CN111738505A

CN111738505A - Bank branch workload prediction method and device, electronic equipment and storage medium

Info

Publication number: CN111738505A
Application number: CN202010563320.3A
Authority: CN
Inventors: 杜姗; 蔡为彬; 周远侠; 唐杰聪
Original assignee: Industrial and Commercial Bank of China Ltd ICBC
Current assignee: Industrial and Commercial Bank of China Ltd ICBC
Priority date: 2020-06-19
Filing date: 2020-06-19
Publication date: 2020-10-02

Abstract

The invention provides a bank outlet workload prediction method and device, electronic equipment and a storage medium, wherein the method comprises the following steps: acquiring workload flow data and auxiliary data of a bank outlet; performing feature extraction according to the workload pipeline data and the auxiliary data to obtain feature data; inputting the characteristic data into a plurality of pre-trained monomer prediction models respectively to obtain corresponding predicted values; processing the predicted value of each monomer prediction model by adopting a non-dominated sorting genetic algorithm with an elite strategy to obtain a bank outlet workload prediction result; the single prediction model is a holt-windows model, a multilayer perceptron model, a decision tree model, a gradient lifting tree regression model and a long-short term memory network model, can efficiently and accurately predict the workload of each post of a bank network in the future for a period of time every day, and guides the bank network to determine the number of workers needed by each post.

Description

Bank branch workload prediction method and device, electronic equipment and storage medium

Technical Field

The invention relates to the technical field of artificial intelligence, in particular to a method and a device for predicting the workload of a bank outlet, electronic equipment and a storage medium.

Background

The bank outlets are places where banks work outside, and are generally divided into branches, branch processing, savings centers and the like. In order to effectively provide high-quality service for customers, the workload of the bank outlets needs to be accurately predicted, and the staff of the outlets are reasonably scheduled, so that the optimization of the combination of the positions and the labor of the outlets and the optimal configuration of the customer service are realized.

The traditional bank branch can predict the transaction amount and workload of each post of the branch usually based on the experience of business personnel and the manual statistical data of historical transaction data, and further guide the scheduling work of each post of the branch according to the relevant policy in the row. Because the transaction amount of the network is influenced by economic environment and peripheral human environment, including external environment change, national festivals and holidays, major events and the like, greater deviation is brought by experience memory and judgment of business personnel, the network human resource allocation is excessive or insufficient, and the bank operation cost is increased.

Disclosure of Invention

Aiming at the problems in the prior art, the invention provides a method, a device, equipment and a medium for predicting the workload of a bank outlet, which can at least partially solve the problems in the prior art.

In order to achieve the purpose, the invention adopts the following technical scheme:

in a first aspect, a method for predicting workload of a banking outlet is provided, which includes:

acquiring workload flow data and auxiliary data of a bank outlet;

performing feature extraction according to the workload pipeline data and the auxiliary data to obtain feature data;

inputting the characteristic data into a plurality of pre-trained monomer prediction models respectively to obtain corresponding predicted values;

processing the predicted value of each monomer prediction model by adopting a non-dominated sorting genetic algorithm with an elite strategy to obtain a bank outlet workload prediction result;

the single prediction model is a holt-windows model, a multilayer perceptron model, a decision tree model, a gradient lifting tree regression model and a long-short term memory network model.

Further, the bank outlet workload prediction method further comprises the following steps:

evaluating the effect of the pre-trained monomer prediction models by using the root mean square error;

and selecting part of monomer prediction models for predicting the workload of the bank outlets according to the evaluation result.

Further, the assistance data comprises: holiday data, weather data, salary date data and repayment date data.

Further, the extracted features are obtained by analyzing single-column operation features, multi-column operation features and aggregation operation features by using principal components.

Further, the single column of operational features includes: workload rate of change characteristics, holiday characteristics, date characteristics, repayment date characteristics, payday characteristics.

Further, the multi-column operating features include: time-based work average and median.

Further, the aggregation operation features include: attribute-based workload mean and median.

In a second aspect, a bank outlet workload prediction apparatus is provided, which includes:

the method comprises the steps that an original data acquisition module acquires workload flow data and auxiliary data of a bank outlet;

the characteristic extraction module is used for extracting characteristics according to the workload pipeline data and the auxiliary data to obtain characteristic data;

the prediction module is used for respectively inputting the characteristic data into a plurality of pre-trained monomer prediction models to obtain corresponding prediction values;

the optimal solution calculation module is used for processing the predicted value of each monomer prediction model by adopting a non-dominated sorting genetic algorithm with an elite strategy to obtain a bank branch workload prediction result;

In a third aspect, an electronic device is provided, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the steps of the bank outlet workload prediction method.

In a fourth aspect, a computer readable storage medium is provided, on which a computer program is stored, which computer program, when being executed by a processor, realizes the steps of the above-mentioned bank outlet workload prediction method.

