Method, device and equipment for predicting cash deposit amount of automatic telling machine, medium and product

Abstract

The invention discloses a method, a device, equipment, a medium and a product for predicting the cash deposit amount of an automatic telling machine, and relates to the technical field of artificial intelligence, and the method comprises the steps: generating a target time enhancement feature according to a target time feature and a target static feature; according to the target initial spatial-temporal characteristics corresponding to the target automatic telling machine and other initial spatial-temporal characteristics corresponding to other automatic telling machines, determining spatial correlation coefficients between other automatic telling machines and the target automatic telling machine; and according to the spatial correlation coefficient and other initial spatial-temporal features, determining a target spatial-temporal fusion feature corresponding to the target automatic telling machine, and according to the target spatial-temporal fusion feature, predicting a target cash storage amount corresponding to the target automatic telling machine at the current time point. According to the invention, the time feature, the space feature and the static feature which influence the cash deposit amount of the automatic telling machine are combined and are jointly used for predicting the cash deposit amount of the automatic telling machine, so that the accuracy and the reliability of predicting the cash deposit amount of the automatic telling machine are ensured, and a bank can reasonably configure cash resources and effectively prevent risks.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method, device, equipment, medium and product for predicting the cash deposit amount of a cash dispenser. Background Art

[0002] In recent years, people's living standards have been improving day by day, deposit and withdrawal transactions at ATMs have become more frequent, and the corresponding ATM market has also been developing rapidly. In order to ensure that users can withdraw cash in real time, banks will predict the cash storage volume of ATMs and put cash into ATMs based on the predicted cash storage volume.

[0003] The existing technology usually predicts the future cash deposit volume by analyzing the historical cash deposit volume trend of the ATM. However, there are many uncertain factors in the cash deposit volume of the ATM. The accuracy and reliability of the cash deposit volume prediction cannot be guaranteed by relying solely on the historical cash deposit volume trend. Summary of the invention

[0004] The present invention provides a method, device, equipment, medium and product for predicting the cash deposit amount of a cash dispenser, so as to solve the problem that the prior art relies on historical cash deposit amount trends to predict the cash deposit amount, and cannot guarantee the accuracy and reliability of the cash deposit amount prediction.

[0005] According to one aspect of the present invention, a method for predicting the amount of cash deposited in a cash dispenser is provided, the method comprising:

[0006] Acquire at least one target time feature and at least one target static feature corresponding to the current time point, and generate a target time enhancement feature according to the target time feature and the target static feature; wherein the target static feature includes at least one of a user behavior feature, a payment method feature, and a withdrawal fee feature;

[0007] According to the target initial spatiotemporal features corresponding to the target cash dispenser and other initial spatiotemporal features corresponding to at least one other cash dispenser, determine the spatial correlation coefficients between each of the other cash dispensers and the target cash dispenser; wherein the target initial spatiotemporal features are determined according to at least one target spatial feature and the target time enhancement feature corresponding to the target cash dispenser, and the other initial spatiotemporal features are determined according to at least one other spatial feature and the target time enhancement feature corresponding to the other cash dispensers;

[0008] According to each of the spatial correlation coefficients and each of the other initial spatiotemporal features, the target spatiotemporal fusion features corresponding to the target ATM are determined, and the target cash deposit amount corresponding to the target ATM at the current time point is predicted according to the target spatiotemporal fusion features.

[0009] According to another aspect of the present invention, a device for predicting the amount of cash deposited in a cash dispenser is provided, the device comprising:

[0010] A time enhancement feature generation module, used to obtain at least one target time feature and at least one target static feature corresponding to the current time point, and generate a target time enhancement feature according to the target time feature and the target static feature; wherein the target static feature includes at least one of a user behavior feature, a payment method feature, and a withdrawal fee feature;

[0011] A spatial correlation coefficient determination module, used to determine the spatial correlation coefficient between each of the other cash dispensers and the target cash dispenser according to the target initial spatiotemporal characteristics corresponding to the target cash dispenser and other initial spatiotemporal characteristics corresponding to at least one other cash dispenser; wherein the target initial spatiotemporal characteristics are determined according to at least one target spatial characteristic and the target time enhancement characteristic corresponding to the target cash dispenser, and the other initial spatiotemporal characteristics are determined according to at least one other spatial characteristic and the target time enhancement characteristic corresponding to the other cash dispenser;

[0012] The cash deposit amount prediction module is used to determine the target spatiotemporal fusion characteristics corresponding to the target ATM according to each of the spatial correlation coefficients and each of the other initial spatiotemporal characteristics, and predict the target cash deposit amount corresponding to the target ATM at the current time point according to the target spatiotemporal fusion characteristics.

[0013] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0014] at least one processor; and

[0015] a memory communicatively connected to the at least one processor; wherein,

[0016] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the method for predicting the cash deposit amount of an ATM according to any one of the present inventions.

[0017] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement any method for predicting the cash deposit amount of an ATM according to the present invention when executed.

[0018] According to another aspect of the present invention, there is provided a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method for predicting the cash deposit amount of a cash dispenser according to any one of the present inventions is implemented.

[0019] The present invention realizes the effect of combining the time characteristics, spatial characteristics and static characteristics that affect the cash deposit amount of the ATM, and jointly using them to predict the cash deposit amount of the ATM, thereby ensuring the accuracy and reliability of the cash deposit amount prediction of the ATM, which is beneficial for the bank to rationally allocate cash resources and effectively prevent risks.

[0020] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 A flowchart of a method for predicting the amount of cash deposited in a cash machine provided in the first embodiment of the present invention;

[0023] Figure 2 A flowchart of a method for predicting the amount of cash deposited in a cash machine provided in Embodiment 2 of the present invention;

[0024] Figure 3 A flowchart of a method for generating initial spatiotemporal features of a target provided in Embodiment 3 of the present invention;

[0025] Figure 4 A flowchart of a method for generating target spatiotemporal fusion features provided in Embodiment 4 of the present invention;

[0026] Figure 5 A schematic diagram of the structure of a device for predicting the amount of cash deposited in a cash machine provided in Embodiment 5 of the present invention;

[0027] Figure 6 The present invention is a schematic diagram of the structure of an electronic device for realizing the method for predicting the cash deposit amount of a cash dispenser according to an embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0029] It should be noted that the terms "candidate", "target", "first", "second", "third", "current", "future", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0030] Embodiment 1

[0031] Figure 1 This is a flowchart of a method for predicting the amount of cash stored in a cash machine provided in the first embodiment of the present invention. This embodiment can be applied to predicting the amount of cash stored in a cash machine by using three types of feature dimensions: time feature, space feature and static feature. This method can be executed by a prediction device for the amount of cash stored in a cash machine. The prediction device for the amount of cash stored in a cash machine can be implemented in the form of hardware and / or software. Figure 1 As shown, the method includes:

[0032] S101. Acquire at least one target time feature and at least one target static feature corresponding to the current time point, and generate a target time enhancement feature according to the target time feature and the target static feature.

