Client payment transaction data abnormity identification system and method based on artificial intelligence

Abstract

The invention relates to the technical field of data exception recognition, and discloses a customer payment transaction data exception recognition system and method based on artificial intelligence. The method comprises the following steps: firstly, performing data filling on an initial customer payment transaction data set based on a clustering distance to obtain a processed customer payment transaction data set, and then performing data expansion on a few types of customer payment transaction data sets in the processed customer payment transaction data set to obtain a processed customer payment transaction data set; secondly, constructing an initial customer payment transaction hybrid prediction model, and optimizing hyper-parameters of the model by using an improved artificial rabbit optimization algorithm; and finally, in combination with the processed customer payment transaction data set, calculating customer payment transaction abnormal characteristics, and identifying customer payment transaction abnormal data. According to the method, the client payment transaction data is analyzed and processed, so that the purpose of client payment transaction data exception identification is achieved, and the method is objective and accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of data anomaly identification, and specifically to a customer payment transaction data anomaly identification system and method based on artificial intelligence. Background Art

[0002] Chinese patent CN111949702B discloses a method, device and equipment for identifying abnormal transaction data. The method specifically includes: using the registration information of the entity object and the hotspot information connected by the user in the transaction data to obtain the characteristic parameters of the entity object in the transaction data, and then using the data model trained with the characteristic label to analyze the user data in the transaction data to obtain the characteristic parameters captured by the user in the entity object; constructing an abnormal recognition model, inputting the characteristic parameters into multi-dimensional features into the abnormal recognition model, and outputting the transaction data recognition result; judging the association between the transaction data recognition result and the entity object, when there is an association between the transaction data recognition result and the entity object, the corresponding transaction data is regarded as abnormal transaction data, otherwise it is not abnormal transaction data. This invention does not pre-process the transaction data, which may lead to problems such as slow recognition speed.

[0003] Traditional methods of identifying abnormal customer payment transaction data take a long time to identify abnormal customer payment transaction data due to the many types of customer payment transaction data and data redundancy, which makes it easy to cause security risks. At the same time, the lack of the use of technologies such as artificial intelligence makes it difficult to monitor customer payment transaction data in real time, making online payment fraud transactions easy to occur. Summary of the invention

[0004] In response to the problems in the related art, the present invention provides an artificial intelligence-based customer payment transaction data anomaly identification system and method to overcome the above-mentioned technical problems existing in the existing related art.

[0005] In order to solve the above technical problems, the present invention is achieved through the following technical solutions: The present invention is a method for identifying abnormalities in customer payment transaction data based on artificial intelligence, comprising the following steps: S1. Acquire the type of customer payment transaction data to obtain an initial customer payment transaction data set, and perform data padding on the initial customer payment transaction data set based on clustering distance to obtain a processed customer payment transaction data set; S2, identifying a few-type customer payment transaction data set and a many-type customer payment transaction data set in the processed customer payment transaction data set, and performing data expansion on the few-type customer payment transaction data set to obtain a processed customer payment transaction data set; S3, integrating the convolutional neural network and the long short-term memory neural network to construct an initial customer payment transaction hybrid prediction model, and introducing an improved artificial rabbit optimization algorithm to optimize the hyperparameters of the customer payment transaction hybrid prediction model to obtain optimized hyperparameters; S4. The initial customer payment transaction hybrid prediction model uses the optimized hyperparameters to obtain the final customer payment transaction hybrid prediction model, and then combines the processed customer payment transaction data set to output the customer payment transaction amount prediction value, and then calculates the abnormal characteristics of the customer payment transaction to identify abnormal customer payment transaction data.

[0006] The invention finds the cluster center of the complete data set, calculates the cluster distance, fills the missing data set according to the cluster distance, and obtains the processed customer payment transaction data set; this method is superior to the traditional data filling method in filling accuracy and filling speed, and uses cluster distance to replace similarity, which effectively improves the data quality; secondly, the data of a few types of customer payment transaction data sets in the processed customer payment transaction data set is expanded; this method only interpolates a few categories of data sets to expand the data set, solves the category imbalance problem in the data set, and facilitates subsequent processing; then the convolutional neural network and the long short-term memory neural network are integrated to construct the initial customer payment transaction hybrid prediction model, and the improved artificial rabbit optimization is used The algorithm optimizes the hyperparameters of the hybrid prediction model for customer payment transactions; the hybrid prediction model uses convolutional neural networks to extract data features, reduces the model's computational complexity and model complexity, and uses long short-term memory neural networks to deal with the problem of gradient expansion disappearance, greatly improving the model's prediction accuracy; the improved artificial rabbit optimization algorithm introduces Levy flight and Monte Carlo methods, which increases the algorithm's search ability and local development ability compared to traditional algorithms, avoids the problem of local optimal solutions, and improves overall performance; finally, the hybrid prediction model for customer payment transactions outputs predicted values, calculates abnormal characteristics of customer payment transactions, and identifies abnormal data of customer payment transactions; this method determines whether there is an anomaly based on time intervals, payment transaction amounts, and payment transaction frequencies, and the identification method is simple and effective.

