Index data prediction method, device and equipment and readable storage medium

Abstract

The invention discloses an index data prediction method, device and equipment and a readable storage medium, and the method comprises the steps: building a tree-type multi-factor system, building and training a dynamic factor model based on a factor data and reference index quantitative screening mechanism, and obtaining a prediction value of a reference index in a prediction period through the iteration of the dynamic factor model, rich predictive factors can be matched and transformed according to requirements of different reference indexes, index prediction in various scenes can be dealt with, meanwhile, a quantitative screening mechanism avoids subjectivity of manual selection, dependence on manpower is reduced, the problem that an existing historical sales trend continuation method ignores external factors possibly influencing sales, and the prediction efficiency is improved is solved. And the technical problems of high information acquisition difficulty, manpower consumption and prediction result deviation easily caused by a microscopic expected summary method are solved.

Description

Technical Field

[0001] The present application relates to the technical field of sales forecasting, and in particular to an indicator data forecasting, device, equipment and readable storage medium. Background Art

[0002] The short-term forecast of the automobile sales market is crucial to all parties involved in the automobile industry. It not only affects the production planning and inventory management of automobile manufacturers, but also has a direct impact on the strategy formulation of automobile retailers and related insurance companies. At present, the two main forecasting methods used in the industry are:

[0003] Historical sales trend continuation method: This method mainly relies on historical sales data and uses statistical models such as linear extrapolation and autoregression to predict future sales. Its advantages are simple operation and easy implementation, but its disadvantages are also obvious, that is, it ignores many external factors that may affect sales, such as economic environment, policy changes, consumption level, car company operating conditions and price factors. This leads to the possibility of systematic deviation in the prediction results.

[0004] Micro-expectation aggregation method: This method collects and aggregates expected data from the front line of sales by investigating 4S stores, sales platforms, etc. Its advantage is that it can capture the most cutting-edge information in the market, but its disadvantage is that it consumes a lot of manpower, and due to the diversity of information sources, it is difficult to achieve the uniformity of quantitative estimates. In addition, the accuracy of the forecast results is highly dependent on the breadth and depth of the survey, and incomplete information may lead to forecast bias.

[0005] Therefore, it is extremely important to provide a method that can integrate multiple information to collaboratively predict automobile sales. Summary of the invention

[0006] The present application provides an indicator data prediction, device, equipment and readable storage medium, which solves the technical problems of the existing historical sales trend continuation method ignoring external factors that may affect sales, and the micro-expectation aggregation method having difficulty in information collection and wasting manpower, both of which easily lead to deviations in the prediction results.

[0007] In view of this, the first aspect of the present application provides an indicator data prediction method, the method comprising:

[0008] Constructing a prediction factor tree, wherein the prediction factor tree includes a plurality of prediction factors, and the prediction factors are connected according to influence modes and conduction chain relationships;

[0009] Determine a benchmark indicator, calculate a first matching degree between the factor data of each of the prediction factors and the benchmark indicator, and screen and generate a total factor library for modeling;

[0010] During the forecast period, the second matching degree between the factor data of each of the forecast factors in the overall modeling factor library and the benchmark indicator is calculated, and the current modeling factor library is screened and generated;

[0011] Constructing a dynamic factor model using the prediction factors and the benchmark indicators contained in the current modeling factor library as variables;

[0012] Iteratively converge the dynamic factor model through an expectation maximization algorithm to obtain model parameters of the dynamic factor model after convergence;

[0013] The predicted values ​​of each of the prediction factors and the benchmark indicators contained in the current modeling factor library at the prediction time are calculated based on the model parameters.

[0014] Optionally, the constructing of a prediction factor tree, wherein the prediction factor tree includes a plurality of prediction factors, wherein the prediction factors are connected according to the influence mode and the conduction chain relationship, further includes:

[0015] Counting the data update frequency and the amount of historical data of the factor data of each of the prediction factors in the prediction factor tree;

[0016] The prediction factors used for predicting indicator data are screened and determined according to the data update frequency and the amount of historical data.

[0017] Optionally, the constructing of a prediction factor tree, wherein the prediction factor tree includes a plurality of prediction factors, wherein the prediction factors are connected according to the influence mode and the conduction chain relationship, further includes:

[0018] Performing data preprocessing on the factor data included in each of the prediction factors;

[0019] The preprocessed factor data is standardized and converted into a target data format.

