Bank ATM inventory cash prediction method and cash adding strategy
By building an ATM inventory cash prediction model based on VRC-XGBoost, combined with multi-dimensional indicators, the accuracy of ATM cash demand forecasting is solved, and more efficient cash management and cost reduction are achieved.
Patent Information
- Application Number
- CN202411991445.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-31
- Publication Date
- 2025-05-30
AI Technical Summary
The existing technology is difficult to accurately predict the cash demand of ATMs in a certain time in the future, which often leads to cash redundancy problems and increases operating costs.
By obtaining the time series data of ATM's historical trunk balance, performing community classification and feature extraction, a cash inventory prediction model based on VRC-XGBoost is constructed, combining multi-dimensional indicators of price levels, climatic conditions, time series and mobile payment to achieve accurate prediction of cash inventory of ATM.
Improves the accuracy of ATM inventory cash forecasts, reduces cash redundancy, reduces operating costs, and provides a more efficient cash-increasing strategy.
Smart Images
Figure CN120069155A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the financial field and relates to a method for predicting the inventory cash of bank ATMs and a cash replenishment strategy. Background Art
[0002] With the rise of mobile payment and online shopping, the cash business of banks has been impacted unprecedentedly, greatly increasing the complexity of ATM inventory management. Banks hope to reduce cash investment while ensuring the service quality of ATMs, and third-party financial service institutions undertaking the ATM cash replenishment business expect to minimize operating costs on the basis of meeting the bank's goals. Therefore, accurately predicting the inventory cash volume of ATMs to formulate the cash replenishment frequency and the optimal cash replenishment volume strategy is crucial for the benefits of banks and third-party financial service institutions. However, due to the fact that the inventory cash of ATMs is affected by numerous factors and has strong time-varying and random characteristics, the current technology has not yet formed a method for accurately predicting the cash demand of each ATM within a certain period in the future. The common practice is to put 40% more cash than the actual demand in ATMs or make subjective judgments based on personal experience, which often leads to the problem of cash redundancy and is not conducive to the efficient use of funds. At the same time, cash delivery requires armored vehicles and manual labor, and excessive cash delivery will also lead to an increase in costs. Summary of the Invention
[0003] To solve the above problems, the technical solution adopted by the present invention is: a method for predicting the inventory cash of bank ATMs, including the following steps:
[0004] S1: Obtain the time series data set of the historical balance of the ATM cash box;
[0005] S2: Mine and process the time series data set of the historical balance of the ATM cash box and conduct community classification;
[0006] S3: Based on multi-dimensional indicators affecting the ATM withdrawal volume, construct the input feature set of the bank ATM inventory cash prediction model;
[0007] S4: Based on the classified community data and the input feature set, respectively construct the VRC-XGBoost prediction models for the inventory cash of ATMs for different community classifications;
[0008] S5: Based on the samples widened from the time series data of the ATM cash box balance, train the prediction models for the inventory cash of ATMs for different community classifications to obtain the trained VRC-XGBoost prediction models for the inventory cash of ATMs for different community classifications;
[0009] S6: Use the trained VRC-XGBoost prediction models for the balance of the ATM cash box for different community classifications to realize the prediction of the inventory cash of bank ATMs.
[0010] Further: The VRC-XGBoost prediction model for the ATM's in-stock cash includes:
[0011] Feature extraction module: It is used to decompose the balance sequence of the cash box into several sub-time series by using the VMD algorithm, and incorporate the several sub-time series into the input feature set;
[0012] Feature screening and dimensionality reduction: It is used to perform feature screening and dimensionality reduction on the feature set by using the RFECV algorithm to obtain the optimal input feature set;
[0013] Prediction module: It is used to realize the prediction of the in-stock cash based on the optimal input feature set.
[0014] Further: The prediction module: adopts an XGBoost model based on the OPTUNA optimization framework.
[0015] Further: The process of processing and mining the historical balance data of the ATM's cash box to realize the community classification of the ATM is as follows:
[0016] Based on the value of the teller number, split it into single-machine data according to the value of the teller number, and draw a grouped box plot of the ATM's cash box balance;
[0017] Based on the ATM grouped box plot, divide the first threshold baseline and the second threshold baseline; the value of the first threshold baseline is less than the value of the second threshold baseline;
[0018] Compare the cash box balance with the first threshold baseline and the second threshold baseline respectively to realize the community classification of the ATM.
