Method for training neural network, method and device for predicting incoming call flow

Abstract

This disclosure provides a method for training a neural network, a method for predicting incoming call traffic, and an apparatus for predicting incoming call traffic, which can be used in the fields of artificial intelligence or finance. The network comprises an input layer, an embedding layer, a decoding layer, and an output layer. The neural network training method includes inputting traffic feature values ​​from the previous time period before the current time period into the neural network, and adjusting the parameters of the neural network so that the predicted traffic for the previous time period approaches the actual traffic for the current time period.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and more specifically, to a method for training a neural network, and a method and device for predicting incoming call traffic. Background Art

[0002] With the rapid development of science and technology, artificial intelligence (AI) has rapidly developed in customer service. Telephone customer service is a crucial channel for businesses to process and communicate with their customers. Related technologies enable the provision of telephone customer service through intelligent interaction.

[0003] In the process of realizing the concept of the present disclosure, the applicant discovered that there are at least the following problems in the related technology: in order to meet the requirements of both intelligent interactive availability and cost control, it is necessary to be able to accurately predict the incoming call traffic in the future. Summary of the Invention

[0004] In view of this, the present disclosure provides a method for training a neural network, a method and an apparatus for predicting incoming call traffic, so as to improve the prediction accuracy of incoming call traffic and meet the requirements of both intelligent interactive usability and cost control.

[0005] One aspect of the present disclosure provides a method for training a neural network, the neural network comprising: an input layer configured to obtain at least the value of the traffic feature of the current period and the value of the traffic feature of the previous period before the current period; an embedding layer configured to determine the first feature weight of each traffic feature, and determine a first predicted traffic based on the value of the traffic feature of the previous period before the current period and the feature weight of each traffic feature; a decoding layer configured to determine the second predicted traffic of the current period based on the value of the traffic feature of the current period, and determine the third predicted traffic based on the value of the traffic feature of the previous period before the current period; and an output layer configured to output a predicted traffic obtained by fusion of the first predicted traffic, the second predicted traffic, and the third predicted traffic. The method for training the neural network may include: inputting the value of the traffic feature of the previous period before the current period into the neural network, and adjusting the parameters of the neural network so that the predicted traffic for the previous period approaches the actual traffic of the current period.

[0006] According to an embodiment of the present disclosure, the embedding layer includes: an attention mechanism network, which is configured to determine the first feature weights of each traffic feature based on the influence of the values ​​of the traffic features in a specified time period on the first predicted traffic; and a long short-term memory network, which is configured to process the values ​​of the traffic features in the previous time period of the current time period after adjustment by the first feature weights to obtain the first predicted traffic.

[0007] According to an embodiment of the present disclosure, traffic features include incoming traffic features; the attention mechanism network includes: a first sub-attention mechanism network, configured to process the traffic features of a specified time period after time sequence sorting for the traffic features of each server in at least one server to determine the traffic sub-feature weight; a second sub-attention mechanism network, configured to determine the server sub-feature weight of each server in at least one server; the long short-term memory network is specifically configured to process the values ​​of the traffic features of the previous time period of the current time period after adjusting the traffic sub-feature weight and the server sub-feature weight, to obtain a first predicted traffic.

[0008] According to an embodiment of the present disclosure, the neural network also includes: a time-based attention mechanism network, which is configured to determine the second feature weights of each traffic feature in adjacent time periods to adjust the value of the traffic feature in the current time period and / or adjust the value of the traffic feature in the previous time period of the current time period.

[0009] According to an embodiment of the present disclosure, the value of the traffic feature is determined in the following manner: unstructured data in the external feature association data is converted into structured data through a pre-built dictionary to obtain external feature data; the external feature data and the incoming traffic data are spliced ​​to obtain spliced ​​data; the spliced ​​data is sorted according to time sequence, and the time dimension is deleted after the sorting is completed to obtain a traffic feature data stream; the traffic feature data stream is normalized to obtain the value of the traffic feature.

[0010] According to an embodiment of the present disclosure, traffic characteristics include: external characteristics and incoming traffic characteristics, and external characteristics include: at least one of weather, holidays, business processing cycle nodes, traffic-related business promotion activities, server identification or time.

[0011] According to an embodiment of the present disclosure, the method for training a neural network further includes: during a single training of the neural network, stopping some nodes of the neural network from outputting processing results.

[0012] One aspect of the present disclosure provides a method for predicting incoming call traffic, comprising: obtaining a value of a traffic feature of a current time period and a value of a traffic feature of a time period before the current time period; and processing the value of the traffic feature of the current time period and the value of the traffic feature of the time period before the current time period using a trained neural network to obtain a predicted traffic; wherein the neural network comprises: an input layer, configured to obtain at least a value of the traffic feature of the current time period and a value of the traffic feature of the time period before the current time period; an embedding layer, configured to determine a first feature weight of each traffic feature, and based on the value of the traffic feature of the time period before the current time period and the feature weights of the traffic characteristics determine the first predicted traffic; a decoding layer is configured to determine the second predicted traffic of the current time period based on the value of the traffic characteristics of the current time period, and determine the third predicted traffic based on the value of the traffic characteristics of the previous time period of the current time period; and an output layer is configured to output the predicted traffic obtained by fusing the first predicted traffic, the second predicted traffic and the third predicted traffic; wherein the neural network is trained in the following manner: the values ​​of the traffic characteristics of the previous time period of the current time period are input into the neural network, and the parameters of the neural network are adjusted so that the predicted traffic for the previous time period is close to the actual traffic of the current time period.

