Queuing scheduling method and device, electronic equipment, medium and program product

Abstract

The invention provides a queuing scheduling method which can be applied to the technical field of big data and the technical field of artificial intelligence. The method comprises the following steps: acquiring service state data of a target website; the service state data is input into a queuing time prediction model, the predicted waiting time is obtained, the queuing time prediction model comprises a first feature extraction layer, a second feature extraction layer and a feature enhancement layer, the first feature extraction layer is used for extracting global context features of the service state data, and the second feature extraction layer is used for extracting global context features of the service state data; the second feature extraction layer is used for extracting time sequence dependent features of the global context features, and the feature enhancement layer is used for performing feature weighting on the time sequence dependent features; and generating a queuing scheduling suggestion based on the predicted waiting time.

Description

Technical Field

[0001] The present disclosure relates to the fields of big data technology and artificial intelligence technology, and more specifically to a queuing scheduling method, apparatus, device, medium, and program product. Background Art

[0002] In financial services, customers often face waiting in long queues when visiting branches. Uncertainty in waiting times not only reduces customer satisfaction but also impacts branch operational efficiency. To alleviate this issue, existing technologies have attempted to use time series forecasting methods to estimate customer waiting times, providing reference information for customers or enabling internal branch operations to optimize staff scheduling and service processes. However, existing forecasting methods are mostly based on traditional models, which present limitations when applied to branch queue data.

[0003] On the one hand, customer behavior at outlets is highly random, and queue data often contains missing values ​​or outliers. Traditional models are less robust to noisy data, prone to large fluctuations in prediction results and high errors. On the other hand, queue times are affected by a variety of factors, such as business type, number of windows, self-service device usage, staffing, and external weather. These variables exhibit dynamic trends over different time periods, making it difficult for traditional models to adapt to complex time-varying relationships given fixed input dimensions. Summary of the Invention

[0004] In view of the above problems, the present disclosure provides a queue scheduling method, apparatus, device, medium and program product.

[0005] According to a first aspect of the present disclosure, a queuing scheduling method is provided, the method comprising: acquiring business status data of a target outlet; inputting the business status data into a queuing time prediction model to obtain an estimated waiting time, wherein the queuing time prediction model comprises a first feature extraction layer, a second feature extraction layer, and a feature enhancement layer, the first feature extraction layer being used to extract global context features of the business status data, the second feature extraction layer being used to extract timing-dependent features of the global context features, the feature enhancement layer being used to perform feature weighting on the timing-dependent features; and generating a queuing scheduling recommendation based on the estimated waiting time.

[0006] According to an embodiment of the present disclosure, extracting the global context features of the business status data based on the first feature extraction layer specifically includes: constructing a first input sequence based on the business status data; obtaining auxiliary variables, and generating an extended input sequence based on the auxiliary variables and the first input sequence; and using the first feature extraction layer to perform a residual stacking operation based on dilated convolution on the extended input sequence to obtain the global context features.

[0007] According to an embodiment of the present disclosure, extracting the temporal dependency feature of the global context feature based on the second feature extraction layer specifically includes: performing forward and backward state propagation on the global context feature based on the second feature extraction layer to generate a forward hidden state sequence and a backward hidden state sequence respectively; and fusing the forward hidden state sequence and the backward hidden state sequence according to time steps to generate the temporal dependency feature.

[0008] According to an embodiment of the present disclosure, feature weighting is performed on the temporal dependent features based on the feature enhancement layer, specifically including: evaluating the importance of different time steps in the temporal dependent features based on the self-attention mechanism in the feature enhancement layer to generate attention weights; and weighted fusion of the attention weights with the temporal dependent features to generate a target feature vector.

[0009] According to an embodiment of the present disclosure, constructing the first input sequence based on the business status data specifically includes: according to the order of customer arrival at the target outlet, splicing the business status data corresponding to each N consecutive customers into a record to construct the first input sequence, where N is a preset value.

[0010] According to an embodiment of the present disclosure, the obtaining of auxiliary variables and the generation of an extended input sequence based on the auxiliary variables and the first input sequence specifically include: obtaining auxiliary variables related to the target outlet business, the auxiliary variables including at least one of the number of service windows, the number of self-service devices, or the number of outlet staff; and aligning the auxiliary variables according to time steps and splicing them to the first input sequence to generate the extended input sequence.

[0011] According to an embodiment of the present disclosure, the use of the first feature extraction layer to perform a residual stacking operation based on dilated convolution on the extended input sequence to obtain the global context feature specifically includes: setting a receptive field parameter in the first feature extraction layer; and based on a preset dilated convolution kernel size and dilation rate, using the receptive field parameter to perform a multi-level dilated convolution residual stacking operation on the extended input sequence to obtain the global context feature.

[0012] According to an embodiment of the present disclosure, generating a queue scheduling suggestion based on the expected waiting time specifically includes: comparing the expected waiting time with at least one preset waiting threshold, generating a queue level of the target network point based on the comparison result, and generating a queue scheduling suggestion based on the queue level, wherein the queue scheduling suggestion includes prompt waiting information and window adjustment instructions; or generating a queue scheduling suggestion based on the expected waiting time and the business status data using a scheduling decision model.

[0013] A second aspect of the present disclosure provides a queuing scheduling device, which includes: a data acquisition module, used to obtain business status data of a target network point; a queuing time prediction module, used to input the business status data into a queuing time prediction model to obtain an estimated waiting time, wherein the queuing time prediction model includes a first feature extraction layer, a second feature extraction layer and a feature enhancement layer, the first feature extraction layer is used to extract global context features of the business status data, the second feature extraction layer is used to extract timing dependency features of the global context features, and the feature enhancement layer is used to perform feature weighting on the timing dependency features; a scheduling suggestion module, used to generate a queuing scheduling suggestion based on the estimated waiting time.

[0014] According to an embodiment of the present disclosure, the queuing time prediction module can also be used to construct a first input sequence based on the business status data; obtain auxiliary variables, generate an extended input sequence based on the auxiliary variables and the first input sequence; and use the first feature extraction layer to perform a residual stacking operation based on dilated convolution on the extended input sequence to obtain the global context feature.

[0015] According to an embodiment of the present disclosure, the queuing time prediction module can also be used to perform forward and backward state propagation on the global context feature based on the second feature extraction layer to generate a forward hidden state sequence and a backward hidden state sequence respectively; and to fuse the forward hidden state sequence and the backward hidden state sequence according to the time step to generate the temporal dependency feature.