[0019] Further: The realizing the community classification of the ATM by comparing the cash box balance with the first threshold baseline and the second threshold baseline respectively includes:
[0020] When the cash box balance is below the first threshold baseline, it is a low-activity group;
[0021] When the cash box balance is between the first threshold baseline and the second threshold baseline, it is a medium-activity group;
[0022] When the cash box balance is above the second threshold baseline, it is a high-activity group.
[0023] Further: The multi-dimensional indicators affecting ATM withdrawal volume are based on four characteristic dimensions of price level, climate condition, time series, and mobile payment, and thirteen representative indicators are determined: consumer price index, average temperature, average precipitation, balance of the cash box three days before the prediction time, balance of the cash box two days before the prediction time, balance of the cash box one day before the prediction time, balance of the cash box on the same day of the previous week of the prediction time, balance of the cash box one day before the previous week of the prediction time, balance of the cash box one day after the previous week of the prediction time, first-order difference, second-order difference, holiday factor, and mobile payment index.
[0024] Further: Based on the multi-dimensional indicators affecting ATM withdrawal volume, an input feature set for the bank ATM cash box balance prediction model is constructed;
[0025] Using the Spearman correlation coefficient method, the correlation between each of the thirteen indicators and the cash box balances of the high-activity group, medium-activity group, and low-activity group is calculated in turn. Considering the correlation calculation results of the thirteen indicators and the three groups comprehensively, each indicator is analyzed one by one to determine whether it is positively correlated, negatively correlated, or insignificantly correlated, and screening is carried out.
[0026] A prediction device for the bank ATM inventory cash includes:
[0027] An acquisition module: used to acquire the historical cash box balance time series data set of the ATM;
[0028] A classification module: used to mine and process the historical cash box balance time series data set of the ATM and perform community classification;
[0029] A construction module I: used to construct an input feature set for the bank ATM inventory cash prediction model based on the multi-dimensional indicators affecting ATM withdrawal volume;
[0030] A VRC-XGBoost prediction model module: used to construct VRC-XGBoost prediction models for ATM inventory cash for different community classifications respectively based on the classified community data and the input feature set;
[0031] A training module: used to train the prediction models for ATM inventory cash for different community classifications based on the samples widened by the ATM cash box balance time series data, and obtain the trained VRC-XGBoost prediction models for ATM inventory cash for different community classifications;
[0032] A prediction module: used to implement the prediction of the bank ATM inventory cash by using the trained VRC-XGBoost prediction models for ATM inventory cash for different community classifications.
[0033] A readable storage medium stores program modules, and when the program modules run in a processor, any of the methods described above can be implemented.
[0034] A readable storage medium stores program modules, characterized in that when the program modules run in a processor, any of the methods described above can be implemented.
[0035] Based on the cash replenishment strategy of a prediction method for the cash in bank ATM inventory as described in any one of them, based on the prediction result of the cash in bank ATM inventory, a cash replenishment method is adopted to realize the prediction of the cash replenishment of bank ATM.
[0036] Further, the cash replenishment method is as follows:
[0037] Taking the demand quantity of the predicted result of the cash in bank ATM inventory floating up by X% or the fixed cash reserve quantity minus the predicted cash inventory quantity floating down by X% as the final cash replenishment strategy.
[0038] A prediction method and a cash replenishment strategy for the cash in bank ATM inventory provided by the present invention have the following advantages: relying on the cooperation with a third-party financial service institution, starting from the actual operation scenario of the ATM in commercial bank branches, analyzing the historical operation data of bank ATM, mining and extracting data features, and proposing a threshold method based on the data distribution characteristics to realize the community classification of ATM. On this basis, innovative features are incorporated to construct a prediction input set with four dimensions of price level, climate condition, time series, and mobile payment. Combining the advantages of variational mode decomposition (VMD) for adaptive decomposition, recursive feature elimination cross-validation (RFECV) for eliminating redundant features, and extreme gradient boosting algorithm (XGBoost) for accurately capturing changes, a cash inventory integrated prediction model based on VRC-XGBoost is proposed, and comparison and effectiveness analysis are carried out, proving that the VRC-XGBoost model has better prediction effect. According to the experience of banking experts, a cash replenishment strategy can be formed based on the prediction result and applied to the formulation of the cash replenishment frequency and the optimal cash replenishment quantity strategy, so as to more efficiently manage the cash flow of ATM, reduce unnecessary cash replenishment operations, save the cash replenishment cost of ATM, and the present invention can be applied to practice to provide business solutions and management inspirations. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.