[0020] Optionally, the calculating of the first matching degree between the factor data of each of the prediction factors and the benchmark indicator and the screening and generation of the overall factor library for modeling are specifically:

[0021] For a plurality of preset evaluation dimensions, a first matching degree between the factor data of each of the prediction factors and the benchmark indicator is calculated respectively by using a plurality of preset models;

[0022] Counting the statistical values ​​of the first matching degree of each of the prediction factors in all the preset evaluation dimensions;

[0023] After sorting all the prediction factors in descending order according to the statistical value of the first matching degree, a target number of the prediction factors are selected in order to generate a modeling overall factor library.

[0024] Optionally, the construction of the dynamic factor model using the prediction factors and the benchmark indicators contained in the current modeling factor library as variables is specifically as follows:

[0025] Taking the prediction factors and the benchmark variables contained in the current modeling factor library as variables, performing data preprocessing on the variables;

[0026] Constructing a dynamic factor model and determining the dynamic relationship and time series between the variables in the dynamic factor model;

[0027] Perform initial random assignment of model parameters to the dynamic factor model.

[0028] Optionally, the dynamic factor model is iteratively converged by an expectation maximization algorithm, and the model parameters of the dynamic factor model after convergence are specifically:

[0029] Calculate the first conditional expectation of the latent factors of the dynamic factor model by Kalman filtering method;

[0030] Calculating a second conditional expectation of a preset error term of the dynamic factor model;

[0031] According to the first conditional expectation and the second conditional expectation, updating the model parameters of the dynamic factor model by maximizing the log-likelihood function of the observed data;

[0032] Based on a preset optimization algorithm, optimizing and adjusting the model parameters of the dynamic factor model;

[0033] If the model parameters converge or the current iteration round reaches the preset number of iterations, the converged model parameters of the dynamic factor model are obtained, otherwise the first conditional expectation of the latent factors of the dynamic factor model is returned to be recalculated.

[0034] Optionally, after obtaining the predicted values ​​of each of the prediction factors and the benchmark indicator contained in the current modeling factor library at the prediction time by calculating according to the model parameters, the method further includes:

[0035] Performing frequency conversion processing on the prediction factor and the prediction value of the benchmark indicator by a preset frequency conversion method to obtain a prediction value corresponding to a target frequency;

[0036] The predicted values ​​of the prediction factors and the benchmark indicators are seasonally restored to obtain corresponding predicted values ​​under the target seasonal pattern.

[0037] A second aspect of the present application provides an indicator data prediction device, the device comprising:

[0038] A factor tree construction unit, used to construct a prediction factor tree, wherein the prediction factor tree includes a plurality of prediction factors, and the prediction factors are connected according to influence modes and conduction chain relationships;

[0039] A first matching degree calculation unit is used to determine a benchmark indicator, calculate a first matching degree between the factor data of each of the prediction factors and the benchmark indicator, and screen and generate a modeling overall factor library;

[0040] A second matching degree calculation unit is used to calculate the second matching degree between the factor data of each of the prediction factors in the overall modeling factor library and the benchmark indicator within the prediction period, and screen and generate the current modeling factor library;

[0041] A model building unit, used to build a dynamic factor model using the prediction factors and the benchmark indicators contained in the current modeling factor library as variables;

[0042] A model iteration unit, used for iteratively converging the dynamic factor model through an expectation maximization algorithm to obtain model parameters of the dynamic factor model after convergence;

[0043] The prediction unit is used to calculate the predicted values ​​of each prediction factor and the benchmark indicator contained in the current modeling factor library at the prediction time according to the model parameters.

[0044] A third aspect of the present application provides an indicator data prediction device, the device comprising a processor and a memory:

[0045] The memory is used to store program code and transmit the program code to the processor;

[0046] The processor is used to execute the steps of the method for predicting indicator data as described in the first aspect according to the instructions in the program code.

[0047] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store program code, and the program code is used to execute the method described in the first aspect above.

