Method and apparatus for predicting number of terminals, computer device, and readable storage medium

Abstract

The present application relates to a method and apparatus for predicting the number of terminals, a computer device, and a readable storage medium. The method comprises: on the basis of the historical number of terminals of a target area corresponding to each historical moment within a historical time period, constructing terminal number regression models corresponding to different candidate time windows, each candidate time window being intercepted from the historical time period; on the basis of the model precision of each terminal number regression model, selecting a target time window from the candidate time windows; analyzing the historical number of terminals corresponding to each historical moment within the target time window, so as to obtain a terminal number feature within the target time window; and inputting the terminal number feature within the target time window and the historical number of terminals corresponding to each historical moment into a terminal number prediction model, so as to obtain a predicted number of terminals of the target region at each future moment within a future time period.

Description

Terminal quantity prediction method and device, computer device, and readable storage medium

[0001] Cross-reference to Related Applications

[0002] This application claims priority to the Chinese patent application No. CN2024112508047, filed on September 6, 2024, and entitled “Terminal quantity prediction method, device, computer device, and readable storage medium”, the content of which is incorporated herein by reference in its entirety. TECHNICAL FIELD

[0003] The present application relates to the field of communication technology, and in particular, to a terminal quantity prediction method, device, computer device, and readable storage medium. BACKGROUND

[0004] Wireless terminal users refer to individuals who subscribe to wireless mobile communication services provided by a specific operator. They use various wireless terminal devices to access a range of services, such as voice communication, messaging, and data transmission. These terminal devices include, but are not limited to, smartphones, feature phones, tablets, laptops, and similar portable small devices. These devices provide different functions and features to meet the individual needs and preferences of wireless terminal users. For operators, analyzing the number of wireless terminal users has great value. SUMMARY

[0005] The present application provides a terminal quantity prediction method, device, computer device, and readable storage medium that can improve the accuracy of terminal quantity prediction.

[0006] In a first aspect, the present application provides a terminal quantity prediction method, comprising:

[0007] constructing terminal quantity regression models corresponding to different candidate time windows according to historical terminal quantities of the target area at historical time points in a historical time period; wherein each candidate time window is obtained by cutting from the historical time period;

[0008] selecting a target time window from the candidate time windows according to the model accuracy of each terminal quantity regression model;

[0009] analyzing the historical terminal quantities of the target time window at the historical time points to obtain terminal quantity features in the target time window;

[0010] inputting the terminal quantity features in the target time window and the historical terminal quantities corresponding to the historical time points into a terminal quantity prediction model to obtain predicted terminal quantities of the target area at future time points in a future time period.

[0011] In one of the embodiments, the target time window is selected from the candidate time windows according to the model accuracy of the terminal quantity regression model, comprising:

[0012] For any terminal quantity regression model, the model accuracy of the terminal quantity regression model is determined according to the Akaike information criterion AIC, Bayesian information criterion BIC and Hannan-Quinn information criterion HQIC corresponding to the terminal quantity regression model;

[0013] The terminal quantity regression model with the highest model accuracy in the terminal quantity regression models is taken as the target regression model;

[0014] The candidate time window corresponding to the target regression model in the candidate time windows is taken as the target time window.

[0015] In one of the embodiments, the terminal quantity characteristics in the target time window are obtained by analyzing the historical terminal quantities corresponding to the historical time points in the target time window, comprising:

[0016] The feature quantity of the terminal quantity characteristics in the target time window is determined according to the window length of the target time window;

[0017] The feature quantity is taken as the row number of the matrix, and the window length is taken as the column number of the matrix to construct an initial matrix;

[0018] The initial matrix is optimized according to the historical terminal quantities corresponding to the historical time points in the target time window to obtain an optimized matrix;

[0019] The terminal quantity characteristics in the target time window are determined according to the optimized matrix and the historical terminal quantities corresponding to the historical time points in the target time window.

[0020] In one of the embodiments, the initial matrix is optimized according to the historical terminal quantities corresponding to the historical time points in the target time window to obtain an optimized matrix, comprising:

[0021] The historical time period is divided into a plurality of adjacent observation sub-periods by using the target time window; the period length of the observation sub-period is the same as the length of the target time window;

[0022] For each observation sub-period, the historical terminal quantities corresponding to the historical time points in the observation sub-period and the historical terminal quantity corresponding to the next historical time point corresponding to the observation sub-period are taken as observation samples; wherein the next historical time point corresponding to the observation sub-period is later than the historical time points in the observation sub-period;

[0023] For each observation sample, the sample neighbor probability of the observation sample is determined by using the initial matrix;

[0024] The initial matrix is optimized to obtain an optimized matrix, aiming at maximizing the sum of sample neighbor probabilities of each observation sample.

[0025] In one of the embodiments, the sample neighbor probability of the observation sample is determined by using the initial matrix, including:

[0026] The number of historical terminals corresponding to the next historical time of the observation sub-period in each observation sample is taken as the target value of each observation sample.

[0027] The other observation samples with the same target value as the observation sample are determined as the neighbor samples of the observation sample.

[0028] The single neighbor probability between the observation sample and the neighbor sample is determined by using the initial matrix for each neighbor sample of the observation sample.

[0029] The sum of the single neighbor probabilities corresponding to each neighbor sample is taken as the sample neighbor probability of the observation sample.

[0030] In one of the embodiments, the single neighbor probability between the observation sample and the neighbor sample is determined by using the initial matrix, including:

[0031] The first distance measure between the observation sample and the neighbor sample is determined by using the initial matrix.

[0032] The second distance measure between the observation sample and the other observation sample is determined by using the initial matrix.

[0033] The sum of the second distance measures is determined as the measure sum value.

[0034] The first distance measure is divided by the measure sum value to obtain the single neighbor probability between the observation sample and the neighbor sample.

[0035] In one of the embodiments, the number of historical terminals corresponding to the next historical time of the observation sub-period in each observation sample is taken as the target value of each observation sample, including:

[0036] The number of historical terminals corresponding to the next historical time of the observation sub-period in each observation sample is taken as the target value of each observation sample, including:

[0037] In one of the embodiments, the method further includes: taking the number of historical terminals corresponding to each historical time in the observation sub-period in each observation sample as the sample feature of each observation sample; and wherein:

[0038] The first distance measure between the observation sample and the neighbor sample is determined by using the initial matrix, including: the first distance measure between the sample feature of the observation sample and the sample feature of the neighbor sample of the observation sample is determined by using the initial matrix.