[0040] Figure 1It is the flowchart of the steps of the intelligent prediction method for the inventory cash of bank ATMs in the embodiments of the present invention;
[0041] Figure 2 It is the ATM community classification diagram based on double baselines in the embodiments of the present invention;
[0042] Figure 3 It is the index system diagram of the influencing factors of the balance of the ATM cash box in the embodiments of the present invention;
[0043] Figure 4 It is the VMD decomposition result diagram of three typical active ATMs in the embodiments of the present invention, where (a) is the VDM decomposition result diagram of the balance of the high-active ATM cash box, (b) is the VDM decomposition result diagram of the balance of the medium-active ATM cash box, and (c) is the VDM decomposition result diagram of the balance of the low-active ATM cash box;
[0044] Figure 5 It is the comparison diagram of the prediction results generated by the intelligent prediction method for the inventory cash of bank ATMs for three typical active ATMs in the embodiments of the present invention; where (a) is the comparison diagram of the inventory cash prediction results of the high-active ATM, (b) is the comparison diagram of the inventory cash prediction results of the medium-active ATM, and (c) is the comparison diagram of the inventory cash prediction results of the low-active ATM;
[0045] Figure 6 It is the schematic diagram of the ATM cash replenishment strategy formed based on the prediction results in the embodiments of the present invention. Detailed implementation manners
[0046] It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other. The present invention will be described in detail below with reference to the drawings and in combination with the embodiments.
[0047] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. The description of at least one exemplary embodiment below is actually only illustrative and in no way restrictive of the present invention and its application or use. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.
[0048] It should be noted that the terms used herein are for the purpose of describing specific embodiments only and are not intended to limit the exemplary embodiments of the present invention. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they specify the presence of features, steps, operations, devices, components, and / or combinations thereof.
[0049] Unless otherwise specifically stated, the relative arrangements of components and steps, numerical expressions, and numerical values set forth in these embodiments do not limit the scope of the present invention. At the same time, it should be clear that, for the sake of convenience of description, the dimensions of the various parts shown in the drawings are not drawn in actual proportional relationships. Technologies, methods, and devices known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, such technologies, methods, and devices should be regarded as part of the authorized specification. In all the examples shown and discussed here, any specific value should be construed as merely exemplary and not as a limitation. Therefore, other examples of the exemplary embodiments may have different values. It should be noted that like reference numerals and letters denote like items in the following drawings, and thus, once an item is defined in one drawing, it does not require further discussion in subsequent drawings.
[0050] In the description of the present invention, it should be understood that the orientation or positional relationships indicated by orientation words such as "front, rear, upper, lower, left, right", "lateral, vertical, perpendicular, horizontal", and "top, bottom" are generally based on the orientation or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description. Without contrary description, these orientation words do not indicate and imply that the device or element referred to must have a specific orientation or be constructed and operated in a specific orientation, and thus should not be construed as limiting the protection scope of the present invention: the orientation words "inside, outside" refer to the inside and outside relative to the contour of each component itself.
[0051] For ease of description, spatial relative terms such as "above", "over", "on the upper surface", "upper" etc. may be used herein to describe the spatial positional relationship of one device or feature to other devices or features as shown in the figures. It should be understood that the spatial relative terms are intended to encompass different orientations in use or operation in addition to the orientation depicted in the figures of the device. For example, if the device in the figures is inverted, the device described as "above" or "over" other devices or structures will then be positioned "below" or "under" other devices or structures. Thus, the exemplary term "above" can include both the orientations of "above" and "below". The device may also be positioned in other different ways (rotated 90 degrees or in other orientations), and corresponding interpretations should be made for the spatial relative descriptions used herein.
[0052] In addition, it should be noted that the use of terms such as "first", "second" etc. to limit components is only for the convenience of differentiating the corresponding components. Without additional statements, the above terms have no special meanings, and thus should not be construed as limiting the protection scope of the present invention.
[0053] Figure 1 is the step flow chart of the intelligent prediction method for the bank ATM inventory cash in the embodiment of the present invention;
[0054] An intelligent cash replenishment method for bank ATMs based on an integrated model, comprising the following steps:
[0055] S1: Obtain the historical balance time series data set of the ATM cash box;
[0056] S2: Mine and process the historical balance time series data set of the ATM cash box and perform community classification;
[0057] S3: Based on multi-dimensional indicators affecting the ATM withdrawal volume, construct the input feature set of the bank ATM inventory cash prediction model;
[0058] S4: Based on the classified community data and the input feature set, respectively construct the VRC-XGBoost prediction models for the ATM inventory cash of different community classifications;
[0059] S5: Based on the samples after widening the ATM cash box balance time series data, train the prediction models for the ATM inventory cash of different community classifications to obtain the trained VRC-XGBoost prediction models for the ATM inventory cash of different community classifications;
[0060] S6: For the test set data, use the trained VRC-XGBoost prediction models for the inventory cash of different community classifications of ATMs to achieve the prediction of the inventory cash of bank ATMs.