[0013] One aspect of the present disclosure provides an inbound traffic prediction device, comprising: a feature value acquisition module, configured to obtain the value of the traffic feature of the current period and the value of the traffic feature of the previous period before the current period; and a model training module, configured to use a trained neural network to process the value of the traffic feature of the current period and the value of the traffic feature of the previous period before the current period to obtain the predicted traffic; wherein the neural network comprises: an input layer, configured to obtain at least the value of the traffic feature of the current period and the value of the traffic feature of the previous period before the current period; an embedding layer, configured to determine the first feature weight of each traffic feature, and based on the first feature weight of the previous period before the current period, the first feature weight of each traffic feature is obtained. The first predicted flow is determined based on the values ​​of the flow characteristics of the segment and the characteristic weights of the respective flow characteristics; the decoding layer is configured to determine the second predicted flow of the current time period based on the values ​​of the flow characteristics of the current time period, and to determine the third predicted flow based on the values ​​of the flow characteristics of the previous time period of the current time period; and the output layer is configured to output the predicted flow obtained by fusion of the first predicted flow, the second predicted flow and the third predicted flow; wherein the neural network is trained in the following manner: the values ​​of the flow characteristics of the previous time period of the current time period are input into the neural network, and the parameters of the neural network are adjusted so that the predicted flow for the previous time period is close to the actual flow of the current time period.

[0014] Another aspect of the present disclosure provides an electronic device comprising one or more processors and a storage device, wherein the storage device is used to store executable instructions, which, when executed by the processor, implement the above-mentioned method for training a neural network and / or method for predicting incoming call traffic.

[0015] Another aspect of the present disclosure provides a computer-readable storage medium storing computer-executable instructions, which, when executed, are used to implement the above method for training a neural network and / or the method for predicting incoming call traffic.

[0016] Another aspect of the present disclosure provides a computer program, which includes computer-executable instructions. When the instructions are executed, they are used to implement the above method for training a neural network and / or the method for predicting inbound traffic.

[0017] The neural network training method and inbound traffic prediction method and apparatus provided in the embodiments of this disclosure incorporate an attention mechanism in time series prediction scenarios, focusing on traffic features that have a significant impact on predicted traffic, thereby achieving more accurate predictions. Furthermore, by obtaining weights for different dimensions, the external factors that most influence inbound traffic can be identified, allowing the marketing strategies that inbound users are most concerned with to be determined.

[0018] The methods for training neural networks, and the methods and devices for predicting inbound call traffic provided by the embodiments of the present disclosure use LSTM neural networks to better extract periodic features from long time series, considering that LSTM neural networks exhibit superior capabilities in predicting time series data.

[0019] The neural network training method and inbound traffic prediction method and apparatus provided in the embodiments of this disclosure utilize a time-based attention mechanism in the decoder stage to focus on the sequence content that should be emphasized at each moment, as the performance of the encoder-decoder architecture decreases with the length of the encoder sequence. This identifies the time periods that have the greatest impact on the predicted traffic flow, focusing on these periods and achieving more accurate predictions. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The above 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:

[0021] Figure 1 Schematically illustrates an exemplary system architecture to which a method for training a neural network, a method for predicting inbound call traffic, and an apparatus according to an embodiment of the present disclosure can be applied;

[0022] Figure 2 A flowchart of a method for training a neural network according to an embodiment of the present disclosure is schematically shown;

[0023] Figure 3Schematically shows a structural diagram of a neural network according to an embodiment of the present disclosure;

[0024] Figure 4 Schematically shows a structural diagram of an attention mechanism network according to an embodiment of the present disclosure;

[0025] Figure 5 A flow chart for determining the value of a flow characteristic according to an embodiment of the present disclosure is schematically shown;

[0026] Figure 6 A flowchart of a method for predicting incoming call traffic according to an embodiment of the present disclosure is schematically shown;

[0027] Figure 7 Schematically shows a flow chart of a method for predicting incoming call traffic according to another embodiment of the present disclosure;

[0028] Figure 8 A block diagram schematically shows a device for predicting inbound traffic according to an embodiment of the present disclosure; and

[0029] Figure 9 The block diagram of an electronic device according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0030] 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.

[0031] 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 features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0032] 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.

[0033] When using expressions such as "at least one of A, B, or C," they should generally be interpreted in accordance with the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, or C" should include but is not limited to systems having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.). The terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly specifying the number of the indicated technical features. Therefore, features defined as "first" or "second" may explicitly or implicitly include one or more features.

[0034] Telephone customer service is a vital channel for many businesses to conduct business and communicate with their customers. Accurately predicting incoming call volume over a specific period of time can help rationally allocate computing resources, achieving intelligent interaction availability while controlling costs.

[0035] For example, in related technologies, enterprises such as banks can set fixed thresholds for CPU and memory load. When the existing system load exceeds these thresholds, the system is upgraded and expanded. This approach can solve the problem in the short term, but it is not very helpful for long-term development and is detrimental to the company's long-term growth. Technical solutions for overall inbound traffic forecasting generally include trend analysis, regression analysis, and exponential smoothing.

[0036] Trend analysis uses existing data to fit a curve to the historical data to reflect its changing trend, and then predicts a future point based on the changing trend. However, trend analysis has large errors and is only suitable for coarse-grained predictions.

[0037] Regression analysis requires first determining the relationship between the predicted value and its influencing factors, establishing an expression to derive the incoming call volume at a certain point in the future. However, regression analysis has difficulty establishing a clear relationship between the predicted value and the influencing factors in the case of incoming call volume.

[0038] Exponential smoothing uses an exponentially weighted combination of historical data to predict future values ​​of a time series. Setting an attenuation factor of 0 < a < 1, it predicts incoming call volume at a specific point in the future based on the principle of weighting near values ​​over far values. However, while exponential smoothing accounts for dependencies over longer time series, it fails to account for the impact of external factors on the predicted value, and determining the appropriate attenuation factor is difficult.

[0039] To solve the complex prediction problem described above, a neural network approach could be considered, treating incoming call traffic as a purely random number. However, if longer time series are considered, hidden patterns may exist within the time series. Based on extensive statistical analysis conducted by the applicant, it has been found that incoming call traffic exhibits a certain degree of periodicity. For example, small peaks may occur on certain days of the month or during certain times of the day. Furthermore, incoming call traffic may be affected by external factors, such as whether it is a holiday, the time of day, or the approaching credit card payment date. This can also be influenced by marketing strategies.