[0039] The second distance measure between the observation sample and other observation samples is determined using the initial matrix, including: the second distance measure between the sample feature of the observation sample and the sample feature of other observation samples is determined using the initial matrix.

[0040] In one of the embodiments, the second distance measure between the observation sample and other observation samples is determined using the initial matrix, including:

[0041] The historical terminal quantity corresponding to each historical moment in the observation sub-period in each observation sample is taken as the sample feature of each observation sample;

[0042] The second distance measure between the sample feature of the observation sample and the sample feature of other observation samples is determined using the initial matrix.

[0043] In one of the embodiments, the historical terminal quantity corresponding to the historical moment of the next moment in the observation sub-period is taken as the target value corresponding to the observation sample, including:

[0044] The historical terminal quantity corresponding to the historical moment of the next moment in the observation sub-period is rounded, and the rounded value is taken as the target value corresponding to the observation sample.

[0045] In one of the embodiments, the terminal quantity feature in the target time window is determined according to the optimization matrix and the historical terminal quantity corresponding to each historical moment in the target time window, including:

[0046] The product between the optimization matrix and the historical terminal quantity corresponding to each historical moment in the target time window is taken as the initial quantity feature in the target time window;

[0047] An interpretative index of the initial quantity feature in the target time window is determined;

[0048] The initial quantity feature satisfying the interpretative threshold of the interpretative index is taken as the terminal quantity feature in the target time window.

[0049] In one of the embodiments, the length of the future period is the same as the window length of the target time window.

[0050] In one of the embodiments, the length of the observation sub-period is the same as the length of the target time window.

[0051] In a second aspect, the application further provides a terminal quantity prediction device, including:

[0052] A model construction module is configured to construct a terminal quantity regression model corresponding to different candidate time windows according to the historical terminal quantity corresponding to each historical moment in a historical time period in a target region; each candidate time window is obtained by cutting from the historical time period.

[0053] a window selection module, configured to select a target time window from the candidate time windows according to model accuracy of the terminal quantity regression models;

[0054] an analysis module, configured to analyze the historical terminal quantities corresponding to the historical time points in the target time window to obtain terminal quantity features in the target time window;

[0055] a prediction module, configured to input the terminal quantity features in the target time window and the historical terminal quantities corresponding to the historical time points into the terminal quantity prediction model to obtain predicted terminal quantities of the target region at future time points in a future time period.

[0056] In a third aspect, the present application also provides a computer device, comprising a memory and a processor, the memory storing a computer program, and the processor implementing the following steps when executing the computer program:

[0057] constructing terminal quantity regression models corresponding to different candidate time windows according to historical terminal quantities of the target region at historical time points in a historical time period, wherein each candidate time window is obtained by cutting from the historical time period;

[0058] selecting a target time window from the candidate time windows according to model accuracy of the terminal quantity regression models;

[0059] analyzing the historical terminal quantities corresponding to the historical time points in the target time window to obtain terminal quantity features in the target time window;

[0060] inputting the terminal quantity features in the target time window and the historical terminal quantities corresponding to the historical time points into the terminal quantity prediction model to obtain predicted terminal quantities of the target region at future time points in a future time period.

[0061] In a fourth aspect, the present application also provides a computer readable storage medium, storing a computer program, and the computer program being executed by a processor to implement the following steps:

[0062] constructing terminal quantity regression models corresponding to different candidate time windows according to historical terminal quantities of the target region at historical time points in a historical time period, wherein each candidate time window is obtained by cutting from the historical time period;

[0063] selecting a target time window from the candidate time windows according to model accuracy of the terminal quantity regression models;

[0064] analyzing the historical terminal quantities corresponding to the historical time points in the target time window to obtain terminal quantity features in the target time window;

[0065] The terminal quantity feature in the target time window and the historical terminal quantity corresponding to each historical time are input into a terminal quantity prediction model to obtain the predicted terminal quantity of the target region at each future time in a future period.

[0066] In a fifth aspect, the present application provides a computer program product comprising a computer program which, when executed by a processor, implements the following steps:

[0067] According to the historical terminal quantity of the target region corresponding to each historical time in a historical period, a terminal quantity regression model corresponding to each candidate time window is constructed, wherein each candidate time window is obtained by cutting from the historical period;

[0068] According to the model accuracy of each terminal quantity regression model, a target time window is selected from each candidate time window;

[0069] The historical terminal quantity corresponding to each historical time in the target time window is analyzed to obtain a terminal quantity feature in the target time window;

[0070] The terminal quantity feature in the target time window and the historical terminal quantity corresponding to each historical time are input into a terminal quantity prediction model to obtain the predicted terminal quantity of the target region at each future time in a future period. BRIEF DESCRIPTION OF DRAWINGS

[0071] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the description of the embodiments of the present application or the related art will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0072] FIG. 1 is a flowchart of a terminal quantity prediction method in an embodiment;

[0073] FIG. 2 is a flowchart of a step of selecting a candidate time window corresponding to a target regression model from each candidate time window in an embodiment;

[0074] FIG. 3 is a flowchart of a step of optimizing an initial matrix in an embodiment;

[0075] FIG. 4 is a flowchart of a step of determining the sample neighborhood probability of an observation sample in an embodiment;

[0076] FIG. 5 is a flowchart of a step of determining the single neighborhood probability between an observation sample and a neighborhood sample in an embodiment;

[0077] FIG. 6 is a flowchart of a step of obtaining a single neighbor probability between an observation sample and a neighbor sample in an embodiment;

[0078] FIG. 7 is a structural block diagram of a terminal quantity prediction device in an embodiment;

[0079] FIG. 8 is an internal structural diagram of a computer device in an embodiment. DETAILED DESCRIPTION

[0080] For the purposes of the present application, the technical solutions and advantages thereof are more clearly apparent, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely intended to explain the present application, and are not intended to limit the present application.

[0081] For the operators, analyzing the number of wireless terminal users has great value. In addition to directly affecting the optimization of the network, improving the performance and efficiency of the network, and providing more stable, high-speed, and reliable communication services for users, it can further help to generate revenue, improve market competitiveness, and better grasp global market opportunities and strategic planning by using the brand value and industry reputation of the operator.

