Business processing time prediction method and device, storage medium and electronic equipment

Abstract

The invention discloses a business processing time prediction method and device, a storage medium and electronic equipment, and relates to the technical field of information, the method comprises the steps of firstly obtaining dynamic feature data and static feature data, the dynamic feature data comprising processing equipment information and processing object information of a to-be-processed task; constructing a time prediction model according to the integrated learning model and the time sequence analysis model, and inputting the dynamic and static feature data into the time prediction model to obtain prediction processing time; and determining the target prediction time of the to-be-processed task according to the M prediction processing times. According to the method, dynamic and static feature data are comprehensively covered through dynamic and static multi-source data fusion, and the time prediction model can analyze the dynamic feature data based on the time sequence and can perform model optimization in real time according to the integrated learning model, so that the service processing time is predicted more accurately, and the service processing efficiency is improved. The prediction precision of the service processing time is improved, and the real-time response capability is enhanced.

Description

Technical Field

[0001] This application relates to the field of information technology, and more specifically, to a method for predicting business processing time, a device for predicting business processing time, a computer-readable storage medium, and an electronic device. Background Technology

[0002] In the financial services industry, predicting customer transaction processing times is crucial for improving user experience, optimizing resource allocation, and enhancing operational efficiency. In recent years, with the acceleration of digital transformation, financial institutions have faced increasing transaction volumes and complexity. The processing time for these transactions is influenced by various factors, such as transaction type, employee efficiency, system load, and customer credit assessment. The accuracy and robustness of transaction processing time prediction technologies are currently poor, leading to unsatisfactory customer experiences. Summary of the Invention

[0003] The main objective of this application is to provide a method, apparatus, computer-readable storage medium, and electronic device for predicting business processing time, in order to solve the problem that the prediction accuracy and robustness of business processing time in related technologies are poor, resulting in a poor customer experience.

[0004] To achieve the above objectives, according to one aspect of this application, a method for predicting business processing time is provided. The method includes: acquiring dynamic feature data and static feature data, wherein the static feature data includes feature information of a task to be processed at the current time, and the dynamic feature data includes processing device information and processing object information of the task to be processed; the processing device information includes operating information of the processing device processing the task to be processed, and the processing object information includes the number and processing speed of the processing objects processing the task to be processed; and the feature information of the task to be processed includes the number, type, and priority of the task to be processed; constructing a time prediction model based on an ensemble learning model and a time series analysis model, and inputting the dynamic feature data and the static feature data into the time prediction model to obtain the predicted processing time of the task to be processed; wherein there are M time prediction models, and at least two of the time prediction models have different input dynamic feature data and static feature data, where M is a positive integer; and determining the target predicted time of the task to be processed based on the M predicted processing times.

[0005] Optionally, acquiring dynamic feature data includes: determining a preset time window; acquiring N initial running datasets of the processing device according to the preset time window, wherein each initial running dataset includes multiple initial running information of the processing device, and each initial running dataset is sampled at a single sampling moment within the preset time window, where N is a positive integer; and performing weighted aggregation processing on the data in the N initial running datasets according to the preset time window to obtain the processing device information of the task to be processed.

[0006] Optionally, constructing a time prediction model based on an ensemble learning model and a time series analysis model includes: obtaining the historical actual processing time and historical predicted processing time of the task to be processed at a historical time, wherein the historical predicted processing time is the processing time of the task to be processed predicted by the ensemble learning model based on historical static feature data at a historical time; obtaining the operating parameters of the ensemble learning model, wherein the operating parameters of the ensemble learning model include the number of training samples, the degree of bias of a single sample, and the number of decision trees, wherein the degree of bias of a single sample characterizes the degree of deviation between the historical actual processing time and the historical predicted processing time; constructing an objective function of the ensemble learning model based at least on the historical actual processing time, the historical predicted processing time, and the operating parameters of the ensemble learning model; determining the ensemble learning model based on the objective function of the ensemble learning model, and fusing the ensemble learning model and the time series analysis model to obtain the time prediction model.

[0007] Optionally, constructing a time prediction model based on an ensemble learning model and a time series analysis model includes: obtaining the operating parameters of the time series analysis model, which include a time step index, a hidden state of the time step, a hidden state of the previous time step, an output layer weight matrix, and an output layer bias term. The time step index represents a moment in the time series data, the hidden state of the time step represents the time series feature information of the corresponding moment, and the hidden state of the previous time step represents the time series feature information resulting from the transition of the moment corresponding to the time step; constructing the objective function of the time series analysis model based on the operating parameters of the time series analysis model; determining the time series analysis model based on the objective function of the time series analysis model, and fusing the ensemble learning model and the time series analysis model to obtain the time prediction model.

[0008] Optionally, inputting the dynamic feature data and the static feature data into the time prediction model to obtain the predicted processing time of the task to be processed includes: obtaining a first weight coefficient and a second weight coefficient; inputting the static feature data into the ensemble learning model of the time prediction model to obtain a first predicted time of the task to be processed, and inputting the dynamic feature data into the time series analysis model of the time prediction model to obtain a second predicted time of the task to be processed; using the product of the first weight coefficient and the first predicted time as a first value, and using the product of the second weight coefficient and the second predicted time as a second value; and determining the sum of the first value and the second value as the predicted processing time of the task to be processed.

[0009] Optionally, obtaining the first weight coefficient and the second weight coefficient includes: obtaining an activation function, the activation function being used to map the first weight coefficient and the second weight coefficient within a preset range; obtaining the deviation parameters of the ensemble learning model and the deviation parameters of the time series analysis model, the deviation parameters of the ensemble learning model representing the degree of deviation between the model prediction value and the true value of the ensemble learning model, and the deviation parameters of the time series analysis model representing the degree of deviation between the model prediction value and the true value of the time series analysis model; determining the first weight coefficient and the second weight coefficient based at least on the activation function, the deviation parameters of the ensemble learning model, and the deviation parameters of the time series analysis model, wherein the first weight coefficient is positively correlated with the deviation parameters of the time series analysis model, and the second weight coefficient is positively correlated with the deviation parameters of the ensemble learning model.

[0010] Optionally, determining the target prediction time of the task to be processed based on the M predicted processing times includes: obtaining multiple calculation weights, each calculation weight corresponding to one of the predicted processing times; and performing a weighted summation of the M predicted processing times based on each predicted processing time and its corresponding calculation weight to obtain the target prediction time of the task to be processed.

[0011] To achieve the above objectives, according to another aspect of this application, a device for predicting business processing time is provided. The device includes: an acquisition unit, configured to acquire dynamic feature data and static feature data, wherein the static feature data includes feature information of a task to be processed at the current time, and the dynamic feature data includes processing device information and processing object information of the task to be processed; the processing device information includes operating information of the processing device processing the task to be processed, and the processing object information includes the number and processing speed of the processing objects processing the task to be processed; and the feature information of the task to be processed includes the number, type, and priority of the task to be processed; a prediction unit, configured to construct a time prediction model based on an ensemble learning model and a time series analysis model, and input the dynamic feature data and the static feature data into the time prediction model to obtain the predicted processing time of the task to be processed; wherein there are M time prediction models, and at least two of the time prediction models input different dynamic feature data and static feature data, where M is a positive integer; and a determination unit, configured to determine the target predicted time of the task to be processed based on the M predicted processing times.