[0007] Preferably, the S1 comprises the following steps: S11. Obtain the data type of customer payment transaction, which includes transaction type, transaction time, payment channel, transaction amount, etc. Set the number of customer payment transaction data types to m , collect customer payment transaction data under each customer payment transaction data type, and quantify the customer payment transaction data to generate an initial customer payment transaction data set ,in Indicates m types of customer payment transaction data; S12, performing data filling on the initial customer payment transaction data set based on the cluster distance to obtain a processed customer payment transaction data set, the specific steps are as follows: S121, judging whether the initial customer payment transaction data set has missing values, when the customer payment transaction data type has missing values, adding the customer payment transaction data in the customer payment transaction data type to the customer payment transaction missing data set, otherwise adding the customer payment transaction data in the customer payment transaction data type to the customer payment transaction complete data set; selecting an initial cluster center point in the customer payment transaction complete data set, traversing the customer payment transaction complete data set, calculating the customer payment transaction data closest to the initial cluster center point in the customer payment transaction complete data set, merging them into the cluster cluster of the initial cluster center point, and then calculating the mean of the square sum of distances of the customer payment transaction data in the cluster cluster of the initial cluster center point, and using it as the second cluster center point; continuously updating the second cluster center point until the cluster center point in the cluster cluster no longer changes, and obtaining the final cluster center point; S122. Using the final cluster center point as the filling value, the missing values ​​in the customer payment transaction missing data set are filled for the first time to obtain a preliminary filled customer payment transaction data set; calculating the distance between the customer payment transaction data and the filling value in the preliminary filled customer payment transaction data set, recorded as the cluster distance, setting a distance threshold, when the distance between the filling value and the minimum cluster distance value is less than the distance threshold, the preliminary filled customer payment transaction data set is not filled for the second time, otherwise the filling value is replaced by the minimum cluster distance value, and the preliminary filled customer payment transaction data set is filled for the second time; until all customer payment transaction data types are filled, a processed customer payment transaction data set is obtained.

[0008] This invention effectively improves data quality by finding the cluster center of the complete data set, calculating the cluster distance, and using the cluster distance to replace the similarity. It fills in the missing data set based on the cluster distance, and is superior to traditional data filling methods in terms of filling accuracy and speed.

[0009] Preferably, S2 comprises the following steps: S21, setting a data number threshold, when the number of customer payment transaction data of the customer payment transaction data type in the processed customer payment transaction data set is less than the data number threshold, the corresponding customer payment transaction data type is recorded as a small-type customer payment transaction data set, otherwise the corresponding customer payment transaction data type is recorded as a large-type customer payment transaction data set; in the small-type customer payment transaction data set, any customer payment transaction data is selected as pending customer payment transaction data, and b nearest neighbor data of the pending customer payment transaction data in the processed customer payment transaction data set are calculated to form a first nearest neighbor data set; When the nearest data in the first nearest neighbor data set is greater than the pending customer payment transaction data, the pending customer payment transaction data is regarded as noise data, and other customer payment transaction data is reselected; when the nearest data in the first nearest neighbor data set is less than or equal to the pending customer payment transaction data and greater than or equal to half of the pending customer payment transaction data, the pending customer payment transaction data is added to the pending customer payment transaction data set; when the nearest data in the first nearest neighbor data set is less than half of the pending customer payment transaction data and greater than or equal to 0, other customer payment transaction data is reselected to obtain a pending customer payment transaction data set; S22: Calculate the nearest neighbor data of the pending customer payment transaction data in the pending customer payment transaction data set in the few-type customer payment transaction data set to obtain a second nearest neighbor data set. ,in Indicates c The nearest neighbor data; Select the pending customer payment transaction data from the pending customer payment transaction data set; , calculate pending customer payment transaction data The distance to the second nearest neighbor dataset is denoted as ,in Indicates pending customer payment transaction data To c The distance between the nearest neighbor data and the payment transaction data of the pending customer For data expansion, the calculation formula is as follows: ; in, Indicates pending customer payment transaction data The extended data of Indicates pending customer payment transaction data To The distance between the nearest neighbor data, ; All pending customer payment transaction data in the pending customer payment transaction data set are expanded in sequence to obtain an expanded customer payment transaction data set, and the processed customer payment transaction data set is generated by combining the multi-type customer payment transaction data sets.