[0082] In the traditional technology, the number of wireless terminals is predicted by questionnaires, big data analysis, and social network analysis, but the above methods still have inaccuracy.

[0083] Based on this, the purpose of the present application is to provide a terminal quantity prediction method, device, computer device, and readable storage medium capable of improving the accuracy of terminal quantity prediction.

[0084] The embodiment provides a terminal quantity prediction method, which comprises steps 101 to 104, and the specific process is as follows:

[0085] Step 101, constructing a terminal quantity regression model corresponding to different candidate time windows according to the historical terminal quantity of the target area at each historical time in a historical time period.

[0086] Specifically, the number of wireless terminal users is obtained from the operator network element system or device (user management system, scheduling and monitoring system, base station, etc.), communication industry reports, and other data sources. According to the specific data collection situation and application purpose, the time interval can be flexibly set (such as day, week, month, and year), and the data collection area scale can be specifically and accurately set as street, district, city, province, country, and continent.

[0087] Each candidate time window is obtained from the historical time period. These time windows can be fixed (such as one hour, one day, one week, etc.), or they can be variable (such as gradually increasing time windows).

[0088] For example, starting from the beginning of the historical time period, data is cut according to a predetermined time window length. For example, if the length of the selected time window is one day, the first time window is from the first day of the historical time period to the end of the first day, the second time window is from the second day to the end of the second day, and so on. If it is necessary to consider overlapping time windows, starting from the beginning of the historical time period, a smaller time unit (such as one hour, one day, etc.) is moved each time, and then a time period with the same length as the predetermined length is cut as a new time window.

[0089] For each cut time window, the historical terminal quantity data in the window is used to construct a regression model. In this way, each time window has a corresponding regression model. These time windows will be used as input features of the regression model. For each candidate time window, the corresponding historical terminal quantity data is used as input and output to construct a regression model. Various regression algorithms can be selected, such as linear regression, decision tree regression, random forest regression, etc., and the most suitable algorithm is selected according to the data characteristics and business requirements.

[0090] For example, Y(t) represents the number of wireless terminal users at time t, the length of the time window is set to p as a to-be-determined parameter, and the upper limit is set to N. Take p = 1, …, N, and construct a regression model of Y(t) about Y(t-1)…Y(t-p):

[0091] Wherein, the value of the coefficient a j is obtained by least squares or MATLAB tools.

[0092] Step 102, according to the model accuracy of each terminal quantity regression model, select the target time window from each candidate time window.

[0093] Optionally, for the terminal quantity regression model constructed for each candidate time window in step 101, the prediction accuracy is evaluated using indicators such as mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), etc.

[0094] Compare the accuracy indicators of the models corresponding to each candidate time window, and select the time window corresponding to the model with the highest accuracy (i.e. the smallest error) as the target time window. This time window is considered to best reflect the trend of terminal quantity changes, and the model has the best fitting effect on the data in this time window.

[0095] Step 103, analyze the historical terminal quantity corresponding to each historical time in the target time window to obtain the terminal quantity characteristics in the target time window.

[0096] Specifically, the terminal quantity data in the target time window is extracted from the collected historical data.

[0097] Then, in-depth analysis is performed on these data to extract features related to the changes in the number of terminals. These features can include periodic changes (such as daily changes, weekly changes, etc.), trend changes (such as long-term growth or reduction trends), and other factors that can affect the number of terminals (such as holiday effects, special events, etc.). The features obtained from these analyses are sorted and summarized to provide input for subsequent prediction models.

[0098] Step 104, input the terminal number features in the target time window and the historical terminal number corresponding to each historical time into the terminal number prediction model to obtain the predicted terminal number of each future time in the future period in the target area.

[0099] Wherein, the period length of the future period is the same as the window length of the target time window.

[0100] The terminal number features in the target time window and the corresponding historical terminal number data are used as input data sets.

[0101] A suitable prediction model (such as a time series prediction model, a machine learning regression model, etc.) is selected and trained using the input data set. This model will be used to predict the number of terminals in the future period. Using the trained model, input the time features of the future period (such as future dates, times, etc.), and other external factors that may affect the number of terminals (such as weather forecasts, special event announcements, etc.) to obtain the predicted number of terminals at each future time in the future period.

[0102] The above terminal number pre-storage method constructs a regression model based on historical data, which can more objectively reflect the trend of the number of terminals; by constructing regression models for different candidate time windows, this process can adapt to the analysis needs of different time scales. This flexibility enables the method to handle a variety of scenarios, whether short-term prediction or long-term trend analysis. Selecting the target time window according to the model accuracy ensures that the data set used for subsequent analysis and prediction is the most representative. Focusing on analyzing the terminal number features in the target time window helps to understand the key factors that affect the number of terminals. Integrating historical data and extracted features; further, using the prediction model to accurately predict the number of terminals in the future period. Since the entire process is based on regression analysis, the results of the model have strong interpretability.

[0103] In an exemplary embodiment, as shown in FIG. 2, according to the model accuracy of each terminal number regression model, a target time window is selected from each candidate time window, including:

[0104] Step 201, for any terminal quantity regression model, according to the Akaike Information Criterion AIC, Bayesian Information Criterion BIC and Hannan-Quinn Information Criterion HQIC corresponding to the terminal quantity regression model, the model precision of the terminal quantity regression model is determined.

[0105] Wherein, AIC (Akaike Information Criterion): used to measure the trade-off between the goodness of fit and the complexity of the statistical model. The smaller the AIC value, the better the model, that is, the more concise the model while fitting the data.

[0106] BIC (Bayesian Information Criterion): also considers the trade-off between the goodness of fit and the complexity of the model, but the penalty for model complexity in the calculation is more strict. The smaller the BIC value, the better the model, and unlike AIC, the penalty term for model complexity is larger.

[0107] HQIC (Hannan-Quinn Information Criterion): similar to AIC and BIC, but the penalty for model complexity in the calculation is slightly different. The smaller the HQIC value, the better the model, and in addition, the penalty term for model complexity is slightly smaller than BIC.

