Time series model determination method and apparatus, device, and storage medium

Abstract

The application provides a time series model determination method and device, equipment and a storage medium, relates to the technical field of big data, and in particular to the technical field of big data analysis, statistical modeling and pattern recognition. The method comprises: obtaining autocorrelation coefficient graphs and partial autocorrelation coefficient graphs of time series data; identifying tail characteristics and truncation characteristics of the autocorrelation coefficient graphs and the partial autocorrelation coefficient graphs respectively; and determining the model order of a time series model according to the tail characteristics and the truncation characteristics, wherein the time series model is used to predict data of the time series data in a future time period. The application can accurately determine the order of the time series model, automatically identify the category of the time series model, improve the fluency of automatic execution of the model component, and avoid inefficient operations of stopping for manual judgment in the middle of the fully automatic process.

Description

Technical Field

[0001] This application relates to the field of big data technology, specifically to the fields of big data analysis, statistical modeling and pattern recognition, and particularly to a method, apparatus, device and storage medium for determining a time series model. Background Technology

[0002] Time series models are one of the most commonly used methods in various real-world business scenarios. Because they do not require explanatory variables to assist in prediction, they can be analyzed simply by providing the target data itself. Therefore, they can achieve relatively good prediction and estimation results in many environments where the data is indeed incomplete.

[0003] There are many types of time series models, each suitable for different data distribution types. However, currently, the model order and type are usually determined manually, which is not only costly and inefficient, but also affects the connection and coherence between the model and other functional components of the system, thus affecting the model's intelligence level and computing speed. Summary of the Invention

[0004] This application provides a method, apparatus, device, and storage medium for determining a time series model, in order to solve the problems of low operational efficiency, reduced intelligence level, and decreased computational speed caused by manually determining the model order and type in the prior art.

[0005] Firstly, this application provides a method for determining a time series model, comprising:

[0006] Obtain the autocorrelation coefficient and partial autocorrelation coefficient plots of time series data;

[0007] Identify the tailing and truncation features of the autocorrelation coefficient plot and the partial autocorrelation coefficient plot, respectively;

[0008] The model order of the time series model is determined based on the tailing and truncation features, wherein the time series model is used to predict the data of the time series data in future time periods.

[0009] Secondly, this application provides an apparatus for determining a time series model, comprising:

[0010] The correlation coefficient plot acquisition module is used to acquire the autocorrelation coefficient plot and partial autocorrelation coefficient plot of time series data;

[0011] The feature recognition module is used to identify the tailing features and truncation features of the autocorrelation coefficient plot and the partial autocorrelation coefficient plot, respectively.

[0012] The model determination module is used to determine the model order of the time series model based on the tailing feature and the truncation feature, wherein the time series model is used to predict the data of the time series data in future time periods.

[0013] Thirdly, embodiments of this application also provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for determining a time series model as described in any of the embodiments of this application.

[0014] Fourthly, embodiments of this application also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for determining a time series model as described in any of the embodiments of this application.

[0015] Fifthly, embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, implements the method for determining a time series model as described in any of the embodiments of this application.

[0016] The method, apparatus, device, and storage medium for determining time series models provided in this application automatically identify tailing and truncation features from autocorrelation coefficient plots and partial autocorrelation coefficient plots, thereby determining the model order of the time series model. This not only accurately determines the model order and model type, but also improves the smoothness of automatic execution of model code components, avoiding the inefficient operation of stopping for manual judgment in the middle of the fully automated process. Especially in the current application scenarios of various intelligent algorithms with extremely high requirements for computational continuity and speed, it can greatly improve the intelligence level and computing speed of time series models. Attached Figure Description

[0017] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0018] Figure 1 A flowchart illustrating a method for determining a time series model provided in an embodiment of this application;

[0019] Figure 2a A flowchart of the tail feature recognition method in the method for determining the time series model provided in the embodiments of this application;

[0020] Figure 2b This is a schematic diagram of data with trailing features provided in an embodiment of this application;

[0021] Figure 3a A flowchart of the truncation feature recognition method in the method for determining the time series model provided in the embodiments of this application;

[0022] Figure 3b This is a schematic diagram of data with truncation features provided in an embodiment of this application;

[0023] Figure 4 A flowchart illustrating a method for determining a time series model provided in an embodiment of this application;

[0024] Figure 5 A flowchart illustrating a method for determining a time series model provided in an embodiment of this application;

[0025] Figure 6 A schematic diagram of the structure of a time series model determination device provided in an embodiment of this application;

[0026] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0027] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concepts of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0028] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the application and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present application, not the entire structure.

[0029] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first" and "second" are used only to distinguish descriptions and should not be construed as indicating or implying relative importance. The acquisition, storage, use, and processing of data in the technical solution of this application all comply with the relevant provisions of national laws and regulations.

[0030] Figure 1 This is a flowchart illustrating a method for determining a time series model provided in an embodiment of this application. This embodiment is applicable to situations where the model order of a time series model is automatically determined, thereby determining the model type based on the model order. It relates to the field of big data technology, specifically to the fields of big data analysis, statistical modeling, and pattern recognition. This method can be executed by a time series model determination device, which can be implemented in software and / or hardware, preferably configured in an electronic device, such as a computer or server. Figure 1 As shown, the method specifically includes:

[0031] S101. Obtain the autocorrelation coefficient plot and partial autocorrelation coefficient plot of the time series data.

[0032] S102. Identify the tailing and truncation features of the autocorrelation coefficient plot and the partial autocorrelation coefficient plot, respectively.

[0033] S103. Determine the model order of the time series model based on the tailing and truncation features, wherein the time series model is used to predict the data of the time series data in the future time period.

[0034] In real-world work scenarios, time series models are generally used directly to predict the future based on existing data. There are many types of time series models, each suitable for different data distributions, such as AR(p) models, MA(q) models, and ARMA(p,q), where p and q represent the model order. Accurately determining the model order and subsequently the model type is crucial for improving the applicability and accuracy of data analysis. However, current technologies typically rely on manual methods based on the correlation characteristics of time series data to determine the model order and type, which is not only inefficient but also easily influenced by subjective perceptions, reducing the accuracy of model determination. The technical solution of this application first obtains the autocorrelation coefficient plot and partial autocorrelation coefficient plot of the time series data, and then automatically identifies whether the data exhibits tailing and truncation characteristics from these plots, thereby determining the model order of the time series model based on these tailing and truncation characteristics.

[0035] Specifically, the time series data can be composed of bank monetary data from different sources. For example, in a scenario where a time series model is used to predict a bank's total year-end revenue, the time series data can be a sequence of monetary data from different areas such as wealth management or deposits, and monetary data from different branches. Therefore, the corresponding time series data can be determined based on the scenario predicted by the time series model, and this time series data has the applicable physical meaning or attributes under that scenario. This application does not limit the applicable scenarios of the time series model. The autocorrelation coefficient and partial autocorrelation coefficient are two types of correlation coefficients used in time series analysis methods. The autocorrelation coefficient plot and partial autocorrelation coefficient plot are images composed of the autocorrelation coefficient and partial autocorrelation coefficient obtained from the time series data. Specific acquisition methods can be referred to in the relevant existing technologies of time series analysis methods, and will not be elaborated here.

