Model input data stability monitoring method and device based on auto-encoder

Abstract

The invention provides a model input data stability monitoring method and device based on an auto-encoder, and the method comprises the steps: extracting reference input data from historical input data, and extracting verification input data from current input data; training a feature extraction model by using an auto-encoder, and processing the reference input data and the verification input data by using the feature extraction model to obtain a reference feature matrix and a verification feature matrix; calculating stability indexes based on the reference feature matrix and the verification feature matrix, wherein the stability indexes comprise year-on-year difference and distribution link difference; and when the stability index is greater than a preset threshold value, generating alarm information. The method solves the problem that in the prior art, only distribution of model output data is concerned in monitoring of the model stability, influences of model input changes on the model stability are ignored, and consequently monitoring of the model performance is not accurate.

Description

Technical Field

[0001] This invention relates to the field of model monitoring technology, and more specifically, to a method, apparatus, computer-readable storage medium, and electronic device for monitoring the stability of model input data based on an autoencoder. Background Technology

[0002] In existing technologies, when validating a model, the distribution of the model's output results is usually monitored. However, the models used in existing technologies are usually multimodal data inputs. If the distributions of multiple input indicators change simultaneously, the model results may have small differences in distribution, which may lead to deviations in the model's validation results.

[0003] In summary, existing model performance monitoring methods have omissions, which makes it impossible to guarantee the accuracy of model monitoring results. Summary of the Invention

[0004] The main objective of this application is to provide a method, apparatus, computer-readable storage medium, and electronic device for monitoring the stability of model input data based on an autoencoder, so as to at least solve the problem in the prior art that the monitoring of model stability only focuses on the distribution of model output data and ignores the impact of model input changes on model stability, resulting in inaccurate monitoring of model performance.

[0005] To achieve the above objectives, according to one aspect of this application, a method for monitoring the stability of model input data based on an autoencoder is provided, comprising: extracting baseline input data from historical input data and extracting validation input data from current input data; training a feature extraction model using an autoencoder, processing the baseline input data and the validation input data using the feature extraction model to obtain a baseline feature matrix and a validation feature matrix; calculating a stability index based on the baseline feature matrix and the validation feature matrix, the stability index including year-on-year difference and month-on-month distribution difference; and generating an alarm message when the stability index exceeds a preset threshold.

[0006] Optionally, extracting baseline input data from historical input data and extracting verification input data from current input data includes: extracting input data for a preset period from the historical input data to obtain first input data; extracting multiple data segments from the first input data with a preset step size to obtain multiple baseline input data; and sequentially extracting corresponding input data from the current input data according to each baseline input data to obtain the verification input data.

[0007] Optionally, an autoencoder is used to train a feature extraction model, and the feature extraction model is used to process the benchmark input data and the verification input data to obtain a benchmark feature matrix and a verification feature matrix. This includes: training multiple feature extraction models using the autoencoder, each feature extraction model corresponding one-to-one with the benchmark input data; extracting features from the corresponding benchmark input data using each feature extraction model to obtain multiple benchmark feature matrices; and extracting features from the corresponding verification input data using each feature extraction model to obtain multiple verification feature matrices.

[0008] Optionally, calculating the stability index based on the benchmark feature matrix and the validation feature matrix includes: extracting year-on-year input data from the historical input data based on the current input data; processing the year-on-year input data using the feature extraction model to obtain a year-on-year feature matrix; calculating the year-on-year difference quantification value based on the validation feature matrix and the benchmark feature matrix; calculating the distribution month-on-month relative difference quantification variance based on the validation feature matrix, the year-on-year feature matrix, and the benchmark feature matrix; and summing the year-on-year difference quantification value and the distribution month-on-month relative difference quantification variance to obtain the stability index.

[0009] Optionally, extracting year-on-year input data from the historical input data based on the current input data includes: extracting input data of the same period from the historical input data based on the period of the current input data to obtain second input data; and sequentially extracting corresponding input data from the second input data according to each current input data to obtain the year-on-year input data.

[0010] Optionally, calculating the year-on-year difference quantification value based on the verification feature matrix and the benchmark feature matrix includes: calculating the difference between the verification feature matrix and the benchmark feature matrix using the maximum average difference to obtain the year-on-year difference quantification value.

[0011] Optionally, calculating the distribution's relative month-on-month difference quantification variance based on the verification feature matrix, the year-on-year feature matrix, and the benchmark feature matrix includes: calculating the year-on-year difference quantification value between each verification feature matrix and the corresponding benchmark feature matrix to obtain multiple first difference values; calculating the year-on-year difference quantification value between each year-on-year feature matrix and the corresponding benchmark feature matrix to obtain multiple second difference values; calculating the absolute difference between each first difference value and the corresponding second difference value to obtain multiple relative difference quantification values; taking the variance of all the relative difference quantification values ​​and calculating the mean variance to obtain the distribution's relative month-on-month difference quantification variance.

[0012] According to another aspect of this application, a model input data stability monitoring device based on an autoencoder is provided, comprising: an acquisition unit, configured to extract baseline input data from historical input data and extract validation input data from current input data; a training unit, configured to train a feature extraction model using an autoencoder, and process the baseline input data and the validation input data using the feature extraction model to obtain a baseline feature matrix and a validation feature matrix; a calculation unit, configured to calculate a stability index based on the baseline feature matrix and the validation feature matrix, the stability index including year-on-year difference and month-on-month distribution difference; and an alarm unit, configured to generate alarm information when the stability index exceeds a preset threshold.

[0013] According to another aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform any of the methods described.

[0014] According to another aspect of this application, an electronic device is provided, comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including methods for performing any one of the methods described.