[0016] According to an embodiment of the present disclosure, the queuing time prediction module can also be used to evaluate the importance of different time steps in the timing-dependent features based on the self-attention mechanism in the feature enhancement layer to generate attention weights; and to weightedly fuse the attention weights with the timing-dependent features to generate a target feature vector.

[0017] According to an embodiment of the present disclosure, the queuing time prediction module can also be used to splice the business status data corresponding to each N consecutive customers into a record based on the order of customer arrival at the target outlet to construct the first input sequence, where N is a preset value.

[0018] According to an embodiment of the present disclosure, the queuing time prediction module can also be used to obtain auxiliary variables related to the target outlet business, wherein the auxiliary variables include at least one of the number of service windows, the number of self-service devices or the number of outlet staff; and align the auxiliary variables according to the time step and splice them into the first input sequence to generate the extended input sequence.

[0019] According to an embodiment of the present disclosure, the queuing time prediction module can also be used to set receptive field parameters in the first feature extraction layer; and based on a preset dilated convolution kernel size and dilation rate, use the receptive field parameters to perform multi-level dilated convolution residual stacking operations on the extended input sequence to obtain the global context feature.

[0020] According to an embodiment of the present disclosure, the scheduling suggestion module may also be configured to compare the estimated waiting time with at least one preset waiting threshold, generate a queue level for the target network point based on the comparison result, and generate a queue scheduling suggestion based on the queue level, wherein the queue scheduling suggestion includes waiting information and window adjustment instructions; or generate a queue scheduling suggestion using a scheduling decision model based on the estimated waiting time and the service status data. A third aspect of the present disclosure provides an electronic device comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the above method.

[0021] The fourth aspect of the present disclosure further provides a computer-readable storage medium having a computer program or instructions stored thereon, which implements the steps of the above method when the computer program or instructions are executed by a processor.

[0022] The fifth aspect of the present disclosure further provides a computer program product, comprising a computer program or instructions, which implement the steps of the above method when executed by a processor.

[0023] According to the embodiments of the present disclosure, by setting a feature extraction and enhancement layer with clear functions in the prediction model, it is possible to achieve modular allocation of tasks at different processing stages, which helps to improve the robust processing capabilities for missing, noisy or sudden values ​​in business data, and enhance the system's adaptability to the highly random nature of customer behavior. In addition, by uniformly inputting a variety of business status data into the model, it supports the processing capabilities of dynamically changing input features, and solves the problem that traditional models have difficulty modeling time-varying data under fixed input dimensions. In addition, queue scheduling suggestions are output in real time based on the estimated waiting time generated by the model, eliminating the need for manual analysis and rule reasoning, and can quickly respond to business changes, thereby improving the system's ability to handle sudden queue pressures and customer service continuity. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The above contents and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0025] Figure 1 Schematically illustrates an application scenario diagram of the queuing scheduling method, apparatus, device, medium, and program product according to an embodiment of the present disclosure;

[0026] Figure 2 The following schematically shows a flow chart of a queuing scheduling method according to an embodiment of the present disclosure;

[0027] Figure 3 Schematically shows a flow chart of a method for obtaining an estimated waiting time based on a queuing time prediction model according to an embodiment of the present disclosure;

[0028] Figure 4 The following schematically shows a neural network structure diagram based on queue time prediction according to an embodiment of the present disclosure;

[0029] Figure 5A The following schematically shows a structural block diagram of a queuing scheduling device according to an embodiment of the present disclosure;

[0030] Figure 5B A schematic diagram schematically illustrates a queuing scheduling system according to an embodiment of the present disclosure; and

[0031] Figure 6 A block diagram of an electronic device suitable for implementing a queuing scheduling method according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0032] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the detailed description below, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0033] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise," "include," etc. used herein indicate the presence of the features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0034] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0035] When expressions such as "at least one of A, B, and C, etc." are used, they should generally be interpreted in accordance with the meaning commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0036] First, the technical terms described in this article are explained and illustrated as follows.

[0037] Temporal Convolutional Network (TCN) is a neural network structure used to process time series data. It is based on one-dimensional convolution and combines dilated convolution with residual connection to capture dependencies over a long period of time.

[0038] Dilated Convolution: Dilated convolution inserts "holes" into a regular convolution operation, sampling the input sequence at intervals. This expands the receptive field while minimizing the number of parameters. For example, a 1D convolution with a dilation rate of 2 skips one element during convolution calculation. This method improves the model's ability to capture long-term dependencies while maintaining computational efficiency.

[0039] Residual connections are a structural design in deep networks. Their core idea is to add the input directly to the output, skipping some nonlinear transformations in intermediate layers. This mechanism can alleviate the vanishing and exploding gradient problems that often occur in deep neural networks, improving training stability and convergence speed.

[0040] Bidirectional Long Short-Term Memory (Bi-LSTM): An extended form of recurrent neural network architecture, Bi-LSTM combines two information paths: a forward LSTM (from the past to the present) and a backward LSTM (from the future to the present). It simultaneously considers historical information and future context at each time step, effectively improving the model's completeness and contextual representation capabilities in time series modeling. It is widely used in speech recognition, sequence labeling, and prediction tasks.

[0041] Forward state / backward state: In a Bi-LSTM, the forward state represents the hidden state accumulated from the start of the sequence to the current time step, reflecting the transfer of historical information; the backward state is the hidden state formed by moving backward from the end of the sequence, reflecting the impact of future context on the current time point. The fusion of the two helps capture symmetric time-dependent features.

[0042] Self-Attention: A neural network attention mechanism that allows the model to dynamically focus on information at other positions in the same sequence as it processes each position in the input sequence. By constructing a query-key-value vector and calculating a weighted average, self-attention can model the degree of correlation between different time steps in the sequence, thereby improving the model's ability to focus on key time segments.

[0043] Multi-head attention is a parallel attention mechanism that uses multiple attention heads to learn sequence dependencies in different subspaces. Each head can focus on different dependency patterns in the sequence, and the outputs of multiple heads are ultimately concatenated or summed, improving the model's ability to capture complex patterns.

[0044] The receptive field represents the range of what the output units of a layer in a neural network can "see" in the input space. In time series modeling, a larger receptive field indicates that the layer can integrate information from a longer time span.