The invention provides a bank outlet workload prediction method, a device, equipment and a medium, wherein the method comprises the following steps: acquiring workload flow data and auxiliary data of a bank outlet; performing feature extraction according to the workload pipeline data and the auxiliary data to obtain feature data; inputting the characteristic data into a plurality of pre-trained monomer prediction models respectively to obtain corresponding predicted values; processing the predicted value of each monomer prediction model by adopting a non-dominated sorting genetic algorithm with an elite strategy to obtain a bank outlet workload prediction result; the single prediction model is a holt-windows model, a multilayer perceptron model, a decision tree model, a gradient lifting tree regression model and a long-short term memory network model. The method can efficiently and accurately predict the workload of each post every day in a future period of time of a bank network, and guide the bank network to determine the number of workers required by each post, thereby realizing the optimization of the combination of the post and labor of the network and the optimal configuration of customer service.

In order to make the aforementioned and other objects, features and advantages of the invention comprehensible, preferred embodiments accompanied with figures are described in detail below.

Drawings

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts. In the drawings:

fig. 1 is a first flowchart of a workload prediction method of a banking outlet in an embodiment of the present invention;

FIG. 2 is a flowchart illustrating a second method for predicting workload of banking outlets in an embodiment of the present invention;

fig. 3 is a block diagram of a workload prediction apparatus of a banking outlet in an embodiment of the present invention;

FIG. 4 is a diagram illustrating a system for accurately predicting workload of a banking outlet according to an embodiment of the present invention;

FIG. 5 is a block diagram illustrating a data processing apparatus in a system for accurately predicting workload of a banking outlet according to an embodiment of the present invention;

FIG. 6 is a diagram illustrating a structure of a feature engineering apparatus in a system for accurately predicting workload of a banking outlet according to an embodiment of the present invention;

FIG. 7 is a diagram illustrating a structure of a modeling apparatus in a system for accurately predicting workload of a banking outlet according to an embodiment of the present invention;

fig. 8 is a structural diagram illustrating a prediction device 4 in a system for accurately predicting the workload of a banking outlet according to an embodiment of the present invention;

fig. 9 shows a flowchart of the data processing apparatus 1 in the accurate workload prediction system of a banking outlet in the embodiment of the present invention;

FIG. 10 is a flowchart of the feature engineering apparatus 2 in the system for accurately predicting the workload of a banking outlet according to an embodiment of the present invention;

fig. 11 shows a flowchart of the modeling apparatus 3 in the accurate workload prediction system of a banking outlet in the embodiment of the present invention;

fig. 12 is a flowchart illustrating the predicting apparatus 4 in the system for accurately predicting the workload of a banking outlet in the embodiment of the present invention;

fig. 13 is a block diagram of an electronic device according to an embodiment of the invention.

Detailed Description

In order to make the technical solutions better understood by those skilled in the art, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only partial embodiments of the present application, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

It should be noted that the terms "comprises" and "comprising," and any variations thereof, in the description and claims of this application and the above-described drawings, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

It should be noted that the embodiments and features of the embodiments in the present application may be combined with each other without conflict. The present application will be described in detail below with reference to the embodiments with reference to the attached drawings.

The embodiment of the invention provides a bank branch workload prediction technology, which comprehensively uses the means of web crawlers, data cleaning, feature engineering, machine learning and the like, efficiently and accurately predicts the workload of each post of a branch every day in a future period of time, plans the number of the shift values of each post in the future period of time for the bank branch, and assists the development of the optimization work of the post and labor combination of the branch, thereby promoting the customer service manager to change to the comprehensive transformation.

Fig. 1 is a first flowchart of a workload prediction method of a banking outlet in an embodiment of the present invention; as shown in fig. 1, the method for predicting the workload of a banking outlet may include the following steps:

step S1000: acquiring workload flow data and auxiliary data of a bank outlet;

wherein the assistance data comprises: holiday data, weather data, payday data and repayment day data, such as domestic legal holiday data, payment days of network points, repayment day data, data of holidays of special festivals in partial areas, weather information (including wind power level, temperature, rain level and the like) of the areas where the network points are located and the like.

It should be noted that the workload pipeline data refers to data of a current cycle, data of a historical synchronization, and the like.

For example, the workload of the next period (next day, next week, next month, etc.) of a network point needs to be predicted, and the workload flow data may be the workload flow data of the current period (current day, current week, current month) and the current week data of the previous year and/or the current month data of the previous year and/or the current day data of the previous year, etc.

Step S2000: performing feature extraction according to the workload pipeline data and the auxiliary data to obtain feature data;

specifically, a parameter that is closely related to the workload is extracted as a characteristic parameter.