[0015] Applying the technical solution of this application, in the aforementioned autoencoder-based model input data stability monitoring method, firstly, benchmark input data is extracted from historical input data, and validation input data is extracted from current input data; then, an autoencoder is used to train a feature extraction model, which processes the benchmark and validation input data to obtain benchmark and validation feature matrices; subsequently, a stability index is calculated based on the benchmark and validation feature matrices, including year-on-year differences and month-on-month distribution differences; finally, an alarm message is generated when the stability index exceeds a preset threshold. This application achieves simultaneous monitoring of model input data stability by comparing benchmark and current input data, enabling timely detection of the impact of input data changes on model performance. This addresses the problem in existing technologies where model stability monitoring only focuses on the distribution of model output data, neglecting the impact of model input changes on model stability, leading to inaccurate model performance monitoring. Attached Figure Description

[0016] Figure 1 A hardware structure block diagram of a mobile terminal for a model input data stability monitoring method based on an autoencoder provided in an embodiment of this application is shown.

[0017] Figure 2A flowchart illustrating a method for monitoring the stability of model input data based on an autoencoder, according to an embodiment of this application, is shown.

[0018] Figure 3 A structural block diagram of a model input data stability monitoring device based on an autoencoder, according to an embodiment of this application, is shown.

[0019] The above figures include the following reference numerals:

[0020] 102. Processor; 104. Memory; 106. Transmission device; 108. Input / output device. Detailed Implementation

[0021] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0022] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0023] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0024] As described in the background section, existing model performance monitoring methods have omissions, resulting in inaccurate model monitoring results. To address the problem that existing methods for monitoring model stability only focus on the distribution of model output data and ignore the impact of model input changes on model stability, leading to inaccurate model performance monitoring, embodiments of this application provide a method, apparatus, computer-readable storage medium, and electronic device for monitoring model input data stability based on an autoencoder.

[0025] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0026] The methods and embodiments provided in this application can be executed on a mobile terminal, computer terminal, or similar computing device. Taking running on a mobile terminal as an example, Figure 1 This is a hardware structure block diagram of a mobile terminal for a model input data stability monitoring method based on an autoencoder, according to an embodiment of the present invention. Figure 1 As shown, a mobile terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 (which may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.) and a memory 104 for storing data are also shown. The mobile terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the mobile terminal described above. For example, the mobile terminal may also include components that are more... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0027] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the autoencoder-based model input data stability monitoring method in this embodiment of the invention. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, thereby implementing the above-described method. The memory 104 may include high-speed random access memory and non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the mobile terminal via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof. The transmission device 106 is used to receive or send data via a network. Specific examples of the aforementioned networks may include wireless networks provided by the mobile terminal's communication provider. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can be connected to other network devices via a base station to communicate with the Internet. In one example, the transmission device 106 may be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0028] This embodiment provides a method for monitoring the stability of model input data based on an autoencoder, which runs on a mobile terminal, computer terminal, or similar computing device. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Also, although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than that shown here.

[0029] Figure 2 This is a flowchart of a model input data stability monitoring method based on an autoencoder, according to an embodiment of this application. Figure 2 As shown, the method includes the following steps:

[0030] Step S201: Extract baseline input data from historical input data and extract verification input data from current input data;

[0031] Understandably, historical data (baseline input data) is data selected from a specific period in the past. This data reflects how the model processes the input data under ideal conditions and is used as a baseline. Current data (validation input data) is input data collected in the current period and is used to monitor the current input stability of the model.

[0032] In practice, data from a certain month in the past year can be selected as the baseline data, and the current input data is the data of this month. For example, for monitoring data in April of year x, the baseline input data includes data from April of year x-1 (year-on-year baseline), as well as data from January to March and May to December of year x-1 (month-on-month baseline).

[0033] Step S202: Train the feature extraction model using an autoencoder, and process the benchmark input data and validation input data using the feature extraction model to obtain the benchmark feature matrix and the validation feature matrix.

[0034] As is understandable, an autoencoder is an unsupervised learning neural network primarily used for feature learning and dimensionality reduction. It learns a compact representation of the input data, i.e., a feature vector, through encoding and decoding processes. In this invention, an autoencoder is used to train a feature extraction model to capture key features in the data.

[0035] Following the above embodiments, an autoencoder can be trained based on data from January to December of year x-1 to obtain 12 feature extraction models M_1 to M_12. Then, the corresponding feature extraction models are used to process the benchmark input data and validation input data to obtain multiple benchmark validation matrices and validation feature matrices.

[0036] Step S203: Calculate stability indices based on the baseline feature matrix and the validation feature matrix. The stability indices include year-on-year differences and month-on-month differences in distribution.

[0037] Understandably, the stability index consists of two parts: year-on-year difference and month-on-month distribution difference, which respectively reflect the difference between the current data and the data of the same period in history, and the relative change compared with other historical data.

[0038] Step S204: If the stability index is greater than the preset threshold, generate an alarm message.

[0039] Specifically, a reasonable Q threshold is set. If the calculated Q value exceeds this threshold, it indicates a problem with the stability of the model's input data. In this case, the system generates an alarm message, suggesting that further inspection of the model's input data is needed, which may require data preprocessing or model retraining.

[0040] In this embodiment, firstly, baseline input data is extracted from historical input data, and validation input data is extracted from current input data. Then, an autoencoder is used to train a feature extraction model, which processes the baseline and validation input data to obtain a baseline feature matrix and a validation feature matrix. Next, a stability index is calculated based on the baseline and validation feature matrices, including year-on-year differences and month-on-month differences in distribution. Finally, an alarm is generated if the stability index exceeds a preset threshold. This application achieves simultaneous monitoring of model input data stability by comparing baseline and current input data, enabling timely detection of the impact of input data changes on model performance. This addresses the problem in existing technologies where model stability monitoring only focuses on the distribution of model output data, ignoring the impact of model input changes on model stability, leading to inaccurate model performance monitoring.

[0041] In order to obtain baseline input data and verification input data, in one optional implementation, step S201 above includes:

[0042] Step S2011: Extract input data for a preset period from historical input data to obtain the first input data;

[0043] Understandably, a preset period refers to a specific period extracted from historical data as a standard benchmark, usually a time period with stable business cycles.