[0045] In financial services, customers often face long queues when visiting branches, especially during peak hours or during specific service cycles. Uncertainty in queue times not only reduces customer experience and increases the risk of churn, but also poses challenges to branch resource allocation, scheduling, and overall service efficiency. To alleviate these issues, several technologies have been developed to predict and guide customer queue times based on time series models. This provides reference information for customers and informs decision-making for internal branch staffing and business process optimization.

[0046] However, existing forecasting methods, mostly based on traditional time series models, face significant limitations in real-world branch scenarios. Firstly, branch customer behavior is significantly impacted by sudden and sporadic factors, such as holiday effects, unexpected policy adjustments, and changes in surrounding traffic conditions. This results in highly random and non-stationary queue data. Traditional models are susceptible to outliers and noise when dealing with this type of data, resulting in unstable forecasts and large errors, making it difficult to meet banks' demand for high-precision and reliable forecasts.

[0047] On the other hand, queue times are influenced by a complex interplay of factors, including not only customer behavior (such as the type of service being processed, age group, and frequency of service), but also system configuration variables (such as the number of service windows enabled, availability of self-service devices, number of on-site staff, and service processing time). They are even influenced by environmental factors (such as weather, temperature, and air quality) and social factors (such as the approaching payroll period and the centralized payment cycle). These factors are not only numerous but also change frequently and have complex correlations. Traditional models are typically limited to processing fixed-length inputs, single time series, or simple regression relationships. They lack the ability to model multidimensional dynamic features and struggle to capture long-term dependencies and local variations in the data.

[0048] Furthermore, many current queue prediction systems utilize a centralized, cloud-based computing architecture, which presents challenges with real-time performance and availability. Network outages, server failures, or data transmission anomalies can render the prediction system inoperable, impacting real-time customer guidance and branch emergency response. This problem is particularly acute in remote locations or locations with unstable network coverage. The lack of edge computing and local fault tolerance also significantly reduces system reliability, making it impossible to provide effective services offline.

[0049] Based on this, an embodiment of the present disclosure provides a queue scheduling method, comprising: obtaining business status data of a target network point; inputting the business status data into a queue time prediction model to obtain an estimated wait time, wherein the queue time prediction model includes a first feature extraction layer, a second feature extraction layer, and a feature enhancement layer, wherein the first feature extraction layer is used to extract global context features of the business status data, the second feature extraction layer is used to extract temporal dependency features of the global context features, and the feature enhancement layer is used to perform feature weighting on the temporal dependency features; and generating a queue scheduling recommendation based on the estimated wait time. By providing a feature extraction and enhancement layer with clear functions in the prediction model, it is possible to achieve modular allocation of tasks at different processing stages, helping to improve the robustness of handling missing, noisy, or sudden values ​​in business data and enhancing the system's adaptability to the highly random nature of customer behavior. Furthermore, by uniformly inputting multiple types of business status data into the model, it supports the processing of dynamically changing input features, solving the problem that traditional models have difficulty modeling time-varying data under fixed input dimensions. In addition, queue scheduling suggestions are output in real time based on the estimated waiting time generated by the model, eliminating the need for manual analysis and rule reasoning, enabling rapid response to business changes and improving the system's ability to handle sudden queue pressures and customer service continuity.

[0050] It should be noted that the queue scheduling method, apparatus, device, medium, and program product identified in this disclosure can be used in the fields of big data technology, artificial intelligence technology, and financial technology, and can also be used in a variety of fields other than big data technology, artificial intelligence technology, and financial technology. The application fields of the queue scheduling method, apparatus, device, medium, and program product provided in the embodiments of this disclosure are not limited.

[0051] In the technical solutions disclosed herein, the user information (including but not limited to user personal information, user image information, user device information, such as location information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0052] In scenarios where personal information is used for automated decision-making, the methods, devices, and systems provided by the embodiments of the present disclosure all provide users with corresponding operation portals for them to choose to agree or reject the automated decision-making results; if the user chooses to reject, the expert decision-making process will be entered. The expression "automated decision-making" here refers to the activity of automatically analyzing and evaluating an individual's behavioral habits, interests and hobbies, or economic, health, credit status, etc. through computer programs and making decisions. The expression "expert decision-making" here refers to the activity of making decisions by people who specialize in a certain field, have specialized experience, knowledge, and skills, and have reached a certain level of professionalism.

[0053] Figure 1 The application scenario diagram of the queuing scheduling method, apparatus, device, medium and program product according to the embodiments of the present disclosure is schematically shown.

[0054] like Figure 1 As shown, the application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 is used as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links or optical fiber cables.

[0055] A user may use a first terminal device 101, a second terminal device 102, or a third terminal device 103 to interact with a server 105 via a network 104 to receive or send messages, etc. Various communication client applications may be installed on the first terminal device 101, the second terminal device 102, or the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (for example only).

[0056] The first terminal device 101 , the second terminal device 102 , and the third terminal device 103 may be various electronic devices having display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0057] The server 105 may be a server that provides various services, such as a background management server (for example only) that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103. The background management server may analyze and process received data such as user requests, and feed back processing results (e.g., web pages, information, or data obtained or generated based on user requests) to the terminal devices.

[0058] It should be noted that the queue scheduling method provided in the embodiment of the present disclosure can generally be executed by the server 105. Accordingly, the queue scheduling device provided in the embodiment of the present disclosure can generally be set in the server 105. The queue scheduling method provided in the embodiment of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105. Accordingly, the queue scheduling device provided in the embodiment of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105.

[0059] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is merely illustrative. Any number of terminal devices, networks and servers may be provided as required.

[0060] The following will be based on Figure 1 The scene described by Figures 2 to 4 The queuing scheduling method of the disclosed embodiment is described in detail.

[0061] Figure 2 The flowchart of the queuing scheduling method according to an embodiment of the present disclosure is schematically shown.

[0062] like Figure 2 As shown, the queue scheduling method 200 of this embodiment includes operations S210 to S230, and the queue scheduling method can be executed by the server 105.

[0063] In operation S210, business status data of a target network point is obtained. The business status data may include indicators of multiple dimensions of the target network point.

[0064] For example, business status data may include queuing data indicators, such as the current number of customers in the queue, customer arrival rate per minute, average customer waiting time, longest waiting time for customers in the queue, number-taking category distribution, etc. Queuing data indicators can be extracted in real time from the number-taking system.