[0033] The current time point refers to the current real-time time point, that is, the time point corresponding to the cash deposit amount of the ATM that needs to be predicted. The target time feature refers to the relevant time feature that can affect the cash deposit amount of the ATM (also known as the cash demand), including but not limited to seasonal factor features, economic factor features, and weather factor features.

[0034] Seasonal factors include, but are not limited to, whether it is a holiday, whether it is a weekend, whether it is a peak tourist season during winter and summer vacations, etc. It is understandable that different seasonal characteristics will lead to different cash usage needs of consumers, which will directly affect the cash storage volume of ATMs. Economic factors include, but are not limited to changes in financial policies, exchange rates, price levels, etc. It is understandable that different economic factors will lead to different degrees of withdrawal needs of consumers, which will directly affect the cash storage volume of ATMs. Weather factors include, but are not limited to temperature, wind speed, precipitation, air pollution index, etc. It is understandable that different weather factors will lead to different frequencies of going out for consumers, further affecting the use rate of ATMs by consumers, which will directly affect the cash storage volume of ATMs. For example, bad weather such as heavy snow and typhoons may cause people to go out less and reduce the use rate of ATMs; on the contrary, after the weather improves, there may be a concentrated rebound in demand.

[0035] The target static features include at least one of user behavior features, payment method features, and withdrawal fee features. User behavior features include, but are not limited to, cash usage habits, consumption types, salary payment dates, etc. It is understandable that different user behavior features will affect consumers' usage rate of ATMs, thereby directly affecting the amount of cash stored in ATMs. Payment method features include, but are not limited to the popularity of electronic payments, etc. It is understandable that different payment method features will affect consumers' reliance on cash, thereby directly affecting the amount of cash stored in ATMs. For example, the popularity of electronic payments has reduced people's reliance on cash, and the amount of cash stored in ATMs has decreased. Withdrawal fee features include, but are not limited to, withdrawal fees, inter-bank fees, withdrawal limits, etc. It is understandable that withdrawal fee features will affect consumers' cash withdrawal habits, thereby directly affecting the amount of cash stored in ATMs.

[0036] It should be noted that the relevant information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for display, data for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data comply with relevant laws, regulations and standards in relevant regions.

[0037] In one embodiment, at least one target time feature and at least one target static feature corresponding to the current time point are obtained. Further, the target static feature is input into a static feature encoder, and the target static feature is encoded using the static feature encoder to generate a context feature, wherein the static feature encoder is constructed according to a gated residual network (Gated Residual Network, GRN). Further, the target time feature and the context feature are jointly input into a pre-generated gated residual network, and the gated residual network is used to process the target time feature and the context feature, and output the target time enhancement feature. It can be understood that, since there is a significant influence between the target static feature and the target time feature, the target time feature is enhanced using the target static feature, which can ensure that the generated target time enhancement feature contains richer feature information.

[0038] S102. Determine the spatial correlation coefficient between each other cash dispenser and the target cash dispenser according to the target initial spatiotemporal characteristics corresponding to the target cash dispenser and other initial spatiotemporal characteristics corresponding to at least one other cash dispenser.

[0039] Among them, the target ATM refers to an ATM (Automated Teller Machine) with a predicted cash storage demand. Other ATMs refer to ATMs other than the target ATM. The target initial spatiotemporal features refer to the features obtained by combining the time features and the spatial features of the target ATM, and the other initial spatiotemporal features refer to the features obtained by combining the time features and the spatial features of other ATMs. Specifically, the target initial spatiotemporal features are determined based on at least one target spatial feature and target time enhancement feature corresponding to the target ATM, and the other initial spatiotemporal features are determined based on at least one other spatial feature and target time enhancement feature corresponding to other ATMs.

[0040] Spatial characteristics include, but are not limited to, geographical location characteristics and traffic condition characteristics. It is understandable that different geographical location characteristics will lead to different usage frequencies and cash demands for ATMs, which will directly affect the amount of cash stored in the ATMs. For example, in areas with large traffic flow, such as commercial centers and tourist attractions, the frequency of ATM usage is high, and the demand for cash will increase accordingly; while in some residential areas or rural areas, the frequency of ATM usage may be closely related to daily life, and the demand for cash is relatively stable. It is understandable that different traffic condition characteristics will lead to different usage frequencies for ATMs, which will directly affect the amount of cash stored in the ATMs. For example, ATMs located in convenient transportation locations, especially in subway stations, bus stations, airports and other places, usually have a higher frequency of use.

[0041] The spatial correlation coefficient reflects the topological relationship between each other cash machine and the target cash machine. For example, the larger the spatial correlation coefficient between any other cash machine and the target cash machine, the greater the importance of the other cash machine to the target cash machine, that is, the closer the topological relationship.

[0042] In one embodiment, feature splicing is performed based on at least one target spatial feature and target time enhancement feature corresponding to the target ATM to generate the target initial spatiotemporal feature corresponding to the target ATM; and feature splicing is performed based on at least one other spatial feature and target time enhancement feature corresponding to each other ATM to generate other initial spatiotemporal features corresponding to each other ATM.

[0043] Feature splicing is performed according to the target initial spatiotemporal features and other initial spatiotemporal features, and the feature splicing results are input into the graph attention network. The output results of the graph attention network are further processed by the activation function to determine the spatial correlation coefficients between each other ATM and the target ATM.

[0044] S103. Determine the target spatiotemporal fusion features corresponding to the target ATM according to the spatial correlation coefficients and other initial spatiotemporal features, and predict the target cash deposit amount corresponding to the target ATM at the current time point according to the target spatiotemporal fusion features.