[0036] Tailing characteristics refer to data that is positive and gradually decreases towards 0, or data that is negative and gradually increases towards 0. When data exhibits tailing characteristics, it indicates that the data has a tailing property. Truncating characteristics refer to data that shows a sudden and rapid drop to near 0 for the first time when it is positive, or a sudden and rapid rise to near 0 for the first time when it is negative. When data exhibits truncating characteristics, it indicates that the data has a truncating property, and the order of the correlation coefficient corresponding to the rapid drop or rise is the truncating order. This application's embodiments can automatically identify the tailing and truncating characteristics of autocorrelation coefficient plots and partial autocorrelation coefficient plots, determine whether the autocorrelation coefficient and partial autocorrelation coefficient each have tailing and truncating characteristics, and thus determine the model order of the time series model based on the tailing and truncating characteristics. For example, for an ARMA(p,q) model, when both the autocorrelation coefficient plot and the partial autocorrelation coefficient plot exhibit tailing characteristics, then both model orders p and q are 0. When both the autocorrelation coefficient plot and the partial autocorrelation coefficient plot exhibit truncation characteristics, then the model orders p and q are the truncation orders determined by the autocorrelation coefficient plot and the partial autocorrelation coefficient plot, respectively. When both the autocorrelation coefficient plot and the partial autocorrelation coefficient plot exhibit tailing characteristics, then the model order p is 0, and the model order q is the truncation order determined by the partial autocorrelation coefficient plot. When both the autocorrelation coefficient plot and the partial autocorrelation coefficient plot exhibit tailing characteristics, then the model order q is 0, and the model order p is the truncation order determined by the autocorrelation coefficient plot. Once the model orders p and q are determined, the type of time series model is naturally determined as AR(p), MA(q), or ARMA(p,q).

[0037] It should be noted that the fundamental requirement of current artificial intelligence is that everything is judged by machines instead of humans. This places high demands on the seamless integration between the model and other functional components. If a time series model embedded in a complex system still needs to pause its process for structural judgment during operation, it undoubtedly severely impacts the model's intelligence level and computational speed. This is completely unacceptable for engineering work. For example, if a process requiring manual judgment occurs at midnight, it would require waiting for over ten hours until the next day when staff return to work, significantly delaying data analysis efficiency. The technical solution of this application, however, automatically identifies tailing and truncation features from autocorrelation and partial autocorrelation coefficient plots to determine the model order of the time series model. This not only accurately determines the model order and type, avoiding the influence of subjective factors, but also improves the smoothness of automatic execution of model code components, avoiding the inefficient operation of stopping for manual judgment midway through a fully automated process. Especially in current intelligent algorithm application scenarios with extremely high requirements for computational continuity and speed, this can greatly improve the intelligence level and computational speed of time series models.

[0038] In one implementation, identifying the tailing and truncation features of the autocorrelation coefficient plot and the partial autocorrelation coefficient plot, respectively, may include: identifying the tailing features of the autocorrelation coefficient plot and the partial autocorrelation coefficient plot; and identifying the truncation features of the autocorrelation coefficient plot or the partial autocorrelation coefficient plot if the latter does not have a tailing feature. That is, first, the autocorrelation coefficient plot and the partial autocorrelation coefficient plot are assessed for tailing; if no tailing is found, then the truncation feature is determined. Specifically, indices T1 and T2 can be constructed, with values ​​of only 0 or 1. T1 = 0 represents tailing, and T1 = 1 represents no tailing; T2 = 0 represents truncation, and T2 = 1 represents no truncation. Initially, it is assumed that the target has tailing but no truncation, i.e., T1 = 0 and T2 = 1. Then, first verify its tailing property. If the verification result is that it has tailing property, the judgment ends and the markings T1=0 and T2=1 are maintained. If the verification result is that it has no tailing property, then T1=1 is changed and T2 is still maintained at 1, and then the truncation property is judged. If the verification result is that it has truncation property, then T2=0 is changed, otherwise T2=1 is maintained.

[0039] If either the autocorrelation coefficient plot or the partial autocorrelation coefficient plot shows no tailing or truncation, it indicates that the data is not suitable for time series analysis. In this case, an empirical analysis method can be used to average the past data as the future forecast. This empirical method can be used as a special case of the time series model when the data is deemed unsuitable for time series analysis, in order to ensure the smooth automated operation of the model.

[0040] Furthermore, prior to model building, data preparation is a necessary step. Time series models process data sequences recorded over time, with each data point having a corresponding chronological time point. This data must be free of errors and significant outliers. Therefore, in this embodiment, for missing values ​​in the NA class, a Monte Carlo algorithm can be used to simulate the overall sample distribution. Then, a random sample generation method is used to fill the missing values, setting an upper limit for outliers. The 5% and 95% quantiles of the overall sample are used. Data exceeding the [5%, 95%] quantile range are processed, with data exceeding the 95% quantile assigned the 95% quantile value and data below the 5% quantile assigned the 5% quantile value. This embodiment does not impose any specific numerical limits on the quantile values; parameters can be configured as needed. This data preparation improves the quality of the time series data, providing a foundation for accurately determining the model order and type of the time series model.

[0041] Figure 2aThis is a flowchart of the tail feature identification method in the time series model determination method provided in this application embodiment. This embodiment further optimizes how to identify tail features based on the above embodiments. Figure 2a As shown, the method includes:

[0042] S201. Perform difference calculations on the correlation coefficients of adjacent orders in the autocorrelation coefficient plot and the partial autocorrelation coefficient plot respectively to obtain the first-order difference sequence of autocorrelation coefficient and the first-order difference sequence of partial autocorrelation coefficient.

[0043] S202. Based on the first-order difference sequence of autocorrelation coefficients and the first-order difference sequence of partial autocorrelation coefficients, identify the tailing characteristics of the autocorrelation coefficient plot and the partial autocorrelation coefficient plot, respectively.