[0040] The present disclosure provides a method for training a neural network, a method for predicting inbound traffic, and an apparatus. The method for training a neural network includes a training data input process and a model parameter adjustment process. The training data input process includes inputting the values ​​of the traffic characteristics of the current period and the values ​​of the traffic characteristics of the previous period into the neural network. After completing the training data input process, the model parameter adjustment process is entered. By adjusting the parameters of the neural network, the predicted traffic for the previous period is closer to the actual traffic of the current period. The neural network includes: an input layer configured to obtain at least the values ​​of the traffic characteristics of the current period and the values ​​of the traffic characteristics of the previous period; an embedding layer configured to determine a first feature weight of each traffic feature and determine a first predicted traffic based on the values ​​of the traffic characteristics of the previous period and the feature weights of each traffic feature; a decoding layer configured to determine a second predicted traffic for the current period based on the values ​​of the traffic characteristics of the current period and a third predicted traffic based on the values ​​of the traffic characteristics of the previous period; and an output layer configured to output a predicted traffic resulting from the fusion of the first predicted traffic, the second predicted traffic, and the third predicted traffic.

[0041] In the disclosed embodiment, considering that the incoming traffic shows a certain periodicity in time, for example, it will show a small peak on certain days of a month or certain periods of a day; at the same time, the incoming traffic may also be affected by certain external characteristics, such as weather conditions, whether it is a holiday, whether it is close to the credit card repayment date, whether there has been a credit card collection recently... then the incoming call volume will also increase, so there is a certain connection between the incoming call volume and the marketing strategy. Each server will collect traffic data at fixed intervals (for example, 1 hour), so each server will generate a series of time series data. Assume that the number of servers is N g The number of external features collected by each server is N1, so we need to use all server data within the time interval T to predict the N iTraffic values ​​in the following time period. This embodiment uses an encoder-decoder architecture. In time series problem prediction, the excellent capabilities of the LSTM neural network are leveraged to better extract periodic features from long time series. On this basis, an attention mechanism is added to focus on the external features that have the greatest impact on the predicted value and the time period (referred to as the time period) that has the greatest impact on the current value, thereby achieving more accurate predictions. In addition, by obtaining the weights of different dimensions, the external factors that have the greatest impact on inbound traffic, namely the marketing strategies that inbound users are most concerned about, can be understood.

[0042] Figure 1 The following schematically illustrates an exemplary system architecture in which a method for training a neural network, a method for predicting incoming call traffic, and an apparatus for predicting incoming call traffic according to an embodiment of the present disclosure can be applied. Figure 1 What is shown is merely an example of a system architecture to which the embodiments of the present disclosure can be applied, to help those skilled in the art understand the technical content of the present disclosure, but it does not mean that the embodiments of the present disclosure cannot be used in other devices, systems, environments or scenarios. It should be noted that the method for training a neural network, the method and apparatus for predicting incoming call traffic provided by the embodiments of the present disclosure can be used in the field of artificial intelligence in aspects related to predicting incoming call traffic, and can also be used in various fields other than the field of artificial intelligence, such as the financial field. The application fields of the method for training a neural network, the method and apparatus for predicting incoming call traffic provided by the embodiments of the present disclosure are not limited.

[0043] like Figure 1 As shown, the system architecture 100 according to this embodiment may include terminal devices 101, 102, and 103, a network 104, and a server 105. The network 104 may include multiple gateways, routers, hubs, network cables, etc., which are used to provide a medium for communication links between the terminal devices 101, 102, and 103 and the server 105. The network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0044] Users can use terminal devices 101, 102, and 103 to interact with other terminal devices and server 105 via network 104 to receive or send information, such as sending model training instructions, incoming traffic prediction instructions, incoming traffic data, and incoming traffic prediction results. Terminal devices 101, 102, and 103 can be installed with various communication client applications, such as artificial intelligence classification applications, banking applications, e-commerce applications, web browser applications, search applications, office applications, instant messaging tools, email clients, social platform software, and other applications (only as examples).

[0045] Terminal devices 101, 102, and 103 include, but are not limited to, smartphones, desktop computers, augmented reality devices, tablet computers, remote video surveillance terminals, laptop computers, and other electronic devices that can support customer service functions. The terminal devices may include a client to transmit information related to incoming call traffic when users use customer service functions.

[0046] Server 105 can receive model training requests, incoming call traffic prediction requests, model download requests, and process the requests. For example, server 105 can be a backend management server, a server cluster, etc. The backend management server can analyze and process received service requests and information requests, and feed back the processing results (such as incoming call traffic prediction results, model parameters obtained from model training, etc.) to the terminal device.

[0047] It should be noted that the methods for training a neural network and predicting incoming call traffic provided in the embodiments of the present disclosure can be performed by terminal devices 101, 102, 103 or server 105. Accordingly, the apparatus for predicting incoming call traffic provided in the embodiments of the present disclosure can be provided in terminal devices 101, 102, 103 or server 105. It should be understood that the number of terminal devices, networks, and servers is merely illustrative. Any number of terminal devices, networks, and servers may be provided as required.

[0048] Figure 2 The flowchart of the method for training a neural network according to an embodiment of the present disclosure is schematically shown.

[0049] like Figure 2 As shown, the method for training a neural network includes operations S210 to S220.

[0050] In operation S210, the values ​​of the traffic characteristics of the previous period of the current period are input into the neural network.

[0051] Among them, the neural network can include an input layer, an embedding layer, a decoding layer and an output layer.

[0052] The input layer is configured to obtain at least the value of the traffic feature of the current period and the value of the traffic feature of the previous period before the current period.

[0053] The embedding layer is configured to determine a first feature weight of each traffic feature, and determine a first predicted traffic based on the values ​​of the traffic features in the previous period of the current period and the feature weights of each traffic feature.