[0108] Three information criteria Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and Hannan-Quinn Information Criterion (HQIC), their calculation principles are shown in formulas (1)~(3) AIC = -2log(L) + 2k (1) BIC = -2log(L) + k logn (2) HQIC = -2log(L) + 2k log(logn) (3)

[0109] Where L represents the maximum likelihood function, used to evaluate the goodness of fit between the model and the observed data, k represents the number of parameters in the model, and n represents the total number of observation samples in the data set. The smaller the value of the three information criteria, the better the fitting effect. In specific applications, built-in functions of Python statsmodels library, R language MASS library or stats library, etc. can be used for calculation.

[0110] Step 202, the terminal quantity regression model with the highest model accuracy is taken as the target regression model.

[0111] From step 201, three kinds of information criteria are obtained when the time window length p is 1, 2, …, N. The three kinds of information criteria are normalized to the interval [0, 1] by using the normalization formula: Z' = (Z-Z min ) / (Z max -Z min ), and the dimension effect is eliminated. Then the normalized information criterion values of each p are added up to obtain the integrated information criterion, denoted as Integrated_IC. The p corresponding to the lowest Integrated_IC value is taken as the optimal time window length, denoted as d.

[0112] Step 203, the candidate time window corresponding to the target regression model in each candidate time window is taken as the target time window.

[0113] It can be understood that a specific time window is determined, and the data in this time window is the most valuable for predicting the terminal quantity. In subsequent analysis and prediction, the data in the target time window is focused on, and how to use these data to improve the prediction accuracy.

[0114] In an exemplary embodiment, as shown in FIG. 3, the historical terminal quantities corresponding to each historical time in the target time window are analyzed to obtain the terminal quantity characteristics in the target time window, including:

[0115] Step 301, the feature quantity of the terminal quantity characteristics in the target time window is determined according to the window length of the target time window.

[0116] Specifically, Y(t-1), Y(t-2), … Y(t-d) are taken as the original feature variables of the wireless terminal users, and the original feature variables are used to generate neighbor component variables, where is the floor operation, and the quantity ratio 0.5 can be adjusted to other positive numbers less than 1, and r>1 needs to be ensured.

[0117] Step 302, the feature quantity is taken as the number of matrix rows, and the window length is taken as the number of matrix columns to construct an initial matrix.

[0118] According to the feature quantity as the number of matrix rows and the window length as the number of matrix columns, an initial matrix A is constructed,

[0119] Step 303: Optimize the initial matrix based on the number of historical terminals corresponding to each historical moment within the target time window to obtain the optimized matrix.

[0120] Step 304: Based on the optimization matrix and the number of historical terminals corresponding to each historical moment within the target time window, determine the terminal quantity characteristics within the target time window. Step 304 will be described in detail later.

[0121] As shown in Figure 4, step 303 optimizes the initial matrix based on the number of historical terminals corresponding to each historical moment within the target time window, resulting in an optimized matrix, including:

[0122] S401 uses a target time window to divide the historical time period into multiple adjacent observation sub-periods.

[0123] The length of the observation sub-period is the same as the length of the target time window.

[0124] For example, the historical time period is divided into {t1, t2, t3, t4, t5, t6, t7, t8, t9…}, where t1, t2, t3, t4, t5, t6, t7, t8, t9… are historical moments. That is, the historical time period includes multiple historical moments such as t1, t2, t3, t4, t5, t6, t7, t8, t9… The length of the target time window is 3 moments; then the first observation sub-period is {t1, t2, t3}, the second observation sub-period is {t2, t3, t4}, the third observation sub-period is {t3, t4, t5}, the fourth observation sub-period is {t4, t5, t6}….

[0125] S402, for each observation sub-period, the number of historical terminals corresponding to each historical moment within the observation sub-period, and the number of historical terminals corresponding to the next historical moment of the observation sub-period, are used as observation samples.

[0126] The next historical moment corresponding to this observation sub-period is later than any historical moment within this observation sub-period.

[0127] For example, taking the first observation sub-period as {t1, t2, t3}, the next historical time corresponding to this observation sub-period is t4.

[0128] In this embodiment, n sets of observation samples {(x i ,y i ), {i = 1, 2, ..., n}. Where, Let be the original feature variables of the i-th group of samples (i.e., the number of historical terminals corresponding to each historical moment within the target time window). This represents the number of historical terminals corresponding to the next moment of the target time window.

[0129] S403, for each observation sample, determine a sample neighbor probability of the observation sample by using the initial matrix.

[0130] S404, optimize the initial matrix to obtain an optimized matrix, by maximizing a sum of the sample neighbor probabilities of the observation samples. S404 will be described in detail later.

[0131] Optionally, as shown in FIG. 5, the step of determining the sample neighbor probability of the observation sample by using the initial matrix in S403 comprises:

[0132] S501, taking a number of historical terminals corresponding to a next historical time of an observation sub-period in each observation sample as a target value of each observation sample.

[0133] Optionally, the number of historical terminals corresponding to the next historical time of the observation sub-period is rounded, and the rounded value is taken as the target value corresponding to the observation sample.

[0134] It can be understood that rounding is because when the target area used in the experiment is large, the statistical unit of the number of historical terminals is million, and there will be very fine numbers such as 847.920599. In order to calculate the probability later, rounding is performed, that is, y i The rounding obtains Y(t).

[0135] S502, determining other observation samples with the same target value as the target value of the observation sample as neighbor samples of the observation sample.

[0136] S503, for each neighbor sample of the observation sample, determining a single neighbor probability between the observation sample and the neighbor sample by using the initial matrix.

[0137] S504, taking a sum of the single neighbor probabilities corresponding to the neighbor samples as the sample neighbor probability of the observation sample. S504 will be described in detail later.

[0138] Specifically, as shown in FIG. 6, the step of determining the single neighbor probability between the observation sample and the neighbor sample by using the initial matrix in S503 comprises:

[0139] S601, determining a first distance measure between the observation sample and the neighbor sample by using the initial matrix.

[0140] Specifically, taking the number of historical terminals corresponding to each historical time in each observation sample in the observation sub-period as a sample feature of each observation sample; and determining a first distance measure between the sample feature of the observation sample and a sample feature of the neighbor sample of the observation sample by using the initial matrix.