[0012] To achieve the above objectives, according to another aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium including a stored executable program, wherein, when the executable program is executed, it controls the device where the computer-readable storage medium is located to perform any of the aforementioned business processing time prediction methods.

[0013] To achieve the above objectives, according to another aspect of this application, an electronic device is provided, comprising: a memory storing an executable program; and a processor for running the program, wherein the program executes any of the aforementioned business processing time prediction methods during runtime.

[0014] In this embodiment, the above-described method for predicting business processing time is adopted. First, dynamic feature data and static feature data are acquired. The static feature data includes the feature information of the task to be processed at the current time, and the dynamic feature data includes the processing device information and processing object information of the task to be processed. The processing device information includes the operating information of the processing device processing the task to be processed, and the processing object information includes the number of tasks and processing speed of the processing objects processing the task to be processed. The feature information of the task to be processed includes the number, type, and priority of the task to be processed. Then, a time prediction model is constructed based on the ensemble learning model and the time series analysis model, and the dynamic feature data and static feature data are input into the time prediction model to obtain the predicted processing time of the task to be processed. There are M time prediction models, and at least two time prediction models have different input dynamic feature data and static feature data. Finally, the target predicted time of the task to be processed is determined based on the M predicted processing times. This method comprehensively covers dynamic and static feature data through the fusion of dynamic and static multi-source data. The time prediction model can not only analyze dynamic feature data based on time series, but also optimize the model in real time according to the ensemble learning model, which can more accurately predict the processing time of business, improve the prediction accuracy of business processing time, enhance real-time response capability and multi-scenario adaptability, effectively reduce operating costs, and thus solve the problem of poor prediction accuracy and robustness of business processing time in related technologies, which leads to poor customer experience. Attached Figure Description

[0015] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0016] Figure 1 A hardware block diagram of a computer terminal for implementing a method for predicting business processing time is shown.

[0017] Figure 2 This is a flowchart of a business processing time prediction method provided according to an embodiment of this application;

[0018] Figure 3 This is a schematic diagram of a business processing time prediction device provided according to an embodiment of this application;

[0019] Figure 4 This is a structural block diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0020] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0021] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0022] First, some nouns or terms that appear in the description of the embodiments of this application shall be interpreted as follows:

[0023] Multi-source data fusion is the process of constructing a unified feature space by integrating heterogeneous data from processing systems, sensors of processing equipment, monitoring devices of the processed objects, and environmental sensing devices, such as passenger flow sensors.

[0024] Dynamic Prediction Model: A hybrid machine learning model based on ensemble learning (eXtremeGradientBoosting, or XGBoost) and time series analysis (Long-Short Trem Memory, or LSTM). It can dynamically adjust parameters based on real-time data to predict the time to process a single task, such as the serving of a single order or batch of orders.

[0025] Real-time incremental learning: A technique in which the model continuously receives new data after deployment and updates its parameters through online learning algorithms, such as stochastic gradient descent, to adapt to changes in the environment in which the processing device is located.

[0026] Federated Learning Framework: A distributed machine learning architecture that allows multiple processing devices to train models locally and share encrypted parameters (rather than raw data) to protect privacy and improve the generalization ability of global models.

[0027] Adaptive Attention Mechanism: A dynamic feature weighting method that assigns variable weights to input features in different contexts, such as peak hours in a restaurant or equipment malfunctions, by calculating feature relevance scores.

[0028] Feedback Optimization Module: This module adjusts model parameters and resource allocation strategies by comparing the error between the prediction results and the processing time of the processing task.

[0029] Edge-Cloud Collaborative Architecture: A computing resource allocation scheme that distributes real-time data processing tasks to local edge nodes, such as restaurant servers, while entrusting model training and global optimization tasks to cloud central nodes.

[0030] Priority Weight Overlay: This is a logic that adds extra weight coefficients to the feature vectors of a specific task to give them higher priority in model prediction.

[0031] Time Series Augmentation: This method generates synthetic time-series data through time warping, random sampling, and other techniques to enhance the model's robustness to extreme scenarios, such as a surge in restaurant orders or equipment failure.

[0032] Embedded finance refers to integrating financial functions such as payment, credit, insurance, and wealth management into the processes of non-financial platforms such as e-commerce, travel, healthcare, and retail. Users can complete financial transactions within the context of their existing usage scenarios without needing to switch to a dedicated financial app.

[0033] In the application process, the embedded financial functions can be further extended to, for example, payment and settlement in the catering industry, dynamic assessment of merchant credit limits (based on serving efficiency and customer flow data), and installment payments for customer consumer credit, so as to achieve a deep integration of special operating methods and financial services.

[0034] It should be noted that the information collected in this application (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data used for analysis, etc.) are information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of this data all comply with relevant laws, regulations, and standards, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding access points are provided for users to choose to authorize or refuse. For example, interfaces are set up between this system and relevant users or organizations, providing users with corresponding access points to choose to agree to or refuse automated decision-making results; if the user chooses to refuse, the process proceeds to the expert decision-making stage.

[0035] For example, this system has an interface with relevant users or organizations. Before obtaining relevant information, it needs to send an acquisition request to the aforementioned users or organizations through the interface. After receiving consent from the aforementioned users or organizations, the relevant information is obtained. Users can view the purpose of data use in real time through the authorization interface and have the right to withdraw authorization or delete data at any time. After the authorization is withdrawn, the system will terminate the relevant data processing within 24 hours.

[0036] Example 1

[0037] According to an embodiment of this application, an embodiment of a method for predicting business processing time is also provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0038] The method embodiment provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1 A hardware block diagram of a computer terminal (or mobile device) for implementing a method to predict business processing time is shown. Figure 1 As shown, the computer terminal 10 (or mobile device) may include one or more processors 102 (shown as 102a, 102b, ..., 102n in the figure) 102 (processor 102 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, and a transmission device 106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of a BUS bus), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 1The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0039] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0040] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the business processing time prediction method in this embodiment. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned business processing time prediction method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0041] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0042] The display may be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0043] The system architecture of this application includes four core modules: a data acquisition layer, a preprocessing layer, a model layer, and an application layer. The data acquisition layer collects multi-source data through Internet of Things (IoT) devices, device processing systems, and device terminals. The preprocessing layer is used for data cleaning, feature engineering, and time series alignment. The model layer is used to integrate learning models and time series analysis hybrid models, combined with a federated learning framework. The application layer is used for real-time prediction, visualization interface, early warning, and scheduling.

[0044] Under the aforementioned operating environment, this application provides the following: Figure 2 The method for predicting business processing time is shown. Figure 2 This is a flowchart of a method for predicting business processing time according to Embodiment 1 of this application.

[0045] Step S201: Obtain dynamic feature data and static feature data. The static feature data includes feature information of the task to be processed at the current time. The dynamic feature data includes processing device information and processing object information of the task to be processed. The processing device information includes the operation information of the processing device that processes the task to be processed. The processing object information includes the number of processing objects and processing speed of the processing objects that process the task to be processed. The feature information of the task to be processed includes the number, type and priority of the task to be processed.

