Web site traffic volume prediction method, device, equipment, medium and program product

Abstract

The disclosure provides a network point passenger flow prediction method, which can be applied to the fields of artificial intelligence and financial technology. The method comprises the following steps: establishing a two-dimensional image matrix of a to-be-tested network point area, the to-be-tested network point area comprising a plurality of network points, and the two-dimensional image matrix comprising spatial features and historical passenger flow features of the plurality of network points; arranging the two-dimensional image matrix in time sequence to obtain a space-time array data set; and training the space-time array data set by using a pre-constructed network point flow prediction model to obtain predicted passenger flow of each network point; wherein the network point flow prediction model is constructed by using a ConvLSTM-Unet algorithm, the ConvLSTM-Unet algorithm fuses a ConvLSTM algorithm and a Unet algorithm, the ConvLSTM algorithm is used for embedding time sequence data in the network point flow prediction model, and the Unet algorithm is used for extracting spatial features. The disclosure further provides a network point passenger flow prediction device, equipment, storage medium and program product.

Description

Technical Field

[0001] This disclosure relates to the fields of artificial intelligence and finance, specifically to a method, apparatus, equipment, medium, and program product for predicting customer traffic at business outlets. Background Technology

[0002] With the rapid development of technologies such as big data and artificial intelligence, traditional banking services have faced strong challenges from internet finance, making the sales-oriented transformation of the financial industry an inevitable trend. Against the backdrop of personalized service experiences and the trend towards O2O (Online to Offline), bank branches, which are most closely connected to customers, not only cannot be weakened but need to be strengthened even more. Using artificial intelligence algorithms to model and accurately predict customer traffic at bank branches can, on the one hand, effectively alleviate pressure on teller counters, allowing for flexible and adaptable staff scheduling and ensuring the most effective use of human resources. On the other hand, customer traffic prediction fully motivates teller staff, effectively improving the operational efficiency of bank branches. Employees can anticipate customer flow in the coming hours and prepare accordingly, meeting customer needs while fully releasing their enthusiasm, reducing customer waiting time, increasing customer satisfaction, and laying a solid foundation for further customer maintenance work at bank branches.

[0003] Existing methods for predicting customer traffic at bank branches are often based on traditional time-series forecasting statistical methods, such as the ARIMA, SARIMA, and LSTM algorithms. On the one hand, these methods rely solely on time-series characteristics, neglecting spatial correlations. However, in reality, a branch's customer traffic at a given moment is not only related to time information but also to its spatial location, such as the spatiotemporal relationships between branches, the branch's location within the city, regional weather changes, traffic conditions, and major regional policy decisions like epidemic control and prevention zones. On the other hand, existing methods require establishing or selecting different customer traffic prediction models for each bank branch within a region. This potentially ignores the spatiotemporal influencing factors between branches. For example, if branch A experiences low customer traffic at a certain moment due to unforeseen circumstances, branch B may experience a surge in customer traffic. Such a single prediction model can only target individual branches and cannot form a network of relationships between different branches within a region.

[0004] The information disclosed in this section is only for understanding the background of the inventive concept of this disclosure, and therefore may include information that does not constitute prior art. Summary of the Invention

[0005] In view of the above problems, this disclosure provides methods, apparatus, equipment, media and program products for predicting customer traffic at outlets to improve the feasibility of prediction, and to at least partially solve the above technical problems.

[0006] According to the first aspect of this disclosure, a method for predicting customer traffic at business outlets is provided, comprising: establishing a two-dimensional image matrix of a business outlet area to be tested, the business outlet area containing multiple business outlets, and the two-dimensional image matrix containing spatial features and historical customer traffic features of the multiple business outlets; arranging the two-dimensional image matrix in a temporal sequence to obtain a spatiotemporal array dataset; training the spatiotemporal array dataset using a pre-constructed business outlet traffic prediction model to obtain the predicted customer traffic for each business outlet; wherein the business outlet traffic prediction model is constructed using the ConvLSTM-Unet algorithm, the ConvLSTM-Unet algorithm integrating the ConvLSTM algorithm and the Unet algorithm, the ConvLSTM algorithm being used to embed temporal data in the business outlet traffic prediction model, and the Unet algorithm being used to extract spatial features.

[0007] According to embodiments of this disclosure, establishing a two-dimensional image matrix of the area to be tested includes: acquiring feature data, which includes spatial information data and historical passenger flow data of each area; dividing the area to be tested into a spatial grid, wherein each grid contains one and only one area; and using the spatial grid as a matrix and the feature data as the pixel values ​​of the matrix to obtain a two-dimensional image matrix.

[0008] According to embodiments of this disclosure, arranging a two-dimensional image matrix in the time dimension to obtain a spatiotemporal array dataset includes: selecting a historical statistical time period for the customer traffic of a business outlet; dividing the historical statistical time period into units of time on average; and stacking the two-dimensional image matrices corresponding to each unit of time in chronological order to obtain a spatiotemporal array dataset.

[0009] According to embodiments of this disclosure, the branch traffic prediction model includes an input layer, multiple downsampling layers, multiple upsampling layers, a fusion layer, and an output layer. The model is trained on a spatiotemporal array dataset using a pre-built branch traffic prediction model to obtain the predicted passenger flow for each branch. This process includes: inputting the spatiotemporal array dataset into the input layer to obtain an input feature map; downsampling the input feature map through multiple downsampling layers to obtain a high-dimensional feature map; upsampling the high-dimensional feature map through multiple upsampling layers to obtain an output feature map; and inputting the output feature map into the output layer to obtain the predicted passenger flow for each branch. The multiple downsampling layers and the multiple upsampling layers have the same number of levels, and the corresponding output feature maps have the same size. Each level of the multiple upsampling layers is fused with the corresponding output feature map from the multiple downsampling layers through a fusion layer.