[0141] In the embodiment, any observation sample is set as x i , x j is the neighbor sample of the observation sample, and D(x i , x j ) is the first distance measurement.

[0142] S602, the initial matrix is used to determine the second distance measurement between the observation sample and other observation samples.

[0143] Specifically, the initial matrix is used to determine the second distance measurement between the sample feature of the observation sample and the sample feature of other observation samples.

[0144] For example, the second distance measurement between the sample feature of the observation sample and the sample feature of other observation samples is determined in the same way as formula (4).

[0145] S603, the sum of each second distance measurement is determined as the measurement sum value.

[0146] Wherein, the sum of each second distance measurement is:

[0147] S604, the first distance measurement is divided by the measurement sum value to obtain the single neighbor probability between the observation sample and the neighbor sample.

[0148] Calculate the sample neighbor probability: given x i , the probability of selecting x j as the neighbor sample is shown in formula (5): ii = 0,

[0149] Wherein, is a kernel function. σ controls the proportion of the kernel function, and adjusts the rate at which the weight assigned to the neighbor sample decreases as the distance between them increases.

[0150] Thereafter, in S504, the sum of the single neighbor probability corresponding to each neighbor sample is taken as the sample neighbor probability of the observation sample.

[0151] Thereafter, in S404, the initial matrix is optimized to obtain the optimized matrix by maximizing the sum of the sample neighbor probabilities of each observation sample.

[0152] Define the objective function: the other observation samples with the same target value as the target value of the observation sample are taken as the neighbor samples corresponding to the observation sample. The i-th group of samples is correctly matched with y i The probability of the numerical value is provided by formula (6)

[0153] Wherein, Ci indices of the nearest neighbors of the ith sample (owning the same y i value as all samples of the ith sample). So in this framework, the approximate expected number of samples matching the y i value can be calculated by formula (7):

[0154] Let h(A) be the objective function.

[0155] Correspondingly, the initial matrix is optimized with the goal of maximizing the function value of the objective function, to obtain an optimized matrix. Gradient-based optimizers [7] or evolutionary algorithms [8] are used to maximize the objective function to update the transformation matrix A.

[0156] Specifically, gradient-based optimizers, such as gradient descent (for minimization problems) or gradient ascent (for maximization problems), use gradient information of the objective function with respect to the parameters to update the parameters. In this case, the gradient of the objective function with respect to the transformation matrix A needs to be calculated, and the gradient information is used to update A. In addition, evolutionary algorithms are a class of optimization algorithms that simulate natural selection and genetic mechanisms. They do not require gradient information, so they are suitable for problems where gradients are difficult to calculate or do not exist.

[0157] Thereafter, in step 304, the terminal quantity feature in the target time window is determined according to the optimized matrix and the historical terminal quantity corresponding to each historical time in the target time window.

[0158] Specifically, the optimized matrix A is multiplied by the original feature variable to obtain A(Y(t-1), Y(t-2), …, Y(t-d)) T is an r-order vector, denoted as (NC_1(t), …, NC_r(t)); each component of the vector is taken as an r-wireless terminal user neighbor component variable. Their advantage is that similar wireless terminal user samples can be closer in the feature space, and dissimilar samples can be farther apart.

[0159] In one exemplary embodiment, as shown in FIG. 6, the determination of the terminal quantity feature in the target time window according to the optimized matrix and the historical terminal quantity corresponding to each historical time in the target time window in step 304 includes:

[0160] Step 601, the product between the optimized matrix and the historical terminal quantity corresponding to each historical time in the target time window is taken as the initial quantity feature in the target time window.

[0161] Where F = {NC_1(t), …, NC_r(t)} represents a set of r neighbor component variables, i.e., the initial quantity feature in the target time window.

[0162] Step 602: Determine the explanatory indicators of the initial quantitative characteristics within the target time window.

[0163] Specifically, the interpretability φ of each NC_j is calculated according to formula (8). j .

[0164] Where v(S) represents the prediction result obtained by applying all nearest-neighbor component variables in set S. The prediction method can be a simple method with short running time, such as linear regression or support vector machine. |S| represents the number of elements in set S.

[0165] Step 603: The initial quantitative characteristics of the explanatory indicators that meet the explanatory thresholds are used as the terminal quantity characteristics within the target time window.

[0166] φ j NC_j(t) variables with excessively small absolute values ​​are removed, such as NC_j(t) with interpretability values ​​in the range [-5, 5]. The remaining variables are treated as highly interpretable variables of the nearest neighbor components, i.e., the number of terminals within the target time window.

[0167] For example, this example provides a method for predicting the number of terminals, specifically applied in the following scenarios:

[0168] 1. Obtain monthly wireless terminal user data for an organization in a certain city, and let Y(t) represent the number of wireless terminal users at time t.

[0169] 2. Set the time window length to p, with an upper limit of 40, and take p = 1, 2, ..., 40. Construct regression models of Y(t) with respect to Y(t-1)...Y(tp) respectively, and calculate the corresponding three information criteria AIC, BIC, and HQIC. Normalize these three information criteria respectively, and then sum the normalized information criterion values ​​of each p to obtain the integrated information criterion, denoted as Integrated_IC. Integrated_IC takes the minimum value when p = 39, and 39 is taken as the optimal time window length.

[0170] 3. Using Y(t-1), Y(t-2), ..., Y(t-39) as the original feature variables of the wireless terminal user, generate 17 nearest neighbor component variables (NC_1(t), ..., NC_r(t)) for the wireless terminal user. Calculate the interpretability φ of each NC_j. j φ was found j The variables with the smallest absolute values ​​and in the range [-5, 5] are NC_7(t), NC_15(t), NC_1(t), NC_5(t), and NC_3(t). These are removed, and the remaining variables are taken as the highly explanatory feature variables of the nearest neighbor components.

[0171] 4. Merge the original feature variables of the wireless terminal users [Y(t-1), Y(t-2), …, Y(t-39)] with the high-explaining feature variables of the nearest neighbor components [NC_2(t), NC_4(t), NC_6(t), NC_8(t), NC_9(t), NC_10(t), NC_11(t), NC_12(t), NC_13(t), NC_14(t), NC_16(t), NC_17(t)], to obtain two types of feature variables, which are taken as inputs of the N-BEATS model, to obtain the estimation of the target variable Y(t), i.e., the evolution trend of the number of wireless terminal users.