[0061] The steps S1 / S2 / S3 / S4 / S5 / S6 are executed in sequence;
[0062] Furthermore: The process of processing and mining the historical balance data of the ATM cash boxes to achieve the community classification of ATMs is as follows:
[0063] S11: Based on the numerical value of the teller number, split it into single-machine data according to the numerical value of the teller number, and draw a grouped box plot of the ATM cash box balance;
[0064] S12: Divide the first threshold baseline and the second threshold baseline based on the ATM grouped box plot; the value of the first threshold baseline is less than the value of the second threshold baseline;
[0065] S13: Compare the cash box balance with the first threshold baseline and the second threshold baseline respectively to achieve the community classification of ATMs.
[0066] The comparison of the cash box balance with the first threshold baseline and the second threshold baseline respectively to achieve the community classification of ATMs includes:
[0067] When the cash box balance of the ATM ≤ the first threshold baseline, then this ATM belongs to the low-activity group;
[0068] When the first threshold baseline < the cash box balance of the ATM < the second threshold baseline, then this ATM belongs to the medium-activity group;
[0069] When the cash box balance of the ATM ≥ the second threshold baseline, then this ATM belongs to the high-activity group.
[0070] Furthermore: The multi-dimensional indicators affecting the ATM withdrawal volume are based on four characteristic dimensions of price level, climate condition, time series, and mobile payment, and thirteen representative indicators are determined: The multi-dimensional indicators affecting the ATM withdrawal volume are based on four characteristic dimensions of price level, climate condition, time series, and mobile payment, and thirteen representative indicators are determined: consumer price index, average temperature, average precipitation, the cash box balance d-3 three days before the prediction time, the cash box balance d-2 two days before the prediction time, the cash box balance d-1 one day before the prediction time (d represents the current day), the cash box balance on the same day of last week of the prediction time, the cash box balance of the day before last week of the prediction time, the cash box balance of the day after last week of the prediction time, first-order difference (time series difference), second-order difference (time series difference), holiday factor, and mobile payment index.
[0071] Construct an input feature set for the bank ATM cash inventory prediction model based on multi-dimensional indicators affecting ATM withdrawal volume;
[0072] Using the Spearman correlation coefficient method, calculate the correlation of each of the thirteen indicators with the cash box balances of the high-activity group, medium-activity group, and low-activity group in turn. For example, the correlation coefficients of the consumer price index with the high, medium, and low-activity groups are 0.172, 0.093, and 0.051 respectively. Considering the correlation calculation results of the thirteen indicators with the three groups comprehensively, analyze whether each indicator is positively correlated, negatively correlated, or insignificantly correlated one by one, and conduct screening.
[0073] When positively correlated, the Spearman correlation coefficient > 0; when completely monotonically positively correlated, the Spearman correlation coefficient = 1;
[0074] When negatively correlated, the Spearman correlation coefficient < 0; when completely monotonically negatively correlated, the Spearman correlation coefficient = -1;
[0075] When there is no tendency at all, the Spearman coefficient = 0;
[0076] Screening method: Through the calculation results of the Spearman correlation coefficient, analyze the correlation of each indicator with the three groups one by one. If the indicator approaches 0 in more than two of the three groups, it is not included in the feature set;
[0077] The VRC-XGBoost prediction model for the ATM cash inventory includes:
[0078] Feature extraction module: Used to decompose the cash box balance sequence into several sub-time series by using the VMD algorithm, and incorporate the several sub-time series into the input feature set;
[0079] Use VMD to decompose the original data (ATM cash box balance time series data) into several sub-time series to solve the constrained variational problem, adaptively decompose the signal into intrinsic mode functions with finite bandwidth, and ensure that the sum of the bandwidths of each mode is minimized.
[0080] Incorporate the decomposed sub-time series and the key influencing indicators screened by the Spearman correlation coefficient into the feature set for use when inputting into the prediction model.