[0045] In one embodiment, a weighted sum is performed based on each spatial correlation coefficient and each other initial spatiotemporal feature, and the target spatiotemporal fusion feature corresponding to the target cash machine is determined based on the weighted summation result. Further, the cash deposit amount is predicted using a pre-trained prediction model based on the target spatiotemporal fusion feature to determine the target cash deposit amount corresponding to the target cash machine at the current time point.

[0046] Optionally, since the cash deposit amount of the ATM fluctuates greatly, a conformal prediction method is introduced to predict the fluctuation range of the cash deposit amount.

[0047] In the training phase, the conditional distribution is constructed using training samples with known labels, and then confidence intervals are calculated based on this distribution. In the prediction phase, the confidence interval of each new sample is calculated and marked as "conform to the training distribution" or "does not conform to the training distribution" based on the confidence level. In time series, additional calibration sets and base models are usually introduced to form a new method called Inductive Conformal Prediction (ICP).

[0048] Specifically, the ICP method operates by splitting the training set into a proper training set of size n and a calibration set of size m. The proper training set is used to train the base model M, while the calibration set is used to calculate the unacceptability score, which is a measure of how abnormal a given sample is compared to previously observed data.

[0049] The resulting empirical failure score distribution is used to calculate the key failure score ε, which corresponds to the [(m+1)(1+α)]th minimum residual. i+1 , target cash deposit amount y i+1 The prediction interval is [y i+1 -ε,y i+1 +ε].

[0050] The embodiment of the present invention combines the time characteristics, spatial characteristics and static characteristics that affect the cash deposit amount of the ATM, and uses them together to predict the cash deposit amount of the ATM, thereby ensuring the accuracy and reliability of the cash deposit amount prediction of the ATM, which is beneficial for the bank to rationally allocate cash resources and effectively prevent risks.

[0051] Embodiment 2

[0052] Figure 2 This is a flowchart of a method for predicting the amount of cash deposited in a cash machine provided by Embodiment 2 of the present invention. This embodiment further optimizes and expands the above embodiment and can be combined with the above optional implementations. Figure 2 As shown, the method includes:

[0053] S201. Input at least one candidate static feature into a first gated residual network, and determine first selection weight values ​​corresponding to each candidate static feature according to an output result of the first gated residual network.

[0054] Among them, the candidate static features refer to the feature set of all static features that may affect the cash deposit amount prediction. The first selection weight value reflects the influence of the candidate static features on the cash deposit amount prediction. It can be understood that the larger the first selection weight value corresponding to any candidate static feature, the greater the influence of the candidate static feature on the cash deposit amount prediction.

[0055] In one implementation, the first selection weight value is determined using the following formula:

[0056] v1=Softmax(GRN1(x1,c1))

[0057] Among them, x1 represents any candidate static feature, GRN1 represents the first gated residual network, v1 represents the first selection weight value corresponding to the candidate static feature, and Softmax is used to standardize the output of GRN1. Since the calculation of the first selection weight value only depends on the candidate static feature, the context feature c1 at this time is zero.

[0058] S202: Input each candidate static feature into a second gated residual network, and determine a first feature variable value corresponding to each candidate static feature according to an output result of the second gated residual network.

[0059] In one embodiment, the first characteristic variable value is determined using the following formula:

[0060] X1=GRN2(x1,c2)

[0061] Wherein, x1 represents any candidate static feature, GRN2 represents the second gated residual network, and X1 represents the first feature variable value corresponding to the candidate static feature. Since the calculation of the first feature variable value only depends on the candidate static feature, the context feature c2 at this time is zero.

[0062] S203: Determine first weighted feature variable values ​​corresponding to each candidate static feature according to each first feature variable value and each first selection weight value, and determine a target static feature from each candidate static feature according to each first weighted feature variable value.

[0063] In one implementation, the following formula is used to determine the first weighted characteristic variable value:

[0064] A1=v1X1

[0065] Among them, v1 represents the first selection weight value corresponding to any candidate static feature, X1 represents the first feature variable value corresponding to the candidate static feature, and A1 represents the first weighted feature variable value corresponding to the candidate static feature.

[0066] After determining the first weighted feature variable values ​​corresponding to each candidate static feature, according to the ranking results of the first weighted feature variable values, select the target number of candidate static features with larger first weighted feature variable values ​​as the target static features. For example, select the top five candidate static features in the ranking of the first weighted feature variable values ​​as the target static features.

[0067] By inputting at least one candidate static feature into the first gated residual network, and determining the first selection weight value corresponding to each candidate static feature according to the output result of the first gated residual network; inputting each candidate static feature into the second gated residual network, and determining the first feature variable value corresponding to each candidate static feature according to the output result of the second gated residual network; determining the first weighted feature variable value corresponding to each candidate static feature according to each first feature variable value and each first selection weight value, and determining the target static feature from each candidate static feature according to each first weighted feature variable value, the candidate static features that are not important for the prediction of the cash deposit amount can be eliminated, that is, the candidate static features that generate unnecessary noise for the prediction of the cash deposit amount can be eliminated, thereby improving the accuracy and efficiency of the cash deposit amount prediction.

[0068] S204: Input at least one candidate time feature and the first context feature into a first gated residual network, and determine second selection weight values ​​corresponding to each candidate time feature according to an output result of the first gated residual network.

[0069] The first context feature is determined according to the target static feature.

[0070] In one implementation, the second selection weight value is determined using the following formula:

[0071] v2=Softmax(GRN1(x2,c3))

[0072] Among them, x2 represents any candidate temporal feature, GRN1 represents the first gated residual network, v2 represents the second selection weight value corresponding to the candidate temporal feature, Softmax is used to standardize the output of GRN1, and c3 represents the target static feature.

[0073] S205: Input each candidate time feature into a third gated residual network, and determine the second feature variable value corresponding to each candidate time feature according to the output result of the third gated residual network.

[0074] In one embodiment, the second characteristic variable value is determined using the following formula:

[0075] X2=GRN3(x2,c4)

[0076] Wherein, x2 represents any candidate time feature, GRN3 represents the third gated residual network, and X2 represents the second feature variable value corresponding to the candidate time feature. Since the calculation of the second feature variable value only depends on the candidate time feature, the context feature c4 at this time is zero.

[0077] S206: Determine the second weighted feature variable values ​​corresponding to the candidate time features according to the second feature variable values ​​and the second selection weight values, and determine the target time feature from the candidate time features according to the second weighted feature variable values.