[0172] 5. The evolution trend of the number of wireless terminal users is shown, and the number of wireless terminal users in the city will increase and remain at a high level for a long time (such as the policy support of Xiong'an New Area, the relocation of large enterprises, etc., which may lead to such growth of the number of wireless terminal users).

[0173] It should be understood that, although each step in the flowchart involved in each of the above embodiments is shown in sequence according to the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, at least part of the steps in the flowchart involved in each of the above embodiments can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or steps or stages in other steps.

[0174] Based on the same concept, the embodiments of the present application also provide a terminal quantity prediction device for implementing the terminal quantity prediction method described above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more terminal quantity prediction device embodiments provided below can refer to the limitations of the terminal quantity prediction method in the above text, which will not be repeated here.

[0175] In an exemplary embodiment, as shown in FIG. 7, a terminal quantity prediction device is provided, which includes a model construction module 11, a window selection module 12, an analysis module 13, and a prediction module 14, wherein:

[0176] The model construction module 11 is configured to construct a terminal quantity regression model corresponding to each candidate time window according to the historical terminal quantity of the target region at each historical time in a historical time period; wherein each candidate time window is obtained by cutting from the historical time period;

[0177] The window selection module 12 is configured to select a target time window from each candidate time window according to model accuracy of each terminal quantity regression model;

[0178] The analysis module 13 is configured to analyze historical terminal quantities corresponding to each historical time in the target time window to obtain terminal quantity features in the target time window.

[0179] The prediction module 14 is configured to input the terminal quantity features in the target time window and the historical terminal quantities corresponding to each historical time into a terminal quantity prediction model to obtain predicted terminal quantities of each future time in a future time period of the target region.

[0180] In one embodiment, the window selection module 12 is configured to: for any terminal quantity regression model, determine model accuracy of the terminal quantity regression model according to Akaike information criterion (AIC), Bayesian information criterion (BIC) and Hannan-Quinn information criterion (HQIC) corresponding to the terminal quantity regression model.

[0181] The terminal quantity regression model with the highest model accuracy among the terminal quantity regression models is taken as a target regression model.

[0182] The candidate time window corresponding to the target regression model in each candidate time window is taken as the target time window.

[0183] In one embodiment, the analysis module 13 is configured to analyze historical terminal quantities corresponding to each historical time in the target time window to obtain terminal quantity features in the target time window, including:

[0184] According to the window length of the target time window, the number of features of the terminal quantity features in the target time window is determined.

[0185] The number of features is taken as the number of rows of a matrix, and the window length is taken as the number of columns of the matrix to construct an initial matrix.

[0186] According to the historical terminal quantities corresponding to each historical time in the target time window, the initial matrix is optimized to obtain an optimized matrix.

[0187] According to the optimized matrix and the historical terminal quantities corresponding to each historical time in the target time window, the terminal quantity features in the target time window are determined.

[0188] In one embodiment, the analysis module 13 is configured to optimize the initial matrix according to the historical terminal quantities corresponding to each historical time in the target time window to obtain an optimized matrix, including:

[0189] adopting a target time window, dividing the historical time period into a plurality of adjacent observation sub-periods; a time period length of the observation sub-periods is the same as a length of the target time window;

[0190] for each observation sub-period, taking a historical terminal quantity corresponding to each historical time in the observation sub-period and a historical terminal quantity corresponding to a next historical time corresponding to the observation sub-period as an observation sample; wherein the next historical time corresponding to the observation sub-period is later than each historical time in the observation sub-period;

[0191] for each observation sample, adopting an initial matrix to determine a sample near neighbor probability of the observation sample;

[0192] maximizing a sum of the sample near neighbor probabilities of each observation sample as a target, optimizing the initial matrix to obtain an optimized matrix.

[0193] In one of the embodiments, the analysis module 13 is configured to adopt the initial matrix to determine the sample near neighbor probability of the observation sample, and includes:

[0194] taking the historical terminal quantity corresponding to the next historical time corresponding to the observation sub-period in each observation sample as a target value of each observation sample;

[0195] determining other observation samples with the same target value as the target value of the observation sample as near neighbor samples of the observation sample;

[0196] for each near neighbor sample of the observation sample, adopting the initial matrix to determine a single near neighbor probability between the observation sample and the near neighbor sample;

[0197] taking a sum of the single near neighbor probabilities corresponding to each near neighbor sample as the sample near neighbor probability of the observation sample.

[0198] In one of the embodiments, the analysis module 13 is configured to adopt the initial matrix to determine the single near neighbor probability between the observation sample and the near neighbor sample, and includes:

[0199] adopting the initial matrix to determine a first distance measure between the observation sample and the near neighbor sample;

[0200] adopting the initial matrix to determine a second distance measure between the observation sample and other observation samples;

[0201] taking a sum of the second distance measures as a measure sum value;

[0202] dividing the first distance measure by the measure sum value to obtain the single near neighbor probability between the observation sample and the near neighbor sample.

[0203] In one of the embodiments, the analysis module 13 takes the historical terminal quantity corresponding to the next historical time corresponding to the observation sub-period in each observation sample as the target value of each observation sample, and includes:

[0204] The historical terminal quantity corresponding to the next time of the observation sub-period is rounded, and the rounded value is taken as the target value corresponding to the observation sample.

[0205] In one of the embodiments, the analysis module 13 takes the historical terminal quantity corresponding to each historical time within the observation sub-period in each observation sample as the sample feature of each observation sample; wherein:

[0206] The first distance measure between the observation sample and the neighbor sample is determined by using the initial matrix, including: the first distance measure between the sample feature of the observation sample and the sample feature of the neighbor sample of the observation sample is determined by using the initial matrix;

[0207] The second distance measure between the observation sample and other observation samples is determined by using the initial matrix, including: the second distance measure between the sample feature of the observation sample and the sample feature of other observation samples is determined by using the initial matrix.

[0208] In one of the embodiments, the analysis module 13 is configured to determine the second distance measure between the observation sample and other observation samples by using the initial matrix, including:

[0209] The historical terminal quantity corresponding to each historical time within the observation sub-period in each observation sample is taken as the sample feature of each observation sample;

[0210] The second distance measure between the sample feature of the observation sample and the sample feature of other observation samples is determined by using the initial matrix.