[0065] For example, business status data may include business processing indicators, including the type of business currently being processed, the average processing time for different types of business, and the number of customers of each business category waiting. Business processing indicators can be obtained through ATMs, business system logs or business process management platforms.

[0066] To improve the system's granularity of business dynamics, business status data collection can be implemented using high-frequency polling or event-triggered mechanisms, ensuring real-time updates within seconds. This collection can be done through local API calls or by supporting simultaneous extraction from heterogeneous systems through various mechanisms, such as database polling.

[0067] In the disclosed embodiments, multi-source data can also be subjected to field standardization, time series alignment, and missing value processing to form a unified, structured input data format, ensuring the stability of subsequent model processing. Partially missing or abnormal fields can be corrected using pre-set strategies (such as default value filling, sliding mean smoothing, and historical mean substitution), thereby enhancing robustness to abnormal data.

[0068] In operation S220, the service status data is input into the queue time prediction model to obtain the estimated waiting time. The queue time prediction model can adopt a multi-layer deep neural network structure, for example, it can include a first feature extraction layer, a second feature extraction layer, and a feature enhancement layer. The first feature extraction layer is used to extract the global context features of the service status data, the second feature extraction layer is used to extract the time-dependent features of the global context features, and the feature enhancement layer is used to perform feature weighting on the time-dependent features. The multiple layers can be connected in the form of data streams to jointly complete the feature extraction, time-series modeling and dynamic weight adjustment of the service status data, thereby improving the prediction accuracy and adaptability of the queue time.

[0069] In some embodiments, the first feature extraction layer can be constructed based on a temporal convolutional network structure. By setting up multiple layers of dilated convolution kernels, this structure can model input sequences with long time spans while using a relatively low number of parameters, thereby effectively extracting long-term trends in business status data. For example, at outlets where customer visits show significant daily or weekly fluctuations, the first feature extraction layer constructed based on the temporal convolutional network structure can automatically identify and utilize this periodic information to predict the trend of queue pressure at subsequent time points.

[0070] In some embodiments, the first feature extraction layer can employ a stacked multi-head convolutional architecture, with each convolutional branch operating on a time window of varying scale, to concurrently extract local peak fluctuation characteristics and overall traffic trends. This architecture is particularly suitable for outlets experiencing sudden peaks in traffic, such as at the end of the month, during bill payment periods, or on special event days.

[0071] In some embodiments, the first feature extraction layer not only processes business status data, but also integrates external auxiliary variables such as weather, traffic, holidays, etc., and jointly models them with the original time series features through the embedding layer, thereby constructing a richer contextual feature representation and improving the model's perception of uncontrollable external influencing factors.

[0072] In some embodiments, the second feature extraction layer can be implemented using a bidirectional long short-term memory network, which can simultaneously model both forward and backward information flows in the input sequence and extract the contextual dependency structure between adjacent time points. This structure can provide more accurate temporal dependency modeling in situations where customer wait times are significantly influenced by the duration of preceding transactions.

[0073] In some embodiments, the second feature extraction layer is constructed based on the gated recurrent unit, taking advantage of its fewer parameters and higher training efficiency. It is suitable for business outlets with higher response speed requirements, especially in edge deployment scenarios, and can provide lower latency and higher processing throughput.

[0074] In some embodiments, the second feature extraction layer may adopt a hierarchical nested structure, for example, setting up multiple stacked bidirectional recurrent units, each layer encoding time features at different levels of abstraction, so that the model can learn the differences between short-term fluctuations and long-term dependencies separately, and enhance the hierarchical modeling capabilities of complex time series data.

[0075] In some embodiments, the feature enhancement layer can be built based on a self-attention mechanism to automatically assess the importance of each time step in a time series feature, learn the weight distribution, and perform weighted processing on the feature vectors. This structure can enhance focus on critical moments during peak business hours and suppress redundant effects during periods of low volatility.

[0076] In some embodiments, the feature enhancement layer can combine the channel attention mechanism and the time attention mechanism to perform double weighting on the feature dimension and the time dimension respectively, thereby achieving simultaneous focus on key indicator dimensions and key time periods, and improving the sensitivity and predictive discrimination of the model.

[0077] In some embodiments, the feature enhancement layer can be constructed using a lightweight attention mechanism, for example, limiting the attention calculation process to sliding aggregation within a local window range to improve computing efficiency, which is suitable for embedded devices or business scenarios with strong lightweight deployment requirements.

[0078] It should be noted that the above embodiments of the first feature extraction layer, the second feature extraction layer, and the feature enhancement layer are merely exemplary descriptions of optional structures of the present disclosure and do not constitute a limitation on the scope of protection of the present disclosure. Those skilled in the art may adjust, replace, or combine the model structures based on the needs of specific business scenarios, computing resource limitations, or target accuracy requirements.

[0079] In operation S230, a queue scheduling suggestion is generated based on the estimated waiting time. The generation process of the queue scheduling suggestion can be flexibly designed according to actual application requirements and system deployment strategies. It can adopt a rule-based policy engine for static judgment or combine machine learning algorithms to optimize historical scheduling results through feedback.

[0080] For example, the system can compare the current customer's estimated wait time with a preset wait threshold to determine whether it exceeds an acceptable range. If the estimated wait time exceeds the threshold, the system can trigger a scheduling suggestion generation module, outputting suggestions including but not limited to: opening additional service windows, prompting customers to use self-service devices, diverting customers to other branches, or prioritizing certain types of services.

[0081] For example, after obtaining the estimated waiting time, the system can further combine it with the current business status data and use a preset scheduling decision model to generate queue scheduling recommendations. The scheduling decision model can be built based on a logical judgment tree, a reinforcement learning model, or other models with decision-making capabilities, and is used to dynamically generate scheduling strategies based on the overall operation of the current branch. When performing this operation, the system can use the key operating indicators of the current branch as model inputs, such as the number of queues and average waiting times for various business types, the number of service windows and their enabled status, the utilization rate of self-service equipment, employee scheduling and load conditions, etc. The scheduling decision model comprehensively analyzes the input information, combined with the estimated customer waiting time for the corresponding time period, to determine whether the queue pressure exceeds the warning threshold, and automatically generates scheduling recommendations accordingly. For example, during business peak hours, when the queuing time prediction model determines that there will be a long waiting time in the next 15 minutes and only some windows are currently enabled, the system can output a scheduling plan of "recommending the temporary opening of one general window" or "switching the designated window to a dedicated window for high-frequency business"; if the queuing pressure is mainly concentrated on a specific type of business, the system can also suggest guiding some customers to switch to self-service equipment through on-site guidance or terminal prompts, thereby realizing automatic diversion of part of the traffic.