[0078] In one embodiment, the second weighted characteristic variable value is determined using the following formula:

[0079] A2=v2X2

[0080] Among them, v2 represents the second weighted feature variable value corresponding to any candidate time feature, X2 represents the second feature variable value corresponding to the candidate time feature, and A2 represents the second weighted feature variable value corresponding to the candidate time feature.

[0081] After determining the second weighted feature variable values ​​corresponding to each candidate time feature, select the target number of candidate time features with larger second weighted feature variable values ​​as the target time features according to the ranking results of the second weighted feature variable values. For example, select the top five candidate time features in the ranking of the second weighted feature variable values ​​as the target time features.

[0082] By inputting at least one candidate time feature and a first context feature into a first gated residual network, and determining the second selection weight value corresponding to each candidate time feature according to the output result of the first gated residual network; wherein the first context feature is determined according to the target static feature; inputting each candidate time feature into a third gated residual network, and determining the second feature variable value corresponding to each candidate time feature according to the output result of the third gated residual network; determining the second weighted feature variable value corresponding to each candidate time feature according to each second feature variable value and each second selection weight value, and determining the target time feature from each candidate time feature according to each second weighted feature variable value, the candidate time features that are not important for the prediction of the cash deposit amount can be eliminated, that is, the candidate time features that generate unnecessary noise for the prediction of the cash deposit amount can be eliminated, thereby improving the accuracy and efficiency of the cash deposit amount prediction.

[0083] S207. Generate target time enhancement features based on target time features and target static features, and determine the spatial correlation coefficients between each other ATM and the target ATM based on the target initial time-space features corresponding to the target ATM and other initial time-space features corresponding to at least one other ATM.

[0084] Optionally, a target time enhancement feature is generated according to the target time feature and the target static feature, including:

[0085] A1. Perform position encoding on each target time feature to generate a first encoding feature, and perform timestamp encoding on each target time feature to generate a second encoding feature.

[0086] Among them, Positional Encoding (PE) is used to obtain the order relationship of time features, and Timestamp Encoding (TE) is used to capture the impact of time features at a specific time point.

[0087] B1. Generate a target coding feature according to the first coding feature and the second coding feature, input the target coding feature and the second context feature into a fourth gated residual network, and determine a target time enhancement feature according to an output result of the fourth gated residual network.

[0088] The second context feature is generated according to the target static feature.

[0089] In one embodiment, the target time enhancement feature is determined using the following formula:

[0090] θ(t)=GRN4(x t ,c ep ,c ek )

[0091] Among them, x t represents the target encoding feature; c ep ,c ek represents the second context feature, which is generated according to the target static feature; GRN4 represents the fourth gated residual network, and GRN4 shares weights within the entire layer; θ(t) represents the target time enhancement feature.

[0092] By performing position encoding on each target time feature to generate a first encoding feature, and performing timestamp encoding on each target time feature to generate a second encoding feature, the sequential relationship of the time features is obtained, as well as the effect of capturing the impact of the time features at a specific time point, thereby achieving the effect of obtaining the time characteristics of the target time features, which is beneficial to improving the accuracy of cash deposit prediction.

[0093] Since there is an influence between static features and time features, a target coding feature is generated based on the first coding feature and the second coding feature, and the target coding feature and the second context feature are input into the fourth gated residual network. According to the output result of the fourth gated residual network, the target time enhancement feature is determined, so that the target time enhancement feature contains both time feature information and static feature information, thereby enhancing the feature information expression of the target time enhancement feature, which is beneficial to improving the accuracy of cash deposit prediction.

[0094] Optionally, according to the target initial spatiotemporal features corresponding to the target cash dispenser and other initial spatiotemporal features corresponding to at least one other cash dispenser, determining the spatial correlation coefficients between each other cash dispenser and the target cash dispenser, respectively, includes:

[0095] A2. Generate a first spatiotemporal weighted feature according to the target ATM weight value corresponding to the target ATM and the target initial spatiotemporal feature, and generate a second spatiotemporal weighted feature according to the other ATM weight values ​​corresponding to each other ATM and each other initial spatiotemporal feature.

[0096] For example, assuming that the target ATM weight value corresponding to the target ATM i is W i , the target initial spatiotemporal feature is x t,i The weight value of other ATMs corresponding to other ATMs j is W j , other initial space-time features are x t,j The first spatiotemporal weighted feature is W i x t,i , the second spatiotemporal weighted feature is W j x t,j .

[0097] B2. Input the feature splicing results of the first spatiotemporal weighted feature and the second spatiotemporal weighted feature into the graph attention network, and determine the spatial correlation coefficient according to the output result of the graph attention network.

[0098] In one embodiment, the spatial correlation coefficient is determined using the following formula:

[0099]

[0100] Among them, W i x t,i is the first spatiotemporal weighted feature, W j x t,j is the second spatiotemporal weighted feature, is the GraphAttentionNetwork (GAT). LeakyReLU is an activation function that enables the model to obtain nonlinear information. t,i,j Represents the spatial correlation coefficient between target ATM i and other ATMs j.

[0101] A first spatiotemporal weighted feature is generated according to the target ATM weight value corresponding to the target ATM and the target initial spatiotemporal feature, and a second spatiotemporal weighted feature is generated according to the weight values ​​of other ATMs corresponding to each other ATM and each other initial spatiotemporal feature; the feature splicing results of the first spatiotemporal weighted feature and the second spatiotemporal weighted feature are input into the graph attention network, and the spatial correlation coefficient is determined according to the output result of the graph attention network, so as to use the graph attention network to capture the topological structure between the target ATM and other ATMs, thereby ensuring the rationality and accuracy of the determination of the spatial correlation coefficient.

[0102] S208. Determine the target spatiotemporal fusion features corresponding to the target ATM based on the spatial correlation coefficients and other initial spatiotemporal features, and predict the target cash deposit amount corresponding to the target ATM at the current time point based on the target spatiotemporal fusion features.

[0103] Embodiment 3

[0104] Figure 3 This is a flowchart of a method for generating initial spatiotemporal features of a target provided by Embodiment 3 of the present invention. This embodiment further optimizes and expands the above embodiment and can be combined with the above optional implementations. Figure 3 As shown, the method includes:

[0105] S301, generating a query matrix according to the target time enhancement feature and the first weight matrix, generating a key matrix according to the target time enhancement feature and the second weight matrix, and generating a value matrix according to the target time enhancement feature and the third weight matrix.