[0046] Specifically, the business processing time prediction system integrates IoT devices with the business system to capture dynamic data in real time. For example, it collects real-time temperature and pressure information from the processing equipment's sensors, obtains the number of tasks each employee is currently handling through the employee management system, and captures static data from the processing equipment system, such as the types of dishes and the number of orders in a restaurant setting. This richness of data significantly improves prediction accuracy and reduces the mean absolute error of task processing time prediction. By capturing real-time equipment status and efficiency fluctuations, the system can make more accurate predictions under different working environments. For example, during peak hours in a restaurant setting, the prediction error rate drops from 18.3% using traditional methods to 6.0%. Furthermore, for task prioritization, a priority weighting mechanism is used, which adds extra weight coefficients to the feature vectors of specific tasks to be processed, giving them higher priority in the model's prediction logic. This embodiment supports manual priority adjustment through the management interface; and the priority effect gradually decreases over time, avoiding long-term resource occupation.

[0047] Step S202: Construct a time prediction model based on the ensemble learning model and the time series analysis model, and input the dynamic feature data and the static feature data into the time prediction model to obtain the predicted processing time of the task to be processed. There are M time prediction models, and at least two of the time prediction models have different dynamic feature data and static feature data input, where M is a positive integer.

[0048] When dynamic and static feature data are input into the time prediction model for calculation, corresponding calculation weights are assigned. These weights are determined based on an attention mechanism, and the feature weight allocation formula in the attention mechanism calculation is as follows:

[0049] ;

[0050] in, It is the output of the attention mechanism, that is, the feature vector after dynamic weighting, which is used for model prediction; It is a query vector, which represents current context information, such as peak period markers, equipment fault status, etc. It is a key vector, which represents an identifier matrix of input features, such as the matrix representation of features like device status and efficiency; It is a value vector, which represents the actual value vector of the input features, and... One-to-one correspondence; It is a key vector The transpose of Q is used to perform matrix multiplication with Q to calculate feature correlation; It is a matrix product of the query vector and the transpose key vector, and the elements in the resulting matrix represent the relevance scores of Q with each K. It is a key vector The dimension is used to normalize the relevance score and avoid excessively large values. It is the square root of the key vector dimension, which is a normalization factor that reduces the impact of dimension on the relevance score. It is the Softmax activation function, which converts the normalized relevance score into a probability distribution to obtain the weight ratio of each feature.

[0051] The weights obtained based on the attention mechanism technology are automatically adjusted according to the operating status of the equipment, such as restaurant customer traffic and equipment load; and based on the query-key matrix dot-product attention.

[0052] Additionally, if a device malfunction is detected, the device state feature weights are adjusted. Increase by 1.5 times, that is ,in, The weights for the improved device status features.

[0053] Specifically, firstly, ensemble learning is used to build a prediction model for handling static features; secondly, time series analysis is used to build a prediction model for handling dynamic features; finally, the prediction results from both models are weighted and combined through model fusion to obtain the final predicted processing time. Model fusion technology ensures the stability and accuracy of the prediction results in the face of changing operating environments by dynamically adjusting the weights of different models. For example, in the case of equipment failure, the weights of equipment state features are adjusted accordingly. The time prediction value has been increased to 1.5 times the original value, which can reflect the impact of changes in equipment status on the processing time of tasks in a timely manner, thus improving the model's generalization ability and prediction accuracy. Among them, the time prediction model is a dynamic prediction model. The dynamic prediction model uses XGBoost to process static features, such as the order volume trend of dishes, and dynamically allocates the importance weight of features through an attention mechanism.

[0054] Step S203: Based on the M predicted processing times mentioned above, determine the target predicted time for the task to be processed.

[0055] Specifically, the system takes a weighted average of all predicted times from M models and dynamically adjusts the weights based on the overall performance of the current models, such as Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE), to determine the final predicted time. Through model fusion, even if a single model is significantly affected by external disturbances, such as a sudden surge in orders in a restaurant setting, the system can still correct its predictions based on the results of other models, ultimately providing a more stable and reliable predicted time, further enhancing the stability and reliability of the prediction results.

[0056] The method for predicting the processing time of the above-mentioned business processes in this application first obtains dynamic feature data and static feature data. The static feature data includes the feature information of the task to be processed at the current moment, and the dynamic feature data includes the processing equipment information and processing object information of the task to be processed. The processing equipment information includes the operating information of the processing equipment processing the task to be processed, and the processing object information includes the number of processing objects and processing speed of the processing objects processing the task to be processed. The feature information of the task to be processed includes the number, type and priority of the task to be processed. A time prediction model is constructed based on an ensemble learning model and a time series analysis model, and the dynamic feature data and static feature data are input into the time prediction model to obtain the predicted processing time of the task to be processed. There are M time prediction models, and at least two time prediction models have different dynamic feature data and static feature data input, where M is a positive integer. Based on the M predicted processing times, the target predicted time of the task to be processed is determined. This method, through multi-source data fusion, dynamic model architecture, and real-time optimization mechanism, more accurately predicts business processing time, thereby significantly improving prediction accuracy, real-time response capability, and multi-scenario adaptability. It effectively reduces operating costs and improves customer satisfaction, thus solving the problem of poor prediction accuracy and robustness of business processing time in related technologies, which leads to poor customer experience.

[0057] In this embodiment, by acquiring dynamic and static feature data and constructing M time prediction models using an ensemble learning model and a time series analysis model, the low accuracy of task processing time prediction caused by single data dimensions and excessive reliance on human experience in related technologies can be effectively solved. Specifically, dynamic feature data covers the operating status of processing equipment and the working state of the processed objects, such as the real-time temperature and pressure of kitchen equipment in a restaurant environment, and the chef's current workload and work efficiency. This real-time changing dynamic feature data is crucial for accurately predicting the time required to process tasks. Static feature data includes the quantity, type, and priority of tasks to be processed, such as the types of dishes ordered and the required preparation time in a restaurant environment, and whether the order is a high-level order. Each of the M time prediction models makes predictions based on different dynamic and static feature data. This parallel processing method enhances the model's adaptability to complex scenarios and improves the robustness of the prediction. Finally, the target prediction time of the task to be processed is determined from the M predicted processing times. By comparing and synthesizing the prediction results of multiple models, a more accurate and reliable prediction value is obtained. Compared to traditional methods that rely solely on a single model for prediction, the technical solution of this application significantly reduces prediction errors and improves the real-time response capability and intelligence level of the prediction system.

[0058] In some embodiments, acquiring dynamic feature data includes the following steps:

[0059] Step S301: Determine the preset time window;

[0060] Step S302: According to the preset time window, obtain N initial running datasets of the processing device. Each initial running dataset includes multiple initial running information of the processing device. Each initial running dataset is obtained by sampling at a single sampling moment within the preset time window. N is a positive integer.

[0061] Step S303: According to the preset time window, perform weighted aggregation processing on the data in the N initial running datasets to obtain the processing device information of the task to be processed.

[0062] The multi-source data fusion and processing flow is as follows: First, input the data:

[0063] Processing task data: ;

[0064] Processing equipment status: ;

[0065] Processing equipment efficiency: .

[0066] Multi-source data fusion technology has the following technical characteristics: heterogeneity: it covers structured (e.g., order tables), semi-structured (e.g., scheduling logs), and unstructured (e.g., customer reviews) data; real-time: the data update frequency ranges from the second level, i.e., equipment status, to the hour level; integrity and supplementation: the fusion process includes sub-steps such as data cleaning, format standardization, and conflict resolution to ensure the consistency and availability of multi-source data and provide a high-quality data foundation for subsequent modeling.