[0081] Feature screening and dimensionality reduction: Used to perform feature screening and dimensionality reduction on the feature set using the recursive feature elimination cross-validation method RFECV algorithm to obtain the optimal input feature set; specifically: Use the RFECCV recursive feature elimination cross-validation method to calculate the importance scores of feature variables with the time series data to be measured one by one and sort them, delete the features with the smallest contribution in the feature set according to the importance scores, and update and save the feature subset. Repeat the above steps until a single target subset, and use ten-fold cross-validation to evaluate the performance of all feature subsets in turn.
[0082] Prediction module: used to predict the ATM inventory cash based on the optimal input feature set.
[0083] The said prediction module: adopts the XGBoost model based on the OPTUNA optimization framework.
[0084] The VRC-XGBoost prediction model is integrated into the VRC-XGBoost integrated model through VMD, RFECV, and XGBoost. The whole process is VRC-XGBoost, and the model in the subsequent prediction part is XGBoost;
[0085] Furthermore: the process of training the prediction models for the ATM inventory cash classified by different communities to obtain the trained VRC-XGBoost prediction models for the ATM inventory cash classified by different communities is as follows:
[0086] Widen the original data based on the optimal input feature set according to the date, and divide the samples; divide the training set and the test set in a ratio of 8:2;
[0087] Train the inventory cash integrated prediction model based on the training set data, and adjust the parameters using the OPTUNA framework to obtain the trained VRC-XGBoost prediction models for the ATM cash box balances classified by different communities;
[0088] Input the test set data into the trained VRC-XGBoost prediction models for the ATM inventory cash classified by different communities, evaluate the performance and generalization ability of the models, and obtain the predicted values of the cash box balances of each ATM.
[0089] A prediction device for the ATM inventory cash of a bank, including:
[0090] Acquisition module: used to acquire the historical time series data set of the ATM cash box balance;
[0091] Classification module: used to mine and process the historical time series data set of the ATM cash box balance and conduct community classification;
[0092] Construction module I: used to construct the input feature set of the bank ATM inventory cash prediction model based on multi-dimensional indicators affecting the ATM withdrawal volume;
[0093] VRC-XGBoost prediction model module: used to construct the VRC-XGBoost prediction models for the ATM inventory cash classified by different communities respectively based on the classified community data and the input feature set;
[0094] Training module: It is used to train the prediction model of the ATM inventory cash for different community classifications based on the widened samples of the ATM cash box balance time series data, and obtain the trained VRC-XGBoost prediction model of the ATM inventory cash for different community classifications;
[0095] Prediction module: It is used to adopt the trained VRC-XGBoost prediction model of the ATM inventory cash for different community classifications to realize the prediction of the bank ATM inventory cash.
[0096] A readable storage medium stores program modules, and the program modules can be run in a processor to implement the method described in any one of the above.
[0097] Based on the cash replenishment strategy of the prediction method of the bank ATM inventory cash described in any one of the above, based on the prediction result of the bank ATM inventory cash, a cash replenishment method is adopted to realize the prediction of the cash replenishment of the bank ATM.
[0098] According to the cash replenishment strategy of the bank ATM inventory cash described above, the cash replenishment method is as follows:
[0099] Taking the demand of the prediction result of the bank ATM inventory cash floating up by X% or the fixed cash reserve amount minus the predicted inventory cash amount floating down by X% as the final cash replenishment strategy.
[0100] The X% adopts 5%;
[0101] A cash replenishment device for bank ATM inventory cash includes:
[0102] Acquisition module: It is used to acquire the historical cash box balance time series data set of the ATM;
[0103] Classification module: It is used to mine and process the historical cash box balance time series data set of the ATM and conduct community classification;
[0104] Construction module I: It is used to construct the input feature set of the bank ATM cash box balance prediction model based on multi-dimensional indicators affecting the ATM withdrawal amount;
[0105] VRC-XGBoost prediction model module: It is used to respectively construct the VRC-XGBoost prediction model of the ATM cash box balance for different community classifications based on the classified community data and the input feature set;
[0106] Training module: It is used to train the prediction model of the ATM cash box balance for different community classifications based on the widened samples of the ATM cash box balance time series data, and obtain the trained VRC-XGBoost prediction model of the ATM cash box balance for different community classifications;
[0107] Prediction module: It is used to adopt the VRC-XGBoost prediction model of the ATM inventory cash classified by different communities that has been trained to realize the prediction of the bank ATM inventory cash;
[0108] Cash replenishment strategy module: Based on the prediction result of the bank ATM inventory cash, a cash replenishment method is adopted to realize the cash replenishment of the bank ATM.