[0010] The invention expands the data set of a few types of customer payment transaction data in the processed customer payment transaction data set, interpolates only in a few categories of data sets, expands the data set and reduces calculation time, solves the problem of category imbalance in the data set, and facilitates subsequent processing.

[0011] Preferably, S3 comprises the following steps: S31. Fusing the convolutional neural network and the long short-term memory neural network, a hybrid prediction model for initial customer payment transactions is constructed. The specific steps are as follows: S311, assigning a time series to the processed customer payment transaction data set according to the transaction time to obtain a customer payment transaction data time series; setting a convolutional neural network to include an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer, adding an attention mechanism after the input layer, and adding a long short-term memory neural network between the pooling layer and the fully connected layer; S312, after the data sample is input from the input layer of the initial customer payment transaction hybrid prediction model, the data is screened through the attention mechanism, input to the convolution layer and the pooling layer for data feature extraction, and then passed through the long short-term memory neural network, combined with the customer payment transaction data time series to obtain the spatiotemporal correlation, output the predicted value of the data sample, and construct the initial customer payment transaction hybrid prediction model; S32, obtaining a data set of customer payment transaction data in previous years, performing data filling and data expansion on the data set of customer payment transaction data in previous years to obtain a sample set of customer payment transaction data, and assigning a time series, inputting the sample set of customer payment transaction data into an initial customer payment transaction hybrid prediction model according to the time series for training, taking the output prediction value accuracy as the fitness function, and using an improved artificial rabbit optimization algorithm to optimize the hyperparameters in the initial customer payment transaction hybrid prediction model to obtain the optimized hyperparameters, the specific steps are as follows: S321. During the training process of the hybrid prediction model for initial customer payment transactions, it is assumed that there is an artificial rabbit population, and the number of artificial rabbit populations is i , the dimension of the artificial rabbit individuals in the artificial rabbit population is j , the position of the artificial rabbit individuals in the artificial rabbit population represents the hyperparameter, the artificial rabbit population is initialized to determine the initial position of the artificial rabbit individuals and the initial fitness function value; the artificial rabbit population is in the process of foraging, and the current number of iterations is set to d , No. d The first iteration eThe position of each artificial rabbit is recorded as , No. d The first iteration The position of each artificial rabbit is recorded as , represents a random number between [0, 1], and the standard normal distribution coefficient is ; Set the running factor, use Levi flight to replace the running factor, the Levi flight coefficient is , and represents a random number between (0, 1), then the Levy flight function , then d +1 iteration e Artificial rabbit individual position The calculation formula is as follows: ; in, represents the rounding function; S322, in the process of random hiding of the artificial rabbit population, the artificial rabbit individuals choose caves to hide, and set the hiding parameters as , represents an integer of 0 or 1, then e The first artificial rabbit h Cave Locations , and then use the location of the cave to calculate the location of the artificial rabbit To update, ; Introduce the Monte Carlo method to update the position of the artificial rabbit again , calculate the average position of the artificial rabbit individuals in the artificial rabbit population at this time , the control parameters are , represents a random number between [0, 1], Represents a random number between (0, 1); when hour, ,when hour, ; S323, calculate the d The first iteration e The fitness function value corresponding to the position of the artificial rabbit and d +1 iteration e The fitness function value corresponding to the position of the artificial rabbit , corresponding to d The iteration and d + Hyperparameters in the hybrid prediction model for initial customer payment transactions at 1 iteration; Comparison of fitness function values And the fitness function value , select a smaller fitness function value as the current optimal fitness function value, and the individual position of the artificial rabbit corresponding to the current optimal fitness function value is the current optimal hyperparameter; set the maximum number of iterations, and when the current number of iterations reaches the maximum number of iterations, stop the iteration and get the global optimal position, which corresponds to the optimized hyperparameter.

[0012] This invention constructs an initial customer payment transaction hybrid prediction model by integrating convolutional neural networks and long short-term memory neural networks, and uses an improved artificial rabbit optimization algorithm to optimize the hyperparameters of the customer payment transaction hybrid prediction model; the hybrid prediction model uses convolutional neural networks to extract data features, reducing the model's calculation amount and model complexity, and the long short-term memory neural network handles the gradient expansion vanishing problem, greatly improving the model's prediction accuracy; the improved artificial rabbit optimization algorithm introduces Levy flight and Monte Carlo methods, which increases the algorithm's search ability and local development ability compared to traditional algorithms, avoids the problem of local optimal solutions, and improves overall performance.