[0067] To unify data from different frequencies to a time window T, such as 1 minute, the following data alignment formula can be used:

[0068] ;

[0069] in, =T is the weighted aggregate value of device status within the time window T. It covers the average level of multiple core indicators of devices. The value range varies depending on the device type. For example, the occupancy status is 0-1. It is used to achieve time sequence consistency of multi-source data. T is the preset data alignment time window. It is usually configured as 1 minute and can be flexibly adjusted to 30 seconds, 5 minutes, etc. It is the basic time unit for data aggregation and balances real-time performance and computational efficiency. t is a single sampling moment within the time window T. For example, when T=1 minute, t can be 0-59 seconds, that is, 1 sampling point per second. It is used to traverse all the original sampled data within the window. It is the original status data of the equipment, which is single-dimensional data and The consistent value range is the basic data source for aggregation calculation; n is the number of sampling points within the time window T, that is, the number of times the original data is collected within the window, which is used to calculate the average value of the data within the window; This is the summation operator, which represents the summation operation applied to all sampling times t within a time window T. Perform cumulative calculations. Feature encoding example:

[0070] ;

[0071] in, It is an embedding function, which is used to convert discrete levels into continuous low-dimensional vectors, making it easier for the model to capture the semantic associations and quantitative differences between experience levels; These are the components of the embedded vector, each corresponding to a specific value in one of the three dimensions of the low-dimensional vector. They are obtained through model training and optimization. Each component represents the quantitative expression of the experience level in a certain feature dimension. The level is a classification of the processing efficiency of the personnel (i.e. the processing object) of the task to be processed, such as "beginner", "intermediate" and "advanced", for example, the speed at which different chefs serve food in a restaurant environment. It is a level embedding vector, a one-dimensional vector consisting of three components, which maps discrete experience levels to continuous numerical vectors and is used as the feature input of the model. It is a 3-dimensional real number space, which represents the range of values ​​of the embedded vector. That is, the vector belongs to the 3-dimensional real number set. 3 is the fixed dimension of the vector, which can be adjusted to other dimensions according to the actual scenario.

[0072] In this embodiment, the acquisition of system dynamic feature data is based on a preset time window to ensure data real-time performance and consistency. N initial running datasets of the processing device are collected, each dataset encompassing multiple initial running information points of the device at a single sampling moment. N represents the number of sampling points collected within the preset time window, ensuring data continuity and integrity. Subsequently, these initial running datasets undergo weighted aggregation to generate processing device information. This process considers the fluctuations in device status over time, using average values ​​to eliminate the impact of instantaneous anomalies on prediction. Weighted aggregation not only improves data quality but also provides a more comprehensive description of device status for subsequent model training, enabling the prediction model to more accurately reflect the real-time operating status of the processing device, thereby improving the accuracy and stability of the task processing time prediction. The system in this embodiment can capture and analyze subtle changes in device status in real time, ensuring high accuracy of prediction results and enhancing the robustness of the model when handling complex dynamic scenarios.

[0073] In some embodiments, a time prediction model is constructed based on an ensemble learning model and a time series analysis model, including the following steps:

[0074] Step S401: Obtain the historical actual processing time and historical predicted processing time of the task to be processed at a historical moment. The historical predicted processing time is the processing time of the task to be processed predicted by the ensemble learning model based on historical static feature data at a historical moment.

[0075] Step S402: Obtain the running parameters of the above-mentioned ensemble learning model. The running parameters of the above-mentioned ensemble learning model include the number of training samples of the above-mentioned ensemble learning model, the degree of bias of a single sample, and the number of decision trees. The degree of bias of a single sample represents the degree of deviation between the above-mentioned historical actual processing time and the above-mentioned historical predicted processing time.

[0076] Step S403: Construct the objective function of the ensemble learning model based at least on the above-mentioned historical actual processing time, the above-mentioned historical predicted processing time, and the above-mentioned running parameters of the ensemble learning model.

[0077] Step S404: Determine the ensemble learning model based on the objective function of the ensemble learning model, and fuse the ensemble learning model and the time series analysis model to obtain the time prediction model.

[0078] The dynamic prediction model includes the XGBoost model and the LSTM model, and the objective function of the ensemble learning model is:

[0079] ;

[0080] in, is the objective function of the XGBoost model, which measures the deviation between the model's predicted values ​​and the true values, and guides the optimization of model parameters; n is the number of samples, specifically the total number of relevant samples participating in the training of the XGBoost model; i is the sample index, which ranges from 1 to n and is used to traverse all training samples. It is the true value of the i-th sample, that is, the actual processing time of the i-th task to be processed; It is the predicted value of the i-th sample, that is, the predicted processing time of the i-th task to be processed output by the model; , is the squared loss term, which is used to quantify the degree of deviation between the true value and the predicted value of a single sample; This is the summation operator, which represents the cumulative summation of the squared loss terms over all n samples; is a hyperparameter used to control the complexity penalty strength of the tree, balancing the model's fitting ability and generalization ability; K is the number of decision trees in the XGBoost model, i.e., the total number of weak classifiers in the model; k is the decision tree index, which ranges from 1 to K and is used to traverse all decision trees. It is the complexity index of the k-th decision tree, which serves as a regularization term for the model to prevent overfitting. This is the summation operator, which represents the cumulative summation of the complexity metrics of all K decision trees.

[0081] In this embodiment, the process of constructing the time prediction model is further refined to improve the model's prediction accuracy in complex dynamic environments. The time prediction model integrates an ensemble learning model and a time series analysis model. The ensemble learning model is used to process historical static feature data, such as the types of tasks to be processed in a restaurant environment; while the time series analysis model focuses on time series analysis to capture time-dependent features, such as the changing trend of tasks over time. In specific implementation, the actual processing time of the tasks to be processed at historical moments and the processing time predicted by the ensemble learning model are first collected. Through comparative analysis, the operating parameters of the ensemble learning model are obtained, including key indicators such as the number of training samples, the degree of bias of individual samples, and the number of decision trees. These parameters not only reflect the current training state of the model but also characterize the degree of deviation between the historical actual processing time and the historical predicted processing time, serving as an important basis for model optimization. Subsequently, based at least on these historical data and model parameters, the objective function of the ensemble learning model is constructed, and the final ensemble learning model is determined by optimizing the objective function. Finally, the optimized XGBoost model is fused with the LSTM model to obtain the time prediction model. This fusion process, through an adaptive weight allocation mechanism, ensures that the model can dynamically adjust its prediction strategy according to real-time environmental changes, effectively handling surges in tasks during peak hours and stable conditions during off-peak hours. Especially when processing high-dimensional dynamic data, the model demonstrates excellent robustness and adaptability, rapidly responding to environmental changes, reducing prediction errors, improving prediction accuracy and real-time performance, and significantly enhancing equipment operating efficiency and customer satisfaction.

[0082] In some embodiments, a time prediction model is constructed based on an ensemble learning model and a time series analysis model, including the following steps:

[0083] Step S501: Obtain the running parameters of the above time series analysis model. The running parameters of the above time series analysis model include time step index, hidden state of time step, hidden state of the previous time step, output layer weight matrix and output layer bias term. The time step index represents the time in the time series data. The hidden state of the above time step represents the time series feature information of the corresponding time step. The hidden state of the previous time step represents the time series feature information caused by the migration of the time corresponding to the above time step.