[0010] According to embodiments of this disclosure, each downsampling layer includes a max pooling layer and multiple feature extraction layers connected in sequence, and the feature extraction layer includes a convolutional kernel, a batch normalization layer and an activation layer connected in sequence; each upsampling layer includes a composite layer of an upsampling function and a convolutional kernel connected in sequence and multiple feature extraction layers.

[0011] According to embodiments of this disclosure, the fusion layer adopts a bidirectional ConvLSTM network model. Each level of the multi-layer upsampling layer fuses the feature maps output by the corresponding level of the multi-layer downsampling layer through the fusion layer, including: copying the first feature map output by each level of the multi-layer downsampling layer and the second feature map output by each level of the multi-layer upsampling layer into the bidirectional ConvLSTM network model; fusing the first and second feature maps of the corresponding level to obtain bidirectional spatiotemporal information; and outputting the bidirectional spatiotemporal information to the multi-layer upsampling layer.

[0012] According to embodiments of this disclosure, the branch traffic prediction model further includes a mask layer. Training the spatiotemporal array dataset using the pre-built branch traffic prediction model to obtain the predicted passenger flow of each branch also includes: smoothing and filtering the output feature map through the mask layer to obtain a secondary output feature map; inputting the secondary output feature map to the output layer to obtain the predicted passenger flow of each branch.

[0013] According to embodiments of this disclosure, the feature data also includes branch attribute data, used to provide feature data for each branch itself.

[0014] According to embodiments of this disclosure, before training a spatiotemporal array dataset using a pre-built branch traffic prediction model, the method further includes: establishing an initial branch traffic prediction model; dividing the spatiotemporal array dataset into a training set and a test set; training the initial branch traffic prediction model using the training set to obtain a trained branch traffic prediction model; testing the trained branch traffic prediction model using the test set to obtain a prediction error; and adjusting the parameters of the trained branch traffic prediction model based on the prediction error to obtain a pre-built branch traffic prediction model.

[0015] According to embodiments of this disclosure, the prediction error includes mean square error, root mean square error, mean absolute error, and smoothed mean absolute error.

[0016] The second aspect of this disclosure provides a device for predicting customer traffic at business outlets, comprising: a building module for building a two-dimensional image matrix of a business outlet area to be tested, the business outlet area containing multiple business outlets, and the two-dimensional image matrix containing spatial features and historical customer traffic features of the multiple business outlets; an arrangement module for arranging the two-dimensional image matrix in a temporal sequence to obtain a spatiotemporal array dataset; and a training module for training the spatiotemporal array dataset using a pre-built business outlet traffic prediction model to obtain the predicted customer traffic for each business outlet; wherein the business outlet traffic prediction model is constructed using the ConvLSTM-Unet algorithm, the ConvLSTM-Unet algorithm integrating the ConvLSTM algorithm and the Unet algorithm, the ConvLSTM algorithm being used to embed temporal data in the business outlet traffic prediction model, and the Unet algorithm being used to extract spatial features.

[0017] A third aspect of this disclosure provides an electronic device comprising: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to perform the branch office traffic prediction method of any of the above embodiments.

[0018] A fourth aspect of this disclosure also provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the branch office traffic prediction method of any of the above embodiments.

[0019] The fifth aspect of this disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the branch office traffic prediction method of any of the above embodiments.

[0020] Compared with existing technologies, the branch customer traffic prediction method, apparatus, electronic device, storage medium, and program product provided in this disclosure have at least the following beneficial effects:

[0021] (1) The branch customer flow prediction method disclosed herein combines the spatial and temporal features of the branch to obtain a spatiotemporal array dataset as the input of the model. At the same time, a branch flow prediction model based on the ConvLSTM-Unet algorithm, which can extract both temporal and spatial features, is used to train the spatiotemporal array dataset. This unifies the customer flow prediction model of each branch in the test area and realizes real-time prediction of customer flow of all bank branches in the area.

[0022] (2) The network traffic prediction model disclosed herein adopts a symmetrical downsampling layer and upsampling layer structure, and fuses the downsampling feature maps and upsampling feature maps of the corresponding layers through a fusion layer. This achieves the combination of abstract data from the deep feature maps in the upsampling layer and detailed information from the shallow feature maps in the downsampling layer, thereby enhancing the features of the training samples. Therefore, it is not easy to overfit even for small samples. Furthermore, since the problem space of the network traffic prediction matrix sample disclosed herein is small, the model can be trained with a small amount of sample data. A relatively accurate prediction result can be obtained with a small number of measurements, which can greatly reduce the consumption of financial, time and human resources during the data collection process.

[0023] (3) The branch traffic prediction model disclosed herein removes the fully connected layers and softmax classifiers found in traditional convolutional neural networks. During training, it does not require classification of feature maps, thus breaking through the limitations of traditional convolutional neural networks on the size of input images and enabling dynamic end-to-end data input and output. The size of the input image can be adjusted according to the number of branches in the area, making it possible to predict customer traffic when the number of bank branches in a certain area is relatively small. Attached Figure Description

[0024] The foregoing contents, as well as other objects, features, and advantages of this disclosure, will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0025] Figure 1 The illustration schematically depicts application scenarios of a method, apparatus, device, medium, and program product for predicting customer traffic at business outlets according to embodiments of this disclosure.

[0026] Figure 2 A flowchart illustrating a method for predicting customer traffic at business outlets according to an embodiment of this disclosure is shown schematically.

[0027] Figure 3 A flowchart illustrating a method for establishing a two-dimensional image matrix according to an embodiment of the present disclosure is shown schematically.

[0028] Figure 4 A flowchart illustrating the acquisition of a spatiotemporal array dataset according to an embodiment of the present disclosure is shown schematically;

[0029] Figure 5 A schematic diagram illustrating the structure of a customer traffic prediction model for business outlets according to an embodiment of the present disclosure is shown.