[0082] The following describes preferred embodiments of the present disclosure. It should be noted that the following examples do not limit the present disclosure.

[0083] Figure 3 The flowchart of the method for obtaining the estimated waiting time based on the queuing time prediction model according to an embodiment of the present disclosure is schematically shown. Figure 4 The figure schematically shows a neural network structure diagram based on queue time prediction according to an embodiment of the present disclosure.

[0084] like Figure 3 As shown, the method for obtaining the estimated waiting time based on the queuing time prediction model in this embodiment includes operations S310 to S330.

[0085] In operation S310, the service status data of the target network point is obtained.

[0086] In operation S320, the service status data is input into a queuing time prediction model to obtain an estimated waiting time.

[0087] According to the embodiments of the present disclosure, different types of feature extraction mechanisms can be combined to improve the accuracy and stability of queue time prediction from three dimensions: global trend modeling, time dependency capture, and key information focus.

[0088] In the embodiment of the present disclosure, in order to achieve high-precision modeling of queuing time, a first input sequence may be constructed based on service status data.

[0089] According to the embodiments of the present disclosure, considering that the amount of data that can be collected by a single outlet in actual operation is relatively limited, and that customers' queuing behavior usually exhibits strong temporal continuity and adjacent dependence, in order to fully exploit this local sequence feature, the system adopts an input method based on adjacent customer combinations to construct a first input sequence.

[0090] For example, based on the order in which customers arrive at the target branch, the business status data corresponding to each N consecutive customers is concatenated into a record to construct the first input sequence, where N is a preset value. The value of N can be flexibly set based on the actual business scenario. For example, when customer visits are frequent and the processing time for individual transactions is short, a smaller N value can be selected to increase sensitivity to fine-grained fluctuations. When branch business exhibits significant cyclicality or customer behavior changes relatively steadily, N can be appropriately increased to incorporate customer behavior characteristics over a wider timeframe, improving the ability to model overall trends.

[0091] Specifically, the business status data of each customer can be arranged in the order of their arrival and processed by the system in a sliding window manner. That is, starting from the first customer, each time a customer is moved backward, the data of N consecutive customers is taken and spliced ​​into a new input record until a complete input sequence set covering the entire queue is constructed.

[0092] Preferably, the service status data of three consecutive customers can be concatenated into a sample record, following the chronological order of their arrival, and used as the first input sequence unit of the model. Each sample thus consists of the service characteristics of three groups of adjacent customers, collectively describing the local dynamic characteristics of queuing behavior over a short period of time.

[0093] Furthermore, auxiliary variables related to the network operation status can be obtained to supplement the description of the current business capabilities and service resource configuration.

[0094] For example, auxiliary variables can include service resource indicators, such as the total number of windows, the number of currently open windows, the type of business handled by each window, the number and usage rate of self-service devices, whether the smart service terminal is online, the number of back-end employees, and other resource configuration conditions, which can be obtained simultaneously through the branch scheduling system and the equipment status monitoring platform.

[0095] For example, auxiliary variables can include environmental context indicators, such as weather conditions, holiday information, traffic flow in nearby areas, etc., and can also be introduced through external service interfaces to enhance the model's perception of sudden or special scenarios.

[0096] For example, statistical auxiliary variables generated based on historical queuing behavior data can be introduced, such as the average queue length in a certain period of time, the incremental change in queues in the previous time window, the volatility of business processing time, etc., to enhance the model's perception of time series stability and trend changes.

[0097] In order to improve the expressiveness of the model, the auxiliary variables can be aligned with the first input sequence by time step and spliced ​​to generate an extended input sequence. This ensures that the data at each time step not only contains the behavioral characteristics of the client side, but also integrates the support capability indicators of the server side.

[0098] Combined with reference Figure 3 and Figure 4 , a Temporal Convolutional Network (TCN) layer can be used to perform feature extraction on the extended input sequence. The TCN layer can be built based on the residual stacking structure of dilated convolution to enhance the ability to model the temporal dependencies of long sequences.

[0099] Specifically, the receptive field parameter can be set within the TCN layer to control the range of historical information the model can perceive in the temporal dimension. A larger receptive field allows the queue time prediction model to incorporate more contextual information, making it more effective at identifying trends, cyclical fluctuations, and temporal shifts in queue behavior. The receptive field is determined by parameters such as the number of layers, kernel size, and dilation rate of the multi-level dilated convolutional layer.

[0100] Furthermore, multi-level dilated convolution can be performed on the expanded input sequence based on a preset dilated convolution kernel size (for example, 3) and a layer-by-layer increasing dilation rate (for example, 1, 2, 4, 8, etc.). Dilated convolution introduces holes between the convolution kernels, allowing each layer to expand its temporal perception range while maintaining the length of the input sequence. This allows for a larger receptive field without increasing the number of convolution layers. This multi-level dilated convolution layer structure is suitable for tasks with limited data volume but strong temporal dependence, such as network queuing data.

[0101] To avoid information attenuation and vanishing gradients that can occur in deep networks, a residual connection mechanism can be introduced after each dilated convolutional layer. This adds the input and output of the TCN layer and feeds them into the next convolutional module, thereby enhancing the model's information flow stability and training efficiency. By stacking multiple levels of residual convolutional blocks, the system can extract features at different time scales and integrate the outputs of multiple layers to form a more powerful global contextual feature representation.

[0102] According to the embodiments of the present disclosure, the global context feature sequence output by the first feature extraction layer not only includes the overall trend of customer behavior in the current input window, but also integrates the long-term impact of service resources and external states, providing a stable input basis with semantic depth for the subsequent temporal dependency modeling layer.

[0103] In an embodiment of the present disclosure, the second feature extraction layer is used to perform further time series modeling processing on the global context features output by the first feature extraction layer to mine dynamic dependencies in the time series.

[0104] Combined with reference Figure 3 and Figure 4 The second feature extraction layer can be constructed using a bidirectional long short-term memory network (Bi-directional Long Short-Term Memory, Bi-LSTM). This network structure has the ability to simultaneously capture forward and backward sequence information, and is suitable for modeling the causal relationship and contextual features across time steps in queuing behavior.