[0013] Preferably, S4 comprises the following steps: S41, the initial customer payment transaction hybrid prediction model uses the optimized hyperparameters until the initial customer payment transaction hybrid prediction model converges to obtain the final customer payment transaction hybrid prediction model; then the processed customer payment transaction data set is used as a time series according to the transaction time to obtain a final customer payment transaction data set, the time series step is set, the final customer payment transaction data set is segmented using the time series step, and then input into the final customer payment transaction hybrid prediction model, and output the customer payment transaction amount prediction value; S42: Output the predicted value of the customer payment transaction amount at each time series step, obtain a set of predicted values ​​of the customer payment transaction amount, calculate the time interval of the predicted values ​​of the customer payment transaction amount, record it as the customer payment transaction time interval, and calculate the time period The average of the predicted transaction amounts of customers within the range is recorded as the average customer payment transaction amount. The calculation time period The average value of the number of predicted values ​​of customer payment transaction amounts within the range is recorded as the customer payment transaction frequency, and the customer payment transaction time interval, the average customer payment transaction amount and the customer payment transaction frequency are combined as the abnormal characteristics of the customer payment transaction; Set the first threshold respectively , the second threshold and the third threshold When the customer payment transaction time interval is less than the first threshold, or the average customer payment transaction amount is greater than the second threshold, or the customer payment transaction frequency is greater than the third threshold, the customer payment transaction data is abnormal. The customer feedbacks the abnormal customer payment transaction data for further processing and investigation to complete the identification of abnormal customer payment transaction data.

[0014] This embodiment also discloses a system for an artificial intelligence-based customer payment transaction data anomaly identification method, which specifically includes: a payment transaction data filling module, a payment transaction data expansion module, a hyperparameter optimization module, and a data anomaly identification module; The payment transaction data filling module is used to fill in the initial customer payment transaction data set based on the cluster distance; The payment transaction data expansion module is used to expand the payment transaction data set of a few types of customers; The hyperparameter optimization module is used to optimize the hyperparameters of the customer payment transaction hybrid prediction model using an improved artificial rabbit optimization algorithm; The data anomaly identification module is used to identify abnormal customer payment transaction data after calculating abnormal characteristics of the customer payment transaction.

[0015] The present invention has the following beneficial effects: 1. The invention finds the cluster center of the complete data set, calculates the cluster distance, and uses the cluster distance to replace the similarity, thereby effectively improving the data quality and filling the missing data set according to the cluster distance. It is superior to the traditional data filling method in filling accuracy and filling speed.

[0016] 2. The invention expands the data set of a few types of customer payment transaction data in the processed customer payment transaction data set, and interpolates only a few categories of data sets, thereby expanding the data set and reducing calculation time, thereby solving the problem of category imbalance in the data set and facilitating subsequent processing.

[0017] 3. This invention constructs a hybrid prediction model for initial customer payment transactions by integrating convolutional neural networks and long short-term memory neural networks. The hybrid prediction model uses convolutional neural networks to extract data features, reducing the amount of model calculation and model complexity. The long short-term memory neural network handles the gradient expansion vanishing problem, greatly improving the model prediction accuracy.

[0018] 4. The invention optimizes the hyperparameters of the customer payment transaction hybrid prediction model by using an improved artificial rabbit optimization algorithm. The improved artificial rabbit optimization algorithm increases the search ability and local development ability of the algorithm compared to the traditional algorithm by introducing Levy flight and Monte Carlo method, thereby avoiding the trouble of local optimal solution and improving the overall performance.

[0019] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the invention, the drawings required for describing the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For ordinary technicians in this field, they can also obtain drawings based on these drawings without paying any creative work.