[0044] Taking a transaction data monitoring model as an example, the data of the past year is traversed in units of years (preset period). If the input data of a certain month is relatively stable, it is selected as the first input data mentioned above.

[0045] Step S2012: Extract multiple data segments from the first input data with a preset step size to obtain multiple reference input data;

[0046] Understandably, the preset step size is used to extract historical data through a sliding window, ensuring that data from different time periods can be compared fairly and comprehensively.

[0047] Specifically, in the first input data mentioned above, multiple data segments are randomly extracted at a length (the preset step size, such as one week) that can reflect the characteristics of the data, so as to analyze the periodic changes of the data in more detail, improve the accuracy of monitoring, and obtain the benchmark input data mentioned above.

[0048] Step S2013: Extract the corresponding input data from the current input data according to each benchmark input data to obtain the verification input data.

[0049] Specifically, for the input data of the current period, the corresponding fields are extracted using the same compensation as the baseline input data to obtain the verification input data, so as to ensure that the periodic changes of the data are consistent.

[0050] Through the above embodiments, by selecting preset periods and step sizes, the periodic changes in data can be analyzed in greater detail, and minute data drifts can be detected in a timely manner. The above implementation not only focuses on a direct comparison between current data and historical data from the same period, but also subdivides historical data and compares it with current data from multiple perspectives to ensure that monitoring covers all possible influencing factors. Furthermore, based on the monitoring results, the model's input preprocessing procedures can be adjusted in a timely manner or the model can be retrained to adapt to changes in data distribution and maintain the model's long-term effectiveness.

[0051] In order to convert the aforementioned baseline input data and validation input data into a language that the model can recognize, in an optional implementation, step S202 includes:

[0052] Step S2021: Train multiple feature extraction models using an autoencoder, with each feature extraction model corresponding one-to-one with the baseline input data.

[0053] Specifically, for each of the aforementioned benchmark input data, a separate feature extraction model is trained using an autoencoder. These models are specifically designed to capture the unique feature distributions found in their corresponding benchmark datasets. An autoencoder is an artificial neural network whose goal is to compress and reconstruct data by learning efficient data encoding and decoding mechanisms. It consists of an encoder and a decoder. The encoder transforms the input data into an intermediate low-dimensional representation, while the decoder reconstructs this low-dimensional representation back into a high-dimensional original data form.

[0054] Step S2022: Each feature extraction model is used to extract features from the corresponding benchmark input data to obtain multiple benchmark feature matrices;

[0055] Specifically, each trained feature extraction model will be used to process its corresponding benchmark input dataset. The original data will be converted into feature vectors through an encoding process, and then all feature vectors will be combined into a feature matrix, which is the benchmark feature matrix.

[0056] Understandably, the intermediate low-dimensional representations learned by the autoencoder are the feature vectors of the data, which can be viewed as key features extracted from the original data. These feature vectors typically contain the inherent structure and patterns of the input data.

[0057] Step S2023: Use each feature extraction model to extract features from the corresponding verification input data to obtain multiple verification feature matrices.

[0058] Specifically, similar to the baseline feature matrix, the same feature extraction model is used to extract features from the validation input data extracted from the current input data to obtain the validation feature matrix.

[0059] In the above embodiments, each autoencoder is trained on a specific historical data period, thus accurately capturing the characteristics of the data during that period, and even subtle changes can be sensitively detected by the model. Although each model focuses on the characteristics of a specific period, the training process of the autoencoder enables the model to learn the general characteristics of the data, allowing it to maintain a certain level of generalization ability when processing validation data. By comparing the benchmark feature matrix and the validation feature matrix, not only can the similarity between the current data and data from a specific historical period be evaluated, but the differences in feature distribution between the current data and data from multiple historical periods can also be comprehensively considered, thereby obtaining a more comprehensive stability assessment.

[0060] To evaluate the stability of the verification input data, in one optional implementation, step S203 includes:

[0061] Step S2031: Extract year-on-year input data from historical input data based on current input data, process the year-on-year input data using a feature extraction model, and obtain the year-on-year feature matrix;

[0062] Specifically, data from the same period (year-on-year comparison) as the current data is found in historical data, and then processed using a specific feature extraction model to obtain the year-on-year feature matrix. Assuming the current input data is from April of year x, data from April of year x-1 will be extracted from historical data as the year-on-year input data. For the data from April of year x-, features are extracted using the feature extraction model M_4 trained in April, resulting in the feature matrix O'_4, which represents the feature distribution of the historical data from the same period.

[0063] Step S2032: Calculate the year-on-year difference quantification value based on the verification feature matrix and the benchmark feature matrix;

[0064] Specifically, the MMD method is used to compare and verify the feature matrix O_4 with the benchmark feature matrix S_4 to obtain the year-on-year difference quantification value Y.

[0065] Step S2033: Calculate the quantitative variance of the relative difference in the distribution month-on-month based on the verification feature matrix, the year-on-year feature matrix, and the benchmark feature matrix;

[0066] Specifically, for the feature matrices of other non-comparative periods (O_i and S_i, i≠4), the feature distribution differences between O_i and O_4 and between S_i and O'_4 are calculated respectively. The variance is calculated after taking the absolute value to obtain the quantitative variance Z of the relative difference of the distribution chain.

[0067] Step S2034: Sum the year-on-year difference quantification value and the distribution month-on-month relative difference quantification variance to obtain the stability index.

[0068] Specifically, the year-on-year difference quantification value Y is added to the distribution month-on-month relative difference quantification variance Z to obtain the final stability index Q. The Q value reflects the comprehensive stability difference between the current data and historical data in terms of characteristic distribution.