[0048] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0049] In the present application, an indicator data prediction, device, equipment and readable storage medium are provided. By building a tree-type multi-factor system, a dynamic factor model is constructed and trained based on the quantitative screening mechanism of factor data and benchmark indicators. By iterating the dynamic factor model, the predicted value of the benchmark indicator is obtained within the prediction period. Among them, the rich prediction factors can be matched and transformed according to the needs of different benchmark indicators, and can cope with indicator prediction in various scenarios. At the same time, the quantitative screening mechanism avoids the subjectivity of human selection and reduces dependence on manual labor. It solves the technical problems of the existing historical sales trend continuation method ignoring external factors that may affect sales, and the micro-expectation summary method, which has the problem of difficult information collection and labor-intensiveness, and is prone to biased prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A method flow chart of the indicator data method in the embodiment of the present application;

[0051] Figure 2 This is a schematic diagram of the structure of the indicator data device in the embodiment of the present application;

[0052] Figure 3 This is a schematic diagram of the structure of the indicator data device in the embodiment of the present application. DETAILED DESCRIPTION

[0053] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0054] The present application designs an indicator data prediction, device, equipment and readable storage medium, which solves the technical problems of the existing historical sales trend continuation method ignoring external factors that may affect sales, and the micro-expectation aggregation method having difficulty in information collection and wasting manpower, both of which easily lead to deviations in the prediction results.

[0055] For easier understanding, see Figure 1 , Figure 1 is a method flow chart of the indicator data prediction method in the embodiment of the present application, such as Figure 1 As shown, specifically:

[0056] S101, constructing a prediction factor tree, wherein the prediction factor tree includes a plurality of prediction factors, and the prediction factors are connected according to the influence mode and the conduction chain relationship;

[0057] It should be noted that predictor identification is to identify all factors that may affect the target indicator data and classify them into different predictors, such as economic conditions, market trends, consumer preferences, etc.

[0058] Factor data collection: Collect data related to each predictive factor, including but not limited to sales data, consumer survey data, macroeconomic data, etc.

[0059] Establish a tree-like framework to organize factors according to their impact mode and transmission chain, specifically:

[0060] 1. Determine the root node

[0061] The root of the tree framework usually represents the final target or benchmark indicator, that is, the variable we want to predict. In the embodiment of the present application, the root node is "car sales".

[0062] 2. Identify the main branches

[0063] Starting from the root node, identify the main factors that directly affect car sales, which form the first level of branches of the tree. For example, these branches may include "economic conditions", "market trends", "consumer preferences", etc.

[0064] 3. Segmentation of secondary factors

[0065] Each major branch is further subdivided into sub-factors, which are more specific factors that affect the major branches. For example, "economic conditions" may be subdivided into "GDP growth rate", "unemployment rate", "consumer confidence index", etc.

[0066] 4. Consider the conduction chain

[0067] The transmission chain describes the causal relationship and impact path between factors. For example, the "GDP growth rate" may directly affect the "consumer confidence index", and the "consumer confidence index" will affect the "consumer purchasing power" and finally affect the "car sales".

[0068] 5. Build a tree structure

[0069] Continue to expand the tree framework until all relevant factors are covered. Each factor should have its position in the tree and its relationship with other factors. This will help us understand how each factor directly or indirectly affects car sales.

[0070] 6. Dynamic updates and adjustments

[0071] The tree framework is not static, it should be dynamically updated and adjusted according to new data and market changes. For example, if there are new market trends or policy changes, it may be necessary to add or remove factors and data from the tree.

[0072] Example

[0073] Suppose we are building a car sales prediction model, the tree structure may look like this:

[0074] Root node: car sales;

[0075] First-level branches: economic conditions, market competition, and consumer preferences;

[0076] The second branch (taking "economic conditions" as an example): GDP growth rate, unemployment rate, consumer confidence index;

[0077] The third level branch (taking the “Consumer Confidence Index” as an example): stock market performance, interest rate level, and employment growth.

[0078] In this way, we can not only identify the key factors that affect car sales, but also understand how these factors ultimately affect sales through a series of transmission chains. This systematic approach helps to improve the accuracy and comprehensiveness of forecasts.

[0079] Furthermore, a prediction factor tree is constructed, which includes several prediction factors. After the prediction factors are connected according to the influence mode and the transmission chain relationship, the following are also included:

[0080] Count the data update frequency and historical data quantity of the factor data of each predictor in the predictor tree;

[0081] Based on the data update frequency and the amount of historical data, screen and determine the predictive factors used to predict indicator data.

[0082] It should be noted that predictors are variables that are considered likely to affect or predict a certain outcome variable (target variable). They are input variables in the model and are used to predict or estimate the output results through statistical methods.

[0083] Predictors can be quantitative (such as age, temperature, height, etc.) or qualitative (such as gender, region, brand, etc.).

[0084] When building a predictive model, predictors are used to discover relationships with the outcome variable.

[0085] Factor data are specific numerical values ​​or observations associated with the predictors. They are

[0086] Instantiation data in different situations.