[0109] Embodiment 1:
[0110] The embodiment of the present invention provides a method for predicting the bank ATM inventory cash, including steps S01-S05:
[0111] S01. Process and mine the historical balance data of the ATM cash boxes to realize the community classification of the ATM. Step S01 can be further subdivided into S011-S014:
[0112] S011. Perform preprocessing on the ATM inventory cash, remove the redundant text parts in the mixed description of numbers and text, and uniformly adjust the time data format of the time series data to: year-month-day, and sort it in ascending order of date for sorting; The ascending order lays the foundation for extracting thirteen feature indicators, such as when extracting indicators such as d-3 and d-7;
[0113] S012. Split it into single-machine data tables according to the teller number. The data is grouped by the unique identifier of the ATM - the teller number, and the operation data of each ATM is saved as a single table using a loop.
[0114] S013. Processing of abnormal ATMs. Delete the ATMs with multiple records at the same time; According to the 3σ principle, delete the ATMs with more abnormal values exceeding 3σ in the balance of the cash box, and for the ATMs with fewer abnormal values exceeding 3σ, if the abnormal values do not occur on holidays, use the upper and lower limits of 3σ for substitution; Delete the ATMs with a zero value in the balance of the cash box for 2 consecutive months or more; Delete the ATMs with a missing duration exceeding one-third of the time statistical length;
[0115] S014. Classify the ATMs into three categories with a double baseline of the box plot. Based on the grouped box plot of the ATM, continuously try to draw multiple baselines to classify according to the distribution characteristics of the balance of the cash box, referring to the appendix Figure 2 , and finally it is found that the effect of classifying the ATMs into three categories with a double baseline is the best. For the ATMs below the first threshold baseline, the overall distribution of the balance of the cash box is relatively low, indicating a high usage frequency. Considering that the middle-aged group generally has a large demand for cash withdrawals and a low acceptance of large-scale mobile payment methods, it is speculated that it is set up in the activity area of the middle-aged population.
[0116] The ATM located between the first threshold baseline and the second threshold baseline has a medium distribution of the balance in its cash box, indicating a moderate usage frequency. Considering that the elderly group has relatively small withdrawal needs, is not familiar with mobile payment methods, and has a small range of activities, it is speculated that it is in the activity area of the elderly population.
[0117] For the ATM above the second threshold baseline, the overall distribution of the balance values in its cash box is relatively high, reflecting a low usage frequency. Given that the young group has withdrawal needs but is more inclined to use mobile payment and withdraw money online, and has a wide range of activities and may withdraw money in multiple places, it is speculated that it is in the activity area of the young population. Thus, the community classification of ATMs is achieved, and the ATMs located at the center of the average lines of their respective categories are selected as representatives of three types of typical active ATMs to accurately reflect the operating conditions of the three types of ATMs.
[0118] S02. Combine the four-dimensional multi-indicators affecting the ATM withdrawal volume to construct the input feature set of the VRC-XGBoost prediction model for the ATM's in-cash. Step S02 can be further divided into S021 - S023:
[0119] S021. The balance in the ATM's cash box is affected by multiple factors. Selecting and constructing appropriate features is crucial for accurately predicting the balance in the ATM's cash box.
[0120] Cash is often used for residents' daily small-scale consumption. Therefore, the level of the price level is very likely to affect the ATM withdrawal volume, and the consumer price index of residents is the main measure of the price level.
[0121] Weather often affects residents' travel plans, so the dimension of climate conditions is also taken into consideration and represented by the daily average temperature and daily precipitation.
[0122] The time series feature is a commonly used feature analysis method in time series prediction, mainly including holidays, historical cash box balances, and time series differences.
[0123] Among them, one-hot variables are constructed for holidays according to legal holidays. The historical cash box balances include the balances of the previous three days, the same day of last week, and the two days nearby. The time series differences include the first-order difference and the second-order difference.
[0124] The popularization of mobile payment has had a great impact on the ATM business of banks. The mobile payment index with the index code 884069 formulated and updated by Wind in the financial market is selected to represent the development degree of the mobile payment industry. Through the thirteen indicators selected by the above analysis, the present invention scientifically constructs an index system for the influencing factors of the ATM's cash box balance in four dimensions of price level, climate conditions, time series, and mobile payment. Refer to the appendix Figure 3 .
[0125] S023. For the indicator system based on the influencing factors of the cash balance in the cash box, a correlation analysis is conducted between thirteen indicators and the cash balance in the cash box. According to the Spearman correlation coefficient, key influencing indicators are selected as the input feature set for the VRC-XGBoost prediction model of the ATM inventory cash.