[0069] Through the above embodiments, the MMD method quantifies the differences between feature matrices, accurately identifying data distribution shifts, especially sensitively capturing minute changes in high-dimensional data. It not only assesses the direct year-on-year difference between current data and historical data from the same period but also considers the relative stability of data across different periods, providing a more comprehensive perspective on data stability. The calculation of stability indicators enables real-time monitoring of the model; once the indicator exceeds a preset threshold, immediate measures can be taken to prevent a decline in model performance.

[0070] In order to obtain year-on-year input data, in one optional implementation, step S2031 above includes:

[0071] Step S20311: Extract input data with the same period from the historical input data according to the period of the current input data to obtain the second input data;

[0072] Specifically, in time series analysis, period matching refers to comparing data from the current period with data from the same past period to better understand the trend of data changes over time. Therefore, by analyzing historical input data and considering the period of the current input data, historical datasets from the same period can be accurately identified.

[0073] Step S20312: Extract the corresponding input data from the second input data according to each current input data to obtain the year-on-year input data.

[0074] Specifically, based on the second input data, the extraction of data subsets is further refined according to the specific period division of the current input data to ensure that the period is exactly the same as that of the current data. A subset with the same period as the current data is selected from historical data as the source of year-on-year input data to ensure the accuracy and reliability of the comparison.

[0075] Through the above embodiments, the periodic segmentation of historical data ensures that the period of the year-on-year input data is completely consistent with that of the current input data, thus improving the accuracy of the analysis. By comparing the current input data with the year-on-year input data one by one, the trend of data changes over time can be observed in detail, and even daily-level data fluctuations can be accurately captured. Furthermore, due to the use of year-on-year data comparison, any data anomalies that deviate from historical periodic patterns can be quickly identified, which is crucial for early warning and prevention of possible model failures.

[0076] In order to calculate the quantified value of the year-on-year difference, in one optional implementation, step S2032 above includes:

[0077] Step S20321: The difference between the verification feature matrix and the benchmark feature matrix is ​​calculated using the maximum average difference to obtain the year-on-year difference quantification value.

[0078] Understandably, Maximum Mean Discrepancy (MMD) is a statistic used to measure the difference between two distributions, and it is particularly suitable for comparing distributions of high-dimensional data. MMD quantifies the distance between two datasets by measuring the expected difference of kernel functions in the feature space, thus providing a parameter-free, non-linear method for comparing distributions.

[0079] Specifically, a kernel function (such as a Gaussian kernel) is chosen to calculate the similarity between data points. The MMD value between the validation feature matrix O_x and the baseline feature matrix S_x is calculated. The MMD value reflects the difference in data distribution between the two matrices. A larger MMD value indicates a greater difference in distribution between the two datasets; conversely, a smaller MMD value indicates a closer similarity in distribution between the two datasets.

[0080] Through the above embodiments, the difference between the verification feature matrix O_x and the benchmark feature matrix S_x is calculated using the maximum mean difference (MMD) method, yielding the year-on-year difference quantification value Y. This technical solution significantly improves the accuracy and timeliness of monitoring the stability of model input data. As a powerful statistical measure, MMD not only enhances the reliability of high-dimensional data distribution comparisons but also makes the detection of data anomalies more sensitive. By monitoring the year-on-year difference quantification value Y in real time, model maintainers can quickly identify possible distribution drifts in the input data and take timely measures to adjust model parameters or perform data preprocessing, thereby ensuring the long-term stability and prediction accuracy of the model.

[0081] In order to calculate the quantified value of the relative difference in the distribution chain ratio, in an optional implementation, step S2033 above includes:

[0082] Step S20331: Calculate the year-on-year difference quantification value between each verification feature matrix and the corresponding benchmark feature matrix to obtain multiple first difference values;

[0083] Specifically, the above steps quantify the differences in the distribution of current input data and historical data in the feature space across different time periods.

[0084] Step S20332: Calculate the quantified value of the year-on-year difference between each year-on-year feature matrix and the corresponding benchmark feature matrix to obtain multiple second difference values;

[0085] Specifically, this step further compares the distribution differences between historical year-on-year data and historical data in the feature space, which helps to assess the stability of the historical data itself, as another context for the comparison.

[0086] Step S20333: Calculate the absolute difference between each first difference value and the corresponding second difference value to obtain multiple relative difference quantification values;

[0087] Specifically, the relative difference quantification value is calculated by comparing the absolute difference between the first difference value Y_i and the second difference value Z_i. This step aims to assess whether the distribution change of the current data relative to historical data significantly exceeds the normal range of change in historical year-on-year data.

[0088] Step S20334: Take the variance of all relative difference quantification values ​​and calculate the mean variance to obtain the distribution chain ratio relative difference quantification variance.

[0089] Specifically, the variance of all relative difference quantified values ​​D_i is calculated, and then the mean variance is calculated to obtain the relative difference quantified variance of the distribution. The mean variance reflects the volatility of the current data distribution relative to historical data; the smaller the value, the better the stability of the data distribution.

[0090] Through the above embodiments, by calculating the relative difference quantification value and its mean variance, we not only focus on the year-on-year changes in data, but also assess the distribution stability of the data relative to other historical periods, enhancing the comprehensiveness and depth of monitoring. The relative difference quantification variance V of the distribution month-on-month can effectively identify anomalies in the data input, especially those changes that may be hidden in the year-on-year changes, providing a strong basis for continuous model optimization and data quality control.

[0091] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0092] This application also provides a model input data stability monitoring device based on an autoencoder. It should be noted that this autoencoder-based model input data stability monitoring device can be used to execute the autoencoder-based model input data stability monitoring method provided in this application. This device is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0093] The following describes the model input data stability monitoring device based on an autoencoder provided in the embodiments of this application.

[0094] Figure 3 This is a structural block diagram of a model input data stability monitoring device based on an autoencoder, according to an embodiment of this application. Figure 3 As shown, the device includes:

[0095] The acquisition unit 10 is used to extract baseline input data from historical input data and to extract verification input data from current input data;

[0096] Understandably, historical data (baseline input data) is data selected from a specific period in the past. This data reflects how the model processes the input data under ideal conditions and is used as a baseline. Current data (validation input data) is input data collected in the current period and is used to monitor the current input stability of the model.