[0030] Figure 6 A flowchart illustrating a method for training a spatiotemporal array dataset according to an embodiment of the present disclosure is shown schematically.

[0031] Figures 7a-7e This diagram schematically illustrates the internal structure of each layer in a customer traffic prediction model for business outlets according to an embodiment of the present disclosure.

[0032] Figure 8 A flowchart illustrating a method for fusing feature maps according to an embodiment of the present disclosure is shown schematically.

[0033] Figure 9 A flowchart illustrating a method for constructing a branch traffic prediction model according to an embodiment of the present disclosure is shown schematically.

[0034] Figure 10 A flowchart illustrating a method for predicting customer traffic at business outlets according to another embodiment of this disclosure is shown schematically.

[0035] Figure 11 A schematic diagram illustrating the structure of a network customer flow prediction device according to an embodiment of the present disclosure is shown; and

[0036] Figure 12 A block diagram schematically illustrates an electronic device suitable for implementing a branch office traffic prediction method according to an embodiment of the present disclosure. Detailed Implementation

[0037] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0038] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0039] 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 are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0040] When using expressions such as "at least one of A, B, and C", they should generally be interpreted in accordance with the meaning that is commonly understood by a person skilled in the art (e.g., "a system having at least one of A, B, and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B, and C, etc.).

[0041] This disclosure provides a method, apparatus, device, medium, and program product for predicting customer traffic at business outlets, which can be used in the financial sector or other fields. It should be noted that the method, apparatus, device, medium, and program product for predicting customer traffic at business outlets disclosed herein can be used in the financial sector, as well as in any other field. The application areas of the method, apparatus, device, medium, and program product for predicting customer traffic at business outlets disclosed herein are not limited.

[0042] In the technical solution disclosed herein, the acquisition, storage, and application of user personal information comply with the provisions of relevant laws and regulations, necessary confidentiality measures have been taken, and there is no violation of public order and good morals.

[0043] In the technical solution disclosed herein, the user's authorization or consent is obtained before acquiring or collecting the user's personal information.

[0044] Figure 1 The illustration schematically depicts application scenarios of the site visitor flow prediction method, apparatus, device, medium, and program product according to embodiments of the present disclosure.

[0045] like Figure 1 As shown, application scenario 100 according to this embodiment may include terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as a medium for providing a communication link between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0046] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0047] Terminal devices 101, 102, and 103 can be various electronic devices with displays and web browsing capabilities, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0048] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using terminal devices 101, 102, and 103 (for example only). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0049] It should be noted that the customer flow prediction method for business outlets provided in this disclosure can generally be executed by server 105. Correspondingly, the customer flow prediction device for business outlets provided in this disclosure can generally be located in server 105. The customer flow prediction method for business outlets provided in this disclosure can also be executed by a server or server cluster that is different from server 105 and capable of communicating with terminal devices 101, 102, 103 and / or server 105. Correspondingly, the customer flow prediction device for business outlets provided in this disclosure can also be located in a server or server cluster that is different from server 105 and capable of communicating with terminal devices 101, 102, 103 and / or server 105.

[0050] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0051] The following will be based on Figure 1 The described scene, through Figures 2 to 10The method for predicting customer traffic at business outlets according to the disclosed embodiments is described in detail.

[0052] Figure 2 A flowchart illustrating a method for predicting customer traffic at business outlets according to an embodiment of this disclosure is shown schematically.

[0053] like Figure 2 As shown, embodiments of this disclosure provide a method for predicting customer traffic at business outlets, including, for example:

[0054] S210, Establish a two-dimensional image matrix of the area to be tested, which contains multiple sites, and the two-dimensional image matrix contains the spatial features and historical passenger flow features of the multiple sites.

[0055] According to embodiments of this disclosure, a branch is, for example, an institution with a physical space, such as a bank, mobile network service hall, library, or other for-profit or non-profit institution. The customer flow prediction method of this disclosure is particularly applicable to branch types with numerous branches within a certain area, where the customer flow of each branch influences the other. A two-dimensional image matrix, for example, is a matrix representation of the regional distribution map of the branches, where the image pixels represent, for example, the characteristic data of customer flow influencing factors for each branch. Spatial features, for example, are the characteristic data of customer flow-related influencing factors that each branch possesses in geographical space. Historical customer flow characteristics, for example, are the historical customer flow characteristic data of each branch within the area to be measured.

[0056] S220: Arrange the two-dimensional image matrix in time sequence to obtain a spatiotemporal array dataset.

[0057] According to embodiments of this disclosure, the prediction of customer traffic at various locations is, for example, based on customer traffic data for each location within a certain historical time period. By arranging the aforementioned two-dimensional image matrices from different time units in chronological order within that historical time period, a spatiotemporal array dataset to be trained can be obtained.

[0058] S230: A pre-built branch traffic prediction model is used to train the spatiotemporal array dataset to obtain the predicted passenger flow for each branch. The branch traffic prediction model is constructed using the ConvLSTM-Unet algorithm, which combines the ConvLSTM and Unet algorithms. The ConvLSTM algorithm is used to embed time-series data into the branch traffic prediction model, while the Unet algorithm is used to extract spatial features.

[0059] According to embodiments of this disclosure, the ConvLSTM algorithm is a deep learning method that can be used to introduce temporal features or process temporally related features in a dataset to be tested. The Unet algorithm is a deep learning network algorithm for image segmentation, which can be used to extract spatial features from feature maps of spatiotemporal array datasets. Embodiments of this disclosure organically combine these two algorithms to design a novel deep learning neural network architecture, achieving accurate prediction of passenger flow over a future period based on historical passenger flow data over a period of time and relevant passenger flow influencing factor feature data.