[0106] In one implementation, the query matrix, key matrix, and value matrix are generated as follows:

[0107] Q=θ(t)W1; K=θ(t)W2; V=θ(t)W3;

[0108] Among them, θ(t) represents the target time enhancement feature, W1 represents the first weight matrix, W2 represents the second weight matrix, W3 represents the third weight matrix, Q represents the query matrix, K represents the key matrix, and V represents the value matrix.

[0109] S302. Divide the query matrix according to the target time step to obtain at least one first sub-matrix, divide the key matrix according to the target time step to obtain at least one second sub-matrix, and divide the value matrix according to the target time step to obtain at least one third sub-matrix.

[0110] Exemplarily, the query matrix, key matrix, and value matrix are divided according to the target time step, which can be expressed as follows:

[0111] Q=[Q'1=Concat(Q1,……,Q s ),Q'2=Concat(Q s+1 ,……,Q 2s ),……,Q' n ];

[0112] K=[K1'=Concat(K1,……,K s ),K2'=Concat(K s+1 ,……,K 2s ),……,K n '];

[0113] V=[V1'=Concat(V1,……,V s ),V2'=Concat(V s+1 ,……,V 2s ),……,V' n ];

[0114] Where s represents the target time step, Q'1,...,Q' n represents the first submatrix, K1', ..., K n ' represents the second sub-matrix, V1', ..., V' n represents the third sub-matrix.

[0115] S303: Perform attention mechanism processing on each first sub-matrix and each second sub-matrix respectively to obtain at least one attention score, and generate at least one segmented time feature according to each attention score and each third sub-matrix.

[0116] In one embodiment, the attention score is determined using the following formula:

[0117] Attn ij =Attention(Q i ',K j ')

[0118] Among them, Q i ' represents the first sub-matrix of the i-th order, K j ' represents the jth second sub-matrix. ij Indicates Q i ' and K j 'Attention score between.

[0119] The segment time features are generated using the following formula:

[0120] Y i =Softmax(Attn i1 ,Attn i2 ,……,Attnin )V j

[0121] Among them, V j represents the jth third submatrix, Attn i1 ,Attn i2 ,……,Attn in Indicates Q i ' and K j 'Each attention score between. i Represents segment time characteristics.

[0122] S304: Generate a target time splicing feature according to the feature splicing results of each segment time feature, and generate a target initial time-space feature according to the target time splicing feature and the target space feature.

[0123] In one implementation, the target time splicing feature is generated using the following formula:

[0124] Y ′ =w s Concat(Y1,Y2,……,Y s )

[0125] Among them, w s represents the learnable scaling factor of each segment time feature determined according to the target time step, Y1,Y2,…,Y s Represents the time characteristics of each segment, Y ′ Represents the target temporal splicing feature.

[0126] Furthermore, the target initial spatiotemporal features are generated according to the target temporal splicing features and the target spatial features.

[0127] By generating a query matrix according to the target time enhancement feature and the first weight matrix, generating a key matrix according to the target time enhancement feature and the second weight matrix, and generating a value matrix according to the target time enhancement feature and the third weight matrix; dividing the query matrix according to the target time step to obtain at least one first sub-matrix, dividing the key matrix according to the target time step to obtain at least one second sub-matrix, and dividing the value matrix according to the target time step to obtain at least one third sub-matrix, it is helpful to better capture the dynamic trend of the cash deposit amount over time, so as to facilitate the fusion of high time granularity global information and low time granularity local information in the feature extraction process.

[0128] By performing attention mechanism processing on each first submatrix and each second submatrix respectively, at least one attention score is obtained, and at least one segmented time feature is generated according to each attention score and each third submatrix; a target time splicing feature is generated according to the feature splicing result of each segmented time feature, and a target initial spatiotemporal feature is generated according to the target time splicing feature and the target spatial feature, thereby realizing the extraction of time correlation from continuous time segments containing richer time feature information, thereby obtaining the effect of corresponding output at different time scales.

[0129] Embodiment 4

[0130] Figure 4 This is a flowchart of a method for generating target spatiotemporal fusion features provided by the fourth embodiment of the present invention. This embodiment further optimizes and expands the above embodiment and can be combined with the above optional implementations. Figure 4 As shown, the method includes:

[0131] S401. Perform weighted summation on other initial spatiotemporal features according to the spatial correlation coefficients, and determine the topological spatiotemporal features according to the weighted summation result.

[0132] In one implementation, the topological spatiotemporal features are determined using the following formula:

[0133] a t,i,j =Softmax(e t,i,j );

[0134] x t ′ ,i =LeakyReLU(∑ j∈N a t,i,j W j x t,j )

[0135] Among them, e t,i,j represents the spatial correlation coefficient between the target ATM i and other ATM j. W j x t,j Represents other initial spatiotemporal features corresponding to other ATMs j. t ′ ,i Represents topological spatiotemporal characteristics.

[0136] S402, inputting the topological spatiotemporal features into the first long short-term memory network so that the first long short-term memory network can capture the time dependency in the topological spatiotemporal features, and determining the time optimization features according to the output results of the first long short-term memory network.

[0137] In one embodiment, the time optimization feature is determined using the following formula:

[0138] H intra,t ,C intra,t =LSTM1(X intra,t ,H intra,t-1 ,C intra,t-1 ,W,b);

[0139] X intra,t =x ′ t,i ;

[0140] Among them, X intra,t is the topological spatiotemporal feature, LSTM1 represents the first long short-term memory network, H intra,t represents the time optimization feature, H intra,t-1 ,C intra,t-1 represents the hidden layer state and memory unit at the previous moment, W, b represent the learnable weights and biases in the input unit, forgetting unit, output unit and memory unit respectively.

[0141] S403: Input the auxiliary spatiotemporal features into the second long short-term memory network so that the second long short-term memory network can capture the spatial dependencies in the auxiliary spatiotemporal features, and determine the spatial optimization features according to the output results of the second long short-term memory network.

[0142] The auxiliary spatiotemporal features are determined according to the topological spatiotemporal features and the target distance value, and the target distance value is determined according to the distance values ​​between the target ATM and other ATMs.