[0097] In practice, data from a certain month in the past year can be selected as the baseline data, and the current input data is the data of this month. For example, for monitoring data in April of year x, the baseline input data includes data from April of year x-1 (year-on-year baseline), as well as data from January to March and May to December of year x-1 (month-on-month baseline).

[0098] Training unit 20 is used to train a feature extraction model using an autoencoder, and to process the benchmark input data and validation input data using the feature extraction model to obtain the benchmark feature matrix and the validation feature matrix.

[0099] As is understandable, an autoencoder is an unsupervised learning neural network primarily used for feature learning and dimensionality reduction. It learns a compact representation of the input data, i.e., a feature vector, through encoding and decoding processes. In this invention, an autoencoder is used to train a feature extraction model to capture key features in the data.

[0100] Following the above embodiments, an autoencoder can be trained based on data from January to December of year x-1 to obtain 12 feature extraction models M_1 to M_12. Then, the corresponding feature extraction models are used to process the benchmark input data and validation input data to obtain multiple benchmark validation matrices and validation feature matrices.

[0101] Calculation unit 30 is used to calculate stability indicators based on the benchmark feature matrix and the verification feature matrix. The stability indicators include year-on-year differences and month-on-month differences in distribution.

[0102] Understandably, the stability index consists of two parts: year-on-year difference and month-on-month distribution difference, which respectively reflect the difference between the current data and the data of the same period in history, and the relative change compared with other historical data.

[0103] Alarm unit 40 is used to generate alarm information when the stability index is greater than a preset threshold.

[0104] Specifically, a reasonable Q threshold is set. If the calculated Q value exceeds this threshold, it indicates a problem with the stability of the model's input data. In this case, the system generates an alarm message, suggesting that further inspection of the model's input data is needed, which may require data preprocessing or model retraining.

[0105] In this embodiment, the acquisition unit extracts baseline input data from historical input data and validation input data from current input data; the training unit trains a feature extraction model using an autoencoder, and processes the baseline and validation input data using the feature extraction model to obtain a baseline feature matrix and a validation feature matrix; the calculation unit calculates a stability index based on the baseline and validation feature matrices, the stability index including year-on-year difference and month-on-month distribution difference; the alarm unit generates an alarm message when the stability index exceeds a preset threshold. This application achieves simultaneous monitoring of model input data stability by comparing baseline input data and current input data, enabling timely detection of the impact of input data changes on model performance. This addresses the problem in existing technologies where model stability monitoring only focuses on the distribution of model output data, ignoring the impact of model input changes on model stability, leading to inaccurate model performance monitoring.

[0106] In an optional implementation, to acquire baseline input data and verification input data, the acquisition unit includes:

[0107] The first acquisition module is used to extract input data from historical input data for a preset period to obtain the first input data;

[0108] Understandably, a preset period refers to a specific period extracted from historical data as a standard benchmark, usually a time period with stable business cycles.

[0109] Taking a transaction data monitoring model as an example, the data of the past year is traversed in units of years (preset period). If the input data of a certain month is relatively stable, it is selected as the first input data mentioned above.

[0110] The second acquisition module is used to extract multiple data segments from the first input data with a preset step size to obtain multiple reference input data.

[0111] Understandably, the preset step size is used to extract historical data through a sliding window, ensuring that data from different time periods can be compared fairly and comprehensively.

[0112] Specifically, in the first input data mentioned above, multiple data segments are randomly extracted at a length (the preset step size, such as one week) that can reflect the characteristics of the data, so as to analyze the periodic changes of the data in more detail, improve the accuracy of monitoring, and obtain the benchmark input data mentioned above.

[0113] The third acquisition module is used to extract the corresponding input data from the current input data in sequence according to each benchmark input data to obtain the verification input data.

[0114] Specifically, for the input data of the current period, the corresponding fields are extracted using the same compensation as the baseline input data to obtain the verification input data, so as to ensure that the periodic changes of the data are consistent.

[0115] Through the above embodiments, by selecting preset periods and step sizes, the periodic changes in data can be analyzed in greater detail, and minute data drifts can be detected in a timely manner. The above implementation not only focuses on a direct comparison between current data and historical data from the same period, but also subdivides historical data and compares it with current data from multiple perspectives to ensure that monitoring covers all possible influencing factors. Furthermore, based on the monitoring results, the model's input preprocessing procedures can be adjusted in a timely manner or the model can be retrained to adapt to changes in data distribution and maintain the model's long-term effectiveness.

[0116] In order to convert the aforementioned baseline input data and validation input data into a language that the model can recognize, in one optional implementation, the training unit includes:

[0117] The training module is used to train multiple feature extraction models using an autoencoder, with each feature extraction model corresponding one-to-one with the baseline input data.

[0118] Specifically, for each of the aforementioned benchmark input data, a separate feature extraction model is trained using an autoencoder. These models are specifically designed to capture the unique feature distributions found in their corresponding benchmark datasets. An autoencoder is an artificial neural network whose goal is to compress and reconstruct data by learning efficient data encoding and decoding mechanisms. It consists of an encoder and a decoder. The encoder transforms the input data into an intermediate low-dimensional representation, while the decoder reconstructs this low-dimensional representation back into a high-dimensional original data form.

[0119] The first processing module is used to extract features from the corresponding benchmark input data using each feature extraction model to obtain multiple benchmark feature matrices.

[0120] Specifically, each trained feature extraction model will be used to process its corresponding benchmark input dataset. The original data will be converted into feature vectors through an encoding process, and then all feature vectors will be combined into a feature matrix, which is the benchmark feature matrix.