[0060] Figure 3 A flowchart illustrating a method for establishing a two-dimensional image matrix according to an embodiment of the present disclosure is shown.

[0061] According to embodiments of this disclosure, such as Figure 3 As shown, for example, a two-dimensional image matrix is ​​established by operating S211 to S213.

[0062] S211, Obtain feature data, which includes spatial information data and historical passenger flow data for each branch.

[0063] According to embodiments of this disclosure, the spatial information data corresponds to the aforementioned spatial characteristics, and includes, for example,: the administrative division and urban / rural level of each outlet, the number of permanent residents in the streets where each outlet is located, spatial location data of all outlets within the region, road network data within the region, historical road condition information such as traffic flow for the roads where each outlet is located, spatial location data of public transportation stations such as buses and subways, road levels such as arterial roads, secondary arterial roads, branch roads, neighborhood roads (community roads), historical weather data for the location of each outlet, and the category of the area where each outlet is located, such as downtown areas, tourist areas, etc. Historical major policy decision information for the jurisdiction where each outlet is located, such as whether it is a controlled area, etc. Historical passenger flow data corresponds to the aforementioned historical passenger flow characteristics.

[0064] Preferably, in order to improve the accuracy of passenger flow prediction, the feature data may also include branch attribute data, which provides the feature data of each branch, such as the attribute description features of each bank branch, such as the branch area size, the number of branch counter windows, the number of branch staff, and the branch's organizational category, such as: first-level branch, second-level branch, sub-branch, etc.

[0065] S212, divide the area of ​​the test points into spatial grids, and each grid of the spatial grid contains one and only one test point.

[0066] According to embodiments of this disclosure, by dividing the area of ​​the test site into a spatial grid, each site can be separately divided into a grid. The spatial characteristics, historical passenger flow characteristics, and relevant characteristics of its own attributes of the site can be used as pixel values ​​of the corresponding grid matrix to obtain a two-dimensional image matrix. Then, deep learning is performed on the two-dimensional image matrix to achieve feature learning of a single pixel (i.e., a single site), thereby predicting the passenger flow in the next time period based on the historical passenger flow data of that site.

[0067] S213 uses a spatial grid as a matrix and the feature data as the pixel values ​​of the matrix to obtain a two-dimensional image matrix.

[0068] Figure 4 A flowchart illustrating the acquisition of a spatiotemporal array dataset according to an embodiment of the present disclosure is shown.

[0069] According to embodiments of this disclosure, such as Figure 4 As shown, for example, a spatiotemporal array dataset is obtained by operating S221 to S223.

[0070] S221, Select the historical statistical period of customer traffic at the outlet.

[0071] According to embodiments of this disclosure, the historical statistical time period is, for example, the past twenty years, the past ten years, or the passenger flow statistics period since the establishment of a certain outlet.

[0072] It is understandable that, since the establishment and operation time of each outlet in the test area varies, when the historical statistical time period selected uniformly is longer than the establishment and operation time of one or more outlets, the passenger flow statistical time period of that outlet shall be calculated from the time of its establishment.

[0073] S222 divides the historical statistical period into average units of time.

[0074] According to embodiments of this disclosure, dividing the aforementioned uniformly set historical statistical time period into average unit time segments facilitates the prediction of future passenger flow based on these unit time segments. A unit time segment can be, for example, a month, a week, a day, or an hour. When predicting on a daily basis, for example, the passenger flow for each day within a historical time period is statistically analyzed, and then the passenger flow for the next day (e.g., tomorrow) is predicted.

[0075] S223, stack the two-dimensional image matrices corresponding to each unit of time in time sequence to obtain a spatiotemporal array dataset.

[0076] According to embodiments of this disclosure, for example, when time is measured in days, the daily passenger flow data of each location in the test area over a historical time period, along with the location's spatial characteristic data and its own attribute data, can constitute a two-dimensional image matrix. Then, by stacking each two-dimensional image matrix of this historical time period chronologically, a spatiotemporal array dataset can be obtained. Finally, by training this spatiotemporal array dataset using a pre-built location traffic prediction model, the predicted passenger flow for each location on a future day can be obtained.

[0077] Figure 5 A schematic diagram illustrating the structure of a customer traffic prediction model for a business outlet according to an embodiment of the present disclosure is shown.

[0078] Figure 6 A flowchart illustrating a method for training a spatiotemporal array dataset according to an embodiment of the present disclosure is shown.

[0079] Figures 7a-7e The diagram schematically illustrates the internal structure of each layer in a customer traffic prediction model for a business outlet according to an embodiment of the present disclosure.

[0080] Figure 8 A flowchart illustrating a method for fusing feature maps according to an embodiment of the present disclosure is shown.

[0081] According to embodiments of this disclosure, such as Figure 5 As shown, the branch traffic prediction model includes, for example, an input layer, multiple downsampling layers, multiple upsampling layers, a fusion layer, and an output layer. The input layer, multiple downsampling layers, multiple upsampling layers, and the output layer together constitute the Unet network structure, used for spatial feature extraction to achieve pixel-level prediction. The fusion layer uses a ConvLSTM network structure to achieve temporal prediction of branch traffic, predicting future traffic based on historical traffic data. The multiple downsampling layers and multiple upsampling layers have the same number of layers, and the feature maps output by corresponding layers have the same size. Each layer of the multiple upsampling layers is fused with the feature maps output by the corresponding layers of the multiple downsampling layers through the fusion layer. Furthermore, as... Figure 6 As shown, a pre-built network traffic prediction model is used to train a spatiotemporal array dataset to obtain the predicted passenger traffic for each network point, including, for example:

[0082] S231, input the spatiotemporal array dataset into the input layer to obtain the input feature map.