[0054] The decoding layer is configured to determine a second predicted flow rate for the current period based on a value of a flow characteristic for the current period, and to determine a third predicted flow rate based on a value of a flow characteristic for a period preceding the current period.

[0055] The output layer is configured to output a predicted flow obtained by fusing the first predicted flow, the second predicted flow, and the third predicted flow.

[0056] In some embodiments, the weight information may be determined through an attention mechanism network.

[0057] For example, the embedding layer may include: an attention mechanism network, configured to determine the first feature weights of each traffic feature based on the influence of the values ​​of the traffic features in a specified time period on the first predicted traffic; and a long short-term memory network, configured to process the values ​​of the traffic features in the previous time period of the current time period after adjustment by the first feature weights to obtain the first predicted traffic.

[0058] The traffic characteristics may include external characteristics and incoming traffic characteristics. For example, the external characteristics include at least one of weather, holidays, service processing cycle nodes, traffic-related service promotion activities, server identification, or time.

[0059] In operation S220 , the parameters of the neural network are adjusted so that the predicted traffic volume for the previous period approaches the actual traffic volume for the current period.

[0060] For example, the parameters of the neural network may include parameters of the attention mechanism network. The actual flow rate of the current period may be determined based on the value of the flow characteristics of the current period.

[0061] In this embodiment, inbound call traffic exhibits a certain degree of periodicity over time. For example, inbound call traffic may exhibit small peaks on certain days of the month or during certain times of the day. Furthermore, inbound call traffic may be affected by certain external factors, such as weather conditions, whether it is a weekend, whether a credit card payment is approaching, and whether credit card collection has been conducted recently. These factors can increase inbound call volume. For example, inbound call volume is also related to marketing strategies.

[0062] In addition, in order to make the obtained weight information more accurate, the weight can be split into sub-weights of multiple dimensions, and then the total weight can be obtained through splicing, fusion, etc.

[0063] In some embodiments, the traffic characteristics include inbound traffic characteristics.

[0064] Accordingly, the attention mechanism network may include: a first sub-attention mechanism network and a second sub-attention mechanism network.

[0065] Among them, the first sub-attention mechanism network is configured to process the traffic features of each server in at least one server in a specified time period after being sorted in time sequence to determine the traffic sub-feature weight.

[0066] The second sub-attention mechanism network is configured to determine a server sub-feature weight for each server of the at least one server.

[0067] The long short-term memory network is specifically configured to process the values ​​of the traffic features of the previous period of the current period after being adjusted by the traffic sub-feature weights and the server sub-feature weights, to obtain a first predicted traffic.

[0068] In this embodiment, each server will collect traffic data at a fixed interval (for example, 1 hour), so each server will generate a series of time series data. Assume that the number of servers is N g The number of external features collected by each server is N l , all server data within the time interval T are needed to predict a server N i Traffic values ​​for the next period of time. This example uses a seq2seq encoder-decoder architecture. LSTM neural networks demonstrate superior performance in predicting time series problems. Therefore, this paper employs an LSTM neural network to better extract periodic features from long time series. Furthermore, an attention mechanism is added to focus on the external features that have the greatest impact on the predicted value, achieving more accurate predictions. Furthermore, by obtaining weights for different dimensions, we can understand the external factors that have the greatest impact on inbound traffic, namely, the marketing strategies that inbound users are most concerned about.

[0069] In some embodiments, a time-based attention mechanism may be added to focus on the time step that has the greatest impact on the current value, thereby achieving more accurate prediction.

[0070] Specifically, the neural network may also include: a time-based attention mechanism network, configured to determine the second feature weights of each traffic feature in adjacent time periods to adjust the value of the traffic feature in the current time period and / or adjust the value of the traffic feature in the previous time period of the current time period.

[0071] Figure 3 The schematic diagram of the structure of the neural network according to the embodiment of the present disclosure is shown schematically.

[0072] like Figure 3 As shown in the figure, the encoder model constructed uses the LSTM network model. The Attention network is added before the input of LSTM. The input of LSTM is composed of [X t 本服务器 , X t 全局] encoding. "This server" refers to the input information encoding of the currently predicted server, and "global" refers to the input information encoding of other servers. Both encoding processes use the Attention mechanism.

[0073] The Attention mechanism for this server information is shown in Equations (1) and (2).

[0074]

[0075]

[0076] Among them, v l , b l , w l , U l It is the parameter that the model needs to train, such as l represents the lth server. i，k Refers to the feature vector collected from the k-th dimension external features of the i-th server in a certain period of time. The same operation is performed on all features of all servers to obtain the weight α vector at time t. For example, if the number of servers is N g The number of external features collected by each server is N l V, W, U, b are all weight parameters for model training. [ht-1, St-1] reflects the time series information, where h t-1 yes t-1 The hidden state of the decoder at time t, S t-1 is the output of LSTM at time t-1.

[0077] So there is in Represents the value of the first dimension external feature of the i-th server at time t.

[0078] The Attention mechanism of the global server is shown in Equations (3) and (4).

[0079] It is necessary to calculate the correlation between this server and other servers in the world. Figure 3 As shown, the correlation can be determined based on topology information of the server network, etc.

[0080]

[0081]

[0082] Among them, V g , b g , w g , U g Is the parameter that the model needs to train. lIt refers to the value of other servers in the inbound traffic dimension at time t, λ is a tuning parameter, P i，j As a prior probability, it represents the similarity between servers i and j (such as capacity, geographical distance). As above, l represents the lth server. For example, the number of servers is N g V, W, U, and b are all weight parameters for model training.

[0083] So there is That is, the weighted coding of the incoming traffic information of other servers.

[0084] In summary, in the encoder stage, the input of the LSTM neural network at each time t is:

[0085] Let f(e) represent a LSTM processing unit, then h t =f e (h t-1 , x t ).