[0121] Understandably, the intermediate low-dimensional representations learned by the autoencoder are the feature vectors of the data, which can be viewed as key features extracted from the original data. These feature vectors typically contain the inherent structure and patterns of the input data.

[0122] The second processing module is used to extract features from the corresponding verification input data using each feature extraction model to obtain multiple verification feature matrices.

[0123] Specifically, similar to the baseline feature matrix, the same feature extraction model is used to extract features from the validation input data extracted from the current input data to obtain the validation feature matrix.

[0124] In the above embodiments, each autoencoder is trained on a specific historical data period, thus accurately capturing the characteristics of the data during that period, and even subtle changes can be sensitively detected by the model. Although each model focuses on the characteristics of a specific period, the training process of the autoencoder enables the model to learn the general characteristics of the data, allowing it to maintain a certain level of generalization ability when processing validation data. By comparing the benchmark feature matrix and the validation feature matrix, not only can the similarity between the current data and data from a specific historical period be evaluated, but the differences in feature distribution between the current data and data from multiple historical periods can also be comprehensively considered, thereby obtaining a more comprehensive stability assessment.

[0125] To evaluate the stability of the verification input data, in one optional implementation, the above-mentioned computing unit includes:

[0126] The fourth acquisition module is used to extract year-on-year input data from historical input data based on the current input data, and to process the year-on-year input data using a feature extraction model to obtain the year-on-year feature matrix.

[0127] Specifically, data from the same period (year-on-year comparison) as the current data is found in historical data, and then processed using a specific feature extraction model to obtain the year-on-year feature matrix. Assuming the current input data is from April of year x, data from April of year x-1 will be extracted from historical data as the year-on-year input data. For the data from April of year x-, features are extracted using the feature extraction model M_4 trained in April, resulting in the feature matrix O'_4, which represents the feature distribution of the historical data from the same period.

[0128] The first calculation module is used to calculate the year-on-year difference quantification value based on the verification feature matrix and the benchmark feature matrix;

[0129] Specifically, the MMD method is used to compare and verify the feature matrix O_4 with the benchmark feature matrix S_4 to obtain the year-on-year difference quantification value Y.

[0130] The second calculation module is used to calculate the quantitative variance of the relative difference in the distribution month-on-month based on the validation feature matrix, the year-on-year feature matrix and the benchmark feature matrix.

[0131] Specifically, for the feature matrices of other non-comparative periods (O_i and S_i, i≠4), the feature distribution differences between O_i and O_4 and between S_i and O'_4 are calculated respectively. The variance is calculated after taking the absolute value to obtain the quantitative variance Z of the relative difference of the distribution chain.

[0132] The third calculation module is used to sum the quantified values ​​of year-on-year differences and the quantified variances of the relative differences in the month-on-month distribution to obtain a stability index.

[0133] Specifically, the year-on-year difference quantification value Y is added to the distribution month-on-month relative difference quantification variance Z to obtain the final stability index Q. The Q value reflects the comprehensive stability difference between the current data and historical data in terms of characteristic distribution.

[0134] Through the above embodiments, the MMD method quantifies the differences between feature matrices, accurately identifying data distribution shifts, especially sensitively capturing minute changes in high-dimensional data. It not only assesses the direct year-on-year difference between current data and historical data from the same period but also considers the relative stability of data across different periods, providing a more comprehensive perspective on data stability. The calculation of stability indicators enables real-time monitoring of the model; once the indicator exceeds a preset threshold, immediate measures can be taken to prevent a decline in model performance.

[0135] In order to obtain year-on-year input data, in one optional implementation, the fourth acquisition module mentioned above includes:

[0136] The first acquisition submodule is used to extract input data with the same period from the historical input data according to the period of the current input data to obtain the second input data;

[0137] Specifically, in time series analysis, period matching refers to comparing data from the current period with data from the same past period to better understand the trend of data changes over time. Therefore, by analyzing historical input data and considering the period of the current input data, historical datasets from the same period can be accurately identified.

[0138] The second acquisition submodule is used to extract the corresponding input data from the second input data according to each current input data in turn, so as to obtain the year-on-year input data.

[0139] Specifically, based on the second input data, the extraction of data subsets is further refined according to the specific period division of the current input data to ensure that the period is exactly the same as that of the current data. A subset with the same period as the current data is selected from historical data as the source of year-on-year input data to ensure the accuracy and reliability of the comparison.

[0140] Through the above embodiments, the periodic segmentation of historical data ensures that the period of the year-on-year input data is completely consistent with that of the current input data, thus improving the accuracy of the analysis. By comparing the current input data with the year-on-year input data one by one, the trend of data changes over time can be observed in detail, and even daily-level data fluctuations can be accurately captured. Furthermore, due to the use of year-on-year data comparison, any data anomalies that deviate from historical periodic patterns can be quickly identified, which is crucial for early warning and prevention of possible model failures.

[0141] In order to calculate the quantified value of the year-on-year difference, in one optional implementation, the first calculation module mentioned above includes:

[0142] The first calculation submodule is used to calculate the difference between the verification feature matrix and the benchmark feature matrix using the maximum average difference, and obtain the year-on-year difference quantification value.

[0143] Understandably, Maximum Mean Discrepancy (MMD) is a statistic used to measure the difference between two distributions, and it is particularly suitable for comparing distributions of high-dimensional data. MMD quantifies the distance between two datasets by measuring the expected difference of kernel functions in the feature space, thus providing a parameter-free, non-linear method for comparing distributions.

[0144] Specifically, a kernel function (such as a Gaussian kernel) is chosen to calculate the similarity between data points. The MMD value between the validation feature matrix O_x and the baseline feature matrix S_x is calculated. The MMD value reflects the difference in data distribution between the two matrices. A larger MMD value indicates a greater difference in distribution between the two datasets; conversely, a smaller MMD value indicates a closer similarity in distribution between the two datasets.