[0084] Step S502: Construct the objective function of the time series analysis model based on the running parameters of the time series analysis model.

[0085] Step S503: Determine the time series analysis model based on the objective function of the time series analysis model, and fuse the ensemble learning model and the time series analysis model to obtain the time prediction model.

[0086] In time series analysis, the formula for calculating the hidden state at time step t is as follows:

[0087] ;

[0088] The final prediction output formula is as follows:

[0089] ;

[0090] in, t is the LSTM hidden state at time step t, which contains the temporal feature information up to time step t and is used to pass on historical dependencies; t is the time step index, which represents the t-th moment in the time series data, such as the t-th minute or the t-th second; It is the computation function of the LSTM network, which is used to receive the current input and the hidden state of the previous time step, and output the hidden state of the current time step. It is the input feature vector at time step t, which contains the multi-source fusion features at that moment, such as device status features, etc. It is the hidden state of the LSTM at time step t-1; it is used to transmit the timing information of the previous time step, reflecting the memory function of the LSTM. It is the final prediction value of the LSTM model, which is the prediction result of task processing time based on the output of all temporal features; It is the output layer weight matrix, which is used to predict the task processing time based on all temporal features. It is the hidden state of the last time step T, which contains comprehensive feature information of the entire time series; It is the output layer bias term, which is used to adjust the baseline of the predicted values ​​and is optimized through training.

[0091] In this embodiment, the method for constructing a time prediction model further includes obtaining the operating parameters of the time series analysis model. These parameters include the time step index t, the hidden state ht of time step t, and the hidden state of the previous time step t-1. The output layer weight matrix W and the output layer bias term b are defined, where the time step index t represents a specific time point in the time series data, the hidden state ht of time step t encapsulates the temporal feature information up to that time point, and the hidden state of the previous time step t-1 is defined. This carries the temporal characteristics of previous time points, ensuring the continuous transmission of information. Based on these operating parameters, an objective function for the time series analysis model was constructed, and the time series analysis model was determined through this objective function. Subsequently, the ensemble learning model and the time series analysis model were combined through weighted fusion to generate the final time prediction model. This fusion model can comprehensively handle static features and dynamic dependencies, providing more accurate prediction results. During implementation, the model can adaptively adjust its parameters to cope with the constantly changing conditions in the environment where the processing equipment is located, such as real-time changes in equipment status, efficiency fluctuations, and unstable trends of pending tasks. Through real-time incremental learning and federated learning frameworks, the model maintains high flexibility and accuracy, significantly improving the timeliness and reliability of predictions, thereby effectively supporting processing equipment scheduling and resource allocation decisions.

[0092] Furthermore, the aforementioned time prediction model employs a real-time incremental learning mechanism. This means that after deployment, the model continuously receives new data and updates its parameters through online learning algorithms, such as stochastic gradient descent, to adapt to changes in the environment of the processing device. For example, the model is fine-tuned every 15 minutes to ensure that the prediction results are consistent with the current state, and a historical data buffer is maintained to prevent new data from excessively overwriting older knowledge.

[0093] In some embodiments, the dynamic feature data and the static feature data are input into the time prediction model to obtain the predicted processing time of the task to be processed, including the following steps:

[0094] Step S601: Obtain the first weight coefficient and the second weight coefficient;

[0095] Step S602: Input the static feature data into the ensemble learning model of the time prediction model to obtain the first predicted time of the task to be processed, and input the dynamic feature data into the time series analysis model of the time prediction model to obtain the second predicted time of the task to be processed.

[0096] Step S603: The product of the first weighting coefficient and the first prediction time is used as the first value, and the product of the second weighting coefficient and the second prediction time is used as the second value.

[0097] Step S604: The sum of the first value and the second value is determined as the predicted processing time of the task to be processed.

[0098] The weighted fusion formula is as follows:

[0099] ;

[0100] in, It is the predicted processing time, that is, the predicted task processing time of the final output; It is the first weighting coefficient, with a value ranging from 0 to 1, used to allocate the importance ratio of the prediction results of XGBoost and LSTM models; The first prediction time is the task processing time prediction result obtained based on static features. This is the second prediction time, which is the task processing time prediction result obtained based on temporal features; It is the second weighting coefficient, which is complementary to α, ensuring that the sum of the weights of the two models is 1.

[0101] In this embodiment, dynamic and static feature data are efficiently fused to more accurately predict the processing time of the task. First, a first weighting coefficient and a second weighting coefficient are calculated, corresponding to the importance of static and dynamic feature data in the prediction model, respectively. Next, the static feature data is input into the ensemble learning model to obtain the first predicted time, reflecting the impact of invariant factors such as equipment location on the processing time of the task. Simultaneously, dynamic feature data, including real-time order volume, equipment operating status, and employee efficiency, is fed into the time series analysis model to derive the second predicted time, capturing rapidly changing elements during operation. Subsequently, the first and second predicted times are multiplied by their respective weighting coefficients to produce a first value and a second value. Finally, these two values ​​are added to determine the final predicted processing time, thus achieving a comprehensive consideration of both static and dynamic features, improving the accuracy and real-time performance of the prediction. This method ensures that the prediction model can utilize the stable characteristics of historical data while responding promptly to environmental changes.

[0102] In some embodiments, obtaining the first weighting coefficient and the second weighting coefficient includes the following steps:

[0103] Step S701: Obtain the activation function, which is used to map the first weight coefficient and the second weight coefficient within a preset range;

[0104] Step S702: Obtain the deviation parameters of the above-mentioned ensemble learning model and the deviation parameters of the above-mentioned time series analysis model. The deviation parameters of the above-mentioned ensemble learning model represent the degree of deviation between the model prediction value and the actual value of the above-mentioned ensemble learning model, and the deviation parameters of the above-mentioned time series analysis model represent the degree of deviation between the model prediction value and the actual value of the above-mentioned time series analysis model.

[0105] Step S703: Determine the first weight coefficient and the second weight coefficient based at least on the activation function, the bias parameters of the ensemble learning model, and the bias parameters of the time series analysis model. The first weight coefficient is positively correlated with the bias parameters of the time series analysis model, and the second weight coefficient is positively correlated with the bias parameters of the ensemble learning model.

[0106] The first weighting coefficient The dynamic adjustment formula is:

[0107] ;

[0108] in, It is the first weighting coefficient, which is the same as the weighted fusion formula. This is used to dynamically adjust the weight ratio of the two models; It is an activation function, which can be written as the Sigmoid activation function. The Sigmoid activation function is a commonly used non-linear activation function in neural networks. It maps the input value to the interval of 0-1, ensuring... It meets the requirements for the value of the weighting coefficient; It is the mean absolute error of the LSTM model, that is, the average degree of deviation between the predicted value and the true value of the LSTM model; It is the mean absolute error of the XGBoost model, that is, the average degree of deviation between the predicted value and the true value of the XGBoost model; It is a smoothing coefficient used to avoid cases where the denominator is 0, thus ensuring the stability of the calculation.

[0109] In other embodiments, the model can be fine-tuned every 15 minutes to ensure that the prediction results are consistent with the current state; and a historical data buffer pool is maintained to prevent new data from overwriting old knowledge.