[0126] S03. Based on the classified community data and the input feature set, VRC-XGBoost prediction models for ATM inventory cash for different community classifications are respectively constructed; the VRC-XGBoost prediction model for the ATM cash box balance is based on the integration of multiple algorithms in machine learning, and data feature extraction, feature engineering analysis and model establishment are carried out. Step S03 can be further divided into S031 - S033:
[0127] S031. Use the VMD algorithm to extract data features. Taking three types of typical active ATMs as examples, through the decomposition results, refer to Appendix Figure 4 , where (a) is the VDM decomposition result diagram of the cash box balance of high - activity ATMs, (b) is the VDM decomposition result diagram of the cash box balance of medium - activity ATMs, and (c) is the VDM decomposition result diagram of the cash box balance of low - activity ATMs; it can be seen that the VMD algorithm can effectively decompose the time series of the cash box balance into multiple intrinsic mode functions (IMFs). Incorporate the intrinsic mode functions (IMFs) and residuals obtained by VMD decomposition into the input feature set of the prediction model to obtain a complete feature set.
[0128] S032. The number of input features is not necessarily the more the better. As the number of features increases, the features subsequently added to the prediction model may instead reduce the prediction accuracy of the cash box balance. In order to eliminate redundant variables and thus improve the model accuracy, the RFECV algorithm is used for feature selection. Taking R^2 as the measurement standard, select the feature subset with the highest R^2 as the optimal feature set.
[0129] S033. Establish a VRC-XGBoost prediction model based on the OPTUNA optimization framework, and select the root mean square error and the coefficient of determination as the performance indicators for evaluating the prediction model;
[0130] S04: Train the prediction models for ATM inventory cash for different community classifications to obtain the trained VRC-XGBoost prediction models for ATM inventory cash for different community classifications;
[0131] S05. After inputting the samples with the original data widened, the VRC-XGBoost prediction model of the ATM inventory cash based on the training number can predict the future cash box balance. Step S04 can be further divided into S041 - S042:
[0132] Taking the data of the cash box balance of three typical active ATMs from February 1, 2016 to March 1, 2018 as an example, the original data is broadened based on the optimal feature set by date to construct samples, and the training set and test set are divided in a ratio of 8:2.
[0133] S052. Substitute the ATM inventory cash of the training number into the VRC-XGBoost prediction for prediction, and the prediction effect is as attached Figure 5 , where (a) is the comparison chart of the predicted results of the high-active ATM inventory cash, (b) is the comparison chart of the predicted results of the medium-active ATM inventory cash, and (c) is the comparison chart of the predicted results of the low-active ATM inventory cash;
[0134] Among them, the OPTUNA framework is used to perform heuristic search and parameter tuning on the XGBoost model. The optimized parameters include the learning rate, the number of iterations, the maximum tree depth, and the subsampling ratio to achieve the model structure with the best combination of parameters and improve the prediction accuracy. Comparisons and effectiveness analyses are carried out with traditional time series methods and some integrated prediction methods, proving that the VRC-XGBoost integrated model has significantly more accurate prediction effects for the three types of ATMs.
[0135] Based on expert experience, a cash replenishment strategy for ATMs is formed based on the predicted results of the inventory cash. Specifically, relying on the actual business scenario of the cooperation, referring to the attachment Figure 6 , taking a single ATM as an example, the bank will determine the fixed cash reserve amount internally and use a 5% floating up and down based on the predicted value of the cash box balance as the error. For the sake of insurance, the bank will set the cash box balance of the ATM to the lowest, so as to derive that the ATM cash replenishment amount is the fixed cash reserve amount minus 5% of the downward floating of the predicted value of the cash box balance, providing data support for the cash replenishment strategy.
[0136] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for predicting cash on hand at a bank ATM, characterized by: The following steps are involved: S1: Get the historical tail box balance time series dataset of ATM; S2: Mining and processing the ATM historical tail box balance time series data set to perform community classification; S3: Based on the multi-dimensional indicators that affect the ATM withdrawal volume, the input feature set of the bank ATM cash inventory prediction model is constructed; S4: Based on the classified community data and input feature set, VRC-XGBoost prediction models for ATM cash on hand for different community classifications are constructed respectively; S5: Based on the widened samples of the ATM tail box balance time series data, the prediction model of ATM cash in stock of different community classifications is trained to obtain the trained VRC-XGBoost prediction model of ATM cash in stock of different community classifications; S6: Use the trained VRC-XGBoost prediction model of ATM cash inventory classified by different communities to predict the cash inventory of bank ATMs.