[0044] Figure 2b This is a schematic diagram of data with tailing characteristics provided in an embodiment of this application. As shown in the figure, when the data is positive and tends towards 0 with a gradual downward trend, or when the data is negative and tends towards 0 with a gradual upward trend, it is considered to have tailing characteristics. In order to intelligently identify whether there is tailing characteristics, it is necessary to formulate a detailed description of the numerical change process of the correlation coefficient in the autocorrelation coefficient plot and the partial autocorrelation coefficient plot. Moreover, the definition range of sudden drop in value and near 0 needs to be more refined. In this embodiment of the application, the absolute value of the correlation coefficient in the range of 0.5-1 is considered a relatively strong correlation, that is, [-1,-0.5] is a strong negative correlation, [0.5,1] is a strong positive correlation, and [-0.1,0.1] is a weak correlation range, that is, near 0 as mentioned above. Since there is no essential difference in the basic form of the autocorrelation coefficient and the partial autocorrelation coefficient plot, the logic for judging the tailing and truncation characteristics of the two is consistent. The same intelligent judgment mechanism can be used to judge the tailing and truncation characteristics of the autocorrelation coefficient plot and the partial autocorrelation coefficient plot.

[0045] In this embodiment, the correlation coefficients of adjacent orders in the autocorrelation coefficient plot and the partial autocorrelation coefficient plot are first differentially calculated to obtain the first-order difference sequence of autocorrelation coefficients and the first-order difference sequence of partial autocorrelation coefficients. For example, the first-order autocorrelation coefficient a1 of the autocorrelation coefficient plot or the partial autocorrelation coefficient plot is taken as the basis, and the correlation coefficients a1 to a1 of the first 10 orders in the plot are calculated. 10 By performing differencing, the first-order difference sequences d1 to d9 of the autocorrelation coefficient and partial autocorrelation coefficient are obtained, where d i =a (i+1) -a iLet i range from 1 to 9. Then, based on the first-order difference sequence of the autocorrelation coefficient and the first-order difference sequence of the partial autocorrelation coefficient, the tailing characteristics of the autocorrelation coefficient plot and the partial autocorrelation coefficient plot are identified respectively. Specifically, it is determined whether the first-order difference sequence of the autocorrelation coefficient or the first-order difference sequence of the partial autocorrelation coefficient has the following characteristics: the data are positive and tend towards 0 with a gradual decreasing trend, or the data are negative and tend towards 0 with a gradual increasing trend. If the above characteristics are present, the autocorrelation coefficient plot or the partial autocorrelation coefficient plot is considered to have a tailing characteristic. In this way, by performing difference calculations on the correlation coefficients and identifying whether they have a tailing characteristic based on the resulting difference sequence, the accuracy of the determination is higher than that of directly determining whether they have a tailing characteristic based on the correlation coefficients in the autocorrelation coefficient plot and the partial autocorrelation coefficient plot.

[0046] In one implementation, for the case of positive decay, tailing characteristics of the autocorrelation coefficient graph and the partial autocorrelation coefficient graph are identified based on the first-order difference sequence of the autocorrelation coefficient and the first-order difference sequence of the partial autocorrelation coefficient, respectively. This includes: identifying the autocorrelation coefficient graph as having tailing characteristics if each difference value in the first-order difference sequence of the autocorrelation coefficient satisfies a first condition; and identifying the partial autocorrelation coefficient graph as having tailing characteristics if each difference value in the first-order difference sequence of the partial autocorrelation coefficient satisfies the first condition. In other words, the difference values ​​are compared with the first condition. If the difference values ​​in the first-order difference sequence of the autocorrelation coefficient satisfy the first condition, the autocorrelation coefficient graph is identified as having tailing characteristics; if the difference values ​​in the first-order difference sequence of the partial autocorrelation coefficient satisfy the first condition, the partial autocorrelation coefficient graph is identified as having tailing characteristics. The first condition includes the following three points:

[0047] (1) The first correlation coefficient in the autocorrelation coefficient plot or partial autocorrelation coefficient plot is greater than the correlation threshold;

[0048] (2) Each difference value belongs to the first numerical range, and the proportion of difference values ​​belonging to the second numerical range does not exceed the first percentage, wherein the first numerical range includes the second numerical range;

[0049] (3) Satisfies the first formula: Where, d i d j Let i and j represent the difference values, where i and j are natural numbers.

[0050] The first correlation coefficient is a1, and the correlation threshold can be configured to 0.5. Therefore, a correlation coefficient greater than the correlation threshold is represented as a1>0.5. Afterwards, it is necessary to observe whether d1 to d9 gradually decrease and decay to near 0, i.e., within the range of [-0.1, 0.1]. If each of the d9 values ​​decreases... i d are all within the first numerical range (e.g., [-0.1, 0.05]) and within the second numerical range (e.g., [0, 0.05]).i If the number of items does not exceed the first percentage of the total (such as one-third), that is, three items out of one-third of 9, and at the same time satisfies the first formula above, then it can be determined that it has a trailing property.

[0051] It's important to note that the condition a1 > 0.5 in the first condition above is to ensure the sequence has correlation: if the correlation is too weak and generally less than 0.5, it may be a white noise sequence, in which case even if it gradually decays to 0, it cannot be considered a tailing; also, if d is in the range [0, 0.05]... i Too many or If the upward trend accounts for too large a proportion, the overall downward trend will not be obvious, and it is still impossible to determine whether it is a tailing pattern.

[0052] In one implementation, for the case of increasing negative values, tailing characteristics of the autocorrelation coefficient graph and the partial autocorrelation coefficient graph are identified based on the first-order difference sequence of the autocorrelation coefficient and the first-order difference sequence of the partial autocorrelation coefficient, respectively. This includes: identifying the autocorrelation coefficient graph as having tailing characteristics if each difference value in the first-order difference sequence of the autocorrelation coefficient satisfies a second condition; and identifying the partial autocorrelation coefficient graph as having tailing characteristics if each difference value in the first-order difference sequence of the partial autocorrelation coefficient satisfies a second condition. The second condition includes the following three points:

[0053] (1) The first correlation coefficient in the autocorrelation coefficient plot or partial autocorrelation coefficient plot is less than a negative value of the correlation threshold;

[0054] (2) All difference values ​​belong to the third numerical range, and the proportion of difference values ​​belonging to the fourth numerical range does not exceed the second percentage, wherein the third numerical range includes the fourth numerical range;

[0055] (3) Satisfies the second formula: Where, d i d j Let i and j represent the difference values, where i and j are natural numbers.

[0056] Similar to the case of positive number decay, in the case of negative number increase, if a1 < -0.5, it is necessary to check whether d1 to d9 gradually rise to near 0. If each d i All values ​​are within the third numerical range (e.g., [-0.05, 0.1]), and also within the fourth numerical range (e.g., [-0.05, 0]). i The number of items cannot exceed the second percent of the total (e.g., one-third), i.e., 3 items. If the above second formula is satisfied, it can be determined that there is a trailing effect.

[0057] The technical solution of this application provides a formulaic description of whether the numerical change process of the correlation coefficient has a tailing feature, which enables intelligent determination of the model order of the time series model, improves the smoothness of automatic execution of the model code components, avoids the inefficient operation of stopping for manual judgment in the middle of the fully automatic process, and thus improves the intelligence level and computing speed of the time series model.