[0087] Factor data are often used to train statistical models, and by analyzing this data, one can assess the extent to which the predictors affect the outcome variable.

[0088] The history and updating frequency of factor data are critical to the accuracy and reliability of the model.

[0089] Data update frequency statistics: record the update frequency of each factor data, such as monthly, quarterly or annually.

[0090] Statistics on the number of historical data: Count the number of historical data contained in each predictor to evaluate its reliability as a predictor for indicator data prediction.

[0091] Furthermore, a prediction factor tree is constructed, which includes several prediction factors. After the prediction factors are connected according to the influence mode and the transmission chain relationship, the following are also included:

[0092] Perform data preprocessing on the factor data contained in each predictor;

[0093] The preprocessed factor data is standardized and converted into the target data format.

[0094] It should be noted that there are large differences in the statistical caliber, update frequency, data quality, and seasonal dependence of different factor data, and factors need to be preprocessed for subsequent screening and modeling. The main methods used are frequency conversion, same-month and year-on-year transformation, outlier removal, null value filling, seasonal adjustment, and Spring Festival effect adjustment to process the data into a cointegration form with the same or adjacent frequencies and no seasonality.

[0095] The specific process may include:

[0096] The steps include outlier processing (eliminating outliers), missing value processing (filling in null values), frequency conversion (unifying data frequency), year-on-year and month-on-month transformation (unifying data caliber), seasonal adjustment (eliminating seasonal effects, which are different for different data and cannot be compared), Spring Festival effect adjustment (eliminating the impact of holidays such as the Spring Festival on data), and standardization (facilitating data comparison). These steps are independent of each other.

[0097] S102, determining a benchmark indicator, calculating a first matching degree between the factor data of each prediction factor and the benchmark indicator, and screening and generating a modeling overall factor library;

[0098] It should be noted that the definition of benchmark indicators, for example, determines automobile sales as a benchmark indicator.

[0099] Calculation of matching degree and leading edge: When calculating matching degree and leading edge, a variety of models can be used to calculate different evaluation dimensions, and comprehensive scores can be given, and quantitative sorting and screening can be carried out according to the scores.

[0100] For example: time difference correlation, KL information volume calculates overall similarity and leadingness; Bry-Boscha algorithm calculates inflection point matching and inflection point leadingness; dynamic time warping algorithm calculates the similarity of curve shape, etc.

[0101] Furthermore, the first matching degree between the factor data of each predictor and the benchmark index is calculated, and the overall factor library for modeling is screened and generated as follows:

[0102] For several preset evaluation dimensions, the first matching degree between the factor data of each prediction factor and the benchmark indicator is calculated respectively through several preset models;

[0103] Count the statistical values ​​of the first matching degree of each prediction factor in all preset evaluation dimensions;

[0104] After sorting all predictors in descending order according to the statistical value of the first matching degree, the target number of predictors are selected in order to generate a modeling overall factor library.

[0105] It should be noted that the evaluation dimensions and models need to be preset first, including:

[0106] Pre-defined evaluation dimensions: These are different aspects or criteria used to evaluate the correlation of the predictor with the benchmark indicator. For example, these dimensions may include statistical significance, explanatory power, predictive power, etc.

[0107] Preset Model: For each evaluation dimension, we select or build an appropriate statistical or machine learning model to assess the correlation. These models may include linear regression, time series analysis, principal component analysis, etc.

[0108] Secondly, the full sample data is used to calculate the first matching degree between the factor data of each predictor and the benchmark indicator, where:

[0109] Matching calculation: For each predictor and benchmark indicator, we use the preset model to calculate the matching degree on each evaluation dimension. For example, the Pearson correlation coefficient is used to calculate the correlation between the factor data and the benchmark indicator.

[0110] Multiple fits: Each predictor will receive a fit score on multiple evaluation dimensions, which reflects the relevance of the factor in different aspects.

[0111] Next, calculate the statistical value of the matching degree, specifically:

[0112] Average Statistics: To comprehensively consider the performance of a predictor on all evaluation dimensions, we calculate the average of its matching degree on all dimensions. This average provides a comprehensive indicator for evaluating the overall relevance of the predictor.

[0113] Likelihood function: Assess the contribution of a predictor to a model by comparing the likelihood value of the model. An increase in the likelihood value indicates that the new predictor improves the model.

[0114] Incremental Value: Evaluates the incremental value of the new predictor relative to the existing model, that is, the improvement in prediction performance brought by the new predictor over the original model.