Step S3000: inputting the characteristic data into a plurality of pre-trained monomer prediction models respectively to obtain corresponding predicted values;

specifically, the plurality of individual prediction models are different machine learning models, such as a holt-windows model, a multilayer perceptron model, a decision tree model, a gradient lifting tree regression model, a long-short term memory network model, and the like.

Step S4000: and processing the predicted value of each monomer prediction model by adopting a non-dominated sorting genetic algorithm with an elite strategy to obtain a bank outlet workload prediction result.

Specifically, a nondominant sorting genetic algorithm (NSGA Π for short) with an elite strategy is adopted to solve pareto solutions for model prediction values generated by each monomer prediction model to obtain a final prediction result.

By adopting the technical scheme, different characteristics of various machine models are utilized to respectively predict, then a non-dominated sorting genetic algorithm with elite strategy is adopted to carry out multi-objective optimization on the predicted value of each monomer prediction model, namely, results of a plurality of monomer prediction models are dynamically combined, the characteristics of different machine learning models are effectively utilized, the workload of a bank outlet is accurately predicted, various factors such as the change of external objective environment, special holiday periods, major incident emergencies and the like are timely memorized and predicted, an artificial intelligence related technology is introduced to mine the business peak-valley rule, the transaction amount and the workload of each post in each day in a future period of time of the outlet are rapidly and accurately predicted, thereby guiding the bank to compile a counter customer service manager quantitative configuration scheme, scientifically guiding the customer service manager to 'go out' to develop the outlet of the outlet for extension of customers and product marketing, fully improves the service efficiency of the bank outlets.

In an alternative embodiment, referring to fig. 2, the method for predicting the workload of a banking outlet may further include the following:

step S5000: the effect of the pre-trained multiple-monomer prediction models is evaluated using the root mean square error.

In particular, the root mean square error, RMSE, method is used to evaluate the model effect. The RMSE values of all models are solved on a test set, the two steps are divided into two steps, firstly, the prediction results of each sample data in the test set on all models are calculated, and the prediction results are stored in a database; and then, after the prediction results of all the test set samples are calculated, respectively calculating the RMSE value of the predicted value and the real value of each model on the test set according to the RMSE formula, wherein the smaller the value is, the better the effect is represented.

Step S6000: and selecting part of monomer prediction models for predicting the workload of the bank outlets according to the evaluation result.

Specifically, after the values of the models RMSE are sorted from small to large, a preset number of models ranked at the top are selected as available models.

By adopting the technical scheme, the monomer prediction model for prediction can be optimized, and the accuracy of prediction is further improved.

In an alternative embodiment, the extracted features are obtained by analyzing single-column operation features, multi-column operation features and aggregation operation features by using principal components.

Specifically, the single-column operation features are obtained by performing operations such as four fundamental operations, square, power, exponent, logarithm and the like on auxiliary data such as the workload of each post of a website in the past 3 years and holidays, and include: workload rate of change characteristics, holiday characteristics, date characteristics, repayment date characteristics, payday characteristics. Such as: the change rate of the workload of each post in each day relative to the previous day, the same week number of the last week and the same period of the last year is selected to obtain 90 change rates of the workload of the last 30 days; the date is divided into 18 characteristics in total, wherein the total is 108 characteristics, namely, the year, the month, the day, the week in the month, the week in the year, the beginning of the month, the middle of the month, the end of the month, whether the day is a legal holiday, whether the day before is a holiday, whether the day after is a holiday, whether the day is a working day, whether the day is a repayment day, whether the day is a paid day, how many days the day is away from the previous paid day, how many days the day is away from the next paid day, and whether the day is a cold-hot holiday (a network point in a school).

In addition, the multi-column operation characteristics are obtained by operations of summing, differencing, averaging, calculating the maximum value/minimum value/median of the workload in a certain period of time in the workload of each post in the last 3 years of the network, and the like, and comprise the following steps: time-based work average and median.

For example, the average value and the median of the workload of each post corresponding to the forecast date in the same week in the last 4 weeks are calculated to obtain 2 features; and calculating the workload average value and the median of 7 legal festivals and holidays in the past year of each post, 1 day before and after 12 repayment days and 1 day before and after 24 paying days to obtain 43 features, and the total number of the 43 features is 45.

Furthermore, the aggregate operation feature is obtained by counting the workload of each site of a website in the past 3 years according to certain attributes (such as time period, website number appearing many times, etc.), and includes: attribute-based workload mean and median.

Such as: calculating the average value and the median of the workload of each post in the past 12 months, and obtaining 24 characteristics; and calculating the variance, the 25 th percentile, the 75 th percentile and the mean value of the workload of each post in the last 30 days every day to obtain 4 characteristics, wherein the total number of the characteristics is 28. Finally, the workload of each post in the past 30 days is put into the feature column.

And performing principal component analysis on the features to obtain feature data. Specifically, Principal Component Analysis (PCA) is adopted to perform principal component analysis on the features, and principal components with accumulated variance larger than a preset threshold are selected to represent feature data of each post of a dot.