[0105] Specifically, the Bi-LSTM layer can include two channels: a forward recurrent unit and a backward recurrent unit. The forward recurrent unit gradually receives the global context feature sequence in chronological order, capturing the influence of the current time step and its historical features. The backward recurrent unit, on the other hand, processes the same input sequence in reverse order, looking back from future time steps to extract subsequent feature information that may affect the current state. This bidirectional structure enables the model to simultaneously utilize the bidirectional semantic information of its context at any point in time, thereby improving the modeling ability of nonlinear dynamic features such as sudden queuing events and window changes.

[0106] After forward and backward propagation are completed, the forward and backward hidden state sequences can be fused on a time-step basis. This fusion can be performed using methods such as time-step concatenation, weighted summation, or element-wise addition. For example, element-wise addition can sum the forward and backward hidden state vectors at the same time step to generate a unified temporal dependency feature representation. This representation not only incorporates the context of the business environment at the current time point but also encodes the fluctuation trends before and after through bidirectional modeling, helping subsequent modules accurately assess customer queue wait times at the current moment.

[0107] By introducing the Bi-LSTM layer, the system can more comprehensively understand the dynamic evolution characteristics of the queuing system when faced with customer behavior sequences with obvious contextual dependencies. In particular, it can maintain stable modeling capabilities in scenarios where customer arrival rhythms are uneven or business processing times fluctuate drastically, thereby effectively improving the timeliness and accuracy of the entire queuing prediction system.

[0108] In an embodiment of the present disclosure, in order to further improve the model's responsiveness to key time points and core feature dimensions, the system introduces a feature enhancement layer based on the temporal dependency features output by the second feature extraction layer.

[0109] Combined with reference Figure 3 and Figure 4 , which can be constructed using a self-attention mechanism. The self-attention mechanism layer evaluates the importance of different time steps in the time-dependent features, generates attention weights, and then weightedly fuses these attention weights with the time-dependent features to achieve dynamic and adjustable feature enhancement, thereby improving the robustness and sensitivity of queue time prediction in complex business scenarios.

[0110] Specifically, temporal dependency features include a sequence of feature vectors that fuse the forward and backward states at each time step. These serve as the input to the self-attention layer. Through a query-key-value algorithm, the attention distribution of each time step across the entire sequence is constructed. After calculating the dependency of each time step on other time steps, the queue time prediction model generates a set of attention weight matrices, which are then weighted and summed across the features in the original sequence, ultimately outputting a set of enhanced representation vectors.

[0111] Through these operations, the system automatically learns which historical time points have the greatest impact on current queuing trends. For example, in certain scenarios, large-scale customer visits within the previous few minutes, or a sudden surge in queues due to equipment failure, will have a lagged impact on subsequent time points. The self-attention mechanism can perceive these delayed correlations and increase the feature weights of these time steps in the output. Conversely, for time periods with less business fluctuations or stable service processes, the feature importance of these time steps will be appropriately suppressed, thereby preventing redundant information from interfering with the final prediction.

[0112] According to the embodiments of this disclosure, the enhanced feature vectors processed by the self-attention mechanism serve as the model's output features for generating the final queue time prediction results, demonstrating enhanced temporal semantics selection and deeper feature expression. This structure is particularly well-suited for real-world network environments characterized by highly nonlinear customer behavior, short-term peak traffic concentrations, and dynamically fluctuating service resources, effectively improving the responsiveness and stability of queue predictions.

[0113] In operation S330 , a queue scheduling suggestion is generated based on the estimated waiting time.

[0114] According to the embodiments of the present disclosure, a three-layer queue time prediction model is constructed. The first feature extraction layer (TCN) effectively models the long-term temporal dependence of customer behavior. The second feature extraction layer (Bi-LSTM) further extracts the contextual relationship between previous and subsequent states. The self-attention mechanism enhances the model's sensitivity to key time points, thereby maintaining stable and highly accurate queue time prediction results even in scenarios with large fluctuations in customer arrivals and diverse business types. Furthermore, the embodiments of the present disclosure not only utilize customer behavior data (such as the number of queue members and business types), but also incorporate auxiliary variables such as the number of windows and self-service device usage rates, and perform unified modeling by expanding the input sequence. This approach enhances the model's depth of perception of actual operating conditions, enabling it to generalize modeling across different outlets, time periods, and business configurations, avoiding the performance degradation of traditional models due to fixed input structures.

[0115] Furthermore, the queue time prediction model's architecture balances inference efficiency and accuracy, making it suitable for deployment on financial institutions' internal servers or edge computing devices. Even in scenarios where network outages and cloud services are inaccessible, queue time predictions and scheduling recommendations can be independently completed, improving the system's continuity and fault tolerance.

[0116] Corresponding to the above-mentioned queue scheduling method, an embodiment of the present disclosure further provides a queue scheduling device and a queue scheduling system.

[0117] Figure 5A The structural block diagram of the queuing scheduling device according to an embodiment of the present disclosure is schematically shown.

[0118] like Figure 5A As shown, the queuing scheduling device 500 of this embodiment includes a data acquisition module 510 , a queuing time prediction module 520 , and a scheduling suggestion module 530 .

[0119] The data acquisition module 510 may be used to acquire the service status data of the target network point. In one embodiment, the data acquisition module 510 may be used to perform the operation S210 described above, which will not be described in detail here.

[0120] The queuing time prediction module 520 can be used to input the service status data into a queuing time prediction model to obtain an estimated waiting time, wherein the queuing time prediction model includes a first feature extraction layer, a second feature extraction layer, and a feature enhancement layer, wherein the first feature extraction layer is used to extract global context features of the service status data, the second feature extraction layer is used to extract temporal dependency features of the global context features, and the feature enhancement layer is used to perform feature weighting on the temporal dependency features. In one embodiment, the queuing time prediction module 520 can be used to perform operation S220 described above, which will not be repeated here.

[0121] The scheduling suggestion module 530 may be used to generate a queue scheduling suggestion based on the estimated waiting time. In one embodiment, the scheduling suggestion module 530 may be used to perform the operation S230 described above, which will not be described in detail here.

[0122] According to an embodiment of the present disclosure, the queuing time prediction module 520 can also be used to construct a first input sequence based on the business status data; obtain auxiliary variables, generate an extended input sequence based on the auxiliary variables and the first input sequence; and use the first feature extraction layer to perform a residual stacking operation based on dilated convolution on the extended input sequence to obtain the global context feature.