[0058] Figure 3a This is a flowchart of the truncated feature identification method in the time series model determination method provided in this application embodiment. This embodiment further optimizes how to identify truncated features based on the above embodiments. Figure 3a As shown, the method includes:

[0059] S301. Determine whether there is at least one cutoff point in the autocorrelation coefficient plot or partial autocorrelation coefficient plot, wherein the cutoff point indicates that the correlation coefficients of other orders after the cutoff point do not exceed the fifth numerical range.

[0060] S302. If there is at least one cutoff point, select the correlation coefficient with the smallest order from the at least one cutoff point as the target cutoff point.

[0061] S303. Determine whether the target cutoff point is a true cutoff point. If it is a true cutoff point, identify it as an autocorrelation coefficient plot or partial autocorrelation coefficient plot with truncation characteristics. The true cutoff point means that the correlation coefficients of other orders before the true cutoff point are all in the strong correlation region.

[0062] Figure 3b This is a schematic diagram of data with truncation characteristics provided in an embodiment of this application. As shown in the figure, a truncation characteristic can be identified when the correlation coefficient changes abruptly. For example, a positive value drops directly to [-0.1, 0.1], and before the drop, the correlation coefficients of each order should remain at a high level, for example, not lower than 0.5; or, a negative value rises directly to [-0.1, 0.1], and before the rise, the correlation coefficients of each order should remain at a low level, for example, not higher than -0.5. In these cases, the truncation characteristic is considered to be present.

[0063] Furthermore, in one implementation, the truncation order can be configured to be within 5. Since truncation models exceeding order 5 would be too complex, even if the judgment is accurate, the excessive number of model parameters would lead to enormous computational difficulties, easily causing problems such as computer crashes during model fitting. Therefore, only truncation possibilities within order 5 are considered, and those exceeding order 5 are considered untruncation.

[0064] Specifically, set an order parameter p, and observe the p-th order correlation coefficient a in the autocorrelation coefficient plot or partial autocorrelation coefficient plot. pLet p vary from 1 to 5, and find at least one potential cutoff point, denoted as a. p* p* can only take a value between 1 and 5 for this potential cutoff point a. p* The criterion is that the absolute value of the correlation coefficients of all other orders following this point does not exceed the fifth numerical range (e.g., [-0.1, 0.1]). If there is a potential cutoff point a that meets the requirements among a1 to a5. p* If a condition is met, the point with the smallest index is selected as the final cutoff point. This is because if a condition p*<5 is found, then the other correlation coefficient points a after p* are selected. p It must also satisfy the judgment condition, so it is sufficient to take the point with the smallest index, that is, the first point that satisfies the judgment condition. If none of a1 to a5 meet the judgment criteria for a potential cutoff point, then it is determined that there is no truncation.

[0065] After finding the target cutoff point a p* Next, it is necessary to determine whether the target cutoff point is a true cutoff point. If it is a true cutoff point, it is identified as having truncated characteristics in the autocorrelation coefficient plot or partial autocorrelation coefficient plot. A true cutoff point indicates that the correlation coefficients of other orders prior to this true cutoff point are all in the strong correlation region.

[0066] In one implementation, determining whether a target cutoff point is a true cutoff point includes:

[0067] If the true cutoff point is greater than zero, then check if the correlation coefficients of all other orders before the true cutoff point are greater than the threshold. If so, then the target cutoff point is determined to be the true cutoff point; or

[0068] If the true cutoff point is less than zero, then it is determined whether the correlation coefficients of other orders before the true cutoff point are all less than the negative value of the judgment threshold. If so, the target cutoff point is determined to be the true cutoff point.

[0069] That is, if a p* If the correlation coefficient is greater than 0, then check whether all correlation coefficient points before p* are greater than the judgment threshold, such as 0.5 (including a). p* If so, then it is determined that there is truncation, and the truncation order is p*; if a p* If <0, then check whether all correlation coefficient points before p* are less than -0.5 (including a). p* If the condition is met, then the condition is considered to have truncation, and the truncation order is p*; if neither of the above two conditions is met, then the condition is considered not to have truncation.

[0070] The technical solution of this application provides a formulaic description of whether the numerical change process of the correlation coefficient has a truncation feature, which enables intelligent determination of the model order of the time series model, improves the smoothness of automatic execution of the model code components, avoids the inefficient operation of stopping for manual judgment in the middle of the fully automatic process, and thus improves the intelligence level and computing speed of the time series model.

[0071] In time series models, the autocorrelation coefficient and partial autocorrelation coefficient determine the model structure. In one implementation, the decision logic is as follows: the candidate model is ARMA(m, n), where m and n are the model's order parameters, and their values ​​do not exceed 5; otherwise, the model would be too complex and lose its modeling significance. Once m and n are determined, the ARMA model is also determined. Its order m is determined by the tailing or truncation of the partial autocorrelation coefficient plot. If the partial autocorrelation coefficient plot shows tailing, then m = 0 is selected; if it shows truncation and the truncation order is p*, then m = p* is selected. Similarly, if the autocorrelation coefficient plot shows tailing, then n = 0; if the autocorrelation coefficient plot is truncation and the truncation order is p*, then n = p* is selected.

[0072] Furthermore, if either the autocorrelation coefficient plot or the partial autocorrelation coefficient plot exhibits characteristics of neither tailing nor truncation, it indicates that the data is not suitable for time series analysis. In this case, an empirical analysis method can be used to average the past data as the future forecast value. This empirical method can be used as a special case of the time series model when it is determined that the data is not suitable for time series analysis, in order to ensure that the automated operation of the model can proceed smoothly.

[0073] Figure 4 This is a flowchart illustrating a method for determining a time series model, provided in an embodiment of this application. This embodiment further optimizes how to modify the model based on the above embodiments. Figure 4 As shown, the method includes:

[0074] S401. Obtain the autocorrelation coefficient plot and partial autocorrelation coefficient plot of the time series data.

[0075] S402. Identify the tailing and truncation features of the autocorrelation coefficient plot and the partial autocorrelation coefficient plot, respectively.

[0076] S403. Determine the model order of the time series model based on the tailing and truncation features, wherein the time series model is used to predict the data of the time series data in the future time period.

[0077] S404. Use the time series model determined according to the model order as the current time series model.

[0078] S405. Obtain the training set and validation set from the time series sample data.

[0079] S406. Fit the current time series model using the training set, and use the fitted current time series model to make predictions to obtain the first prediction result.

[0080] S407. Calculate the first mean square error based on the first prediction result and the validation set.

[0081] S408. The second prediction result is obtained by using the optimal time series model determined based on the optimal model selection criterion.

[0082] S409. Calculate the second mean square error based on the second prediction result and the validation set.

[0083] S410. Based on the comparison results of the first mean square error and the second mean square error, the prediction results of the current time series model for the time series data in the future time period are corrected.