[0110] In this embodiment, the process of obtaining the first and second weight coefficients includes using an activation function to ensure that the weight values ​​fall within a preset range, while considering the deviation parameters of the ensemble learning model and the time series analysis model. Specifically, the deviation parameters of the ensemble learning model reflect the accuracy of static feature prediction, while the deviation parameters of the time series analysis model measure the error of dynamic time series prediction. By analyzing the deviation parameters of the two models, this embodiment can dynamically adjust the weight coefficients to achieve adaptive fusion of prediction results. The first weight coefficient is positively correlated with the deviation parameters of the time series analysis model; when the deviation is large, the weight coefficient increases, indicating a greater reliance on real-time dynamic data. The second weight coefficient is positively correlated with the deviation parameters of the ensemble learning model; when the deviation increases, the weight coefficient increases, enhancing the importance of static features. This weight coefficient adjustment mechanism ensures the prediction accuracy of the model under different operating scenarios, especially in extreme cases such as equipment failure and a surge in pending tasks, enabling more accurate prediction of task processing time and improving equipment operating efficiency.

[0111] In some embodiments, determining the target prediction time for the task to be processed based on the above M predicted processing times includes the following steps:

[0112] Step S801: Obtain multiple calculation weights, each of which corresponds to a prediction processing time.

[0113] Step S802: Based on each of the above-mentioned prediction processing times and their corresponding calculation weights, perform a weighted summation of the above-mentioned M prediction processing times to obtain the above-mentioned target prediction time for the task to be processed.

[0114] The cloud aggregation formula, also known as the Federated Learning (FedAvg) algorithm, is as follows:

[0115] ;

[0116] in To handle the amount of data from the devices, ; These are the parameters of the global model in the cloud, which are used to obtain the general model parameters after fusing the local model parameters of each processing device, and are used for global prediction optimization; N is the number of processing devices participating in federated learning, that is, the total number of edge nodes participating in parameter aggregation; i is the index of each processing device, which ranges from 1 to N, and is used to traverse all participating processing devices. These are the local model parameters of the i-th processing device, which are model parameters trained based on the data of each processing device and then uploaded to the cloud after encryption. It is the amount of training data for the i-th processing device, that is, the total number of samples used by the processing device for local model training; It is the total amount of training data across all processing devices; It is the parameter weight of the i-th processing device, which allocates the importance of the parameters according to the proportion of data volume of the branch; the larger the data volume, the higher the weight. It is the summation operator, which means to sum the weighted parameters of all N branches to obtain the global parameter.

[0117] The aforementioned processing devices are edge devices. Multiple predicted processing times obtained from these devices are sent to the cloud, where they are used to calculate a target predicted time. This process utilizes a federated learning framework, combining edge nodes with parameter aggregation, i.e., a cloud server. Differential privacy technology is employed to ensure that data from multiple processing devices is irreversible. This algorithm can further improve prediction accuracy. Its technical principle involves relearning the times predicted by each processing device based on the dynamic prediction model, similar to a weighted average, resulting in higher accuracy and a stronger overall model learning capability.

[0118] Among them, the edge-cloud collaborative architecture technology has the following technical characteristics: low latency response, that is, edge nodes process real-time prediction (latency <200ms); global model aggregation, that is, the cloud integrates parameters of multiple processing devices to generate a general prediction model.

[0119] In this embodiment, the target predicted time for the task is determined by weighting and summing the M predicted processing times using computational weights. The core principle is that these computational weights are dynamically adjusted based on the real-time environment and task characteristics, ensuring that the contribution of each predicted processing time in the overall evaluation matches its actual impact. For example, during peak periods, the predicted processing times related to equipment status and resource allocation may have a greater impact on the final processing time. In this case, the corresponding computational weights will be automatically increased, giving them a higher proportion in the calculation of the target predicted time. In this way, the system can more accurately reflect the impact of the current operational status on task processing time, improving prediction accuracy. This solution significantly reduces prediction errors and achieves more efficient and intelligent resource scheduling when handling multi-tasking environments. Its advantages are particularly evident in predicting food serving times in the catering industry, providing strong support for optimizing customer experience and business processes.

[0120] The following table compares the effectiveness of this application with traditional methods after actual testing in a restaurant environment:

[0121] Table 1. Comparison of the effects of traditional methods and this embodiment

[0122]

[0123] In some embodiments, there may also be functional modules that adjust model parameters and resource allocation strategies in reverse by comparing the error between the prediction result and the processing time of the processing task. For example, updating model weights based on the loss function; or automatically pushing processing equipment scheduling suggestions when the prediction time exceeds a threshold, such as activating backup equipment.

[0124] In other embodiments, time-series augmentation techniques, such as time warping and random sampling, can be used to generate synthetic time-series data to enhance the model's robustness to extreme scenarios, such as surges in restaurant orders or equipment failures. This can simulate the overlap of equipment failures and peak periods of pending tasks, covering different operational modes across various scenarios and time periods.

[0125] The above embodiments combine the integrated learning model with time series analysis for the first time to solve the problem of joint modeling of static and dynamic features; and achieve a balance between low-latency prediction and global optimization through edge-cloud collaboration; in addition, the attention mechanism and federated learning enable the prediction of business processing time to adapt to changing operating scenarios.

[0126] This embodiment significantly improves technical performance and user experience through multi-source data fusion, dynamic model architecture, and real-time optimization mechanisms. In terms of technical performance, prediction accuracy is significantly enhanced, along with real-time response capabilities and multi-scenario adaptability. Regarding business value, automated early warning and resource allocation reduce the frequency of manual intervention, effectively lowering operating costs and improving customer satisfaction. Customers can view prediction times in real-time through the application (APP), increasing transparency and automatically prioritizing urgent tasks.

[0127] This application achieves a comprehensive improvement in both technical indicators and commercial value through data-driven decision-making and dynamic adaptive models, providing a complete technical path. It also offers valuable insights for other similar financial scenarios, such as customer transaction time prediction, and embedded financial scenarios, including hotel check-out time prediction and scenic spot visitor departure time prediction. Its core effects can be summarized as follows: reduced prediction error rate; minute-level model updates and millisecond-level response; multi-scenario adaptation and automated decision-making.

[0128] The business processing time prediction method provided in this application, through multi-source data fusion, dynamic model architecture and real-time optimization mechanism, more accurately predicts the business processing time, thereby significantly improving prediction accuracy, real-time response capability and multi-scenario adaptability, effectively reducing operating costs and improving customer satisfaction. It also solves the problem of poor prediction accuracy and robustness of business processing time in related technologies, which leads to poor customer experience.

[0129] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0130] Example 2

[0131] This application also provides a business processing time prediction device. It should be noted that the business processing time prediction device of this application can be used to execute the business processing time prediction method provided in this application. The business processing time prediction device provided in this application is described below.

[0132] According to an embodiment of this application, an apparatus for implementing the above-described method for predicting business processing time is also provided, such as... Figure 3 As shown, the device includes: an acquisition unit 100, a prediction unit 200, and a determination unit 300. The acquisition unit 100 is used to acquire dynamic feature data and static feature data. The static feature data includes feature information of the task to be processed at the current time. The dynamic feature data includes processing device information and processing object information of the task to be processed. The processing device information includes the operation information of the processing device processing the task to be processed. The processing object information includes the number of processing objects and the processing speed of the processing objects processing the task to be processed. The feature information of the task to be processed includes the number, type, and priority of the task to be processed. The prediction unit 200 is used to construct a time prediction model based on an ensemble learning model and a time series analysis model, and input the dynamic feature data and the static feature data into the time prediction model to obtain the predicted processing time of the task to be processed. There are M time prediction models, and at least two of the time prediction models have different dynamic feature data and static feature data input, where M is a positive integer. The determination unit 300 is used to determine the target predicted time of the task to be processed based on the M predicted processing times.