[0086] About the decoder stage — time-based attention

[0087] Since the performance of the encoder-decoder architecture decreases with the length of the encoder sequence, a time-based attention mechanism should be added to the decoder stage to focus on the sequence content that should be emphasized at each moment. The formulas for calculating the attention weight at time t in the decoder process are shown in Equations (5) and (6).

[0088]

[0089]

[0090]

[0091] Among them, W d , v d , b d are the training parameters of the model.

[0092] In summary, in the decoder stage, another LSTM network will be used for modeling, and its iterative formula is shown in formula (8).

[0093]

[0094] Among them, ex t It uses the embedding strategy to encode the server's time information and external features for feature mapping. It is the traffic prediction value of server i at time t-1, and its calculation formula is shown in formula (9).

[0095]

[0096] Among them, W m , v y , b m , b y are the training parameters of the model.

[0097] It should be noted that, regarding the input, the encoder's output corresponds to the decoder's output at position i-1. Therefore, the middle attention is not self-attention. Its K (key) and V (value) come from the encoder, and Q (encoder content) comes from the output of the previous decoder.

[0098] Decoding: Encoding can be calculated in parallel and all encoded at once. However, decoding does not decode all sequences at once. Instead, it decodes them one by one like a recurrent neural network (RNN), because the input of the previous position is used as the attention query.

[0099] The attention mechanism network is used to determine the weights of at least two traffic features, the weights of at least two servers, and the weights of at least two time periods.

[0100] Figure 4 The structural diagram of the attention mechanism network according to an embodiment of the present disclosure is schematically shown.

[0101] like Figure 4 As shown in the figure, the attention mechanism network includes: a global pooling layer (Global polling), a first fully connected layer (FC), a normalization layer (BN) and a second fully connected layer (FC).

[0102] Among them, the global average pooling layer is used to obtain the values ​​of at least two traffic features.

[0103] The first fully connected layer is used to determine the weights of at least two traffic features based on an activation function (such as a ReLU function).

[0104] It should be noted that the attention mechanism network can also include a normalization layer and a second fully connected layer.

[0105] The normalization layer is used to normalize the weights of at least two traffic features.

[0106] The second fully connected layer is used to determine the weights of the normalized weights of at least two traffic features based on the loss function (Softmax).

[0107] A weighted attention mechanism is used to select features. Different convolution kernels (e.g., Conv_1 and Conv_2) are used to extract features. The information between different convolution kernels is learned, and the weights of different kernels between channels are assigned for representation. The specific operation is as follows.

[0108] Each element is added while ensuring that the output size and number of channels of different convolution kernels are the same. Global average pooling is used to convert all the flow feature values ​​into a 1×1 feature map to obtain the global information of each channel. The global information S is obtained. c , as shown in formula (10).

[0109]

[0110] Then, output S c The input is fed into the fully connected network, with the goal of finding the information weight of each channel by activating the unit. The nonlinear selection is achieved by using the ReLU activation function to achieve a more accurate weight distribution, while also reducing the number of output neurons, i.e., dimensionality reduction operation. is batch normalization, δ is the ReLU activation function, r is the reduction ratio, L is the minimum length, and Z is the weight, as shown in Equation (11). d is the size after dimensionality reduction, as shown in Equation (12).

[0111]

[0112] d=max(C / r,L) Formula (12)

[0113] Next, the weighted, dimensionality-reduced feature maps obtained by fusion are expanded to their original one-dimensional size through a fully connected layer to represent the weight of each channel. The weight of each feature map is regressed using the softmax function. The sum of the weights of the feature maps of the same channel formed by different convolution kernels is 1, as shown in Equation (13). Finally, the weights are multiplied by the values ​​of the original traffic features and then fused to form the values ​​of the fused traffic features.

[0114]

[0115] The disclosed embodiment adopts a weighted attention mechanism to select features, extracts features by using different convolution kernels, learns the information between different convolution kernels, and then distributes weights to different kernels between channels for characterization, thereby improving the quality of the extracted traffic features.

[0116] The following is an exemplary description of the process of determining the value of the flow characteristic.

[0117] Figure 5 A flowchart for determining the value of a flow characteristic according to an embodiment of the present disclosure is schematically shown.

[0118] like Figure 5 As shown, the method for determining the value of the flow characteristic includes operations S501 to S502.

[0119] In operation S501 , unstructured data in the external feature association data is converted into structured data using a pre-built dictionary to obtain external feature data.

[0120] In operation S502, the external feature data and the incoming traffic data are spliced ​​to obtain spliced ​​data.

[0121] In operation S503 , the spliced ​​data is sorted according to time sequence, and the time dimension is deleted after the sorting is completed to obtain a traffic feature data stream.

[0122] In operation S504, the traffic characteristic data stream is normalized to obtain the value of the traffic characteristic.

[0123] In a specific embodiment, first, raw data is obtained. The data obtained includes incoming call traffic data and external feature data. External feature data may include weather conditions, whether it is a holiday, whether a credit card payment is approaching, whether credit card collection has been carried out recently, and promotion plans for a specific scenario.

[0124] For example, external feature data includes: (moderate rain, holidays, approaching credit card payment deadline, actively promoting credit card limit adjustment scenarios, server ID, 2020-04-01 13:00). Inbound call traffic data (150,000 / h).

[0125] Because some of the external feature data is unstructured, which is not conducive to analysis and processing, it is converted into structured data and a corresponding data dictionary is established to convert it into structured data.

[0126] Example: External feature data: (3, 1, 4, 46, 051, 2020-04-01 13:00) corresponds to (weather, whether it is a holiday, whether it is close to the credit card repayment date, a promotion plan for a certain scenario, server ID, time...)

[0127] Then, through feature processing, highly influential features are obtained from the original data. In this embodiment, time-based feature sequence data is constructed, and the time interval of each data is 1 hour. In order to ensure the accuracy of the prediction, a sufficient amount of data set is obtained and divided into two parts: training data and test data.