It is worth to be noted that the step of determining the features by using principal component analysis is executed before modeling, and the machine learning model learns the association relationship between each feature and the workload of the website by using the extracted feature characterization data, so as to finally realize the function of model prediction.

By adopting the technical scheme, the dimension reduction can be realized, the complex problem is simplified, and meanwhile, the relevant information among the constructed features can be eliminated.

In an optional embodiment, the bank outlet workload prediction method may further include the following:

step I: acquiring historical workload flow data and auxiliary data of the bank outlets;

step II: performing feature extraction according to the historical workload flow data and the auxiliary data to obtain historical feature data;

step IV: and taking the historical characteristic data as a training sample and a testing sample, training and testing each pre-established monomer prediction model until the testing result meets the preset requirement, and obtaining the trained monomer prediction model for predicting the workload of the bank outlets.

Specifically, the historical feature data is sorted from large to small according to dates and is divided into a training set and a test set according to a preset proportion.

Wherein, the training process is as follows: inputting the training samples into a pre-established monomer prediction model, comparing the output result of the monomer prediction model with the labels of the training samples, and reversely adjusting the parameters of the monomer prediction model based on the comparison result to realize model training.

The test process comprises the following steps: inputting a test sample into the trained monomer prediction model, comparing the label of the test sample with the output of the model, judging whether the label meets the preset requirement, if so, successfully training, and if not, optimizing the current model and/or applying the updated training sample set to perform model training again.

Those skilled in the art will appreciate that the bank outlet workload prediction method may further comprise: and constructing a monomer prediction model.

Specifically, the single prediction model is constructed by transforming a machine learning algorithm to a certain extent and packaging the single prediction model into a standardized interface and unified input and output. The method mainly comprises the following steps: firstly, inputting a data format and defining the parameter range of each machine learning algorithm; adding methods such as grid search, random search or Bayesian optimization and the like to each machine learning algorithm to realize automatic parameter adjustment of the machine learning method; and thirdly, outputting the format and the content of the model file.

In addition, modeling is performed by using a single machine learning algorithm, and the modeling can be specifically called in Python.

In an optional embodiment, the bank branch workload prediction method may further include:

and evaluating the quality of the current workload pipeline data and the auxiliary data or the historical workload pipeline data and the auxiliary data.

Specifically, the quality of the current workload pipeline data and the auxiliary data or the quality of the historical workload pipeline data and the auxiliary data are evaluated, and if the evaluation result is poor, the data are indicated to be unsuitable for prediction or training samples.

For example, the evaluation may be: first, it is determined whether the last month of mesh point data is missing, and if so, the quality is evaluated as poor. Then, calculating the data missing proportion of the network point working day, and if the missing proportion is less than or equal to 20%, evaluating the quality of the network point data to be good; and if the deletion ratio is more than 20%, recalculating the deletion ratio of the 10 data with the largest removal date until the deletion ratio is less than or equal to 20% or the remaining data volume of the network point is less than 1 month, and if the final data volume is less than one month, evaluating the quality of the network point data as poor, otherwise, evaluating as good.

and cleaning the current workload pipeline data and the auxiliary data or the historical workload pipeline data and the auxiliary data.

Specifically, the data cleansing includes: removal of duplicate data, missing data supplementation and tagging, noisy data smoothing and tagging, outlier data repair and tagging, and the like.

When filling missing values, the data to be supplemented are divided into two types: the data missing in the working days are supplemented, and the data of the non-working days are supplemented. Data on weekdays: and selecting one of three modes of a front-and-back-periphery mean value, a front-and-back-periphery median value or a front-and-back-periphery same-periphery mean value for filling according to the change amplitude of the data of one week before and after the missing value. Non-workday data: each post workload complement is 0. The supplemental data needs to be marked.

It should be noted that the noise data refers to interference data in a data set, and the noise data is obtained when the workload is negative in a website scheduling scene. And (3) adopting a normal distribution 3 sigma principle when judging the noise data, and setting a point which is three times of the standard deviation of the data set as the noise data. And smoothing the noise data by adopting a smoothdata method, and marking the smoothed data.

Furthermore, when outlier data is repaired and flagged: determining outlier data through the upper and lower edges of the boxplot, and then selecting one of three modes of a front and back four-week average value, a front and back four-week median value or a front and back four-week average value according to the variation amplitude of the data of a circle before and after an outlier to repair. The repaired data needs to be marked.

and preprocessing the current workload pipeline data and the auxiliary data or the historical workload pipeline data and the auxiliary data.

Specifically, the preprocessing may include: data integration, data specification and data transformation.

The data sources acquired by the system are various, and attributes representing the same concept can have different names or units in different data sources, so that inconsistency and redundancy can be caused.