[0133] The aforementioned business processing time prediction device of this application includes an acquisition unit, a prediction unit, and a determination unit. The acquisition unit is used to acquire dynamic feature data and static feature data. The static feature data includes feature information of the task to be processed at the current time. The dynamic feature data includes processing equipment information and processing object information of the task to be processed. The processing equipment information includes the operating information of the processing equipment processing the task to be processed. The processing object information includes the number of processing objects and processing speed of the processing objects processing the task to be processed. The feature information of the task to be processed includes the number, type, and priority of the task to be processed. The prediction unit is used to construct a time prediction model based on an ensemble learning model and a time series analysis model, and input the dynamic feature data and static feature data into the time prediction model to obtain the predicted processing time of the task to be processed. There are M time prediction models, and at least two time prediction models have different dynamic feature data and static feature data input, where M is a positive integer. The determination unit is used to determine the target predicted time of the task to be processed based on the M predicted processing times. This device, through multi-source data fusion, dynamic model architecture, and real-time optimization mechanisms, more accurately predicts business processing time, thereby significantly improving prediction accuracy, real-time response capability, and multi-scenario adaptability. It effectively reduces operating costs and improves customer satisfaction, thus solving the problem of poor prediction accuracy and robustness of business processing time in related technologies, which leads to poor customer experience.

[0134] In some embodiments, the acquisition unit includes: a first determining module, a first acquisition module, and a second acquisition module. The first determining module is used to determine a preset time window; the first acquisition module is used to acquire N initial running datasets of the processing device according to the preset time window, wherein each initial running dataset includes multiple initial running information of the processing device, and each initial running dataset is sampled at a single sampling moment within the preset time window, where N is a positive integer; the second acquisition module is used to perform weighted aggregation processing on the data in the N initial running datasets according to the preset time window to obtain the processing device information of the task to be processed. Through this technical solution, the system can capture and analyze subtle changes in the device state in real time, ensuring high accuracy of the prediction results, and also enhancing the robustness of the model when handling complex dynamic scenarios.

[0135] In some embodiments, the prediction unit further includes: a third acquisition module, a fourth acquisition module, a first prediction module, and a second determination module. The third acquisition module is used to acquire the historical actual processing time and historical predicted processing time of the task to be processed at a historical time. The historical predicted processing time is the processing time of the task to be processed predicted by the ensemble learning model based on historical static feature data at a historical time. The fourth acquisition module is used to acquire the operating parameters of the ensemble learning model. The operating parameters of the ensemble learning model include the number of training samples, the degree of bias of a single sample, and the number of decision trees. The degree of bias of a single sample represents the degree of deviation between the historical actual processing time and the historical predicted processing time. The first prediction module is used to construct the objective function of the ensemble learning model based at least on the historical actual processing time, the historical predicted processing time, and the operating parameters of the ensemble learning model. The second determination module is used to determine the ensemble learning model based on the objective function of the ensemble learning model and fuse the ensemble learning model and the time series analysis model to obtain the time prediction model. The model exhibits excellent robustness and adaptability, can quickly respond to environmental changes, reduce prediction errors, and improve prediction accuracy and real-time performance.

[0136] In some embodiments, the prediction unit includes a fifth acquisition module, a second prediction module, and a third determination module. The fifth acquisition module is used to acquire the operating parameters of the time series analysis model, which include a time step index, a hidden state of the time step, a hidden state of the previous time step, an output layer weight matrix, and an output layer bias term. The time step index represents a moment in the time series data, the hidden state of the time step represents the temporal feature information of the corresponding moment, and the hidden state of the previous time step represents the temporal feature information resulting from the transition of the moment corresponding to the time step. The second prediction module is used to construct the objective function of the time series analysis model based on the operating parameters of the time series analysis model. The third determination module is used to determine the time series analysis model based on the objective function of the time series analysis model, and fuse the ensemble learning model and the time series analysis model to obtain the time prediction model. Through real-time incremental learning and federated learning frameworks, the model maintains high flexibility and accuracy, significantly improving the timeliness and reliability of prediction.

[0137] In some embodiments, the prediction unit further includes: a sixth acquisition module, a third prediction module, a fourth prediction module, and a fourth determination module. The sixth acquisition module is used to acquire a first weight coefficient and a second weight coefficient. The third prediction module is used to input the static feature data into the ensemble learning model of the time prediction model to obtain a first predicted time of the task to be processed, and to input the dynamic feature data into the time series analysis model of the time prediction model to obtain a second predicted time of the task to be processed. The fourth prediction module is used to take the product of the first weight coefficient and the first predicted time as a first value, and the product of the second weight coefficient and the second predicted time as a second value. The fourth determination module is used to determine the sum of the first value and the second value as the predicted processing time of the task to be processed. This ensures that the prediction model can utilize the stable characteristics of historical data and respond to environmental changes in a timely manner.

[0138] In some embodiments, the sixth acquisition module further includes a seventh acquisition module, an eighth acquisition module, and a fifth determination module. The seventh acquisition module is used to acquire an activation function, which maps the first weight coefficient and the second weight coefficient within a preset range. The eighth acquisition module is used to acquire the deviation parameters of the ensemble learning model and the time series analysis model. The deviation parameters of the ensemble learning model characterize the degree of deviation between the model prediction value and the true value, and the deviation parameters of the time series analysis model characterize the degree of deviation between the model prediction value and the true value. The fifth determination module is used to determine the first weight coefficient and the second weight coefficient based at least on the activation function, the deviation parameters of the ensemble learning model, and the deviation parameters of the time series analysis model. The first weight coefficient is positively correlated with the deviation parameters of the time series analysis model, and the second weight coefficient is positively correlated with the deviation parameters of the ensemble learning model. This ensures the prediction accuracy of the model under different operating scenarios, especially in extreme cases such as equipment failure and a surge in pending tasks, enabling more accurate prediction of task processing time and improving equipment operating efficiency.

[0139] In some embodiments, the determining unit includes a ninth acquisition module and a fifth prediction module. The ninth acquisition module is used to acquire multiple calculation weights, which correspond one-to-one with the prediction processing time. The fifth prediction module is used to perform weighted summation on the M prediction processing times according to each prediction processing time and its corresponding calculation weight to obtain the target prediction time of the task to be processed. In a multi-task environment, this can significantly reduce prediction errors and achieve more efficient and intelligent resource scheduling.

[0140] The business processing time prediction device provided in this application embodiment, through multi-source data fusion, dynamic model architecture and real-time optimization mechanism, more accurately predicts the business processing time, solving the problem of poor prediction accuracy and robustness of business processing time in related technologies, which leads to poor customer experience. Thus, it achieves the technical effect of significantly improving prediction accuracy, significantly enhancing real-time response capability and multi-scenario adaptability, effectively reducing operating costs and improving customer satisfaction.