[0084] After identifying the model order and category, model fitting based on historical data can be used for prediction. However, any model will gradually become ineffective over time, requiring model correction. Traditional correction methods typically involve refitting the model with new sample data. This method requires periodic model updates at a fixed frequency, making real-time monitoring impossible. Furthermore, it's important to note that in real-world scenarios, the model's failure cycle is not fixed. This periodic correction approach can easily lead to the model being neglected when correction is needed, while being repeatedly corrected when no correction is required. This application provides an intelligent correction method that can monitor the model's predictive performance in real time and perform real-time corrections based on the degree of improvement using artificial intelligence methods.

[0085] Specifically, in one implementation, for the current time T, time series sample data X = {X} from the past year are collected. T-i+1 Given the dataset X {i = 1, ..., 365}, divide the dataset X into two parts, X1 and X2, where X1 represents the time series data from one year ago to one hundred days ago, i.e., X = {X1, ..., 365}. T-i+1 X1, i = 101, ..., 365; while X2 represents the time series data of the most recent 100 days, i.e., X = {X1, i = 101, ..., 365}; and X2 represents the time series data of the most recent 100 days, i.e., X = {X2, i = 101, ..., 365}. T-i+1 Let the model be a set of elements i = 1, ..., 100. We can use X1 as the training set and X2 as the validation set to refine the model. The specific steps for refinement are as follows:

[0086] (1) The time series model determined according to the time series model determination method in the above embodiment is denoted as Model*. The data of X1 is input into Model* for fitting. After the fitting is completed, the data of the most recent 100 days is predicted, and the first prediction result is compared with the actual data of X2. The mean squared error (MSE) is calculated and the first mean squared error is denoted as MSE0. The MSE is calculated by: square the difference between each prediction result and the actual value, and then averaging the squared terms.

[0087] (2) The second prediction result is obtained by using the optimal time series model determined based on the optimal model selection criterion. For example, the order parameters m and n of the ARMA(m,n) model are traversed from 0 to 5, resulting in a total of 36 models:

[0088] ARMA(0,0),ARMA(0,1),…,ARMA(0,5)

[0089] ARMA(1,0),ARMA(1,1),…,ARMA(1,5)

[0090] ...

[0091] ARMA(5,0),ARMA(5,1),…,ARMA(5,5)

[0092] For each of the 36 models above, the data input to X1 is fitted, and their BIC (Bayesian Information Criterion) values ​​are calculated. The model with the smallest BIC value is selected as the optimal time series model, denoted as Model0. Model0 is then used to predict the data for the most recent 100 days, and the second prediction result is compared with the actual data of X2 to calculate the second mean squared error (MSE), denoted as MSE1.

[0093] Finally, based on the comparison between the first mean square error and the second mean square error, the prediction results of the current time series model for the time series data in future time periods are corrected.

[0094] In one implementation, based on the comparison between the first mean squared error and the second mean squared error, the prediction results of the current time series model for the time series data in future time periods are corrected, including:

[0095] Calculate the difference between the second mean square error and the first mean square error, and calculate the ratio of the difference to the second mean square error. Use the ratio as the deviation value.

[0096] If the deviation value is greater than zero and less than or equal to 1, based on the prediction result of the time series data for the future time period obtained by the current time series model, multiply it by the coefficient {1-[ln(1+R)] 2}, and take the product result as the corrected prediction result, wherein R represents the deviation value.

[0097] In addition, if the deviation value is greater than 1, the optimal time series model is used to predict the time series data for the future time period; if the deviation value is less than or equal to 0, the prediction result of model Model* is directly adopted.

[0098] That is, compare MSE0 and MSE1, record The result of next-day prediction obtained by inputting data set X into model Model* is X T+1 , according to the value of R, the X T+1 is corrected as follows:

[0099]

[0100] wherein, is the final prediction result for T+1, and X 0,T+1 is the prediction result of model Model0.

[0101] The specific correction logic is:

[0102] If R≤0, it proves that MSE0≤MSE1, that is, the model Model* determined by the method of the embodiment of the present application is superior to the model Model0 selected by the BIC traversal method in terms of prediction effect, so it is not necessary to correct the prediction result of Model*;

[0103] If 0<R≤1, it proves that MSE0 is slightly larger than MSE1, and the prediction effect of Model* is slightly worse than that of Model0, so it is necessary to correct Model* according to Model0. The correction method is to perform secondary correction based on the effect deviation R of the two models. The reason why the coefficient of {1-[ln(1+R)] 2} is used for scaling is that model Model0 will also generate errors, and such errors are also unstable. If X is directly (T+1)While using *(1+R) as the correction result is simple in calculation, it may exacerbate the deviation from the actual data results. The range of 1+R is within [1,2]. After the In transformation, it becomes a decimal between [0,1]. Further squaring can flatten the impact of the difference, essentially performing a secondary compression of the correction effect of the difference between the two models on the prediction results. This relatively conservative operation avoids overly aggressive scaling, which could lead to extreme cases of large deviations in the prediction results under certain circumstances. Furthermore, the In transformation and the square transformation are highly malleable curves from a geometric perspective, well-suited to various data point distributions in engineering. Therefore, {1-[In(1+R)]} is adopted. 2} is a very suitable choice as the compression factor;

[0104] If R > 0, it proves that MSE0 > MSE1, and the difference is quite large. Model* performs much worse than Model0 in prediction. In this case, simple fine-tuning is meaningless, and Model0 should be used directly for prediction again. X in the formula 0,T+1 This is the prediction result of Model 0.

[0105] The technical solution of this application embodiment, after determining the current time series model, compares it with the optimal time series model determined based on the optimal model selection criterion. It calculates the mean square error between the prediction results and the actual results of the current time series model and the optimal time series model, respectively. By comparing the two mean square errors, the prediction results are corrected. In the correction, a specific coefficient is used to scale the prediction results of the current time series model. In this way, dynamic and accurate model tuning is achieved. Based on the principle of optimal target benefit, a benefit evaluation index and corresponding intelligent correction rules that are more in line with the actual business scenario are constructed, thereby improving the stability and adaptability of the model.

[0106] Figure 5 This is a flowchart illustrating a method for determining a time series model provided in this application. As shown, the method first assesses the tailing and truncation properties of the autocorrelation coefficient plot and the partial autocorrelation coefficient plot. Based on the assessment results, the model type and order parameter are determined, thus determining the model *. A time series model is selected based on the BIC criterion, determining model *0. The MSE obtained from both is compared, and the prediction results of model * are corrected based on the comparison results to determine the final prediction result of the model. Specifically, this includes the following aspects:

[0107] I. Judgment of trailing and truncation

[0108] 1. When the data is positive and tends to 0 in a gradual downward trend, or when the data is negative and tends to 0 in a gradual downward trend, it can be determined that there is a tailing effect.