[0021] Figure 1 The present invention provides a flow chart of anomaly identification of customer payment transaction data by an artificial intelligence-based customer payment transaction data anomaly identification system. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] Example 1: Please refer to Figure 1 This implementation discloses an artificial intelligence-based method for identifying abnormalities in customer payment transaction data, which specifically includes the following contents: S1. Acquire the type of customer payment transaction data to obtain an initial customer payment transaction data set, and perform data padding on the initial customer payment transaction data set based on clustering distance to obtain a processed customer payment transaction data set; The S1 comprises the following steps: S11. Obtain the data type of customer payment transaction, which includes transaction type, transaction time, payment channel, transaction amount, etc. Set the number of customer payment transaction data types to m , collect customer payment transaction data under each customer payment transaction data type, and quantify the customer payment transaction data to generate an initial customer payment transaction data set ,in Indicates m types of customer payment transaction data; S12, performing data filling on the initial customer payment transaction data set based on the cluster distance to obtain a processed customer payment transaction data set, the specific steps are as follows: S121, judging whether the initial customer payment transaction data set has missing values, when the customer payment transaction data type has missing values, adding the customer payment transaction data in the customer payment transaction data type to the customer payment transaction missing data set, otherwise adding the customer payment transaction data in the customer payment transaction data type to the customer payment transaction complete data set; selecting an initial cluster center point in the customer payment transaction complete data set, traversing the customer payment transaction complete data set, calculating the customer payment transaction data closest to the initial cluster center point in the customer payment transaction complete data set, merging them into the cluster cluster of the initial cluster center point, and then calculating the mean of the square sum of distances of the customer payment transaction data in the cluster cluster of the initial cluster center point, and using it as the second cluster center point; continuously updating the second cluster center point until the cluster center point in the cluster cluster no longer changes, and obtaining the final cluster center point; S122, using the final cluster center point as the filling value, performing the first filling of the missing values ​​in the customer payment transaction missing data set to obtain a preliminary filled customer payment transaction data set; calculating the distance between the customer payment transaction data and the filling value in the preliminary filled customer payment transaction data set, recording it as the cluster distance, setting a distance threshold, when the distance between the filling value and the minimum cluster distance value is less than the distance threshold, not performing the second filling on the preliminary filled customer payment transaction data set, otherwise using the minimum cluster distance value to replace the filling value, and performing the second filling on the preliminary filled customer payment transaction data set; until all customer payment transaction data types are filled, and a processed customer payment transaction data set is obtained; S2, identifying a few-type customer payment transaction data set and a many-type customer payment transaction data set in the processed customer payment transaction data set, and performing data expansion on the few-type customer payment transaction data set to obtain a processed customer payment transaction data set; The S2 comprises the following steps: S21, setting a data number threshold, when the number of customer payment transaction data of the customer payment transaction data type in the processed customer payment transaction data set is less than the data number threshold, the corresponding customer payment transaction data type is recorded as a small-type customer payment transaction data set, otherwise the corresponding customer payment transaction data type is recorded as a large-type customer payment transaction data set; in the small-type customer payment transaction data set, any customer payment transaction data is selected as pending customer payment transaction data, and b nearest neighbor data of the pending customer payment transaction data in the processed customer payment transaction data set are calculated to form a first nearest neighbor data set; When the nearest data in the first nearest neighbor data set is greater than the pending customer payment transaction data, the pending customer payment transaction data is regarded as noise data, and other customer payment transaction data is reselected; when the nearest data in the first nearest neighbor data set is less than or equal to the pending customer payment transaction data and greater than or equal to half of the pending customer payment transaction data, the pending customer payment transaction data is added to the pending customer payment transaction data set; when the nearest data in the first nearest neighbor data set is less than half of the pending customer payment transaction data and greater than or equal to 0, other customer payment transaction data is reselected to obtain a pending customer payment transaction data set; S22: Calculate the nearest neighbor data of the pending customer payment transaction data in the pending customer payment transaction data set in the few-type customer payment transaction data set to obtain a second nearest neighbor data set. ,in Indicates c The nearest neighbor data; Select the pending customer payment transaction data from the pending customer payment transaction data set; , calculate pending customer payment transaction data The distance to the second nearest neighbor dataset is denoted as ,in Indicates pending customer payment transaction data To c The distance between the nearest neighbor data and the payment transaction data of the pending customer For data expansion, the calculation formula is as follows: ; in, Indicates pending customer payment transaction data The extended data of Indicates pending customer payment transaction data To The distance between the nearest neighbor data, ; sequentially expanding all pending customer payment transaction data in the pending customer payment transaction data set to obtain an expanded customer payment transaction data set, and combining the multi-type customer payment transaction data sets to generate a processed customer payment transaction data set; S3, integrating the convolutional neural network and the long short-term memory neural network to construct an initial customer payment transaction hybrid prediction model, and introducing an improved artificial rabbit optimization algorithm to optimize the hyperparameters of the customer payment transaction hybrid prediction model to obtain optimized hyperparameters; The S3 comprises the following steps: S31. Fusing the convolutional neural network and the long short-term memory neural