[0143] In one embodiment, the spatial optimization feature is determined using the following formula:

[0144] H inter,t ,C inter,t =LSTM2(X inter,t ,H inter,t-1 ,C inter,t-1 ,W,b);

[0145] X inter,t =x ′ t,i ·S;

[0146] Among them, X inter,t is the auxiliary spatiotemporal feature, S is the inverse of the target distance. LSTM2 represents the second long short-term memory network, H inter,t represents the spatial optimization feature, H inter,t-1 ,C inter,t-1 represents the hidden layer state and memory unit at the previous moment, W, b represent the learnable weights and biases in the input unit, forgetting unit, output unit and memory unit respectively.

[0147] S404: Determine target spatiotemporal fusion features according to a weighted summation result between the time optimization features and the space optimization features.

[0148] In one implementation, the target spatiotemporal fusion feature is determined using the following formula:

[0149] H fusion,t =θ intra °H intra,t +θ inter °H inter,t

[0150] Among them, H fusion,t represents the target spatiotemporal fusion feature, H intra,t represents the time optimization feature, H inter,t represents the spatial optimization feature, θ intra and θ inter They represent learnable weights respectively, and “°” represents the operation of multiplying matrix elements one by one.

[0151] By performing weighted summation on other initial spatiotemporal features according to each spatial correlation coefficient, and determining the topological spatiotemporal features according to the weighted summation result; inputting the topological spatiotemporal features into the first long short-term memory network for the first long short-term memory network to capture the time dependency in the topological spatiotemporal features, and determining the time optimization features according to the output results of the first long short-term memory network; inputting the auxiliary spatiotemporal features into the second long short-term memory network for the second long short-term memory network to capture the spatial dependency in the auxiliary spatiotemporal features, and determining the space optimization features according to the output results of the second long short-term memory network; determining the target spatiotemporal fusion features according to the weighted summation result between the time optimization features and the space optimization features, achieving the effect of capturing the dependency between time and space, avoiding the problem of weakening the spatiotemporal relationship due to the separate learning of time features and space features, improving the feature accuracy of the target spatiotemporal fusion features, and further improving the accuracy of the cash deposit prediction.

[0152] Optionally, the method further includes:

[0153] The target spatiotemporal fusion features are input into the target fully connected neural network, and based on the output results of the target fully connected neural network, the predicted spatiotemporal fusion features corresponding to the target ATM at at least one future time point are determined; and the predicted cash deposit amount corresponding to the target ATM at each future time point is predicted based on the predicted spatiotemporal fusion features.

[0154] In one implementation, the target spatiotemporal fusion feature is determined using the following formula:

[0155]

[0156] Y t+k =ReLU(WFC H fusion,t +b FC );

[0157] Among them, H fusion,t is the output of multi-scale spatiotemporal fusion features, θ output,s W is a different learnable weight for each time scale s, Flatten is the unfolding operation, and "°" represents the operation of multiplying matrix elements one by one. FC and b FC is the learnable weight and bias matrix for the target fully connected neural network, ReLU is the activation function, Y t+k Represents the predicted spatiotemporal fusion features corresponding to the target ATM at at least one future time point.

[0158] By inputting the target spatiotemporal fusion features into the target fully connected neural network, and based on the output results of the target fully connected neural network, the predicted spatiotemporal fusion features corresponding to the target ATM at at least one future time point are determined; the predicted cash deposit amount corresponding to the target ATM at each future time point is predicted based on the predicted spatiotemporal fusion features, thereby achieving the effect of predicting the cash deposit amount of the target ATM at future moments, which is conducive to continuously ensuring that the bank reasonably allocates cash resources and effectively prevents risks.

[0159] Embodiment 5

[0160] Figure 5 The schematic diagram of the structure of a device for predicting the amount of cash stored in a cash machine provided in the fifth embodiment of the present invention is applicable to the situation where the amount of cash stored in a cash machine is predicted by using three feature dimensions, namely, time feature, space feature and static feature. Figure 5 As shown, the device comprises:

[0161] The time enhancement feature generation module 51 is used to obtain at least one target time feature and at least one target static feature corresponding to the current time point, and generate a target time enhancement feature according to the target time feature and the target static feature; wherein the target static feature includes at least one of a user behavior feature, a payment method feature, and a withdrawal fee feature;

[0162] The spatial correlation coefficient determination module 52 is used to determine the spatial correlation coefficient between each of the other cash dispensers and the target cash dispenser according to the target initial spatiotemporal characteristics corresponding to the target cash dispenser and other initial spatiotemporal characteristics corresponding to at least one other cash dispenser; wherein the target initial spatiotemporal characteristics are determined according to at least one target spatial characteristic and the target time enhancement characteristic corresponding to the target cash dispenser, and the other initial spatiotemporal characteristics are determined according to at least one other spatial characteristic and the target time enhancement characteristic corresponding to the other cash dispenser;

[0163] The cash deposit amount prediction module 53 is used to determine the target spatiotemporal fusion characteristics corresponding to the target ATM according to each of the spatial correlation coefficients and each of the other initial spatiotemporal characteristics, and predict the target cash deposit amount corresponding to the target ATM at the current time point according to the target spatiotemporal fusion characteristics.

[0164] Optionally, the time enhancement feature generation module 51 is specifically used to:

[0165] Inputting at least one candidate static feature into a first gated residual network, and determining first selection weight values ​​corresponding to each of the candidate static features according to an output result of the first gated residual network;

[0166] Inputting each of the candidate static features into a second gated residual network, and determining first feature variable values ​​corresponding to each of the candidate static features according to an output result of the second gated residual network;

[0167] According to each of the first feature variable values ​​and each of the first selection weight values, the first weighted feature variable values ​​corresponding to each of the candidate static features are determined, and according to each of the first weighted feature variable values, the target static feature is determined from the candidate static features.

[0168] Optionally, the time enhancement feature generating module 51 is further configured to:

[0169] Inputting at least one candidate time feature and a first context feature into the first gated residual network, and determining second selection weight values ​​corresponding to each of the candidate time features according to an output result of the first gated residual network; wherein the first context feature is determined according to the target static feature;

[0170] Inputting each of the candidate time features into a third gated residual network, and determining second feature variable values ​​corresponding to each of the candidate time features according to an output result of the third gated residual network;

[0171] According to each of the second feature variable values ​​and each of the second selection weight values, the second weighted feature variable values ​​corresponding to each of the candidate time features are determined, and according to each of the second weighted feature variable values, the target time feature is determined from each of the candidate time features.