[0211] In one of the embodiments, the analysis module 13 is configured to take the historical terminal quantity corresponding to the next time of the observation sub-period as the target value corresponding to the observation sample, including:

[0212] The historical terminal quantity corresponding to the next time of the observation sub-period is rounded, and the rounded value is taken as the target value corresponding to the observation sample.

[0213] In one of the embodiments, the analysis module 13 is configured to determine the terminal quantity feature within the target time window according to the optimization matrix and the historical terminal quantity corresponding to each historical time within the target time window, including:

[0214] The product between the optimization matrix and the historical terminal quantity corresponding to each historical time within the target time window is taken as the initial quantity feature within the target time window;

[0215] An interpretive index of the initial quantity feature within the target time window is determined;

[0216] The initial quantity feature satisfying the explanatory threshold is taken as the terminal quantity feature in the target time window.

[0217] In one of the embodiments, the time period length for the future time period is the same as the window length of the target time window.

[0218] In one of the embodiments, the time period length of the observation time period is the same as the length of the target time window.

[0219] The modules in the terminal quantity prediction apparatus can be implemented by software, hardware, or a combination thereof. The modules can be embedded in or independent of the processor in the computer device in hardware form, or stored in the memory in the computer device in software form, so as to be invoked and executed by the processor to perform the operations corresponding to the modules.

[0220] The features described in the embodiments of the terminal quantity prediction method are applicable to the embodiments of the terminal quantity prediction apparatus. The various implementation manners of the embodiments of the terminal quantity prediction apparatus can be referred to the related descriptions in the embodiments of the terminal quantity prediction method, which will not be described herein again.

[0221] In one exemplary embodiment, a computer device is provided, which can be a server. An internal structure diagram of the computer device can be shown in FIG. 8. The computer device includes a processor, a memory, an input / output interface (I / O), and a communication interface. The processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The database of the computer device is configured to store data. The input / output interface of the computer device is configured to exchange information between the processor and external devices. The communication interface of the computer device is configured to communicate with external terminals through a network connection. The computer program is executed by the processor to implement a terminal quantity prediction method.

[0222] Those skilled in the art can understand that the structure shown in FIG. 8 is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. Specifically, the computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0223] In one example embodiment, a computer device is provided, comprising a memory and a processor, the memory storing a computer program, and the processor implementing the following steps when executing the computer program:

[0224] According to the historical terminal quantity corresponding to each historical time of the target region in the historical time period, a terminal quantity regression model corresponding to each candidate time window is constructed, wherein each candidate time window is obtained by intercepting the historical time period;

[0225] According to the model accuracy of each terminal quantity regression model, a target time window is selected from each candidate time window;

[0226] The historical terminal quantity corresponding to each historical time in the target time window is analyzed to obtain terminal quantity features in the target time window;

[0227] The terminal quantity features in the target time window and the historical terminal quantity corresponding to each historical time are input into a terminal quantity prediction model to obtain the predicted terminal quantity of each future time of the target region in the future time period.

[0228] The features described in the embodiments of the above terminal quantity prediction method are applicable to the embodiments of the computer device, and various implementation manners of the embodiments of the computer device can be referred to the related description in the embodiments of the aforementioned terminal quantity prediction method, which will not be described here.

[0229] In one embodiment, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by a processor to implement the following steps:

[0230] According to the historical terminal quantity corresponding to each historical time of the target region in the historical time period, a terminal quantity regression model corresponding to each candidate time window is constructed, wherein each candidate time window is obtained by intercepting the historical time period;

[0231] According to the model accuracy of each terminal quantity regression model, a target time window is selected from each candidate time window;

[0232] The historical terminal quantity corresponding to each historical time in the target time window is analyzed to obtain terminal quantity features in the target time window;

[0233] The terminal quantity features in the target time window and the historical terminal quantity corresponding to each historical time are input into a terminal quantity prediction model to obtain the predicted terminal quantity of each future time of the target region in the future time period.

[0234] The features described in the embodiments of the terminal quantity prediction method apply to the embodiments of the computer readable storage medium, and various implementation manners of the embodiments of the computer readable storage medium can be referred to the related description in the foregoing embodiments of the terminal quantity prediction method, and will not be described here again.

[0235] In one embodiment, a computer program product is provided, comprising a computer program which, when executed by a processor, implements the following steps:

[0236] According to the historical terminal quantities of the target region at each historical time in the historical time period, a terminal quantity regression model corresponding to each candidate time window is constructed; each candidate time window is obtained by cutting from the historical time period;

[0237] According to the model accuracy of each terminal quantity regression model, a target time window is selected from each candidate time window;

[0238] The historical terminal quantities corresponding to each historical time in the target time window are analyzed to obtain terminal quantity features in the target time window;

[0239] The terminal quantity features in the target time window and the historical terminal quantities corresponding to each historical time are input into the terminal quantity prediction model to obtain the predicted terminal quantities of the target region at each future time in the future time period.

[0240] The features described in the embodiments of the terminal quantity prediction method apply to the embodiments of the computer program product, and various implementation manners of the embodiments of the computer program product can be referred to the related description in the foregoing embodiments of the terminal quantity prediction method, and will not be described here again.

[0241] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0242] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. The non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. The volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, the RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., without being limited thereto.

[0243] The technical features of the above embodiments can be combined in any manner. To make the description concise, not all possible combinations of the technical features in the above embodiments are described, but as long as the combinations of the technical features do not exist contradictions, they should be considered as the scope of the present application.

[0244] The above-described embodiments are merely illustrative of several embodiments of the present application, and the description is relatively specific and detailed, but should not be understood as a limitation on the scope of the patent. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of the present application. Therefore, the scope of protection of the present application should be subject to the appended claims.

Claims

1. A terminal quantity prediction method, comprising: constructing terminal quantity regression models corresponding to different candidate time windows according to historical terminal quantities of a target area at historical time points in a historical time period; wherein each candidate time window is obtained by cutting from the historical time period; selecting a target time window from the candidate time windows according to model accuracies of the terminal quantity regression models; analyzing the historical terminal quantities of the target time window at the historical time points to obtain terminal quantity features in the target time window; inputting the terminal quantity features in the target time window and the historical terminal quantities at the historical time points into a terminal quantity prediction model to obtain predicted terminal quantities of the target area at future time points in a future time period.