2. A method for predicting cash on hand at a bank ATM according to claim 1, characterized in that: The VRC-XGBoost prediction model for ATM cash on hand includes: Feature extraction module: used to decompose the tail box balance sequence into several sub-time series using the VMD algorithm, and incorporate the several sub-time series into the input feature set; Feature screening and dimensionality reduction: used to perform feature screening and dimensionality reduction on the feature set using the RFECV algorithm to obtain the optimal input feature set; Prediction module: used to predict cash on hand based on an optimal input feature set, the prediction module: adopts an XGBoost model based on the OPTUNA optimization framework.
3. The method for predicting cash on hand at a bank ATM according to claim 1, characterized in that: The process of processing and mining the historical tail box balance data of ATM to realize the community classification of ATM is as follows: Based on the value of the teller number, split the data into single machine data according to the value of the teller number, and draw a grouped box plot of the ATM tail box balance; A first threshold baseline and a second threshold baseline are divided based on the ATM group box plot; the value of the first threshold baseline is less than the value of the second threshold baseline; The tail box balance is compared with the first threshold baseline and the second threshold baseline respectively to achieve community classification of the ATM.
4. The method for predicting cash on hand at a bank ATM according to claim 1, characterized in that: The method of comparing the tail box balance with the first threshold baseline and the second threshold baseline respectively to implement community classification of the ATM includes: When the trunk balance is below the first threshold baseline, it is a low-activity group; When the trunk balance is between the first threshold baseline and the second threshold baseline, it is a moderately active group; When the tail box balance is above the second threshold baseline, it is a high-activity group.
5. The method for predicting cash on hand at a bank ATM according to claim 1, characterized in that: The multi-dimensional indicators that affect the ATM withdrawal volume are based on the four characteristic dimensions of price level, climate conditions, time series and mobile payment, and thirteen representative indicators are determined: consumer price index, average temperature, average precipitation, tail box balance three days before the forecast time, tail box balance two days before the forecast time, tail box balance one day before the forecast time, tail box balance on the same day last week before the forecast time, tail box balance one day before the forecast time last week, tail box balance one day after the forecast time last week, first-order difference, second-order difference, holiday factors and mobile payment index.
6. A method for predicting cash on hand at a bank ATM according to claim 3, characterized in that: The input feature set of the bank ATM tail box balance prediction model is constructed based on the multi-dimensional indicators that affect the ATM withdrawal volume; Using the Spearman correlation coefficient method, the correlation between each of the thirteen indicators and the trunk balance of the high-activity group, the medium-activity group, and the low-activity group was calculated respectively. The correlation calculation results of the thirteen indicators and the three groups were comprehensively considered, and each indicator was analyzed to see whether it was positively correlated, negatively correlated, or not significantly correlated, and then screened.
7. A cash prediction device for bank ATMs, characterized by: include: Acquisition module: used to obtain the historical tail box balance time series dataset of ATM; Classification module: used to mine the ATM historical tail box balance time series data set and perform community classification; Construction module I: used to construct the input feature set of the bank ATM cash inventory prediction model based on the multi-dimensional indicators that affect the ATM withdrawal volume; VRC-XGBoost prediction model module: used to build VRC-XGBoost prediction models for ATM cash on hand for different community classifications based on the classified community data and input feature sets; Training module: used to train the prediction model of ATM cash in stock of different community classifications based on the samples widened by the ATM tail box balance time series data, and obtain the trained VRC-XGBoost prediction model of ATM cash in stock of different community classifications; Prediction module: It is used to predict the cash in bank ATMs using the VRC-XGBoost prediction model of ATM cash in stock classified by different communities that has been trained.
8. A readable storage medium storing a program module, characterized in that: The program module is executed in a processor to implement the method according to any one of claims 1 to 6.
9. A banknote adding strategy based on a bank ATM cash inventory prediction method according to any one of claims 1 to 6, characterized in that: Based on the prediction results of the cash inventory of the bank ATM, the banknote adding method is adopted to realize the prediction of the bank ATM cash adding.
10. A bank ATM cash replenishment strategy according to claim 9, characterized in that: The method of adding banknotes is as follows: The final cash replenishment strategy is to increase the predicted demand for bank ATM cash by X% or to reduce the fixed cash reserve by X% from the predicted cash inventory.