward calculated, and residual connection and regularization are performed again to finally generate a feature vector for linear layer input. After the linear layer is calculated by multi-layer neural network, the prediction result is output for specific tasks such as time prediction or risk assessment. This mechanism combines the advantages of conventional attention and feature-driven attention, which can capture richer relationships between input features, while enhancing the model's attention to key features, significantly improving the accuracy and robustness of predictions.

[0041] After obtaining the initial time-consuming parameter information related to the execution link of the order, step S3 is executed to generate a preset number of order execution alternative plans based on the initial time-consuming parameter information. In order to optimize the time planning and risk control of the order and flexibly adapt to the changes in actual production, the original order execution alternative plan needs to be adjusted before the order risk assessment to generate multiple order execution alternative plans.

[0042] In this embodiment, the step of generating a preset number of order execution alternatives according to the initial time-consuming parameter information includes: adjusting the time consumption of the initial time-consuming parameter information according to preset rules to obtain a preset number of different time-consuming parameter information; combining each time-consuming parameter information with the pre-condition information to obtain the corresponding order execution alternative. Among them, the initial time-consuming parameter information is one of the preset number of time-consuming parameter information. By adjusting the initial time-consuming parameter information according to preset rules, various uncertain factors in the actual order execution process can be taken into account. For example, the production link may encounter delays in the supply of raw materials, equipment failures, etc., and the transportation link may be affected by traffic conditions, weather and other factors. Generating multiple different time-consuming parameter information can cover these possible changes, allowing enterprises to prepare response plans in advance. This method can flexibly generate order execution alternatives suitable for different scenarios.

[0043] Preferably, the step of adjusting the time consumption of the initial time consumption parameter information according to preset rules includes: adjusting the time consumption to increase or decrease each time consumption parameter in the initial time consumption parameter information. By increasing or decreasing each time consumption parameter, a variety of different time consumption combinations can be generated relatively quickly. This adjustment method can cover the most basic time changes in the order execution process. In actual business, the time consumption of each link of an order may become longer or shorter due to various factors. For example, the production link may be shortened due to equipment upgrades, or extended due to workers taking leave. By adjusting each time consumption parameter, these actual possible situations can be simulated to a certain extent.

[0044] In one example, the time-consuming parameters include the backend experiment time, the data analysis time, and the result conversion time. When the initial time-consuming parameter information is adjusted, multiple alternative time plans are generated by reducing the backend experiment time, the data analysis time, and the result conversion time by a certain range of hours. For example, assuming that in the initial time-consuming parameter information, the backend experiment time is 54 hours, the data analysis time is 19 hours, and the result conversion time is 85 hours, then the following three alternative time-consuming parameter information will be generated when the plan is adjusted: Plan 1, the backend experiment time is reduced by 5 hours and adjusted to 49 hours, and the other time remains unchanged; Plan 2, the data analysis time is reduced by 10 hours and adjusted to 9 hours, and the other time remains unchanged; Plan 3, the result conversion time is reduced by 12 hours and adjusted to 73 hours, and the other time remains unchanged. According to the above adjustment, four time-consuming parameter information including the initial time-consuming parameter information can be obtained. Combining each time-consuming parameter information with the pre-information, four order execution alternative plans can be obtained.

[0045] It should be noted that, according to the results of historical order data analysis, high-quality completion of a single order usually requires avoiding the simultaneous increase or decrease of all time-consuming parameters. Therefore, only the time of a single link is adjusted, and the adjusted time range is provided by the statistical analysis results of historical data to ensure the rationality and feasibility of the adjustment.

[0046] After generating a preset number of order execution alternatives, execute step S4, use the preset default risk assessment model to predict the default risk of all order execution alternatives, and output the default risk information corresponding to each order execution alternative. After obtaining the prediction plan, it is necessary to determine whether the prediction plan has a default risk. Therefore, it is necessary to predict the default risk of all order execution alternatives in order to obtain the default risk information of each order execution alternative.