[0145] Through the above embodiments, the difference between the verification feature matrix O_x and the benchmark feature matrix S_x is calculated using the maximum mean difference (MMD) method, yielding the year-on-year difference quantification value Y. This technical solution significantly improves the accuracy and timeliness of monitoring the stability of model input data. As a powerful statistical measure, MMD not only enhances the reliability of high-dimensional data distribution comparisons but also makes the detection of data anomalies more sensitive. By monitoring the year-on-year difference quantification value Y in real time, model maintainers can quickly identify possible distribution drifts in the input data and take timely measures to adjust model parameters or perform data preprocessing, thereby ensuring the long-term stability and prediction accuracy of the model.

[0146] In order to calculate the quantified value of the relative difference in the distribution chain, in one optional implementation, the second calculation module includes:

[0147] The second calculation submodule is used to calculate the year-on-year difference quantification value between each verification feature matrix and the corresponding benchmark feature matrix, and obtain multiple first difference values.

[0148] Specifically, the above steps quantify the differences in the distribution of current input data and historical data in the feature space across different time periods.

[0149] The third calculation submodule is used to calculate the quantified value of the year-on-year difference between each year-on-year feature matrix and the corresponding benchmark feature matrix, and obtain multiple second difference values.

[0150] Specifically, this step further compares the distribution differences between historical year-on-year data and historical data in the feature space, which helps to assess the stability of the historical data itself, as another context for the comparison.

[0151] The fourth calculation submodule is used to calculate the absolute difference between each first difference value and the corresponding second difference value to obtain multiple relative difference quantification values;

[0152] Specifically, the relative difference quantification value is calculated by comparing the absolute difference between the first difference value Y_i and the second difference value Z_i. This step aims to assess whether the distribution change of the current data relative to historical data significantly exceeds the normal range of change in historical year-on-year data.

[0153] The fifth calculation submodule is used to take the variance of all relative difference quantification values ​​and calculate the mean variance to obtain the distribution chain ratio relative difference quantification variance.

[0154] Specifically, the variance of all relative difference quantified values ​​D_i is calculated, and then the mean variance is calculated to obtain the relative difference quantified variance of the distribution. The mean variance reflects the volatility of the current data distribution relative to historical data; the smaller the value, the better the stability of the data distribution.

[0155] Through the above embodiments, by calculating the relative difference quantification value and its mean variance, we not only focus on the year-on-year changes in data, but also assess the distribution stability of the data relative to other historical periods, enhancing the comprehensiveness and depth of monitoring. The relative difference quantification variance V of the distribution month-on-month can effectively identify anomalies in the data input, especially those changes that may be hidden in the year-on-year changes, providing a strong basis for continuous model optimization and data quality control.

[0156] The aforementioned autoencoder-based model input data stability monitoring device includes a processor and a memory. The acquisition unit, training unit, calculation unit, and alarm unit are all stored as program units in the memory, and the processor executes these program units to achieve their respective functions. All of the above modules reside in the same processor; alternatively, the modules may be located in different processors in any combination.

[0157] The processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and adjusting kernel parameters can improve the accuracy of model performance monitoring.

[0158] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0159] This invention provides a computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device containing the computer-readable storage medium to perform the autoencoder-based model input data stability monitoring method.

[0160] Specifically, the methods for monitoring the stability of model input data based on autoencoders include:

[0161] Step S201: Extract baseline input data from historical input data and extract verification input data from current input data;

[0162] Step S202: Train the feature extraction model using an autoencoder, and process the benchmark input data and validation input data using the feature extraction model to obtain the benchmark feature matrix and the validation feature matrix.

[0163] Step S203: Calculate stability indices based on the baseline feature matrix and the validation feature matrix. The stability indices include year-on-year differences and month-on-month differences in distribution.

[0164] Step S204: If the stability index is greater than the preset threshold, generate an alarm message.

[0165] This invention provides a processor for running a program, wherein the program executes the above-described method for monitoring the stability of model input data based on an autoencoder.

[0166] Specifically, the methods for monitoring the stability of model input data based on autoencoders include:

[0167] Step S201: Extract baseline input data from historical input data and extract verification input data from current input data;

[0168] Step S202: Train the feature extraction model using an autoencoder, and process the benchmark input data and validation input data using the feature extraction model to obtain the benchmark feature matrix and the validation feature matrix.

[0169] Step S203: Calculate stability indices based on the baseline feature matrix and the validation feature matrix. The stability indices include year-on-year differences and month-on-month differences in distribution.

[0170] Step S204: If the stability index is greater than the preset threshold, generate an alarm message.

[0171] This invention provides an electronic device, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs at least the following steps:

[0172] Step S201: Extract baseline input data from historical input data and extract verification input data from current input data;

[0173] Step S202: Train the feature extraction model using an autoencoder, and process the benchmark input data and validation input data using the feature extraction model to obtain the benchmark feature matrix and the validation feature matrix.

[0174] Step S203: Calculate stability indices based on the baseline feature matrix and the validation feature matrix. The stability indices include year-on-year differences and month-on-month differences in distribution.

[0175] Step S204: If the stability index is greater than the preset threshold, generate an alarm message.

[0176] This application also provides a computer program product, which, when executed on a data processing device, is suitable for executing an initialization program having at least the following method steps:

[0177] Step S201: Extract baseline input data from historical input data and extract verification input data from current input data;

[0178] Step S202: Train the feature extraction model using an autoencoder, and process the benchmark input data and validation input data using the feature extraction model to obtain the benchmark feature matrix and the validation feature matrix.

[0179] Step S203: Calculate stability indices based on the baseline feature matrix and the validation feature matrix. The stability indices include year-on-year differences and month-on-month differences in distribution.

[0180] Step S204: If the stability index is greater than the preset threshold, generate an alarm message.

[0181] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. They can be implemented using computer-executable program code, and thus can be stored in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those described herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0182] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can 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, etc.) containing computer-usable program code.