[0109] 2. When a positive number suddenly drops rapidly to near 0 for the first time, or a negative number suddenly rises rapidly to near 0 for the first time, it can be determined as truncation. The order corresponding to the point of rapid decline is the truncation order (order parameter).

[0110] Based on the above two criteria, an algorithm is constructed to determine the tailing and truncation of the autocorrelation coefficient and the partial autocorrelation coefficient, respectively.

[0111] II. Model Type and Order Selection

[0112] The model type and order parameters of ARMA(m,n) are determined based on the tailing and truncation properties of the autocorrelation coefficient and the partial autocorrelation coefficient.

[0113] If the partial autocorrelation coefficient plot has a tailing property, then m = 0; if the partial autocorrelation coefficient plot has a truncation property and the truncation order is k, then m = k.

[0114] If the autocorrelation coefficient plot has a tailing property, then n = 0; if the autocorrelation coefficient plot has a truncation property and the truncation order is k, then n = k.

[0115] If there is neither tailing nor truncation, it indicates that the model is not suitable for time series analysis methods, and empirical analysis can be used as a special time series model for predictive analysis.

[0116] III. Intelligent Correction of Model Prediction Results

[0117] After selecting Model*, the optimal time series model, Model0, is chosen based on the BIC criterion. The MSEs of the two models are compared, and the final prediction results of the model are adjusted based on the comparison results.

[0118] Figure 6 This is a schematic diagram of the device for determining a time series model provided in an embodiment of this application. This embodiment is applicable to situations where the model order of a time series model is automatically determined, thereby determining the model type based on the model order. It relates to the field of big data technology, specifically to the fields of big data analysis, statistical modeling, and pattern recognition. Figure 6 As shown, the time series model determining device 60 includes:

[0119] The correlation coefficient plot acquisition module 601 is used to acquire the autocorrelation coefficient plot and partial autocorrelation coefficient plot of time series data;

[0120] The feature recognition module 602 is used to identify the tailing features and truncation features of the autocorrelation coefficient graph and the partial autocorrelation coefficient graph, respectively.

[0121] The model determination module 603 is used to determine the model order of the time series model based on the tailing feature and the truncation feature, wherein the time series model is used to predict the data of the time series data in future time periods.

[0122] In some embodiments, the feature recognition module 602 includes a trailing feature recognition submodule, the trailing feature recognition submodule comprising:

[0123] The difference calculation unit is used to perform difference calculations on the correlation coefficients of adjacent orders in the autocorrelation coefficient graph and the partial autocorrelation coefficient graph respectively, to obtain the first-order difference sequence of autocorrelation coefficients and the first-order difference sequence of partial autocorrelation coefficients.

[0124] The tailing feature recognition unit is used to identify the tailing features of the autocorrelation coefficient graph and the partial autocorrelation coefficient graph respectively based on the first-order difference sequence of the autocorrelation coefficient and the first-order difference sequence of the partial autocorrelation coefficient.

[0125] In some embodiments, the trailing feature recognition unit is specifically used for:

[0126] In response to each difference value in the first-order difference sequence of the autocorrelation coefficients satisfying a first condition, the autocorrelation coefficient plot is identified as having a tailing characteristic; and

[0127] If each difference value in the first difference sequence of the partial autocorrelation coefficients satisfies the first condition, then the partial autocorrelation coefficient plot is identified as having a tailing feature.

[0128] The first condition includes:

[0129] The first correlation coefficient in the autocorrelation coefficient plot or partial autocorrelation coefficient plot is greater than the correlation threshold; and

[0130] Each difference value belongs to a first numerical range, and the proportion of difference values ​​belonging to a second numerical range does not exceed a first percentage, wherein the first numerical range includes the second numerical range; and

[0131] Satisfies the first formula: Where, d i d j Let i and j represent the difference values, where i and j are natural numbers.

[0132] In some embodiments, the trailing feature recognition unit is further configured to:

[0133] In response to each difference value in the first-order difference sequence of the autocorrelation coefficients satisfying the second condition, the autocorrelation coefficient plot is identified as having a tailing characteristic; and

[0134] If each difference value in the first-order difference sequence of the partial autocorrelation coefficient satisfies the second condition, then the partial autocorrelation coefficient plot is identified as having a tailing feature.

[0135] The second condition includes:

[0136] The first correlation coefficient in the autocorrelation coefficient plot or partial autocorrelation coefficient plot is less than a negative value of the correlation threshold; and

[0137] All the difference values ​​belong to the third numerical range, and the proportion of difference values ​​belonging to the fourth numerical range does not exceed the second percentage, wherein the third numerical range includes the fourth numerical range; and

[0138] Satisfies the second formula: Where, d i d j Let i and j represent the difference values, where i and j are natural numbers.

[0139] In some embodiments, the feature recognition module 602 includes a truncated feature recognition submodule, the truncated feature recognition submodule comprising:

[0140] The cutoff point determination unit is used to determine whether there is at least one cutoff point in the autocorrelation coefficient graph or partial autocorrelation coefficient graph, wherein the cutoff point indicates that the correlation coefficients of other orders after the cutoff point do not exceed the fifth numerical range.

[0141] The target cutoff point selection unit is used to select the correlation coefficient with the smallest order from the at least one cutoff point as the target cutoff point if the at least one cutoff point exists.

[0142] The true cutoff point determination unit is used to determine whether the target cutoff point is a true cutoff point. If it is a true cutoff point, it is identified that the autocorrelation coefficient plot or partial autocorrelation coefficient plot has a truncated feature. The true cutoff point means that the correlation coefficients of other orders before the true cutoff point are all in the strong correlation region.

[0143] In some embodiments, the true cutoff point determination unit includes:

[0144] The first judgment subunit is configured to, if the true cutoff point is greater than zero, determine whether the correlation coefficients of other orders before the true cutoff point are all greater than a judgment threshold; if so, determine that the target cutoff point is the true cutoff point; or

[0145] The second judgment subunit is used to determine whether the correlation coefficients of other orders before the true cutoff point are all less than the negative value of the judgment threshold if the true cutoff point is less than zero. If so, the target cutoff point is determined to be the true cutoff point.

[0146] In some embodiments, the feature recognition module 602 is specifically used for:

[0147] Identify the tailing characteristics of the autocorrelation coefficient plot and the partial autocorrelation coefficient plot;

[0148] In response to the autocorrelation coefficient plot or partial autocorrelation coefficient plot not having a tailing feature, the truncation feature of the autocorrelation coefficient plot or partial autocorrelation coefficient plot is identified.

[0149] In some embodiments, the time series model is an ARMA model, which has a first-order parameter and a second-order parameter.

[0150] In some embodiments, the model determination module 603 is specifically used for:

[0151] If the partial autocorrelation coefficient plot has a tailing feature, then the first order parameter is 0; if the partial autocorrelation coefficient plot has a truncation feature, then the first order parameter is the target cutoff point of the partial autocorrelation coefficient plot.