[0047] In this embodiment, the preset delinquency risk assessment model is a delinquency risk assessment model that adopts a feature-driven attention mechanism. Preferably, the preset delinquency risk assessment model is a delinquency risk assessment model that adopts a feature-driven multi-head attention mechanism. The delinquency risk assessment model of the feature-driven attention mechanism has the same working principle as the time prediction model of the feature-driven attention mechanism, which will not be repeated here.

[0048] In addition, before using the preset delinquency risk assessment model to predict the delinquency risk of all order execution alternatives, a preset encoder can be used to process the data of the order execution alternatives to obtain data that conforms to the preset delinquency risk assessment model.

[0049] In this embodiment, the default risk information includes the default probability and the default duration. By complementing each other, the default probability and the default duration provide a more comprehensive assessment of the order default risk, which helps enterprises formulate more sophisticated response strategies.

[0050] After obtaining the default risk information corresponding to each order execution alternative, step S5 is executed to output the optimal order execution plan and solution suggestions among the preset number of order execution alternatives according to all the default risk information. By outputting the optimal order execution plan and solution suggestions, clear guidance is provided for the order management decision-making of the enterprise. The enterprise does not need to blindly choose from a large number of alternatives, but can make high-quality decisions quickly based on scientific evaluation methods and targeted suggestions, reasonably arrange the order execution process, and improve resource utilization efficiency.

[0051] When outputting the optimal order execution plan based on all default risk information, since the default risk information includes the default probability and default duration, these two indicators can be comprehensively considered to evaluate the pros and cons of the plan. For example, a risk function can be set to weight the default probability and default duration. The weight reflects the importance that the company attaches to the possibility and duration of default. For example, for some companies with extremely high credit requirements, the weight of the default probability may be increased; and for cost-sensitive companies, the impact of the default duration on the cost may be more valued, and its weight may be increased accordingly. The plan with the smallest risk value is the optimal plan initially screened out.

[0052] When generating proposals, you can provide suggestions for dealing with default risks. If the optimal solution still has a certain default risk, you need to give suggestions based on the source and characteristics of the risk. If the default risk mainly comes from the time-consuming uncertainty of a certain link, such as the transportation link, the suggestion may be to cooperate with a more reliable transporter, increase transportation insurance, or plan alternative transportation routes in advance. If the risk is caused by the production link, you can recommend increasing equipment maintenance, training employees to improve production efficiency, or preparing emergency production equipment to deal with sudden failures.

[0053] In addition to direct responses to the risk of default, it is also necessary to give suggestions based on the overall business scenario. For example, if an order involves the production of a new product, and technical difficulties may arise during the production of the new product, leading to an increased risk of default, it is recommended to conduct technical pre-research in advance, organize a cross-departmental technical team to tackle the problem, and reserve a certain buffer time in the production plan. For orders where customers are extremely sensitive to delivery time, it is recommended to strengthen communication with customers, provide timely feedback on the progress of the order, and negotiate solutions with customers in advance if there is a possible default, so as to reduce the impact on customer satisfaction.

[0054] From the above, it can be seen that the dual-branch order information prediction method of the present invention introduces a dual-branch structure of a preset time prediction model and a preset default risk assessment model, which are responsible for order time prediction and default risk assessment respectively. Each branch independently carries out the prediction task, which can maximize the processing efficiency and accuracy of each task. Specifically, the time prediction branch is responsible for predicting the initial time-consuming parameter information related to the execution link of the order, and dynamically integrates the initial time-consuming parameter information to obtain multiple order execution alternatives. The risk assessment branch focuses on the potential risks in the order execution process. Based on the time prediction results and the existing order information data, combined with the execution difficulty and uncertainty of different links, the default risk is evaluated, and the optimal solution is selected from multiple alternatives, so that it has strong adaptability when processing diversified orders, and adapts to the dynamically changing market environment and production scheduling needs.

[0055] Computer device embodiment: The computer device of this embodiment includes a controller, and when the controller executes a computer program, the steps in the above-mentioned dual-branch order information prediction method embodiment are implemented.