network, a hybrid prediction model for initial customer payment transactions is constructed. The specific steps are as follows: S311, assigning a time series to the processed customer payment transaction data set according to the transaction time to obtain a customer payment transaction data time series; setting a convolutional neural network to include an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer, adding an attention mechanism after the input layer, and adding a long short-term memory neural network between the pooling layer and the fully connected layer; S312, after the data sample is input from the input layer of the initial customer payment transaction hybrid prediction model, the data is screened through the attention mechanism, input to the convolution layer and the pooling layer for data feature extraction, and then passed through the long short-term memory neural network, combined with the customer payment transaction data time series to obtain the spatiotemporal correlation, output the predicted value of the data sample, and construct the initial customer payment transaction hybrid prediction model; S32, obtaining a data set of customer payment transaction data in previous years, performing data filling and data expansion on the data set of customer payment transaction data in previous years to obtain a sample set of customer payment transaction data, and assigning a time series, inputting the sample set of customer payment transaction data into an initial customer payment transaction hybrid prediction model according to the time series for training, taking the output prediction value accuracy as the fitness function, and using an improved artificial rabbit optimization algorithm to optimize the hyperparameters in the initial customer payment transaction hybrid prediction model to obtain the optimized hyperparameters, the specific steps are as follows: S321. During the training process of the hybrid prediction model for initial customer payment transactions, it is assumed that there is an artificial rabbit population, and the number of artificial rabbit populations is i , the dimension of the artificial rabbit individuals in the artificial rabbit population is j , the position of the artificial rabbit individuals in the artificial rabbit population represents the hyperparameter, the artificial rabbit population is initialized to determine the initial position of the artificial rabbit individuals and the initial fitness function value; the artificial rabbit population is in the process of foraging, and the current number of iterations is set to d , No. d The first iteration e The position of each artificial rabbit is recorded as , No. d The first iteration The position of each artificial rabbit is recorded as , represents a random number between [0, 1], and the standard normal distribution coefficient is ; Set the running factor, use Levi flight to replace the running factor, the Levi flight coefficient is , and represents a random number between (0, 1), then the Levy flight function , then d +1 iteration e Artificial rabbit individual position The calculation formula is as follows: ; in, represents the rounding function; S322, in the process of random hiding of the artificial rabbit population, the artificial rabbit individuals choose caves to hide, and set the hiding parameters as , represents an integer of 0 or 1, then e The first artificial rabbit h Cave Locations , and then use the location of the cave to calculate the location of the artificial rabbit To update, ; Introduce the Monte Carlo method to update the position of the artificial rabbit again , calculate the average position of the artificial rabbit individuals in the artificial rabbit population at this time , the control parameters are , represents a random number between [0, 1], Represents a random number between (0, 1); when hour, ,when hour, ; S323, calculate the d The first iteration e The fitness function value corresponding to the position of the artificial rabbit and d +1 iteration e The fitness function value corresponding to the position of the artificial rabbit , corresponding to d The iteration and d + Hyperparameters in the hybrid prediction model for initial customer payment transactions at 1 iteration; Comparison of fitness function values And the fitness function value , select a smaller fitness function value as the current optimal fitness function value, and the artificial rabbit individual position corresponding to the current optimal fitness function value is the current optimal hyperparameter; set the maximum number of iterations, and when the current number of iterations reaches the maximum number of iterations, stop the iteration and get the global optimal position, which corresponds to the optimized hyperparameter; S4, the initial customer payment transaction hybrid prediction model uses the optimized hyperparameters to obtain the final customer payment transaction hybrid prediction model, and then combines the processed customer payment transaction data set to output the customer payment transaction amount prediction value, and then calculates the customer payment transaction abnormal features to identify the customer payment transaction abnormal data; The S4 comprises the following steps: S41, the initial customer payment transaction hybrid prediction model uses the optimized hyperparameters until the initial customer payment transaction hybrid prediction model converges to obtain the final customer payment transaction hybrid prediction model; then the processed customer payment transaction data set is used as a time series according to the transaction time to obtain a final customer payment transaction data set, the time series step is set, the final customer payment transaction data set is segmented using the time series step, and then input into the final customer payment transaction hybrid prediction model, and output the customer payment transaction amount prediction value; S42: Output the predicted value of the customer payment transaction amount at each time series step, obtain a set of predicted values ​​of the customer payment transaction amount, calculate the time interval of the predicted values ​​of the customer payment transaction amount, record it as the customer payment transaction time interval, and calculate the time period The average of the predicted transaction amounts of customers within the range is recorded as the average customer payment transaction amount. The calculation time period The average value of the number of predicted values ​​of customer payment transaction amounts within the range is recorded as the customer payment transaction frequency, and the customer payment transaction time interval, the average customer payment transaction amount and the customer payment transaction frequency are combined as the abnormal characteristics of the customer payment transaction; Set the first threshold respectively , the second threshold and the third threshold When the customer payment transaction time interval is less than the first threshold, or the average customer payment transaction amount is greater than the second threshold, or the customer payment transaction frequency is greater than the third threshold, the customer payment transaction data is abnormal. The customer feedbacks the abnormal customer payment transaction data for further processing and investigation to complete the identification of abnormal customer payment transaction data.