[0172] Optionally, the time enhancement feature generating module 51 is further configured to:

[0173] Performing position coding on each of the target time features to generate a first coding feature, and performing timestamp coding on each of the target time features to generate a second coding feature;

[0174] A target coding feature is generated according to the first coding feature and the second coding feature, and the target coding feature and the second context feature are input into the fourth gated residual network, and the target time enhancement feature is determined according to the output result of the fourth gated residual network; wherein the second context feature is generated according to the target static feature.

[0175] Optionally, the device further includes a target initial spatiotemporal feature generation module, which is specifically used to:

[0176] Generate a query matrix according to the target time enhancement feature and a first weight matrix, generate a key matrix according to the target time enhancement feature and a second weight matrix, and generate a value matrix according to the target time enhancement feature and a third weight matrix;

[0177] Dividing the query matrix according to the target time step to obtain at least one first sub-matrix, dividing the key matrix according to the target time step to obtain at least one second sub-matrix, and dividing the value matrix according to the target time step to obtain at least one third sub-matrix;

[0178] Performing attention mechanism processing on each of the first sub-matrices and each of the second sub-matrices respectively to obtain at least one attention score, and generating at least one segmented time feature according to each of the attention scores and each of the third sub-matrices;

[0179] A target time splicing feature is generated according to the feature splicing results of each of the segmented time features, and the target initial time-space feature is generated according to the target time splicing feature and the target space feature.

[0180] Optionally, the spatial correlation coefficient determination module 52 is specifically configured to:

[0181] Generate a first spatiotemporal weighted feature according to the target cash machine weight value corresponding to the target cash machine and the target initial spatiotemporal feature, and generate a second spatiotemporal weighted feature according to the other cash machine weight values ​​corresponding to each of the other cash machines and each of the other initial spatiotemporal features;

[0182] The feature splicing result of the first spatiotemporal weighted feature and the second spatiotemporal weighted feature is input into a graph attention network, and the spatial correlation coefficient is determined according to the output result of the graph attention network.

[0183] Optionally, the cash deposit amount prediction module 53 is specifically used for:

[0184] Performing weighted summation on each of the other initial spatiotemporal features according to each of the spatial correlation coefficients, and determining a topological spatiotemporal feature according to the weighted summation result;

[0185] Inputting the topological spatiotemporal features into a first long short-term memory network so that the first long short-term memory network can capture the time dependency in the topological spatiotemporal features, and determining the time optimization features according to the output results of the first long short-term memory network;

[0186] Inputting the auxiliary spatiotemporal features into the second long short-term memory network so that the second long short-term memory network can capture the spatial dependencies in the auxiliary spatiotemporal features, and determining the spatial optimization features according to the output results of the second long short-term memory network; wherein the auxiliary spatiotemporal features are determined according to the topological spatiotemporal features and the target distance value, and the target distance value is determined according to the distance values ​​between the target ATM and each of the other ATMs;

[0187] The target spatiotemporal fusion feature is determined according to a weighted summation result between the time optimization feature and the space optimization feature.

[0188] Optionally, the device further includes a spatiotemporal fusion feature prediction module, which is specifically used to:

[0189] Inputting the target spatiotemporal fusion features into a target fully connected neural network, and determining the predicted spatiotemporal fusion features corresponding to the target cash machine at at least one future time point according to the output result of the target fully connected neural network;

[0190] The predicted cash deposit amount of the target ATM at each future time point is predicted based on the predicted spatiotemporal fusion features.

[0191] The device for predicting the cash deposit amount of an ATM provided in the embodiment of the present invention can execute the method for predicting the cash deposit amount of an ATM provided in any embodiment of the present invention, and has the functional modules and beneficial effects corresponding to the execution method.

[0192] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0193] Embodiment 6

[0194] Figure 6A schematic diagram of the structure of an electronic device 60 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0195] like Figure 6 As shown, the electronic device 60 includes at least one processor 61, and a memory connected to the at least one processor 61, such as a read-only memory (ROM) 62, a random access memory (RAM) 63, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 61 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 62 or the computer program loaded from the storage unit 68 to the random access memory (RAM) 63. In the RAM 63, various programs and data required for the operation of the electronic device 60 can also be stored. The processor 61, the ROM 62, and the RAM 63 are connected to each other via a bus 64. An input / output (I / O) interface 65 is also connected to the bus 64.

[0196] A number of components in the electronic device 60 are connected to the I / O interface 65, including: an input unit 66, such as a keyboard, a mouse, etc.; an output unit 67, such as various types of displays, speakers, etc.; a storage unit 68, such as a disk, an optical disk, etc.; and a communication unit 69, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 69 allows the electronic device 60 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0197] The processor 61 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 61 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 61 executes the various methods and processes described above, such as a method for predicting the amount of cash deposited in an ATM.

[0198] In some embodiments, the method for predicting the amount of cash deposited at an ATM can be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 68. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 60 via the ROM 62 and / or the communication unit 69. When the computer program is loaded into the RAM 63 and executed by the processor 61, one or more steps of the method for predicting the amount of cash deposited at an ATM described above can be performed. Alternatively, in other embodiments, the processor 61 can be configured to execute the method for predicting the amount of cash deposited at an ATM by any other appropriate means (e.g., by means of firmware).

[0199] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0200] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0201] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, 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), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0202] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0203] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0204] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0205] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0206] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting the amount of cash deposited in a cash machine, characterized in that: The method comprises: Acquire at least one target time feature and at least one target static feature corresponding to the current time point, and generate a target time enhancement feature according to the target time feature and the target static feature; wherein the target static feature includes at least one of a user behavior feature, a payment method feature, and a withdrawal fee feature; According to the target initial spatiotemporal features corresponding to the target cash dispenser and other initial spatiotemporal features corresponding to at least one other cash dispenser, determine the spatial correlation coefficients between each of the other cash dispensers and the target cash dispenser; wherein the target initial spatiotemporal features are determined according to at least one target spatial feature and the target time enhancement feature corresponding to the target cash dispenser, and the other initial spatiotemporal features are determined according to at least one other spatial feature and the target time enhancement feature corresponding to the other cash dispensers; According to each of the spatial correlation coefficients and each of the other initial spatiotemporal features, the target spatiotemporal fusion features corresponding to the target ATM are determined, and the target cash deposit amount corresponding to the target ATM at the current time point is predicted according to the target spatiotemporal fusion features.