[0141] It should be noted that the first acquisition module, the first prediction module, and the first determination module mentioned above correspond to steps S201 to S203 in Embodiment 1. The three modules and their corresponding steps implement the same instances and application scenarios, but are not limited to the content disclosed in Embodiment 1. It should be noted that the above modules or units can be hardware or software components stored in memory (e.g., memory 104) and processed by one or more processors (e.g., processors 102a, 102b, ..., 102n). The above modules can also be part of a device and run in the computer terminal 10 provided in Embodiment 1.

[0142] Example 3

[0143] Embodiments of this application may provide an electronic device. Figure 4 This is a structural block diagram of an electronic device according to an embodiment of this application. Figure 4 As shown, the electronic device may include: one or more ( Figure 4 (Only one is shown) processor 1002, memory 1004, memory controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0144] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the methods and apparatus in the embodiments of this application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby implementing the above-described methods. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0145] Those skilled in the art will understand that Figure 4The structure shown is for illustrative purposes only. Electronic devices can also be smartphones, tablets, handheld computers, mobile internet devices (MIDs), PADs, and other terminal devices. Figure 4 This does not limit the structure of the aforementioned electronic device. For example, electronic devices may also include components that are more... Figure 4 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 4 The different configurations shown.

[0146] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0147] Example 4

[0148] Embodiments of this application also provide a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the business processing time prediction method provided in Embodiment 1.

[0149] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.

[0150] This application also provides a computer program product, which, when executed on a data processing device, is suitable for performing steps of a method for predicting business processing time.

[0151] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0152] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0153] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection between units or modules, and may be electrical or other forms.

[0154] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0155] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0156] If the aforementioned integrated units are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to related technologies, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0157] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for predicting business processing time, characterized in that, include: Acquire dynamic feature data and static feature data. The static feature data includes feature information of the task to be processed at the current time. The dynamic feature data includes processing device information and processing object information of the task to be processed. The processing device information includes the operation information of the processing device that processes the task to be processed. The processing object information includes the number of processing objects and the processing speed of the processing objects that process the task to be processed. The feature information of the task to be processed includes the number, type and priority of the task to be processed. A time prediction model is constructed based on an ensemble learning model and a time series analysis model. The dynamic feature data and the static feature data are input into the time prediction model to obtain the predicted processing time of the task to be processed. There are M time prediction models, and at least two of the time prediction models have different input dynamic feature data and static feature data. M is a positive integer. Based on the M predicted processing times, the target predicted time for the task to be processed is determined.

2. The method according to claim 1, characterized in that, Acquire dynamic feature data, including: Define the preset time window; According to the preset time window, N initial running datasets of the processing device are obtained. Each initial running dataset includes multiple initial running information of the processing device. Each initial running dataset is obtained by sampling at a single sampling moment within the preset time window. N is a positive integer. According to the preset time window, the data in the N initial running datasets are weighted and aggregated to obtain the processing device information of the task to be processed.

3. The method according to claim 1, characterized in that, A time prediction model is constructed based on an ensemble learning model and a time series analysis model, including: The historical actual processing time and historical predicted processing time of the task to be processed at a historical moment are obtained. The historical predicted processing time is the processing time of the task to be processed predicted by the ensemble learning model based on historical static feature data at a historical moment. Obtain the operating parameters of the ensemble learning model, which include the number of training samples, the degree of bias of a single sample, and the number of decision trees. The degree of bias of a single sample represents the degree of deviation between the historical actual processing time and the historical predicted processing time. The objective function of the ensemble learning model is constructed based at least on the historical actual processing time, the historical predicted processing time, and the running parameters of the ensemble learning model. The ensemble learning model is determined based on the objective function of the ensemble learning model, and the time series analysis model is fused with the ensemble learning model to obtain the time prediction model.

4. The method according to claim 1, characterized in that, A time prediction model is constructed based on an ensemble learning model and a time series analysis model, including: The running parameters of the time series analysis model are obtained. The running parameters of the time series analysis model include the time step index, the hidden state of the time step, the hidden state of the previous time step, the output layer weight matrix, and the output layer bias term. The time step index represents the time in the time series data, the hidden state of the time step represents the time series feature information of the corresponding time step, and the hidden state of the previous time step represents the time series feature information caused by the transition of the time corresponding to the time step. Construct the objective function of the time series analysis model based on the operating parameters of the time series analysis model; The time series analysis model is determined based on the objective function of the time series analysis model, and the time prediction model is obtained by fusing the ensemble learning model and the time series analysis model.

5. The method according to claim 1, characterized in that, The dynamic feature data and the static feature data are input into the time prediction model to obtain the predicted processing time of the task to be processed, including: Obtain the first weighting coefficient and the second weighting coefficient; The static feature data is input into the ensemble learning model of the time prediction model to obtain the first predicted time of the task to be processed, and the dynamic feature data is input into the time series analysis model of the time prediction model to obtain the second predicted time of the task to be processed. The product of the first weighting coefficient and the first prediction time is used as the first value, and the product of the second weighting coefficient and the second prediction time is used as the second value. The sum of the first value and the second value is determined as the predicted processing time of the task to be processed.

6. The method according to claim 5, characterized in that, To obtain the first and second weighting coefficients, the following steps are taken: Obtain an activation function, which is used to map the first weight coefficient and the second weight coefficient within a preset range; Obtain the bias parameters of the ensemble learning model and the bias parameters of the time series analysis model. The bias parameters of the ensemble learning model characterize the degree of deviation between the model predictions and the actual values ​​of the ensemble learning model, and the bias parameters of the time series analysis model characterize the degree of deviation between the model predictions and the actual values ​​of the time series analysis model. The first weight coefficient and the second weight coefficient are determined based at least on the activation function, the bias parameters of the ensemble learning model, and the bias parameters of the time series analysis model. The first weight coefficient is positively correlated with the bias parameters of the time series analysis model, and the second weight coefficient is positively correlated with the bias parameters of the ensemble learning model.

7. The method according to claim 1, characterized in that, Based on the M predicted processing times, the target predicted time for the task to be processed is determined, including: Multiple computational weights are obtained, and each computational weight corresponds one-to-one with the prediction processing time; Based on each of the predicted processing times and their corresponding calculation weights, the M predicted processing times are weighted and summed to obtain the target predicted time for the task to be processed.

8. A device for predicting business processing time, characterized in that, include: The acquisition unit is used to acquire dynamic feature data and static feature data. The static feature data includes feature information of the task to be processed at the current time. The dynamic feature data includes processing device information and processing object information of the task to be processed. The processing device information includes the operation information of the processing device that processes the task to be processed. The processing object information includes the number of processing objects and the processing speed of the processing objects that process the task to be processed. The feature information of the task to be processed includes the number, type and priority of the task to be processed. The prediction unit is used to construct a time prediction model based on the ensemble learning model and the time series analysis model, and input the dynamic feature data and the static feature data into the time prediction model to obtain the predicted processing time of the task to be processed. There are M time prediction models, and at least two of the time prediction models have different input dynamic feature data and static feature data, where M is a positive integer. A determining unit is used to determine the target prediction time of the task to be processed based on the M predicted processing times.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored executable program, wherein, when the executable program is executed, it controls the device on which the computer-readable storage medium is located to perform the business processing time prediction method according to any one of claims 1 to 7.

10. An electronic device, characterized in that, include: Memory, which stores executable programs; A processor for running the program, wherein the program, when running, executes the business processing time prediction method according to any one of claims 1 to 7.

Images