In addition, the simplified representation of the data is obtained through the reduction technology, the occupied space of the simplified data is reduced, but nearly the same analysis result can be generated, and the efficiency of the whole system can be improved.

Moreover, data is more suitable for the system to carry out data mining through data transformation. The same category is represented by the same number, so that the text data is converted into discrete numerical data.

Based on the same inventive concept, the embodiment of the present application further provides a device for predicting the workload of a bank outlet, which can be used to implement the method described in the above embodiment, as described in the following embodiments. Because the principle of the device for predicting the workload of the bank outlets for solving the problems is similar to that of the method, the implementation of the device for predicting the workload of the bank outlets can be referred to the implementation of the method, and repeated details are not repeated. As used hereinafter, the term "unit" or "module" may be a combination of software and/or hardware that implements a predetermined function. Although the means described in the embodiments below are preferably implemented in software, an implementation in hardware, or a combination of software and hardware is also possible and contemplated.

Fig. 3 is a block diagram of a bank outlet workload prediction apparatus in an embodiment of the present invention. As shown in fig. 3, the device for predicting the workload of a banking outlet specifically includes: the system comprises an original data acquisition module 1a, a feature extraction module 1b, a prediction module 1c and an optimal solution calculation module 1 d.

The original data acquisition module 1a acquires the workload flow data and auxiliary data of a bank outlet;

the feature extraction module 1b performs feature extraction according to the workload pipeline data and the auxiliary data to obtain feature data;

the prediction module 1c inputs the characteristic data into a plurality of pre-trained monomer prediction models respectively to obtain corresponding prediction values;

the optimal solution calculation module 1d adopts a non-dominated sorting genetic algorithm with an elite strategy to process the predicted value of each monomer prediction model to obtain a bank branch workload prediction result;

In order to make the present invention better understood by those skilled in the art, the following examples are given to illustrate the steps of the present invention:

fig. 4 shows a structure diagram of a bank outlet workload accurate prediction system in an embodiment of the present invention, and as shown in fig. 4, the bank outlet workload accurate prediction system based on machine learning includes a data processing device 1, a feature engineering device 2, a modeling device 3, and a prediction device 4, where the data processing device 1 is connected to the feature engineering device 2, the feature engineering device 2 is connected to the modeling device 3, and the modeling device 3 is connected to the prediction device 4.

Specifically, the method comprises the following steps:

the data processing device 1 is responsible for acquiring original data such as historical workload data and auxiliary data of each post of a network point, and preprocessing the acquired data, and mainly comprises: obtaining original data, evaluating data quality, cleaning data and reducing data. After passing through the data processing device 1, high quality data with accuracy, integrity and consistency can be obtained.

The feature engineering device 2 extracts features from the raw data obtained by the data processing device 1 to the maximum extent for use in algorithms and models, and mainly comprises: and (4) feature construction and feature extraction. After passing through the feature engineering device 2, data which can be identified by a machine learning algorithm and can be used for training can be obtained, better training data features can be obtained, and the performance of a machine learning model is improved.

And the modeling device 3 is responsible for training and selecting a plurality of machine learning models with better prediction effects. The device mainly includes: the method comprises the steps of establishing a machine learning algorithm pool, training a single machine learning algorithm model and evaluating the prediction effect of the model.

Specifically, the proper machine learning model parameters are adaptively searched by using the characteristic data, so that the machine learning model can well fit the working volume change rule under a certain parameter combination, and the overfitting condition can not occur.

And the prediction device 4 dynamically combines and predicts the daily workload of each post in a future period of time by using the trained model. Predicting future workload by using a single model with better prediction effect acquired from the modeling device 3; and then, carrying out weighted average on the predicted value of the single model by using a multi-objective optimization algorithm to obtain a combined prediction result of the net point workload.

Fig. 5 is a structural diagram of a data processing apparatus in a system for accurately predicting workload of banking outlets in an embodiment of the present invention, and as shown in fig. 5, the data processing apparatus 1 includes: the system comprises a raw data acquisition unit 11, a data quality evaluation unit 12, a data cleaning unit 13 and a data preprocessing unit 14, wherein:

the original data acquisition unit 11 is responsible for acquiring daily workload data, network operating modes, domestic legal holiday data and network characteristic data of each post of the network in the last three years, wherein the characteristic data mainly comprises network payday data, repayment date data and part of regional special holiday data.

The data quality evaluation unit 12 is responsible for evaluating the quality of the original data (mainly historical workload data) of the network point, and if the quality evaluation result of the network point data is poor, the network point is not suitable for modeling.

The data cleaning unit 13 is responsible for cleaning the original data of the mesh points, and includes: removing repeated data, supplementing missing data, smoothing noise points and repairing outliers.

The data preprocessing unit 14 is responsible for preprocessing the original data, and the processing mode mainly includes: data integration, data reduction and data transformation.