[0083] According to embodiments of this disclosure, Figure 5 The leftmost dashed arrow represents the input layer. The spatiotemporal array dataset is input into the input layer for feature extraction, which yields an input feature map of size 64, for example, as output from the input layer.

[0084] Specifically, such as Figure 7aAs shown, the input layer includes, for example, two interconnected feature extraction layers. Each feature extraction layer consists of, for example, a filter (i.e., a convolutional kernel, such as a 3×3 convolutional kernel (Conv)), a batch normalization layer, and an activation function (i.e., an activation layer, such as ReLU, a linear rectified function) connected in sequence.

[0085] Understandably, depending on the needs of feature extraction, the size of the convolution kernel can also be other sizes, such as 7×7, 5×5, etc., and the activation function can also be other activation functions, such as the Sigmoid function, Tanh function, Maxout function, etc., which will not be listed here.

[0086] S232 downsamples the input feature map through multiple downsampling layers to obtain a high-dimensional feature map.

[0087] According to embodiments of this disclosure, the overall network framework integrates the U-net deep network model from the field of computer vision to achieve feature extraction under spatial characteristics and pixel-level prediction. The input feature map output from the input layer undergoes several downsampling iterations to obtain an abstracted high-dimensional feature map. The downsampling layer is, for example, a continuously convolutionally shrinking compressed channel, also known as encoding, primarily used to capture contextual information in the image matrix. It is, for example, based on a typical traditional convolutional neural network (CNN) downsampling method, where the image passes through several convolutional and pooling layers to obtain a low-resolution high-dimensional feature map. Since the fully connected layers in the network have fixed vector lengths, the size of the input image is often limited. However, in the feature map learning process of this disclosure, it proceeds from two-dimensional feature map to two-dimensional feature map, without requiring image classification. Therefore, it does not require fully connected layers or classifier structures such as softmax, and thus there is no limitation on the size of the input image. This allows for the prediction of customer traffic at network points even when the number of points distributed within a certain area is small, even with a small input image size.

[0088] Specifically, such as Figure 7b As shown, each downsampling layer includes, for example, a series of connected max-pooling layers (e.g., 2×2) and multiple (e.g., two) of the aforementioned feature extraction layers. Each time downsampling is performed, the size of the output feature map is halved, while the number of feature maps doubles. Here, we use... This represents the output feature map after each downsampling. Here, F... l W represents the number of feature maps in the l-th layer. l ×H lThis indicates the size of the feature map in layer l. For example, a network traffic prediction model includes three downsampling layers. By compressing the input feature map three times, feature maps with sizes of, for example, 128, 256, and 1024 are obtained. The feature map with a size of 1024 is a highly abstract, high-dimensional feature map.

[0089] S233 upsamples the high-dimensional feature map through multiple upsampling layers to obtain the output feature map.

[0090] According to embodiments of this disclosure, after obtaining the high-dimensional feature map, each pixel in the image matrix needs to be accurately located, and therefore it is input into an upsampling layer for decoding. The upsampling layer is, for example, an extended channel that uses an amplification process symmetrical to the aforementioned compression channel to achieve pixel-level prediction. The extended channel, also known as the decoding process, is used to accurately locate each pixel in the image matrix. It involves inversely upsampling the previously obtained high-dimensional feature map through a series of deconvolution operations corresponding to the downsampling convolution operation, ultimately obtaining an output image with the same layer size as the input image.

[0091] Specifically, such as Figure 7c As shown, each upsampling layer includes, for example, a composite layer of upsampling functions and convolutional kernels connected in sequence (Upsampling+merge (ConvLSTM layer)) and multiple (e.g., two) of the aforementioned feature extraction layers. Here, we use... This represents the feature map copied from the downsampling layer (compressed channel). This represents the high-dimensional feature map output from the previous convolutional layer. Here, F... l+1 =2F l , The network traffic prediction model includes, for example, three upsampling layers. By expanding the high-dimensional feature map three times, feature maps with sizes of, for example, 256, 128, and 64 are obtained. The feature map with a size of 64 is the output feature map after upsampling.

[0092] According to embodiments of this disclosure, when upsampling the high-dimensional feature map, each level also incorporates the feature map output by the corresponding level of the downsampling layer through a fusion layer. This not only includes the abstract data of the deep feature map (e.g., the high-dimensional feature map and the feature map output thereafter), but also embeds the detailed feature information of the shallow feature map (e.g., the input feature map and the feature maps output by the first two downsampling levels), which is equivalent to performing a feature enhancement operation on the training abstract dataset.

[0093] Specifically, such as Figure 7dAs shown, the fusion layer (BConvLSTM) employs a bidirectional ConvLSTM network model, including forward ConvLSTM and backward ConvLSTM. It fully considers spatial correlation and utilizes convolutional operations for input-to-state and state-to-state transitions, rather than using fully connected layers for state transitions in traditional LSTMs. χ d The feature maps output from the fusion layer are first subjected to an upconvolution operation (upsampling function and 2×2 convolution operation) in a composite layer, which doubles the size of each feature map and halves the number of feature channels, resulting in the output. Make χ d With χ e The feature maps have the same size. Then χ d With χ e Feature maps are merged, then passed through two 3×3 filters, batch normalization, and ReLU for subsequent convolution operations. Each ConvLSTM network model consists of, for example, an input gate i. t One output gate O t A forgetting gate t and a memory unit C t The composition can be described as follows:

[0094] i t =σ(W xi *χ t +W hi *H t-1 +W ci *C t-1 +b i )

[0095] f t =σ(W xf *χ t +W hf *H t-1 +W cf *C t-1 +b f )

[0096]

[0097]

[0098]

[0099] Among them, * and χ represents convolution and the Hadamard function, respectively. t It is the input vector (χ in this embodiment) e and ), i tIt is the input gate vector, f t It is the forget gate vector, H t It is the hidden state vector, C t It is a memory unit vector, o t It is the output gate vector, W x* W h* and W c* These are two-dimensional convolutional kernels corresponding to the input, hidden, and memory states, respectively, b i b f b o and b c σ is the bias term, and σ is the Sigmoid activation function.