[0183] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can 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, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0184] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0185] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0186] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0187] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0188] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, 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 technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic tape, disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0189] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0190] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0191] As can be seen from the above description, the embodiments of this application achieve the following technical effects:

[0192] 1) The autoencoder-based model input data stability monitoring method of this application first extracts benchmark input data from historical input data and validation input data from current input data. Then, an autoencoder is used to train a feature extraction model, which processes the benchmark and validation input data to obtain benchmark and validation feature matrices. Next, a stability index is calculated based on the benchmark and validation feature matrices, including year-on-year differences and month-on-month distribution differences. Finally, an alarm is generated when the stability index exceeds a preset threshold. This application achieves simultaneous monitoring of model input data stability by comparing benchmark and current input data, enabling timely detection of the impact of input data changes on model performance. This addresses the problem in existing technologies where model stability monitoring only focuses on the distribution of model output data, ignoring the impact of model input changes on model stability, leading to inaccurate model performance monitoring.

[0193] 2) The model input data stability monitoring device based on an autoencoder of this application comprises: an acquisition unit that extracts baseline input data from historical input data and validation input data from current input data; a training unit that trains a feature extraction model using an autoencoder and processes the baseline and validation input data using the feature extraction model to obtain a baseline feature matrix and a validation feature matrix; a calculation unit that calculates a stability index based on the baseline and validation feature matrices, the stability index including year-on-year difference and month-on-month distribution difference; and an alarm unit that generates an alarm message when the stability index exceeds a preset threshold. This application achieves simultaneous monitoring of model input data stability by comparing baseline input data and current input data, enabling timely detection of the impact of input data changes on model performance. This addresses the problem in existing technologies where model stability monitoring only focuses on the distribution of model output data, ignoring the impact of model input changes on model stability, leading to inaccurate model performance monitoring.

[0194] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for monitoring the stability of model input data based on an autoencoder, characterized in that, include: Baseline input data is extracted from historical input data, and validation input data is extracted from current input data; An autoencoder is used to train a feature extraction model, and the feature extraction model is used to process the benchmark input data and the verification input data to obtain a benchmark feature matrix and a verification feature matrix. Based on the benchmark feature matrix and the verification feature matrix, a stability index is calculated, which includes year-on-year differences and month-on-month differences in distribution. If the stability index exceeds a preset threshold, an alarm message is generated.

2. The method according to claim 1, characterized in that, Baseline input data is extracted from historical input data, and validation input data is extracted from current input data, including: The first input data is obtained by extracting input data for a preset period from the historical input data; Multiple data segments are extracted from the first input data with a preset step size to obtain multiple reference input data; The corresponding input data is extracted from the current input data according to each of the aforementioned benchmark input data to obtain the verification input data.

3. The method according to claim 1, characterized in that, An autoencoder is used to train a feature extraction model, which is then used to process the benchmark input data and the validation input data to obtain a benchmark feature matrix and a validation feature matrix, including: Multiple feature extraction models are trained using the autoencoder, and each feature extraction model corresponds one-to-one with the baseline input data; Each of the aforementioned feature extraction models is used to extract features from the corresponding benchmark input data to obtain multiple benchmark feature matrices. Each of the aforementioned feature extraction models is used to extract features from the corresponding verification input data to obtain multiple verification feature matrices.

4. The method according to claim 1, characterized in that, The stability index is calculated based on the baseline feature matrix and the verification feature matrix, including: Based on the current input data, year-on-year input data is extracted from the historical input data, and the year-on-year input data is processed using the feature extraction model to obtain a year-on-year feature matrix; Calculate the year-on-year difference quantification value based on the verification feature matrix and the benchmark feature matrix; The distribution month-on-month relative difference quantification variance is calculated based on the verification feature matrix, the year-on-year feature matrix, and the benchmark feature matrix. The stability index is obtained by summing the quantified value of the year-on-year difference and the quantified variance of the relative difference in the month-on-month distribution.

5. The method according to claim 4, characterized in that, Extracting year-on-year input data from the historical input data based on the current input data includes: Based on the period of the current input data, input data with the same period of the same time is extracted from the historical input data to obtain the second input data; The corresponding input data is extracted from the second input data according to each current input data to obtain the year-on-year input data.

6. The method according to claim 4, characterized in that, The year-on-year difference quantification value is calculated based on the verification feature matrix and the benchmark feature matrix, including: The difference between the verification feature matrix and the benchmark feature matrix is ​​calculated using the maximum average difference to obtain the quantified value of the year-on-year difference.

7. The method according to claim 4, characterized in that, The distribution relative difference quantification variance is calculated based on the verification feature matrix, the year-on-year feature matrix, and the benchmark feature matrix, including: Calculate the year-on-year difference quantification value between each of the verification feature matrices and the corresponding benchmark feature matrix to obtain multiple first difference values; Calculate the year-on-year difference quantification value between each of the year-on-year feature matrices and the corresponding benchmark feature matrix to obtain multiple second difference values; Calculate the absolute difference between each of the first difference values ​​and the corresponding second difference values ​​to obtain multiple relative difference quantification values; The variance of all the relative difference quantification values ​​is taken and the mean variance is calculated to obtain the relative difference quantification variance of the distribution.

8. A model input data stability monitoring device based on an autoencoder, characterized in that, The device includes: The acquisition unit is used to extract baseline input data from historical input data and to extract verification input data from current input data. The training unit is used to train a feature extraction model using an autoencoder, and to process the benchmark input data and the verification input data using the feature extraction model to obtain a benchmark feature matrix and a verification feature matrix. A calculation unit is used to calculate a stability index based on the benchmark feature matrix and the verification feature matrix, wherein the stability index includes year-on-year difference and month-on-month distribution difference; An alarm unit is used to generate alarm information when the stability index exceeds a preset threshold.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform the method according to any one of claims 1 to 7.

10. An electronic device, characterized in that, include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs comprising methods for performing any one of claims 1 to 7.

Images