[0056] For example, a computer program may be divided into one or more modules, one or more modules are stored in a memory and executed by a controller to implement the present invention. One or more modules may be a series of computer program instruction segments capable of implementing specific functions, and the instruction segments are used to describe the execution process of the computer program in a computer device.

[0057] The computer device may include, but is not limited to, a controller and a memory. Those skilled in the art will appreciate that the computer device may include more or fewer components, or a combination of certain components, or different components, for example, the computer device may also include input and output devices, network access devices, buses, etc.

[0058] For example, the controller may be a central processing unit (CPU), other general controllers, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general controller may be a microcontroller or any conventional controller, etc. The controller is the control center of the computer device, and uses various interfaces and lines to connect the various parts of the entire computer device.

[0059] The memory can be used to store computer programs and / or modules. The controller realizes various functions of the computer device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. For example, the memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound receiving function, a sound conversion to text function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, text data, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (SmartMediaCard, SMC), a secure digital (SecureDigital, SD) card, a flash card (FlashCard), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0060] Computer readable storage medium embodiment: If the module integrated in the computer device of the above embodiment is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the implementation of all or part of the process in the above dual-branch order information prediction method embodiment can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the controller, the steps of the above dual-branch order information prediction method embodiment can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The storage medium may include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electric carrier signals and telecommunication signals.

[0061] It should be noted that the above are only preferred embodiments of the present invention, but the design concept of the invention is not limited thereto, and any non-substantial modifications made to the present invention using this concept also fall within the protection scope of the present invention.

Claims

1. A dual-branch order information prediction method, characterized in that: include: Create an order and enter the pre-requisite information of the order; Using a preset time prediction model to make a prediction based on the prior information, obtaining initial time-consuming parameter information related to the execution link of the order; Generate a preset number of order execution alternative plans according to the initial time-consuming parameter information; Using a preset default risk assessment model to predict the default risk of all the order execution alternatives, and outputting the default risk information corresponding to each of the order execution alternatives; Outputting the optimal order execution plan among the preset number of order execution alternative plans according to all the default risk information; Among them, the step of generating a preset number of order execution alternative plans based on the initial time-consuming parameter information includes: adjusting the time consumption of the initial time-consuming parameter information according to preset rules to obtain the preset number of different time-consuming parameter information; combining each of the time-consuming parameter information with the pre-condition information to obtain the corresponding order execution alternative plan.

2. The dual-branch order information prediction method according to claim 1, characterized in that: The step of adjusting the initial time-consuming parameter information according to a preset rule includes: An increase or decrease in the time consumption is performed on each time consumption parameter in the initial time consumption parameter information.

3. The dual-branch order information prediction method according to claim 1 or 2, characterized in that: The default risk information includes the default probability and the default duration.

4. The dual-branch order information prediction method according to claim 1 or 2, characterized in that: The preset time prediction model is a time prediction model using a feature-driven attention mechanism; and / or The preset delinquency risk assessment model is a delinquency risk assessment model that adopts a feature-driven attention mechanism.

5. The dual-branch order information prediction method according to claim 1 or 2, characterized in that: Before the step of using a preset time prediction model to make a prediction based on the pre-information, the method further includes: The pre-information is processed using a preset encoder to obtain data that conforms to the preset time prediction model.

6. The dual-branch order information prediction method according to claim 5, characterized in that: The preset encoder includes a BERT model for processing text data and a multi-layer perceptron for processing numerical data.

7. The dual-branch order information prediction method according to claim 1 or 2, characterized in that: When the optimal order execution plan among the preset number of order execution alternative plans is output based on all the default risk information, a plan suggestion is also output.

8. A computer device comprising a processor and a memory, characterized in that: The memory stores a computer program, and when the computer program is executed by the processor, the steps of the dual-branch order information prediction method as described in any one of claims 1 to 7 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the controller, the steps of the dual-branch order information prediction method as described in any one of claims 1 to 7 are implemented.

Images