[0100] In the fusion layer, two ConvLSTMs are used to process the input vector χ from the forward and backward directions respectively. e and Encoding is performed. The final output of the fusion layer is bidirectional spatiotemporal information Y. t It can be represented as:

[0101]

[0102] in, and Let these represent the hidden state vectors for the forward and backward states, respectively. and Let be a one-dimensional parameter, and b be a bias term.

[0103] Furthermore, such as Figure 8 As shown, each level of the upsampling layer fuses the feature maps output by the corresponding level of the downsampling layer through the fusion layer, including, for example:

[0104] S2331, copy the first feature map output from each level of the multi-layer downsampling layer and the second feature map output from each level of the multi-layer upsampling layer into the bidirectional ConvLSTM network model.

[0105] S2332, the first feature map and the second feature map of the corresponding level are fused to obtain bidirectional spatiotemporal information.

[0106] S2333 outputs bidirectional spatiotemporal information to a multi-layer upsampling layer.

[0107] According to embodiments of this disclosure, the network model simultaneously embeds a ConvLSTM network during the aforementioned expansion channel process to achieve temporal prediction of customer traffic at network points. During each downsampling process in the compressed channel, for example, the output feature map is input into a bidirectional ConvLSTM network model, and its output is merged with the feature map from the corresponding upsampling process in the expanded channel. These merged features are then uniformly incorporated into subsequent convolutional operations. This embeds shallow temporal data into the network model and performs temporal data augmentation.

[0108] S234, input the output feature map to the output layer to obtain the predicted passenger flow of each network point.

[0109] According to embodiments of this disclosure, Figure 5 The rightmost dotted-line arrow represents the output layer. The output feature map, obtained after upsampling and fusing shallow detail features, is then input into the output layer for feature extraction, resulting in the final two-dimensional feature map, which is the predicted passenger flow of each point in the test area within the next unit of time.

[0110] Specifically, such as Figure 7e As shown, the output layer includes, for example, multiple (e.g., two) interconnected feature extraction layers, followed by a 1×1 convolutional kernel (filter), batch normalization, and an activation function (e.g., ReLU, the rectified linear function). Convolving each pixel in the feature map with a 1×1 kernel enhances the detail features in the final output two-dimensional feature map.

[0111] Preferably, to prevent non-zero pixel values ​​in non-dot regions from causing abnormal output results during feature extraction, an additional filter (also called a mask layer) can be added before the final output of the output layer to smooth the output. Specifically, the feature map output from the last upsampling layer is filtered through a network mask layer (N) using a pixel-level AND operation, setting all non-zero predicted values ​​outside the road network (i.e., pixel values ​​corresponding to the absence of dots in the road network) to 0.

[0112] Figure 9 A flowchart illustrating a method for constructing a site traffic prediction model according to an embodiment of the present disclosure is shown.

[0113] According to embodiments of this disclosure, such as Figure 9 As shown, before training a spatiotemporal array dataset using a pre-built branch traffic prediction model, branch traffic prediction methods may include, for example:

[0114] S241, Establish an initial network traffic prediction model.

[0115] S242 divides the spatiotemporal array dataset into a training set and a test set.

[0116] According to embodiments of this disclosure, the constructed spatiotemporal array dataset is divided into a training set and a test set in a certain ratio (e.g., 8:2, 7:3, etc.).

[0117] S243, the initial network traffic prediction model is trained using the training set to obtain a trained network traffic prediction model.

[0118] S244: The trained network traffic prediction model is tested using a test set to obtain the prediction error.

[0119] According to embodiments of this disclosure, prediction errors include, but are not limited to, mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), and smoothed mean absolute error (SMAE).

[0120] S245, Adjust the parameters of the trained network traffic prediction model based on the prediction error to obtain the pre-constructed network traffic prediction model.

[0121] Figure 10 A flowchart illustrating a method for predicting customer traffic at business outlets according to another embodiment of this disclosure is shown schematically.

[0122] Based on the embodiments of this disclosure, and to facilitate a further understanding of the technical solutions of this disclosure, a more specific embodiment is now provided, such as... Figure 10 As shown, methods for predicting customer traffic at retail outlets include, for example:

[0123] Step S1: Obtain spatial information data of all bank branches within the region, which constitutes M1 feature data.

[0124] Step S2: Obtain historical customer traffic data for all branches within the region and generate M2 feature data for bank branches.

[0125] Step S3: Obtain the feature description data of the network points, forming a total of M3 network point feature description data.

[0126] Step S4: Take the spatial location data of all bank branches from Step S1 and draw a spatial grid according to a preset range (e.g., 1 km x 1 km). The preset range standard is to ensure that there is exactly one bank branch within the divided grid. If the M feature data, such as the passenger flow at a certain moment for the corresponding bank branch in Steps S1, S2, and S3, are used as the pixel values ​​of the spatial grid matrix, then the original spatial data of the bank branch is converted into an M-channel image matrix S. N×N×M(Here, N represents the length or width of the final image matrix, and M = M1 + M2 + M3). Given a time series T = {t1, t2...t...} k The data is divided equally according to time interval t. The grid matrix S at each time point is stacked over time to form a T×S spatiotemporal array, which constitutes the dataset.

[0127] Step S5: Divide the dataset constructed in Step S4 into a training set and a test set according to a certain ratio. The training set serves as the input and output for training the model, while the test set is used to make predictions on the trained model. By comparing the prediction error with the actual results, the model training stopping criteria are measured, and the best training model is selected.