[0152] If the autocorrelation coefficient plot has a tailing characteristic, then the second order parameter is 0; if the autocorrelation coefficient plot has a truncation characteristic, then the second order parameter is the target cutoff point of the autocorrelation coefficient plot.

[0153] In some embodiments, the apparatus further includes a correction module, the correction module comprising:

[0154] A determining unit is used to take the time series model determined according to the model order as the current time series model;

[0155] The training set acquisition unit is used to acquire training and validation sets from time series sample data;

[0156] The first prediction result acquisition unit is used to fit the current time series model with the training set, and to make a prediction using the fitted current time series model to obtain the first prediction result.

[0157] The first calculation unit is used to calculate the first mean square error based on the first prediction result and the validation set.

[0158] The second prediction result acquisition unit is used to make predictions using the optimal time series model determined based on the optimal model selection criterion, and obtain the second prediction result.

[0159] The second calculation unit is used to calculate the second mean square error based on the second prediction result and the validation set;

[0160] The correction unit is used to correct the prediction results of the current time series model for the time series data in future time periods based on the comparison results of the first mean square error and the second mean square error.

[0161] In some embodiments, the correction unit includes:

[0162] The deviation value acquisition subunit is used to calculate the difference between the second mean square error and the first mean square error, and to calculate the ratio of the difference to the second mean square error, and use the ratio as the deviation value.

[0163] The first correction subunit is configured to, if the deviation value is greater than zero and less than or equal to 1, adjust the current time series model's prediction of the time series data for future time periods based on the coefficient {1-[In(1+R)]}. 2 The values ​​are multiplied, and the result of the multiplication is used as the corrected prediction result, where R represents the deviation value.

[0164] In some embodiments, the correction unit further includes:

[0165] The second correction subunit is used to predict the data of the time series data in future time periods using the optimal time series model if the deviation value is greater than 1.

[0166] In some embodiments, the time-series data consists of bank amount data from different sources.

[0167] The time series model determination apparatus provided in this application embodiment can be used to execute the technical solution of the time series model determination method in the above embodiment. Its implementation principle and technical effect are similar, and will not be described again here.

[0168] It should be noted that the division of the various modules in the above device is merely a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, these modules can be implemented entirely in software via processing element calls; they can be fully implemented in hardware; or some modules can be implemented in software via processing element calls, while others are implemented in hardware. For example, the feature recognition module 602 can be a separate processing element, or it can be integrated into a chip in the above device. Alternatively, it can be stored as program code in the memory of the above device, and its functions can be called and executed by a processing element of the device. The implementation of other modules is similar. Moreover, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each of the above modules can be completed through integrated logic circuits in the hardware of the processor element or through software instructions.

[0169] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution of this application all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0170] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 7 As shown, the electronic device may include: a transceiver 721, a processor 722, and a memory 723.

[0171] Processor 722 executes computer execution instructions stored in memory, causing processor 722 to perform the scheme in the above embodiments. Processor 722 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0172] The memory 723 is connected to the processor 722 via the system bus and completes communication between them. The memory 723 is used to store computer program instructions.

[0173] The transceiver 721 can be used to obtain the task to be run and its configuration information.

[0174] The system bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The system bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus. Transceivers are used to enable communication between database access devices and other computers (e.g., clients, read-write libraries, and read-only libraries). Memory may include random access memory (RAM) and may also include non-volatile memory.

[0175] The electronic device provided in this application embodiment can be the terminal device described in the above embodiments.

[0176] This application also provides a computer-readable storage medium storing computer instructions that, when executed on a computer, cause the computer to perform the method for determining the time series model described in the above embodiments.

[0177] This application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium. When the at least one processor executes the computer program, it can implement the technical solution of the time series model determination method in the above embodiments.

[0178] In the implementation of the computer program product, computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0179] Note that the above description is merely a preferred embodiment and the technical principles employed in this application. Those skilled in the art will understand that this application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments. Many other equivalent embodiments may be included without departing from the concept of this application, and the scope of this application is determined by the scope of the appended claims.

Claims

1. A method for determining a time series model, characterized in that, include: Obtain the autocorrelation coefficient plot and partial autocorrelation coefficient plot of time series data, which consists of bank amount data from different sources; Identify the tailing characteristics of the autocorrelation coefficient plot and the partial autocorrelation coefficient plot; In response to the autocorrelation coefficient plot or partial autocorrelation coefficient plot not having a tailing feature, the truncation feature of the autocorrelation coefficient plot or partial autocorrelation coefficient plot is identified. Among them, the trailing feature refers to the data being positive and trending towards 0 in a gradual downward direction, or the data being negative and trending towards 0 in a gradual upward direction. The truncation feature refers to the data being positive for the first time suddenly dropping rapidly to near 0, or the data being negative for the first time suddenly rising rapidly to near 0. The process of identifying the tailing features of the autocorrelation coefficient plot and the partial autocorrelation coefficient plot includes: The correlation coefficients of adjacent orders in the autocorrelation coefficient plot and the partial autocorrelation coefficient plot are calculated by difference to obtain the first-order difference sequence of autocorrelation coefficient and the first-order difference sequence of partial autocorrelation coefficient; Based on the first-order difference sequence of the autocorrelation coefficient and the first-order difference sequence of the partial autocorrelation coefficient, the tailing characteristics of the autocorrelation coefficient plot and the partial autocorrelation coefficient plot are identified respectively. The identification of the truncation features of the autocorrelation coefficient plot and the partial autocorrelation coefficient plot includes: Determine whether there is at least one cutoff point in the autocorrelation coefficient plot or partial autocorrelation coefficient plot, wherein the cutoff point indicates that the correlation coefficients of other orders after the cutoff point do not exceed the fifth numerical range; If at least one cutoff point exists, the correlation coefficient with the smallest order among the at least one cutoff point is selected as the target cutoff point; Determine whether the target cutoff point is a true cutoff point. If it is a true cutoff point, then the autocorrelation coefficient plot or partial autocorrelation coefficient plot is identified as having a truncated tail feature. The true cutoff point indicates that the correlation coefficients of other orders before the true cutoff point are all in the strong correlation region. The model order of the time series model is determined based on the tailing and truncation features, wherein the time series model is used to predict the data of the time series data in future time periods.

2. The method according to claim 1, characterized in that, The step of identifying the tailing features of the autocorrelation coefficient plot and the partial autocorrelation coefficient plot based on the first-order difference sequence of the autocorrelation coefficient and the first-order difference sequence of the partial autocorrelation coefficient, respectively, includes: In response to each difference value in the first-order difference sequence of the autocorrelation coefficients satisfying a first condition, the autocorrelation coefficient plot is identified as having a tailing characteristic; and If each difference value in the first difference sequence of the partial autocorrelation coefficients satisfies the first condition, then the partial autocorrelation coefficient plot is identified as having a tailing feature. The first condition includes: The first correlation coefficient in the autocorrelation coefficient plot or partial autocorrelation coefficient plot is greater than the correlation threshold; and Each difference value belongs to a first numerical range, and the proportion of difference values ​​belonging to a second numerical range does not exceed a first percentage, wherein the first numerical range includes the second numerical range; and Satisfies the first formula: , where d i d j Let i and j represent the difference values, where i and j are natural numbers.