Fig. 6 shows a structural diagram of a feature engineering apparatus in a system for accurately predicting workload of a banking outlet in an embodiment of the present invention, and as shown in fig. 6, the feature engineering apparatus includes: a feature construction unit 21 and a feature extraction unit 22, wherein:

the feature construction unit 21 is configured to mine feature data from the website workload data, the website pattern data, the holiday data, and the like.

The feature extraction unit 22 is responsible for mapping the feature space constructed by the feature construction unit 21 to obtain a new feature space, so that feature dimensions are reduced, and calculation efficiency is improved.

Fig. 7 is a structural diagram illustrating a modeling apparatus in a system for accurately predicting workload of a banking outlet according to an embodiment of the present invention, where as shown in fig. 7, the modeling apparatus includes: a machine learning algorithm pool unit 31, a single machine learning algorithm modeling unit 32 and a single algorithm model effect evaluation unit 33 are constructed, wherein:

the build machine learning algorithm pool unit 31 packages the following machine learning algorithms into an algorithm pool: the method comprises a holt-windows algorithm, a multi-layer perceptron (MLPRegersion) algorithm, a Decision Tree (Decision Tree) algorithm, a Gradient Boosting Decision Tree (GBDT) regression algorithm and a Long Short-term memory network (LSTM) algorithm, and is used for providing a single machine learning algorithm which can be selected when the system models all the mesh points.

And the single machine learning algorithm modeling unit 32 is responsible for training the feature data output by the feature engineering device 2 by using the algorithms in the algorithm pool to obtain a single machine learning algorithm model.

And the single algorithm model effect evaluation unit 33 is responsible for evaluating the model effect trained by the single machine learning algorithm modeling unit 32 and selecting 3 algorithm models with the best prediction effect.

Fig. 8 is a structural diagram of a prediction apparatus 4 in a system for accurately predicting workload of banking outlets in an embodiment of the present invention, and as shown in fig. 8, the prediction apparatus includes: a current day feature generation unit 41, a single model prediction unit 42, a combined model prediction unit 43, wherein:

the present day feature generation unit 41 is responsible for generating feature data for predicting daily workload of each post in a future period of time.

And the single model prediction unit 42 is used for predicting the workload of each post every day in a future period of time by using 3 algorithm models selected by the modeling device respectively, so that 3 model prediction values can be obtained.

And the combined model prediction unit 43 is responsible for performing weighted average on the 3 model prediction values generated by the single model prediction unit 42 by using a multi-objective optimization algorithm to obtain a final prediction result.

Fig. 9 shows a flowchart of a data processing apparatus 1 in a system for accurately predicting workload of a bank outlet in an embodiment of the present invention, which includes the specific steps of:

step S101: raw data is acquired.

Firstly, acquiring workload flow data of each post of a network point every day in the last three years, monthly payoff date and repayment date data of the network point in the last three years and special holiday data (particularly minority nationalities) of the network point in the last three years by using an inline system; then, since the non-workday network points have no data records, the working mode of the network points in each quarter of the last three years can be inferred from the workload flow data, and the working modes of the network points are divided into the following 4 types: working day of 5 days + Saturday rest, working day of 5 days + Saturday work + Sunday rest, working day of 5 days + Saturday rest + Sunday work, seven days of a week are all on duty; and finally, the web crawler acquires national legal holiday data of the three years and weather information (including wind power level, temperature, rain level and the like) of the region where the network point is located.

Step S102: and (6) evaluating the data quality.

And evaluating the quality of the daily workload pipeline data of each post of the network in the last three years, modeling the network with good quality, and not modeling the network with poor quality. The evaluation method is as follows: first, it is determined whether the last month of mesh point data is missing, and if so, the quality is evaluated as poor. Then, calculating the data missing proportion of the network point working day, and if the missing proportion is less than or equal to 20%, evaluating the quality of the network point data to be good; and if the deletion ratio is more than 20%, recalculating the deletion ratio of the 10 data with the largest removal date until the deletion ratio is less than or equal to 20% or the remaining data volume of the network point is less than 1 month, and if the final data volume is less than one month, evaluating the quality of the network point data as poor, otherwise, evaluating as good.

Step S103: missing data supplementation and marking.

When filling missing values, the data that needs to be supplemented fall into two categories: the data missing in the working days are supplemented, and the data of the non-working days are supplemented. Data on weekdays: and selecting one of three modes of a front-and-back-periphery mean value, a front-and-back-periphery median value or a front-and-back-periphery same-periphery mean value for filling according to the change amplitude of the data of one week before and after the missing value. Non-workday data: each post workload complement is 0. The supplemental data needs to be marked.

Step S104: noise data smoothing and labeling.