[0128] Step S6: Construct a customer flow prediction model for business outlets using the ConvLSTM-Unet algorithm. Train the network model on the training set data, and use the test set to validate and tune the network parameters to achieve optimal performance. The model input is an M-channel spatiotemporal array T×S. N×N×M The model output is S N×N A two-dimensional image matrix is ​​represented as the predicted passenger flow data of all outlets within the jurisdiction at the next time step t in a spatial grid matrix.

[0129] Step S7: Train the optimal spatial network passenger flow prediction model through step S6. In a real-world scenario, this model can be used to predict passenger flow and obtain the passenger flow prediction results for all network points within the jurisdiction at the next time step.

[0130] Based on the above-mentioned method for predicting customer traffic at retail outlets, this disclosure also provides a device for predicting customer traffic at retail outlets. The following will be combined with... Figure 8 The device is described in detail.

[0131] Figure 11 A schematic block diagram of a network customer flow prediction device according to an embodiment of the present disclosure is shown.

[0132] like Figure 11 As shown, the customer flow prediction device 1100 of this embodiment includes a setup module 1110, an arrangement module 1120 and a training module 1130.

[0133] The establishment module 1110 is used to establish a two-dimensional image matrix of the area to be tested, which includes multiple network points. The two-dimensional image matrix includes the spatial features and historical passenger flow features of the multiple network points. In one embodiment, the establishment module 1110 can be used to perform the operation S210 described above, which will not be repeated here.

[0134] The arrangement module 1120 is used to arrange the two-dimensional image matrix in a temporal sequence to obtain a spatiotemporal array dataset. In one embodiment, the arrangement module 1120 can be used to perform the operation S220 described above, which will not be repeated here.

[0135] The training module 1130 is used to train the spatiotemporal array dataset using a pre-built branch traffic prediction model to obtain the predicted passenger flow for each branch. The branch traffic prediction model is constructed using the ConvLSTM-Unet algorithm, which combines the ConvLSTM and Unet algorithms. The ConvLSTM algorithm is used to embed time-series data into the branch traffic prediction model, while the Unet algorithm is used to extract spatial features. In one embodiment, the training module 1130 can be used to perform the operation S230 described above, which will not be repeated here.

[0136] According to embodiments of this disclosure, any plurality of modules among the establishment module 1110, arrangement module 1120, and training module 1130 can be combined into one 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 one module. According to embodiments of this disclosure, at least one of the establishment module 1110, arrangement module 1120, and training module 1130 can be at least partially implemented as hardware circuitry, 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-package, an application-specific integrated circuit (ASIC), or implemented in hardware or firmware by any other reasonable means of integrating or packaging circuitry, or implemented in any one of software, hardware, and firmware methods, or in a suitable combination of any of these. Alternatively, at least one of the establishment module 1110, arrangement module 1120, and training module 1130 can be at least partially implemented as a computer program module, which, when run, can perform corresponding functions.

[0137] Figure 12 A block diagram schematically illustrates an electronic device suitable for implementing a branch office traffic prediction method according to an embodiment of the present disclosure.

[0138] like Figure 12As shown, an electronic device 1200 according to an embodiment of the present disclosure includes a processor 1201, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1202 or a program loaded from a storage portion 1208 into a random access memory (RAM) 1203. The processor 1201 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1201 may also include onboard memory for caching purposes. The processor 1201 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.

[0139] RAM 1203 stores various programs and data required for the operation of electronic device 1200. Processor 1201, ROM 1202, and RAM 1203 are interconnected via bus 1204. Processor 1201 performs various operations of the method flow according to embodiments of the present disclosure by executing programs in ROM 1202 and / or RAM 1203. It should be noted that the programs may also be stored in one or more memories other than ROM 1202 and RAM 1203. Processor 1201 may also perform various operations of the method flow according to embodiments of the present disclosure by executing programs stored in said one or more memories.

[0140] According to embodiments of this disclosure, the electronic device 1200 may further include an input / output (I / O) interface 1205, which is also connected to the bus 1204. The electronic device 1200 may also include one or more of the following components connected to the I / O interface 1205: an input section 1206 including a keyboard, mouse, etc.; an output section 1207 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1208 including a hard disk, etc.; and a communication section 1209 including a network interface card such as a LAN card, modem, etc. The communication section 1209 performs communication processing via a network such as the Internet. A drive 1210 is also connected to the I / O interface 1205 as needed. A removable medium 1211, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 1210 as needed so that computer programs read from it can be installed into the storage section 1208 as needed.

[0141] This disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs that, when executed, implement the method according to the embodiments of this disclosure.

[0142] According to embodiments of this disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, such as including, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the 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 embodiments of this disclosure, the computer-readable storage medium may include ROM 1202 and / or RAM 1203 and / or one or more memories other than ROM 1202 and RAM 1203 described above.

[0143] Embodiments of this disclosure also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to cause the computer system to implement the item recommendation method provided in the embodiments of this disclosure.

[0144] When the computer program is executed by the processor 1201, it performs the functions defined in the system / apparatus of this disclosure embodiments. According to embodiments of this disclosure, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0145] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 1209, and / or installed from the removable medium 1211. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0146] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 1209, and / or installed from the removable medium 1211. When the computer program is executed by the processor 1201, it performs the functions defined in the system of this disclosure embodiment. According to embodiments of this disclosure, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0147] According to embodiments of this disclosure, program code for executing the computer programs provided in embodiments of this disclosure can be written in any combination of one or more programming languages. Specifically, these computational 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 execute entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0148] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0149] Those skilled in the art will understand that the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in this disclosure. In particular, the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or combinations fall within the scope of this disclosure.

[0150] The embodiments of this disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of this disclosure. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. The scope of this disclosure is defined by the appended claims and their equivalents. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this disclosure, and all such substitutions and modifications should fall within the scope of this disclosure.