[0128] Among them, regarding data preprocessing. The data is analyzed, invalid data is removed, and then the external feature data is spliced ​​with the inbound traffic data. For example, the input data: (3, 1, 4, 46, 2020-04-01 13:00, 051, 150000...) corresponds to ((weather, whether it is a holiday, whether it is close to the credit card repayment date, a certain scenario promotion plan, server ID, time...)), and then constructs a chronological sequence data. Inbound traffic is calculated in hours and sorted in chronological order. After sorting, the time dimension is deleted. For example, the input data a1 2020-04-01 13:00 (3, 1, 4, 46, 51, 150000...) corresponds to the traffic feature a2 2020-04-01 14:00 (2, 1, 4, 34, 51, 120000...).

[0129] In addition, since the magnitudes of different dimensions in the structured data vary significantly, we performed z-score normalization on the data to eliminate the impact of different magnitudes. This involves subtracting the median of the group of observations from the observed value and then dividing it by the standard deviation. This helps improve model training speed and prediction accuracy.

[0130] In some embodiments, the method for training a neural network may further include the following operation: during a single training of the neural network, stopping some nodes of the neural network from outputting processing results.

[0131] In this example, the input data features are processed by the embedding network to reduce their dimensionality. To prevent overfitting of the prediction model due to low-dimensional training data, dropout technology is employed. The embedding layer primarily utilizes a one-dimensional convolutional layer (to extract features), namely the Conv1D layer, a max pooling layer (to reduce the dimensionality of the input data), and a dropout layer (to prevent overfitting). Finally, a Softmax layer is used to obtain the processed important features.

[0132] Specifically, as the number of network layers increases, considering the increasing difficulty of model training, slow convergence, overfitting and other problems, Dropout and BatchNormalization strategies are used to solve these problems. The principle of Dropout is to stop the output of neural units with a pre-set probability when training the network. This "strike" of some neural units means that only a part of the data features are involved in each network training, thereby preventing the network from learning too much data features of the training set and achieving the purpose of preventing overfitting. The principle of BatchNormalization is to use certain normalization methods to convert the distribution of the input value of any neuron in each layer of the neural network into a standard normal distribution with a mean of 0 and a variance of 1. This makes the gradient larger, avoids the occurrence of the gradient vanishing problem, and speeds up the convergence speed.

[0133] Another aspect of the present disclosure provides a method for predicting incoming call traffic.

[0134] Figure 6 The flowchart of the method for predicting incoming call traffic according to an embodiment of the present disclosure is schematically shown.

[0135] like Figure 6 As shown, the method for predicting incoming call traffic may include operations S610 to S620.

[0136] In operation S610, the value of the traffic characteristic of the current period and the value of the traffic characteristic of the previous period before the current period are obtained.

[0137] In operation S620, the trained neural network is used to process the values ​​of the traffic characteristics of the current period and the values ​​of the traffic characteristics of the previous period before the current period to obtain predicted traffic.

[0138] Among them, the neural network includes: input layer, embedding layer, decoding layer and output layer.

[0139] The input layer is configured to obtain at least the value of the traffic feature of the current period and the value of the traffic feature of the previous period before the current period.

[0140] The embedding layer is configured to determine a first feature weight of each traffic feature, and determine a first predicted traffic based on the values ​​of the traffic features in the previous period of the current period and the feature weights of each traffic feature.

[0141] The decoding layer is configured to determine a second predicted flow rate for the current period based on a value of a flow characteristic for the current period, and to determine a third predicted flow rate based on a value of a flow characteristic for a period preceding the current period.

[0142] The output layer is configured to output a predicted flow obtained by fusing the first predicted flow, the second predicted flow, and the third predicted flow.

[0143] For example, the neural network is trained in the following manner: the values ​​of the traffic characteristics of the previous period of the current period are input into the neural network, and the parameters of the neural network are adjusted so that the predicted traffic for the previous period approaches the actual traffic of the current period.

[0144] Figure 7 The flowchart of the method for predicting incoming call traffic according to another embodiment of the present disclosure is schematically shown.

[0145] like Figure 7 As shown, the incoming traffic information can be stored in the system database, and then the data in the system database can be called for model training to obtain a trained Attention-Based LSTM model.

[0146] To assemble hourly inbound traffic and its corresponding external features into a single piece of data, we need to use multi-dimensional time series data to predict inbound traffic. We collect all historical data for a period of time as a dataset.

[0147] Perform structured encoding on the input data, then normalize the data and process the incoming traffic data into normalized sequence data;

[0148] Input the processed multi-dimensional input data into the encoder-decoder structure with attention mechanism to obtain the output;

[0149] If the predicted incoming call traffic exceeds the system's maximum load, the control system automatically limits traffic and sends an alert. For example, if the predicted traffic at the next time point exceeds the system load, the system sends a variable to the server. When the server receives the value of this variable, it automatically executes the traffic limiting code to implement the limit. It can also send an alert.

[0150] Figure 8 The block diagram of the device for predicting inbound flow according to an embodiment of the present disclosure is schematically shown.

[0151] like Figure 8 As shown, the inbound traffic prediction device 800 may include: a feature value acquisition module 810 and a model training module 820.

[0152] The characteristic value acquisition module 810 is used to obtain the value of the flow characteristic of the current period and the value of the flow characteristic of the previous period before the current period.

[0153] The model training module 820 is used to use the trained neural network to process the values ​​of the traffic characteristics of the current period and the values ​​of the traffic characteristics of the previous period before the current period to obtain the predicted traffic.

[0154] Among them, the neural network includes: input layer, embedding layer, decoding layer and output layer.

[0155] The input layer is configured to obtain at least the value of the traffic feature of the current period and the value of the traffic feature of the previous period before the current period.

[0156] The embedding layer is configured to determine a first feature weight of each traffic feature, and determine a first predicted traffic based on the values ​​of the traffic features in the previous period of the current period and the feature weights of each traffic feature.