2. The method according to claim 1, characterized in that The obtaining of at least one static feature of the target corresponding to the current time point includes: Inputting at least one candidate static feature into a first gated residual network, and determining first selection weight values ​​corresponding to each of the candidate static features according to an output result of the first gated residual network; Inputting each of the candidate static features into a second gated residual network, and determining first feature variable values ​​corresponding to each of the candidate static features according to an output result of the second gated residual network; According to each of the first feature variable values ​​and each of the first selection weight values, the first weighted feature variable values ​​corresponding to each of the candidate static features are determined, and according to each of the first weighted feature variable values, the target static feature is determined from the candidate static features.

3. The method according to claim 2, characterized in that The obtaining of at least one target time feature corresponding to the current time point includes: Inputting at least one candidate time feature and a first context feature into the first gated residual network, and determining second selection weight values ​​corresponding to each of the candidate time features according to an output result of the first gated residual network; wherein the first context feature is determined according to the target static feature; Inputting each of the candidate time features into a third gated residual network, and determining second feature variable values ​​corresponding to each of the candidate time features according to an output result of the third gated residual network; According to each of the second feature variable values ​​and each of the second selection weight values, the second weighted feature variable values ​​corresponding to each of the candidate time features are determined, and according to each of the second weighted feature variable values, the target time feature is determined from each of the candidate time features.

4. The method according to claim 1, characterized in that: The step of generating a target time enhancement feature according to the target time feature and the target static feature includes: Performing position coding on each of the target time features to generate a first coding feature, and performing timestamp coding on each of the target time features to generate a second coding feature; A target coding feature is generated according to the first coding feature and the second coding feature, and the target coding feature and the second context feature are input into the fourth gated residual network, and the target time enhancement feature is determined according to the output result of the fourth gated residual network; wherein the second context feature is generated according to the target static feature.

5. The method according to claim 1, further comprising: Generate a query matrix according to the target time enhancement feature and a first weight matrix, generate a key matrix according to the target time enhancement feature and a second weight matrix, and generate a value matrix according to the target time enhancement feature and a third weight matrix; Dividing the query matrix according to the target time step to obtain at least one first sub-matrix, dividing the key matrix according to the target time step to obtain at least one second sub-matrix, and dividing the value matrix according to the target time step to obtain at least one third sub-matrix; Performing attention mechanism processing on each of the first sub-matrices and each of the second sub-matrices respectively to obtain at least one attention score, and generating at least one segmented time feature according to each of the attention scores and each of the third sub-matrices; A target time splicing feature is generated according to the feature splicing results of each of the segmented time features, and the target initial spatiotemporal feature is generated according to the target time splicing feature and the target spatial feature.

6. The method according to claim 1, characterized in that The step of determining the spatial correlation coefficient between each of the other cash dispensers and the target cash dispenser according to the target initial spatiotemporal features corresponding to the target cash dispenser and other initial spatiotemporal features corresponding to at least one other cash dispenser comprises: Generate a first spatiotemporal weighted feature according to the target cash machine weight value corresponding to the target cash machine and the target initial spatiotemporal feature, and generate a second spatiotemporal weighted feature according to the other cash machine weight values ​​corresponding to each of the other cash machines and each of the other initial spatiotemporal features; The feature splicing result of the first spatiotemporal weighted feature and the second spatiotemporal weighted feature is input into a graph attention network, and the spatial correlation coefficient is determined according to the output result of the graph attention network.

7. The method according to claim 1, characterized in that Determining the target spatiotemporal fusion feature corresponding to the target ATM according to each of the spatial correlation coefficients and each of the other initial spatiotemporal features includes: Performing weighted summation on each of the other initial spatiotemporal features according to each of the spatial correlation coefficients, and determining a topological spatiotemporal feature according to the weighted summation result; Inputting the topological spatiotemporal features into a first long short-term memory network so that the first long short-term memory network can capture the time dependency in the topological spatiotemporal features, and determining the time optimization features according to the output results of the first long short-term memory network; Inputting the auxiliary spatiotemporal features into the second long short-term memory network so that the second long short-term memory network can capture the spatial dependencies in the auxiliary spatiotemporal features, and determining the spatial optimization features according to the output results of the second long short-term memory network; wherein the auxiliary spatiotemporal features are determined according to the topological spatiotemporal features and the target distance value, and the target distance value is determined according to the distance values ​​between the target ATM and each of the other ATMs; The target spatiotemporal fusion feature is determined according to a weighted summation result between the time optimization feature and the space optimization feature.

8. The method according to claim 7, further comprising: Inputting the target spatiotemporal fusion features into a target fully connected neural network, and determining the predicted spatiotemporal fusion features corresponding to the target cash machine at at least one future time point according to the output result of the target fully connected neural network; The predicted cash deposit amount of the target ATM at each future time point is predicted based on the predicted spatiotemporal fusion features.

9. A device for predicting the amount of cash stored in a cash machine, characterized in that: The device comprises: A time enhancement feature generation module, used to obtain at least one target time feature and at least one target static feature corresponding to the current time point, and generate a target time enhancement feature according to the target time feature and the target static feature; wherein the target static feature includes at least one of a user behavior feature, a payment method feature, and a withdrawal fee feature; A spatial correlation coefficient determination module, used to determine the spatial correlation coefficient between each of the other cash dispensers and the target cash dispenser according to the target initial spatiotemporal characteristics corresponding to the target cash dispenser and other initial spatiotemporal characteristics corresponding to at least one other cash dispenser; wherein the target initial spatiotemporal characteristics are determined according to at least one target spatial characteristic and the target time enhancement characteristic corresponding to the target cash dispenser, and the other initial spatiotemporal characteristics are determined according to at least one other spatial characteristic and the target time enhancement characteristic corresponding to the other cash dispenser; The cash deposit amount prediction module is used to determine the target spatiotemporal fusion characteristics corresponding to the target ATM according to each of the spatial correlation coefficients and each of the other initial spatiotemporal characteristics, and predict the target cash deposit amount corresponding to the target ATM at the current time point according to the target spatiotemporal fusion characteristics.

10. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the method for predicting the cash deposit amount of an ATM according to any one of claims 1-8.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to execute the method for predicting the cash deposit amount of an ATM according to any one of claims 1-8.

12. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method for predicting the cash deposit amount of a cash dispenser according to any one of claims 1 to 8 is implemented.

Images