Claims

1. A method for predicting customer traffic at a business outlet, characterized in that, include: A two-dimensional image matrix is ​​established for the network area to be tested, wherein the network area to be tested contains multiple network points, and the two-dimensional image matrix contains the spatial features and historical passenger flow features of the multiple network points; The two-dimensional image matrix is ​​arranged in time sequence to obtain a spatiotemporal array dataset; The spatiotemporal array dataset is trained using a pre-built site traffic prediction model to obtain the predicted passenger traffic for each site. The network traffic prediction model is constructed using the ConvLSTM-Unet algorithm, which integrates the ConvLSTM and Unet algorithms. The ConvLSTM algorithm is used to embed time-series data into the network traffic prediction model, and the Unet algorithm is used to extract the spatial features. The network traffic prediction model includes an input layer, multiple downsampling layers, multiple upsampling layers, a fusion layer, and an output layer; the input layer, the multiple downsampling layers, the multiple upsampling layers, and the output layer form a Unet network structure for extracting spatial features; the fusion layer adopts a bidirectional ConvLSTM network model. The training process includes: The spatiotemporal array dataset is input into the input layer to obtain the input feature map; The input feature map is downsampled through the multi-layer downsampling layer to obtain a high-dimensional feature map. The high-dimensional feature map is upsampled through the multi-layer upsampling layer to obtain the output feature map; The output feature map is input into the output layer to obtain the predicted passenger flow for each of the network points. Each downsampling layer includes a max pooling layer and multiple feature extraction layers connected in sequence. The feature extraction layer includes a convolutional kernel, a batch normalization layer and an activation layer connected in sequence. Each of the upsampling layers includes a composite layer of an upsampling function and a convolutional kernel connected in sequence, and multiple feature extraction layers; The number of layers in the multi-layer downsampling layer and the multi-layer upsampling layer are the same, and the size of the feature maps output by the corresponding layers is the same; Each level of the multi-layer upsampling layer is fused with the feature map output of the corresponding level of the multi-layer downsampling layer through the fusion layer, including: Copy the first feature map output from each level of the multi-layer downsampling layer and the second feature map output from each level of the multi-layer upsampling layer into the bidirectional ConvLSTM network model; The first feature map and the second feature map of the corresponding level are fused to obtain bidirectional spatiotemporal information; The bidirectional spatiotemporal information is output to the multi-layer upsampling layer.

2. The method for predicting customer traffic at service outlets according to claim 1, characterized in that, The two-dimensional image matrix for establishing the area of ​​dots to be tested includes: Acquire feature data, which includes spatial information data and historical passenger flow data for each of the network points; The area to be tested is divided into spatial grids, and each grid of the spatial grid contains one and only one of the network points. Using the spatial grid as a matrix and the feature data as the pixel values ​​of the matrix, the two-dimensional image matrix is ​​obtained.

3. The method for predicting customer traffic at service outlets according to claim 1, characterized in that, The step of arranging the two-dimensional image matrix along the time dimension to obtain the spatiotemporal array dataset includes: Select the historical statistical time period for the customer traffic of the aforementioned outlets; The historical statistical time period is divided into average units of time; The two-dimensional image matrices corresponding to each unit of time are stacked in time sequence to obtain the spatiotemporal array dataset.

4. The method for predicting customer traffic at service outlets according to claim 1, characterized in that, The branch network traffic prediction model further includes a mask layer, and the step of training the spatiotemporal array dataset using the pre-built branch network traffic prediction model to obtain the predicted passenger flow for each branch network further includes: The output feature map is smoothed and filtered by the mask layer to obtain a secondary output feature map; The secondary output feature map is input into the output layer to obtain the predicted passenger flow of each of the network points.

5. The method for predicting customer traffic at service outlets according to claim 2, characterized in that, The feature data also includes branch attribute data, which is used to provide feature data for each branch itself.

6. The method for predicting customer traffic at service outlets according to claim 1, characterized in that, Before training the spatiotemporal array dataset using a pre-built network traffic prediction model, the following steps are also included: Establish an initial network traffic prediction model; The spatiotemporal array dataset is divided into a training set and a test set; The initial network traffic prediction model is trained using the training set to obtain a trained network traffic prediction model. The trained network traffic prediction model was tested using the test set to obtain the prediction error; The parameters of the trained branch traffic prediction model are adjusted based on the prediction error to obtain the pre-built branch traffic prediction model.

7. The method for predicting customer traffic at service outlets according to claim 6, characterized in that, The prediction errors include mean square error, root mean square error, mean absolute error, and smoothed mean absolute error.

8. A device for predicting customer traffic at a business outlet, characterized in that, The apparatus for implementing the method according to any one of claims 1 to 7 comprises: A module is established to create a two-dimensional image matrix of the area to be tested, wherein the area to be tested contains multiple outlets, and the two-dimensional image matrix contains the spatial features and historical passenger flow features of the multiple outlets. An arrangement module is used to arrange the two-dimensional image matrix in a temporal sequence to obtain a spatiotemporal array dataset; and The training module is used to train the spatiotemporal array dataset using a pre-built branch traffic prediction model to obtain the predicted passenger flow for each branch; wherein the branch traffic prediction model is constructed using the ConvLSTM-Unet algorithm, which integrates the ConvLSTM algorithm and the Unet algorithm, the ConvLSTM algorithm is used to embed time-series data into the branch traffic prediction model, and the Unet algorithm is used to extract the spatial features.

9. An electronic device, characterized in that, include: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors execute the branch customer flow prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores executable instructions, which, when executed by a processor, cause the processor to perform the branch customer flow prediction method according to any one of claims 1 to 7.

11. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the branch customer flow prediction method according to any one of claims 1 to 7.

Images