[0157] The decoding layer is configured to determine a second predicted flow rate for the current period based on a value of a flow characteristic for the current period, and to determine a third predicted flow rate based on a value of a flow characteristic for a period preceding the current period.

[0158] The output layer is configured to output a predicted flow obtained by fusing the first predicted flow, the second predicted flow, and the third predicted flow.

[0159] The neural network is trained in the following way: the values ​​of the traffic characteristics of the previous period of the current period are input into the neural network, and the parameters of the neural network are adjusted so that the predicted traffic for the previous period approaches the actual traffic of the current period.

[0160] It should be noted that the implementation methods, technical problems solved, functions realized, and technical effects achieved of each module / unit in the device part embodiment are the same or similar to the implementation methods, technical problems solved, functions realized, and technical effects achieved of each corresponding step in the method part embodiment, and will not be repeated here.

[0161] According to the modules and units of the embodiments of the present invention, any multiple or at least part of the functions of any multiple thereof can be implemented in one module. According to the modules and units of the embodiments of the present invention, any one or more can be split into multiple modules for implementation. According to the modules and units of the embodiments of the present invention, any one or more 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 by hardware or firmware of any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware and firmware or in an appropriate combination of any several thereof. Alternatively, according to the modules and units of the embodiments of the present invention, one or more can be at least partially implemented as a computer program module, which can perform the corresponding function when the computer program module is run.

[0162] For example, any multiple of the feature value acquisition module 810 and the model training module 820 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present disclosure, at least one of the feature value acquisition module 810 and the model training module 820 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 by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware and firmware or in an appropriate combination of any of them. Alternatively, at least one of the feature value acquisition module 810 and the model training module 820 can be at least partially implemented as a computer program module, which can perform the corresponding function when the computer program module is run.

[0163] Figure 9 The block diagram of an electronic device according to an embodiment of the present disclosure is schematically shown. Figure 9 The electronic device shown is only an example and should not limit the functions and scope of use of the embodiments of the present disclosure.

[0164] like Figure 9 As shown, the electronic device 900 according to an embodiment of the present disclosure includes a processor 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage part 908 into a random access memory (RAM) 903. The processor 901 may, for example, include a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a dedicated microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 901 may also include an onboard memory for caching purposes. The processor 901 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.

[0165] Various programs and data required for the operation of the electronic device 900 are stored in the RAM 903. The processor 901, ROM 902, and RAM 903 are communicatively connected to each other via a bus 904. The processor 901 executes the programs in the ROM 902 and / or RAM 903 to perform various operations of the method flow according to the embodiment of the present disclosure. It should be noted that the programs may also be stored in one or more memories other than the ROM 902 and RAM 903. The processor 901 may also execute the programs stored in one or more memories to perform various operations of the method flow according to the embodiment of the present disclosure.

[0166] According to an embodiment of the present disclosure, the electronic device 900 may further include an input / output (I / O) interface 905, which is also connected to the bus 904. The electronic device 900 may further include one or more of the following components connected to the I / O interface 905: an input portion 906 including a keyboard, a mouse, etc.; an output portion 907 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker; a storage portion 908 including a hard disk, etc.; and a communication portion 909 including a network interface card such as a LAN card or a modem. The communication portion 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the I / O interface 905 as needed. A removable medium 911, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed in the drive 910 as needed, so that a computer program read therefrom can be installed into the storage portion 908 as needed.

[0167] According to an embodiment of the present disclosure, the method flow according to an embodiment of the present disclosure can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable storage medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 909, and / or installed from the removable medium 911. When the computer program is executed by the processor 901, the above-mentioned functions defined in the system of the embodiment of the present disclosure are executed. According to an embodiment of the present disclosure, the system, equipment, device, module, unit, etc. described above can be implemented by a computer program module.

[0168] 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.

[0169] According to an embodiment of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, it may include but is 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 containing or storing a program that can be used by or in combination 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 902 and / or RAM 903 described above and / or one or more memories other than ROM 902 and RAM 903.

[0170] An embodiment of the present disclosure also includes a computer program product, which includes a computer program containing program code for executing the method provided by the embodiment of the present disclosure. When the computer program product is run on an electronic device, the program code is used to enable the electronic device to implement the image model training method or the method for predicting incoming traffic provided by the embodiment of the present disclosure.

[0171] When the computer program is executed by the processor 901, the above functions defined in the system / device of the embodiment of the present disclosure are performed. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0172] 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 909, and / or installed from a removable medium 911. 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.

[0173] 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).

[0174] Those skilled in the art will understand that the features described in the various embodiments and / or claims of the present disclosure may be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in the present disclosure. These embodiments are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Although the various embodiments have been described above separately, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. The scope of the present disclosure is defined by the appended claims and their equivalents. 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 method for training a neural network, the neural network comprising: An input layer configured to obtain at least a value of a flow characteristic of a current period and a value of a flow characteristic of a period before the current period; an embedding layer configured to determine a first feature weight of each of the traffic features, and determine a first predicted traffic based on values ​​of the traffic features in a previous period of the current period and the feature weights of each of the traffic features; A decoding layer, configured to determine a second predicted flow rate for a current period based on a value of a flow characteristic for the current period, and to determine a third predicted flow rate based on a value of a flow characteristic for a period previous to the current period; as well as an output layer, configured to output a predicted flow obtained by fusing the first predicted flow, the second predicted flow, and the third predicted flow; The method for training a neural network comprises: Inputting the values ​​of the traffic characteristics of the previous period of the current period into the neural network, adjusting the parameters of the neural network so that the predicted traffic for the previous period approaches the actual traffic of the current period, and outputting the predicted traffic obtained by fusing the first predicted traffic, the second predicted traffic, and the third predicted traffic. Wherein, the embedding layer includes: an attention mechanism network configured to determine a first feature weight of each of the traffic features based on an influence of a value of the traffic feature in a specified time period on the first predicted traffic; and A long short-term memory network is configured to process the value of the traffic feature of the previous period of the current period after the first feature weight is adjusted to obtain the first predicted traffic, And wherein, the traffic characteristics include: external characteristics and incoming traffic characteristics, and the external characteristics include: at least one of weather, holidays, business processing cycle nodes, traffic-related business promotion activities, server identification or time.