2. The method of claim 1, wherein, The selecting a target time window from the candidate time windows according to model accuracies of the terminal quantity regression models comprises: for any terminal quantity regression model, determining the model accuracy of the terminal quantity regression model according to Akaike information criterion (AIC), Bayesian information criterion (BIC) and Hannan-Quinn information criterion (HQIC) corresponding to the terminal quantity regression model; selecting a terminal quantity regression model with the highest model accuracy from the terminal quantity regression models as a target regression model; selecting a candidate time window corresponding to the target regression model from the candidate time windows as the target time window.

3. The method of claim 1, wherein, The analyzing the historical terminal quantities of the target time window at the historical time points to obtain terminal quantity features in the target time window comprises: determining the number of features of the terminal quantity features in the target time window according to the window length of the target time window; constructing an initial matrix by taking the number of features as the number of rows and taking the window length as the number of columns; optimizing the initial matrix according to the historical terminal quantities at the historical time points in the target time window to obtain an optimized matrix; determining the terminal quantity features in the target time window according to the optimized matrix and the historical terminal quantities at the historical time points in the target time window.

4. The method of claim 3, wherein, The optimizing the initial matrix according to the historical terminal quantities at the historical time points in the target time window to obtain an optimized matrix comprises: dividing the historical time period into a plurality of adjacent observation sub-periods by using the target time window; the length of the observation sub-period is the same as the length of the target time window; for each observation sub-period, taking the historical terminal quantities at the historical time points in the observation sub-period and the historical terminal quantity at the next historical time point corresponding to the observation sub-period as an observation sample; wherein the next historical time point corresponding to the observation sub-period is later than the historical time points in the observation sub-period; for each observation sample, determining the sample neighbor probability of the observation sample by using the initial matrix; optimizing the initial matrix to obtain an optimized matrix by maximizing the sum of the sample neighbor probabilities of the observation samples.

5. The method of claim 4, wherein, The determining the sample neighbor probability of the observation sample by using the initial matrix comprises: The historical terminal quantity corresponding to the next historical time of the observation sub-period in each observation sample is taken as a target value of each observation sample; Other observation samples with the same target value as the observation sample are determined as neighbor samples of the observation sample; For each neighbor sample of the observation sample, a single neighbor probability between the observation sample and the neighbor sample is determined by using the initial matrix; The sum of the single neighbor probabilities corresponding to each neighbor sample is taken as a sample neighbor probability of the observation sample.

6. The method of claim 5, wherein, The single neighbor probability between the observation sample and the neighbor sample is determined by using the initial matrix, and includes: A first distance measure between the observation sample and the neighbor sample is determined by using the initial matrix; A second distance measure between the observation sample and other observation samples is determined by using the initial matrix; The sum of each second distance measure is determined as a measure sum value; The first distance measure is divided by the measure sum value to obtain the single neighbor probability between the observation sample and the neighbor sample.

7. The method of claim 5, wherein, The historical terminal quantity corresponding to the next historical time of the observation sub-period in each observation sample is taken as a target value of each observation sample, and includes: The historical terminal quantity corresponding to the next historical time of the observation sub-period is rounded, and the rounded value is taken as the target value corresponding to the observation sample.

8. The method of claim 6, further comprising: The historical terminal quantity corresponding to each historical time within the observation sub-period in each observation sample is taken as a sample feature of each observation sample; wherein: The first distance measure between the observation sample and the neighbor sample is determined by using the initial matrix, and includes: the first distance measure between the sample feature of the observation sample and the sample feature of the neighbor sample of the observation sample is determined by using the initial matrix; The second distance measure between the observation sample and other observation samples is determined by using the initial matrix, and includes: the second distance measure between the sample feature of the observation sample and the sample feature of other observation samples is determined by using the initial matrix.

9. The method of claim 6, wherein, The second distance measure between the observation sample and other observation samples is determined by using the initial matrix, and includes: The historical terminal quantity corresponding to each historical time within the observation sub-period in each observation sample is taken as a sample feature of each observation sample; The second distance measure between the sample feature of the observation sample and the sample feature of other observation samples is determined by using the initial matrix.

10. The method of claim 5, wherein, The historical terminal quantity corresponding to the next historical time of the observation sub-period is rounded, and the rounded value is taken as the target value corresponding to the observation sample. The terminal quantity feature within the target time window is determined according to the optimization matrix and the historical terminal quantity corresponding to each historical time within the target time window, and includes:

11. The method of claim 3, wherein, The product between the optimization matrix and the historical terminal quantity corresponding to each historical time within the target time window is taken as an initial quantity feature within the target time window; An interpretive index of the initial quantity feature within the target time window is determined; ​ The initial quantity feature satisfying the explanatory threshold is taken as the terminal quantity feature in the target time window.

12. The method of any one of claims 1-11, wherein, The time length of the future time period is the same as the window length of the target time window.

13. The method of any one of claims 4-10, wherein, The time length of the observation time period is the same as the length of the target time window.

14. A terminal quantity prediction device, wherein, The device comprises: a model construction module, configured to construct terminal quantity regression models corresponding to different candidate time windows according to historical terminal quantities corresponding to historical time points in a historical time period of a target region; each candidate time window is obtained by intercepting the historical time period; a window selection module, configured to select a target time window from the candidate time windows according to model accuracies of the terminal quantity regression models; an analysis module, configured to analyze historical terminal quantities corresponding to historical time points in the target time window to obtain a terminal quantity feature in the target time window; a prediction module, configured to input the terminal quantity feature in the target time window and the historical terminal quantities corresponding to the historical time points into a terminal quantity prediction model to obtain predicted terminal quantities of the target region at future time points in a future time period.

15. A computer device comprising a memory and a processor, the memory storing a computer program, wherein, The processor implements the steps of the method of any one of claims 1 to 13 when executing the computer program.

16. A computer readable storage medium having stored thereon a computer program, wherein, The computer program, when executed by the processor, implements the steps of the method of any one of claims 1 to 13.

17. A computer program product comprising a computer program, wherein, The computer program, when executed by the processor, implements the steps of the method of any one of claims 1 to 13.

Images