The noise data refers to interference data in a data set, and the noise data is obtained when the workload is negative in a website scheduling scene. And (3) adopting a normal distribution 3 sigma principle when judging the noise data, and setting a point which is three times of the standard deviation of the data set as the noise data. And smoothing the noise data by adopting a smoothdata method, and marking the smoothed data.

Step S105: outlier data repair and tagging.

Determining outlier data through the upper and lower edges of the boxplot, and then selecting one of three modes of a front and back four-week average value, a front and back four-week median value or a front and back four-week average value according to the variation amplitude of the data of a circle before and after an outlier to repair. The repaired data needs to be marked.

Step S106: and (6) integrating data.

Due to the fact that the data acquired by the system are various in source, attributes representing the same concept can have different names or units in different data sources, inconsistency and redundancy can be caused, and the data are integrated by adopting a correlation analysis method.

Step S107: and (5) data reduction.

The simplified representation of the data is obtained through the reduction technology, the occupied space of the simplified data is reduced, but nearly the same analysis result can be generated, and the efficiency of the whole system can be improved.

Step S108: and (5) data transformation.

And the data is more suitable for the system to carry out data mining through data transformation. The same category is represented by the same number, so that the text data is converted into discrete numerical data.

Fig. 10 shows a flowchart of a feature engineering apparatus 2 in a system for accurately predicting workload of a bank outlet in an embodiment of the present invention, which specifically includes the following steps:

step S201: single-row operation of feature construction.

The method mainly comprises the steps of carrying out four-fundamental operation, square operation, evolution operation, power operation, exponent operation, logarithm operation and the like on auxiliary data such as the workload of each post of a network point in the past 3 years, holidays and the like. Calculating the change rate of the workload of each post per day relative to the previous day, the same week number of the last week and the same period of the last year, and selecting 90 change rates of the workload of the last 30 days; the date is divided into 18 characteristics in total, namely, the year, the month, the day, the week in the month, the week in the year, the beginning of the month, the middle of the month, the end of the month, whether the day is a legal holiday, whether the day before is a holiday, whether the day after is a holiday, whether the day is a working day, whether the day is a repayment day, whether the day is a paid day, how many days the day is away from the last paid day, how many days the day is away from the next paid day, and whether the day (a branch in a school) is a summer holiday. There are a total of 108 features.

Step S202: multi-row operation of feature construction

The method mainly comprises the steps of summing, differencing and averaging the workload of a certain period of time in the workload of each post of the network in the last 3 years, and calculating the maximum value/minimum value/median of the workload of the certain period of time. Calculating the average value and median of the workload of each post corresponding to the predicted date in the same week in the last 4 weeks to obtain 2 characteristics; and calculating the workload average value and the median of 7 legal festivals and holidays in the past year of each post, 1 day before and after 12 repayment days and 1 day before and after 24 paying days to obtain 43 features, and the total number of the 43 features is 45.

Step S203: grouping/aggregation operations for feature building.

The workload of each post of the website in the past 3 years is mainly counted according to certain attributes (such as time period, website number appearing for multiple times and the like). Calculating the average value and the median of the workload of each post in the past 12 months, and obtaining 24 characteristics; and calculating the variance, the 25 th percentile, the 75 th percentile and the mean value of the workload of each post in the last 30 days every day to obtain 4 characteristics, wherein the total number of the characteristics is 28. Finally, the workload of each post in the past 30 days is put into the feature column.

Step S204: and (5) feature extraction.

And (3) carrying out principal component analysis on the feature data constructed in the steps S201-S203 by adopting a Principal Component Analysis (PCA), and selecting principal components with the accumulated variance larger than 85% to represent the feature data of each post of the net point, so that the dimension reduction is realized, the complex problem is simplified, and meanwhile, the relevant information among the constructed features can be eliminated.

Fig. 11 shows a flowchart of a modeling apparatus 3 in a system for accurately predicting workload of a bank outlet in an embodiment of the present invention, which includes the specific steps of:

step S301: the data set is split into a training set and a test set.

And sorting the feature data generated by the feature engineering device 3 from big to small according to the date, and splitting the feature data into a training set and a testing set in a ratio of 4: 1.

Step S302: and constructing a machine learning algorithm pool.

The machine learning algorithm provided by the system is transformed to a certain extent, and is packaged into a standardized interface for unified input and output. The method mainly comprises the following steps: firstly, inputting a data format; secondly, defining the parameter range of each machine learning algorithm, and then adding methods such as grid search, random search or Bayesian optimization and the like to each machine learning algorithm to realize automatic parameter adjustment of the machine learning method; and thirdly, outputting the format and the content of the model file.

Step S303: modeling was performed using a single machine learning algorithm.

Model training is performed on the training set split in step S301 by using the algorithms in the machine learning algorithm pool constructed in step S302, and a model of each algorithm can be obtained.

Step S304: evaluating the effect of a single machine learning algorithm model.