3. The method according to claim 1, characterized in that, The step of identifying the tailing features of the autocorrelation coefficient plot and the partial autocorrelation coefficient plot based on the first-order difference sequence of the autocorrelation coefficient and the first-order difference sequence of the partial autocorrelation coefficient, respectively, includes: In response to each difference value in the first-order difference sequence of the autocorrelation coefficients satisfying the second condition, the autocorrelation coefficient plot is identified as having a tailing characteristic; and If each difference value in the first-order difference sequence of the partial autocorrelation coefficient satisfies the second condition, then the partial autocorrelation coefficient plot is identified as having a tailing feature. The second condition includes: The first correlation coefficient in the autocorrelation coefficient plot or partial autocorrelation coefficient plot is less than a negative value of the correlation threshold; and All the difference values ​​belong to the third numerical range, and the proportion of difference values ​​belonging to the fourth numerical range does not exceed the second percentage, wherein the third numerical range includes the fourth numerical range; and Satisfies the second formula: , where d i d j Let i and j represent the difference values, where i and j are natural numbers.

4. The method according to claim 1, characterized in that, The step of determining whether the target cutoff point is a true cutoff point includes: If the true cutoff point is greater than zero, then it is determined whether the correlation coefficients of other orders before the true cutoff point are all greater than a judgment threshold. If so, the target cutoff point is determined to be the true cutoff point; or If the true cutoff point is less than zero, then it is determined whether the correlation coefficients of other orders before the true cutoff point are all less than the negative value of the judgment threshold. If so, the target cutoff point is determined to be the true cutoff point.

5. The method according to claim 1, characterized in that, The time series model is an ARMA model, which has a first-order parameter and a second-order parameter.

6. The method according to claim 5, characterized in that, The step of determining the model order of the time series model based on the tailing and truncation features includes: If the partial autocorrelation coefficient plot has a tailing feature, then the first order parameter is 0; if the partial autocorrelation coefficient plot has a truncation feature, then the first order parameter is the target cutoff point of the partial autocorrelation coefficient plot. If the autocorrelation coefficient plot has a tailing characteristic, then the second order parameter is 0; if the autocorrelation coefficient plot has a truncation characteristic, then the second order parameter is the target cutoff point of the autocorrelation coefficient plot.

7. The method according to claim 1, characterized in that, The method further includes: The time series model determined according to the model order is used as the current time series model; Obtain training and validation sets from time series sample data; The training set is used to fit the current time series model, and the fitted current time series model is used to make a prediction to obtain a first prediction result; Calculate the first mean square error based on the first prediction result and the validation set; The second prediction result is obtained by using the optimal time series model determined based on the optimal model selection criterion. Calculate the second mean square error based on the second prediction result and the validation set; Based on the comparison results of the first mean square error and the second mean square error, the prediction results of the current time series model for the time series data in future time periods are corrected.

8. The method according to claim 7, characterized in that, The step of correcting the prediction results of the current time series model for the time series data in future time periods based on the comparison results of the first mean square error and the second mean square error includes: Calculate the difference between the second mean square error and the first mean square error, and calculate the ratio of the difference to the second mean square error, and use the ratio as the deviation value; If the deviation value is greater than zero and less than or equal to 1, then based on the current time series model's prediction result for the time series data in future time periods, and with the coefficient... Multiply the results and use the result of the multiplication as the corrected prediction, where R represents the deviation value.

9. The method according to claim 8, characterized in that, The step of correcting the prediction results of the current time series model for the time series data in future time periods based on the comparison results of the first mean square error and the second mean square error further includes: If the deviation value is greater than 1, the optimal time series model is used to predict the data of the time series data in future time periods.

10. A device for determining a time series model, characterized in that, include: The correlation coefficient plot acquisition module is used to acquire the autocorrelation coefficient plot and partial autocorrelation coefficient plot of time series data; the time series data consists of bank amount data from different sources; The feature recognition module is used to identify the tailing features and truncation features of the autocorrelation coefficient plot and the partial autocorrelation coefficient plot, respectively. Among them, the trailing feature refers to the data being positive and trending towards 0 in a gradual downward direction, or the data being negative and trending towards 0 in a gradual upward direction. The truncation feature refers to the data being positive for the first time suddenly dropping rapidly to near 0, or the data being negative for the first time suddenly rising rapidly to near 0. The feature recognition module is specifically used for: Identify the tailing characteristics of the autocorrelation coefficient plot and the partial autocorrelation coefficient plot; In response to the autocorrelation coefficient plot or partial autocorrelation coefficient plot not having a tailing feature, the truncation feature of the autocorrelation coefficient plot or partial autocorrelation coefficient plot is identified. The feature recognition module includes a trailing feature recognition submodule, which includes: The difference calculation unit is used to perform difference calculations on the correlation coefficients of adjacent orders in the autocorrelation coefficient graph and the partial autocorrelation coefficient graph respectively, to obtain the first-order difference sequence of autocorrelation coefficients and the first-order difference sequence of partial autocorrelation coefficients. The tail feature recognition unit is used to identify the tail features of the autocorrelation coefficient graph and the partial autocorrelation coefficient graph respectively based on the first-order difference sequence of the autocorrelation coefficient and the first-order difference sequence of the partial autocorrelation coefficient. The feature recognition module includes a truncated feature recognition submodule, which includes: The cutoff point determination unit is used to determine whether there is at least one cutoff point in the autocorrelation coefficient graph or partial autocorrelation coefficient graph, wherein the cutoff point indicates that the correlation coefficients of other orders after the cutoff point do not exceed the fifth numerical range. The target cutoff point selection unit is used to select the correlation coefficient with the smallest order from the at least one cutoff point as the target cutoff point if the at least one cutoff point exists. The true cutoff point determination unit is used to determine whether the target cutoff point is a true cutoff point. If it is a true cutoff point, it is identified that the autocorrelation coefficient plot or partial autocorrelation coefficient plot has a truncated feature. The true cutoff point means that the correlation coefficients of other orders before the true cutoff point are all in the strong correlation region. The model determination module is used to determine the model order of the time series model based on the tailing feature and the truncation feature, wherein the time series model is used to predict the data of the time series data in future time periods.

11. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method for determining a time series model as described in any one of claims 1-9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method for determining a time series model as described in any one of claims 1-9.

13. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the method for determining the time series model according to any one of claims 1-9.

Images