[0024] Embodiment 2: This embodiment also discloses a system for an abnormal customer payment transaction data identification method based on artificial intelligence, which specifically includes: a payment transaction data filling module, a payment transaction data expansion module, a hyperparameter optimization module and a data abnormality identification module; The payment transaction data filling module is used to fill in the initial customer payment transaction data set based on the cluster distance; The payment transaction data expansion module is used to expand the payment transaction data set of a few types of customers; The hyperparameter optimization module is used to optimize the hyperparameters of the customer payment transaction hybrid prediction model using an improved artificial rabbit optimization algorithm; The data anomaly identification module is used to identify abnormal customer payment transaction data after calculating abnormal characteristics of the customer payment transaction.

[0025] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0026] The preferred embodiments of the invention disclosed above are only used to help explain the invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the invention, so that those skilled in the art can understand and use the invention well.

Claims

1. A method for identifying abnormalities in customer payment transaction data based on artificial intelligence, characterized in that: The steps include: S1. Acquire the type of customer payment transaction data to obtain an initial customer payment transaction data set, and perform data padding on the initial customer payment transaction data set based on clustering distance to obtain a processed customer payment transaction data set; S2, identifying a few-type customer payment transaction data set and a many-type customer payment transaction data set in the processed customer payment transaction data set, and performing data expansion on the few-type customer payment transaction data set to obtain a processed customer payment transaction data set; S3, integrating the convolutional neural network and the long short-term memory neural network to construct an initial customer payment transaction hybrid prediction model, and optimizing the hyperparameters of the customer payment transaction hybrid prediction model to obtain optimized hyperparameters; S4. The initial customer payment transaction hybrid prediction model uses the optimized hyperparameters to obtain the final customer payment transaction hybrid prediction model, and then combines the processed customer payment transaction data set to output the customer payment transaction amount prediction value, and then calculates the abnormal characteristics of the customer payment transaction to identify abnormal customer payment transaction data.

2. The method for identifying abnormal customer payment transaction data based on artificial intelligence according to claim 1 is characterized in that: The S1 comprises the following steps: S11, obtaining the customer payment transaction data type and generating an initial customer payment transaction data set; S12. Performing data filling on the initial customer payment transaction data set based on clustering distance to obtain a processed customer payment transaction data set.

3. The method for identifying abnormal customer payment transaction data based on artificial intelligence according to claim 2 is characterized in that: The S12 comprises the following steps: S121. Divide the initial customer payment transaction data set into a customer payment transaction missing data set and a customer payment transaction complete data set according to whether there are missing values; Performing clustering processing on the complete data set of the customer payment transaction to obtain a final cluster center point; S122. Using the final cluster center point as a filling value, the missing values ​​in the customer payment transaction missing data set are filled for the first time, and then the cluster distance is calculated. By comparing the cluster distance and the filling value, a second filling is performed to obtain a processed customer payment transaction data set.

4. The method for identifying abnormal customer payment transaction data based on artificial intelligence according to claim 3 is characterized in that: The S2 comprises the following steps: S21, dividing the processed customer payment transaction data set into a small number of customer payment transaction data set and a large number of customer payment transaction data set; selecting pending customer payment transaction data from the small number of customer payment transaction data set to obtain a pending customer payment transaction data set; S22, calculating the nearest neighbor data of the pending customer payment transaction data set, performing data expansion on the pending customer payment transaction data set according to the nearest neighbor data to obtain an expanded customer payment transaction data set, and combining the multi-type customer payment transaction data sets to generate a processed customer payment transaction data set.

5. The method for identifying abnormal customer payment transaction data based on artificial intelligence according to claim 4 is characterized in that: The S3 comprises the following steps: S31, integrating convolutional neural network and long short-term memory neural network to construct a hybrid prediction model for initial customer payment transactions; S32. Obtain a data set of customer payment transaction data from previous years, perform data filling and data expansion on the data set to obtain a sample set of customer payment transaction data, assign a time series, input the sample set of customer payment transaction data into an initial customer payment transaction hybrid prediction model according to the time series for training, use the output prediction value accuracy as the fitness function, use the improved artificial rabbit optimization algorithm to optimize the hyperparameters in the initial customer payment transaction hybrid prediction model, and obtain the optimized hyperparameters.