[0123] According to an embodiment of the present disclosure, the queuing time prediction module 520 can also be used to perform forward and backward state propagation on the global context features based on the second feature extraction layer to generate a forward hidden state sequence and a backward hidden state sequence respectively; and to fuse the forward hidden state sequence and the backward hidden state sequence according to the time step to generate the temporal dependency feature.

[0124] According to an embodiment of the present disclosure, the queuing time prediction module 520 can also be used to evaluate the importance of different time steps in the timing-dependent features based on the self-attention mechanism in the feature enhancement layer to generate attention weights; and to perform weighted fusion of the attention weights with the timing-dependent features to generate a target feature vector.

[0125] According to an embodiment of the present disclosure, the queuing time prediction module 520 can also be used to concatenate the business status data corresponding to each N consecutive customers into a record based on the order of customer arrival at the target outlet to construct the first input sequence, where N is a preset value.

[0126] According to an embodiment of the present disclosure, the queuing time prediction module 520 can also be used to obtain auxiliary variables related to the business of the target outlet, wherein the auxiliary variables include at least one of the number of service windows, the number of self-service devices or the number of outlet staff; and align the auxiliary variables by time step and splice them into the first input sequence to generate the extended input sequence.

[0127] According to an embodiment of the present disclosure, the queuing time prediction module 520 can also be used to set receptive field parameters in the first feature extraction layer; and based on a preset dilated convolution kernel size and dilation rate, use the receptive field parameters to perform multi-level dilated convolution residual stacking operations on the extended input sequence to obtain the global context feature.

[0128] According to an embodiment of the present disclosure, the scheduling recommendation module 530 can also be used to compare the expected waiting time with at least one preset waiting threshold, generate a queuing level of the target network point based on the comparison result, and generate a queuing scheduling recommendation based on the queuing level, wherein the queuing scheduling recommendation includes prompt waiting information and window adjustment instructions; or based on the expected waiting time and the business status data, generate a queuing scheduling recommendation using a scheduling decision model.

[0129] According to embodiments of the present disclosure, any multiple modules among the data acquisition module 510, the queue time prediction module 520, and the scheduling suggestion module 530 can be combined into a single module, or any one of these modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in a single module. According to embodiments of the present disclosure, at least one of the data acquisition module 510, the queue time prediction module 520, and the scheduling suggestion module 530 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented in hardware or firmware through any other reasonable means of circuit integration or packaging, or can be implemented in any one of the three implementation methods of software, hardware, and firmware, or any appropriate combination of any of these. Alternatively, at least one of the data acquisition module 510, the queue time prediction module 520, and the scheduling suggestion module 530 can be at least partially implemented as a computer program module that, when executed, can perform the corresponding functionality.

[0130] Figure 5B A schematic diagram of a queuing scheduling system according to an embodiment of the present disclosure is schematically shown.

[0131] like Figure 5B As shown, the queue scheduling system 700 can be deployed as a standalone server or edge computing node, located locally at a network point or integrated into a regional management platform, to perform inference tasks for the queue time prediction model. The queue scheduling system 700 includes computing resources for model execution, such as a graphics processing unit (GPU), a central processing unit (CPU), and memory (RAM), providing the hardware foundation for deploying and efficiently running deep learning models. The queue scheduling system 700 also includes a front-end collection and statistics module 710 for queue time prediction, and a back-end processing module 720 for outputting results. This device allows for the embedding of multi-layer predictive network models, enabling dynamic prediction of queue times based on incoming real-time data.

[0132] The front-end data collection and statistics module 710 includes a branch ticket machine 711 installed at each branch and a counter service completion time statistics unit 712. The queue scheduling system 700 interfaces with the branch ticket machine 711 and the counter service processing system to enable real-time data collection of service status. After a customer enters their service request information at the branch ticket machine 711, this information, along with the service completion time recorded by the counter system, is sent to the back-end processing module 720, which constructs a first input sequence and triggers the execution of the queue time prediction model.

[0133] To further enhance the accuracy and adaptability of the queue time prediction model, in addition to the basic front-end data collection and statistics module 710, the queue scheduling system 700 may also include functional components such as an image acquisition module and an external interface module. Image recognition technology can be used to collect real-time statistics on visual features such as the number of customers at the current branch and the degree of congestion in the waiting area. Furthermore, by integrating external environmental variables such as weather information, road conditions, and holiday information, the queue time prediction model's ability to perceive customer waiting behavior can be enhanced, leading to more accurate estimates of queue times. This collected information can not only be used for online inference of the queue time prediction model, but also serve as continuous training data for the queue time prediction model, achieving dynamic improvement and continuous iteration of system performance.

[0134] After data collection is complete, it can be sent to the backend processing module 720 via a data interface or other means. The backend processing module 720 can include a queue time prediction model and a result output module. The queue time prediction model integrates the customer's business information input with the branch's operational status to predict the current customer's expected wait time. The prediction results can be displayed or distributed through the result output module and can be used to display the current queue time on the number collection terminal, client application, or branch scheduling system, assisting customers in making decisions or supporting operations personnel in adjusting scheduling strategies.

[0135] Through the deployment structure of the queue scheduling system, this disclosure enables real-time queue time predictions triggered by customer visits, enhancing the automation and intelligence of queue management. The queue scheduling system also supports modular decoupled deployment, offering excellent scalability and adaptability, allowing for flexible deployment in the operational environments of different types of outlets.

[0136] Figure 6 A block diagram of an electronic device suitable for implementing a queuing scheduling method according to an embodiment of the present disclosure is schematically shown.

[0137] like Figure 6As shown, an electronic device 600 according to an embodiment of the present disclosure includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage unit 606 into a random access memory (RAM) 603. The processor 601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0138] Various programs and data required for the operation of the electronic device 600 are stored in the RAM 603. The processor 601, ROM 602, and RAM 603 are connected to each other via a bus 604. The processor 601 executes the various operations of the method flow according to the embodiment of the present disclosure by executing the programs in the ROM 602 and / or RAM 603. It should be noted that the programs may also be stored in one or more memories other than the ROM 602 and RAM 603. The processor 601 may also execute the various operations of the method flow according to the embodiment of the present disclosure by executing the programs stored in the one or more memories.