2. The method according to claim 1, wherein The traffic characteristics include inbound traffic characteristics; The attention mechanism network includes: The first sub-attention mechanism network is configured to process the traffic features of each server in the at least one server in a specified time period after being sorted in time sequence to determine a traffic sub-feature weight; a second sub-attention mechanism network configured to determine a server sub-feature weight for each server of the at least one server; and The long short-term memory network is specifically configured to process the values ​​of the traffic features of the previous period of the current period adjusted by the traffic sub-feature weights and the server sub-feature weights respectively, to obtain the first predicted traffic.

3. The method according to claim 1 or 2, wherein: The neural network further comprises: The time-based attention mechanism network is configured to determine the second feature weights of each of the traffic features in adjacent time periods to adjust the value of the traffic feature of the current time period and / or adjust the value of the traffic feature of the previous time period of the current time period.

4. The method according to claim 1 or 2, wherein: The value of the flow characteristic is determined as follows: Converting unstructured data in the external feature association data into structured data using a pre-built dictionary to obtain external feature data; Splicing the external feature data and the incoming traffic data to obtain spliced ​​data; Sorting the spliced ​​data in time sequence, and deleting the time dimension after the sorting is completed, to obtain a traffic feature data stream; as well as Normalization is performed on the flow characteristic data stream to obtain the value of the flow characteristic.

5. The method according to claim 1 or 2, wherein the method for training a neural network further comprises: During a single training of the neural network, some nodes of the neural network are stopped from outputting processing results.

6. A method for predicting incoming call traffic, comprising: Obtaining the value of the flow characteristic of the current period and the value of the flow characteristic of the period before the current period; as well as Using a trained neural network to process the values ​​of the flow characteristics of the current period and the values ​​of the flow characteristics of the period before the current period to obtain a predicted flow; Wherein, the neural network includes: An input layer configured to obtain at least a value of a flow characteristic of a current period and a value of a flow characteristic of a period before the current period; an embedding layer configured to determine a first feature weight of each of the traffic features, and determine a first predicted traffic based on values ​​of the traffic features in a previous period of the current period and the feature weights of each of the traffic features; a decoding layer configured to determine a second predicted flow rate for a current period based on a value of a flow characteristic for the current period, and to determine a third predicted flow rate based on a value of a flow characteristic for a period preceding the current period; and an output layer, configured to output a predicted flow obtained by fusing the first predicted flow, the second predicted flow, and the third predicted flow; The neural network is trained in the following way: Inputting the values ​​of the flow characteristics of the previous period of the current period into the neural network, and adjusting the parameters of the neural network so that the predicted flow for the previous period approaches the actual flow of the current period, Outputting a predicted flow obtained by fusing the first predicted flow, the second predicted flow, and the third predicted flow, Wherein, the embedding layer includes: an attention mechanism network configured to determine a first feature weight of each of the traffic features based on an influence of a value of the traffic feature in a specified time period on the first predicted traffic; and A long short-term memory network is configured to process the value of the traffic feature of the previous period of the current period after the first feature weight is adjusted to obtain the first predicted traffic, And wherein, the traffic characteristics include: external characteristics and incoming traffic characteristics, and the external characteristics include: at least one of weather, holidays, business processing cycle nodes, traffic-related business promotion activities, server identification or time.

7. A device for predicting incoming call flow, comprising: A characteristic value acquisition module, used to obtain the value of the flow characteristic of the current period and the value of the flow characteristic of the previous period before the current period; as well as A model training module, configured to process the values ​​of the flow characteristics of the current period and the values ​​of the flow characteristics of the period before the current period using a trained neural network to obtain a predicted flow; Wherein, the neural network includes: An input layer configured to obtain at least a value of a flow characteristic of a current period and a value of a flow characteristic of a period before the current period; an embedding layer configured to determine a first feature weight of each of the traffic features, and determine a first predicted traffic based on values ​​of the traffic features in a previous period of the current period and the feature weights of each of the traffic features; a decoding layer configured to determine a second predicted flow rate for a current period based on a value of a flow characteristic for the current period, and to determine a third predicted flow rate based on a value of a flow characteristic for a period preceding the current period; and an output layer, configured to output a predicted flow obtained by fusing the first predicted flow, the second predicted flow, and the third predicted flow; The neural network is trained in the following way: Inputting the values ​​of the traffic characteristics of the previous period of the current period into the neural network, adjusting the parameters of the neural network so that the predicted traffic for the previous period approaches the actual traffic of the current period, and outputting the predicted traffic obtained by fusing the first predicted traffic, the second predicted traffic, and the third predicted traffic. Wherein, the embedding layer includes: an attention mechanism network configured to determine a first feature weight of each of the traffic features based on an influence of a value of the traffic feature in a specified time period on the first predicted traffic; and A long short-term memory network is configured to process the value of the traffic feature of the previous period of the current period after the first feature weight is adjusted to obtain the first predicted traffic, And wherein, the traffic characteristics include: external characteristics and incoming traffic characteristics, and the external characteristics include: at least one of weather, holidays, business processing cycle nodes, traffic-related business promotion activities, server identification or time.

8. An electronic device comprising: one or more processors; A storage device for storing executable instructions, wherein when the executable instructions are executed by the processor, the method for training a neural network according to any one of claims 1 to 5 is implemented, or the method for predicting incoming call traffic according to claim 6 is implemented.

9. A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, implement the method for training a neural network according to any one of claims 1 to 5, or implement the method for predicting inbound traffic according to claim 6.

Images