6. The method for identifying abnormal customer payment transaction data based on artificial intelligence according to claim 5 is characterized in that: The S31 comprises the following steps: S311, assigning a time series to the processed customer payment transaction data set according to the transaction time to obtain a customer payment transaction data time series; setting a convolutional neural network to include an input layer, a convolutional layer, a pooling layer, a fully connected layer and an output layer, adding an attention mechanism after the input layer, and adding a long short-term memory neural network between the pooling layer and the fully connected layer; S312: After inputting the data sample from the input layer of the initial customer payment transaction hybrid prediction model, the predicted value of the data sample is output to construct the initial customer payment transaction hybrid prediction model.

7. The method for identifying abnormal customer payment transaction data based on artificial intelligence according to claim 6 is characterized in that: The S32 comprises the following steps: S321. During the training process of the hybrid prediction model for initial customer payment transactions, it is assumed that there is an artificial rabbit population, and the number of artificial rabbit populations is i , the dimension of the artificial rabbit individuals in the artificial rabbit population is j , the position of the artificial rabbit individuals in the artificial rabbit population represents the hyperparameter, and the artificial rabbit population is initialized to determine the initial position of the artificial rabbit individuals and the initial fitness function value; The artificial rabbit population is in the process of foraging. Set the current number of iterations to d , No. d The first iteration e The position of each artificial rabbit is recorded as , No. d The first iteration The position of each artificial rabbit is recorded as , represents a random number between [0, 1], and the standard normal distribution coefficient is ; Set the running factor, use Levi flight to replace the running factor, the Levi flight coefficient is , and represents a random number between (0, 1), then the Levy flight function , then d +1 iteration e Artificial rabbit individual position The calculation formula is as follows: ； in, represents the rounding function; S322, in the process of random hiding of the artificial rabbit population, the artificial rabbit individuals choose caves to hide, and set the hiding parameters as , represents an integer of 0 or 1, then e The first artificial rabbit h Cave Locations , and then use the location of the cave to calculate the location of the artificial rabbit To update, ; Introduce the Monte Carlo method to update the position of the artificial rabbit again , calculate the average position of the artificial rabbit individuals in the artificial rabbit population at this time , the control parameters are , represents a random number between [0, 1], Represents a random number between (0, 1); when hour, ,when hour, ; S323, calculate the d The first iteration e The fitness function value corresponding to the position of the artificial rabbit and d +1 iteration e The fitness function value corresponding to the position of the artificial rabbit , corresponding to d The iteration and d + Hyperparameters in the hybrid prediction model for initial customer payment transactions at 1 iteration; Comparison of fitness function values And the fitness function value , select a smaller fitness function value as the current optimal fitness function value, and the individual position of the artificial rabbit corresponding to the current optimal fitness function value is the current optimal hyperparameter; set the maximum number of iterations, and when the current number of iterations reaches the maximum number of iterations, stop the iteration and get the global optimal position, which corresponds to the optimized hyperparameter.

8. The method for identifying abnormal customer payment transaction data based on artificial intelligence according to claim 7 is characterized in that: The S4 comprises the following steps: S41, the initial customer payment transaction hybrid prediction model uses the optimized hyperparameters until the initial customer payment transaction hybrid prediction model converges to obtain the final customer payment transaction hybrid prediction model; then the processed customer payment transaction data set is used as a time series according to the transaction time to obtain a final customer payment transaction data set, the time series step is set, the final customer payment transaction data set is segmented using the time series step, and then input into the final customer payment transaction hybrid prediction model, and output the customer payment transaction amount prediction value; S42. Calculate the customer payment transaction time interval, average customer payment transaction amount and customer payment transaction frequency according to the customer payment transaction amount prediction value, and obtain abnormal characteristics of the customer payment transaction; Set a threshold, and by comparing the abnormal characteristics of customer payment transactions with the threshold, determine whether there are any abnormalities in the customer payment transaction data, and complete the abnormal identification of customer payment transaction data.

9. A system for implementing the method for identifying abnormal customer payment transaction data based on artificial intelligence as claimed in any one of claims 1 to 8, characterized in that: Specifically include: Payment transaction data filling module, payment transaction data expansion module, hyperparameter optimization module and data anomaly identification module; The payment transaction data filling module is used to fill in the initial customer payment transaction data set based on the cluster distance; The payment transaction data expansion module is used to expand the payment transaction data set of a few types of customers; The hyperparameter optimization module is used to optimize the hyperparameters of the customer payment transaction hybrid prediction model using an improved artificial rabbit optimization algorithm; The data anomaly identification module is used to identify abnormal customer payment transaction data after calculating abnormal characteristics of the customer payment transaction.

Images