[0139] According to an embodiment of the present disclosure, electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to bus 604. Electronic device 600 may also include one or more of the following components connected to I / O interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including devices such as a cathode ray tube (CRT), liquid crystal display (LCD), and speakers; a storage section 608 including a hard disk; and a communication section 609 including a network interface card such as a LAN card or modem. Communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to I / O interface 605 as needed. Removable media 611, such as a magnetic disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed in drive 610 as needed, so that computer programs read from the removable media can be installed into storage section 608 as needed.

[0140] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments, or may exist independently and not be incorporated into the device / apparatus / system. The computer-readable storage medium carries one or more programs, and when executed, implements the method according to the embodiments of the present disclosure.

[0141] According to an embodiment of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, and may include, for example, but not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to an embodiment of the present disclosure, a computer-readable storage medium may include the ROM 602 and / or RAM 603 described above, and / or one or more memories other than ROM 602 and RAM 603.

[0142] The embodiments of the present disclosure also include a computer program product, which includes a computer program containing program code for executing the method shown in the flowchart. When the computer program product is run in a computer system, the program code is used to enable the computer system to implement the queue scheduling method provided by the embodiments of the present disclosure.

[0143] The computer program executes the above functions defined in the system / device of the embodiment of the present disclosure when the processor 601 executes the computer program. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0144] In one embodiment, the computer program may be stored on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may be transmitted and distributed in the form of a signal on a network medium, downloaded and installed via the communication portion 609, and / or installed from a removable medium 611. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0145] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 609, and / or installed from a removable medium 611. When the computer program is executed by the processor 601, the above-described functions defined in the system of the embodiment of the present disclosure are performed. According to the embodiment of the present disclosure, the systems, devices, means, modules, units, etc. described above can be implemented by computer program modules.

[0146] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computer programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C" or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect via the Internet).

[0147] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the above-mentioned module, program segment, or a part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0148] Those skilled in the art will appreciate that the features described in the various embodiments of the present disclosure may be combined and / or coupled in various ways, even if such combinations or couplings are not explicitly described in the present disclosure. In particular, the features described in the various embodiments of the present disclosure may be combined and / or coupled in various ways without departing from the spirit and teachings of the present disclosure. All such combinations and / or couplings fall within the scope of the present disclosure.

[0149] The above describes the embodiments of the present disclosure. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Although each embodiment has been described separately above, this does not mean that the measures in each embodiment cannot be advantageously used in combination. Without departing from the scope of the present disclosure, those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A queuing scheduling method, characterized in that: The method comprises: Obtain business status data of target outlets; Inputting the service status data into a queuing time prediction model to obtain an estimated waiting time, wherein the queuing time prediction model includes a first feature extraction layer, a second feature extraction layer, and a feature enhancement layer, wherein the first feature extraction layer is used to extract global context features of the service status data, the second feature extraction layer is used to extract temporal dependency features of the global context features, and the feature enhancement layer is used to perform feature weighting on the temporal dependency features; and Based on the expected waiting time, a queue scheduling suggestion is generated.

2. The method according to claim 1, characterized in that Extracting the global context features of the service status data based on the first feature extraction layer specifically includes: constructing a first input sequence based on the business status data; obtaining auxiliary variables, and generating an extended input sequence based on the auxiliary variables and the first input sequence; and The first feature extraction layer is used to perform a residual stacking operation based on dilated convolution on the extended input sequence to obtain the global context feature.

3. The method according to claim 1 or 2, characterized in that Extracting the temporal dependency feature of the global context feature based on the second feature extraction layer specifically includes: Performing forward and backward state propagation on the global context features based on the second feature extraction layer to generate a forward hidden state sequence and a backward hidden state sequence, respectively; and The forward hidden state sequence and the backward hidden state sequence are fused according to time steps to generate the temporal dependency feature.

4. The method according to claim 3, characterized in that Performing feature weighting on the temporal dependency feature based on the feature enhancement layer specifically includes: Based on the self-attention mechanism in the feature enhancement layer, evaluating the importance of different time steps in the temporal dependent features and generating attention weights; and The attention weight is weightedly fused with the temporal dependency feature to generate a target feature vector.

5. The method according to claim 2, characterized in that The constructing of the first input sequence based on the service status data specifically includes: According to the order in which customers arrive at the target outlet, the business status data corresponding to each N consecutive customers are spliced ​​into a record to construct the first input sequence, where N is a preset value.

6. The method according to claim 2 or 5, characterized in that The acquiring of auxiliary variables and generating an extended input sequence based on the auxiliary variables and the first input sequence specifically includes: Acquiring auxiliary variables related to the target branch's business, the auxiliary variables including at least one of the number of service windows, the number of self-service devices, or the number of branch staff; and The auxiliary variables are aligned according to time steps and then spliced ​​to the first input sequence to generate the extended input sequence.

7. The method according to claim 2 or 5, characterized in that The step of using the first feature extraction layer to perform a dilated convolution-based residual stacking operation on the extended input sequence to obtain the global context feature specifically includes: Setting receptive field parameters in the first feature extraction layer; and Based on a preset dilated convolution kernel size and dilation rate, the receptive field parameters are used to perform a multi-level dilated convolution residual stacking operation on the extended input sequence to obtain the global context feature.

8. The method according to claim 1, characterized in that Generating a queue scheduling suggestion based on the estimated waiting time specifically includes: Comparing the estimated waiting time with at least one preset waiting threshold, generating a queue level of the target network point based on the comparison result, and generating a queue scheduling suggestion according to the queue level, wherein the queue scheduling suggestion includes a waiting prompt information and a window adjustment instruction; or Based on the expected waiting time and the service status data, a queuing scheduling suggestion is generated using a scheduling decision model.

9. A queuing scheduling device, characterized in that: The device comprises: The data acquisition module is used to: acquire the business status data of the target outlet; A queuing time prediction module is configured to input the service status data into a queuing time prediction model to obtain an estimated waiting time, wherein the queuing time prediction model includes a first feature extraction layer, a second feature extraction layer, and a feature enhancement layer, wherein the first feature extraction layer is configured to extract global context features of the service status data, the second feature extraction layer is configured to extract temporal dependency features of the global context features, and the feature enhancement layer is configured to perform feature weighting on the temporal dependency features; The scheduling suggestion module is used to generate a queue scheduling suggestion based on the expected waiting time.

10. An electronic device comprising: one or more processors; a memory for storing one or more computer programs, It is characterized in that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program or instruction stored thereon, characterized in that: When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

12. A computer program product comprising a computer program or instructions, characterized in that When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images