The effect of the model was evaluated using the root mean square error, RMSE, method. The RMSE values of all models are solved on a test set, the two steps are divided into two steps, firstly, the prediction results of each sample data in the test set on all models are calculated, and the prediction results are stored in a database; and then, after the prediction results of all the test set samples are calculated, respectively calculating the RMSE value of the predicted value and the real value of each model on the test set according to the RMSE formula, wherein the smaller the value is, the better the effect is represented. And finally, sorting the RMSE values of the models from small to large, and selecting the first 3 models as available models.

Fig. 12 shows a flowchart of a prediction device 4 in a system for accurately predicting workload of a bank outlet in an embodiment of the present invention, which includes the specific steps of:

step S401: and generating the characteristic data of the current day.

And calling the feature engineering device to generate feature data of the day for predicting future workload.

Step S402: predicting using a single model;

in particular, a single machine learning model is used to predict future workloads.

And 3 machine learning models selected by the modeling device are called to predict the future workload of each post of the network, so that 3 prediction results exist in each post every day in a future period of time.

Step S403: the individual model predictions are combined.

The method adopts a non-dominated sorting genetic algorithm (NSGA Π for short) with elite strategies to solve pareto solutions for the prediction results of 3 machine learning models, namely the result of dynamic combination of the prediction results of a single model.

The apparatuses, modules or units illustrated in the above embodiments may be implemented by a computer chip or an entity, or implemented by a product with certain functions. A typical implementation device is an electronic device, which may be, for example, a personal computer, a laptop computer, a cellular telephone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

In a typical example, the electronic device specifically includes a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor implements the predicting step when executing the program:

referring now to FIG. 13, shown is a schematic diagram of an electronic device 600 suitable for use in implementing embodiments of the present application.

As shown in fig. 13, the electronic apparatus 600 includes a Central Processing Unit (CPU)601 that can perform various appropriate works and processes according to a program stored in a Read Only Memory (ROM)602 or a program loaded from a storage section 608 into a Random Access Memory (RAM)) 603. In the RAM603, various programs and data necessary for the operation of the system 600 are also stored. The CPU601, ROM602, and RAM603 are connected to each other via a bus 604. An input/output (I/O) interface 605 is also connected to bus 604.

The following components are connected to the I/O interface 605: an input portion 606 including a keyboard, a mouse, and the like; an output portion 607 including a display such as a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), and the like, and a speaker; a storage section 608 including a hard disk and the like; and a communication section 609 including a network interface card such as a LAN card, a modem, or the like. The communication section 609 performs communication processing via a network such as the internet. The driver 610 is also connected to the I/O interface 605 as needed. A removable medium 611 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is mounted on the drive 610 as necessary, so that a computer program read out therefrom is mounted as necessary on the storage section 608.

In particular, according to an embodiment of the present invention, the processes described above with reference to the flowcharts may be implemented as computer software programs. For example, an embodiment of the invention comprises a computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the above-mentioned method steps.

In such an embodiment, the computer program may be downloaded and installed from a network through the communication section 609, and/or installed from the removable medium 611.

Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.

For convenience of description, the above devices are described as being divided into various units by function, and are described separately. Of course, the functionality of the units may be implemented in one or more software and/or hardware when implementing the present application.

The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.

As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The application may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The application may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.

The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the system embodiment, since it is substantially similar to the method embodiment, the description is simple, and for the relevant points, reference may be made to the partial description of the method embodiment.

The above description is only an example of the present application and is not intended to limit the present application. Various modifications and changes may occur to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A bank outlet workload prediction method is characterized by comprising the following steps:

acquiring workload flow data and auxiliary data of a bank outlet;

2. The banking outlet workload prediction method according to claim 1, further comprising:

3. A banking outlet workload prediction method according to claim 1, characterised in that said auxiliary data comprises: holiday data, weather data, salary date data and repayment date data.

4. The bank branch workload prediction method according to claim 1, wherein the extracted features are obtained by analyzing single-row operation features, multi-row operation features and aggregate operation features by using principal components.

5. The banking outlet workload prediction method according to claim 4, characterized in that said single column of operational features comprises: workload rate of change characteristics, holiday characteristics, date characteristics, repayment date characteristics, payday characteristics.

6. The banking outlet workload prediction method according to claim 4, wherein the multi-column operational features comprise: time-based work average and median.

7. The banking outlet workload prediction method according to claim 6, wherein the aggregate operational characteristics comprise: attribute-based workload mean and median.

8. A bank outlet workload prediction apparatus, comprising:

9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor when executing the program performs the steps of the bank outlet workload prediction method of any one of claims 1 to 7.

10. A computer-readable storage medium, having stored thereon a computer program, wherein the computer program, when executed by a processor, performs the steps of the bank branch workload prediction method of any one of claims 1 to 7.