[0115] Model fit: Use R 2 (coefficient of determination), AIC (Akaike information criterion), BIC (Bayesian information criterion) and other indicators are used to evaluate the overall fit of the model.

[0116] Discrimination: Use C statistic (Concordance Statistic) or AUC (Area Under the Curve) to measure the model's ability to distinguish different results.

[0117] Calibration: Assessing the consistency of model predictions with observed values, such as using the Hosmer-Lemeshow test.

[0118] Reclassification table: Evaluate the prediction accuracy of the model by comparing the risk categories predicted by the model with the risk categories that actually occurred.

[0119] Cross-validation: Divide the dataset into multiple subsets and evaluate the performance of the model on different subsets to test the stability and generalization ability of the model.

[0120] Stepwise regression: includes forward selection, backward elimination, and stepwise selection methods to optimize the model by gradually adding or removing predictors.

[0121] Factor analysis: Factor analysis is used to identify potential relationships between variables, reduce data dimensions, and improve the interpretability of the model.

[0122] Bootstrap method: Use resampling technology to estimate the prediction interval of the model and evaluate the uncertainty of the model prediction.

[0123] Weight consideration: In some cases, some evaluation dimensions may be more important than other dimensions. Therefore, we can assign different weights to different evaluation dimensions and then calculate the weighted average matching degree.

[0124] Finally, sort and filter in descending order, where:

[0125] Sort in descending order: Sort all predictors in descending order based on their statistical values. This way, the most relevant factors will be at the top.

[0126] Screening factors: Based on the needs of model building, we select a certain number of top-ranked predictors. This number may be based on the complexity, explanatory power, or predictive performance requirements of the model.

[0127] Assume that we have the following preset evaluation dimensions and models:

[0128] Evaluation dimensions:

[0129] Statistical significance (using t-test)

[0130] Coefficient size (linear regression coefficient)

[0131] Predictive power (based on cross-validation R 2 )

[0132] Preset Models:

[0133] Linear Regression Model

[0134] Time Series Forecasting Models

[0135] Principal Component Analysis (PCA)

[0136] For each predictor, we use the above model to calculate the matching degree on each evaluation dimension. Then, we calculate the average matching degree of each factor on all dimensions and sort all factors in descending order according to these averages. Finally, we select the top 10 factors to build our car sales prediction model.

[0137] This approach ensures that the selected predictors are highly correlated with the benchmark indicator in multiple ways, thus improving the accuracy and reliability of the model.

[0138] S103, within the forecast period, calculating the second matching degree between the factor data of each forecast factor in the overall modeling factor library and the benchmark indicator, and screening and generating the current modeling factor library;

[0139] It should be noted that during the forecast period, the predictive factors in the modeling overall factor library are re-evaluated to determine their current correlation with the benchmark indicators. This process is called calculating the second match. This step is to ensure that the selected factors are still relevant under the latest data and market conditions, thereby improving the accuracy of the forecast model. The following is a specific explanation:

[0140] 1. Define the forecast period

[0141] Forecast period setting: Determine the time frame used to evaluate the second match, which could be the past month, quarter, or year, depending on business needs and data availability.

[0142] 2. Calculate the second matching degree

[0143] Select an evaluation model: Choose a statistical or machine learning model appropriate for the current forecast period to evaluate the match and lead between the predictors and the benchmark indicators.

[0144] Correlation analysis: Using sample data of each predictor during the forecast period, the selected model is used to calculate the matching and leading degree of each predictor with the benchmark indicator during the forecast period. This may include time difference correlation, KL information volume to calculate overall similarity and leading degree, Bry-Boscha algorithm to calculate inflection point matching and inflection point leading degree, dynamic time warping algorithm to calculate the similarity of curve shape, etc.

[0145] 3. Evaluation and screening

[0146] Fit and Lead Assessment: Evaluate the secondary fit of each predictor to determine how much they affect the benchmark indicator.

[0147] Screening factors: Based on the results of the second matching degree, the most relevant predictive factors are selected. A threshold may be set, and only factors exceeding this threshold will be selected into the current modeling factor library.

[0148] S104, constructing a dynamic factor model using the prediction factors and benchmark indicators contained in the current modeling factor library as variables;

[0149] Furthermore, the dynamic factor model is constructed by taking the prediction factors and benchmark indicators contained in the current modeling factor library as variables:

[0150] Taking the predictive factors and benchmark variables contained in the current modeling factor library as variables, data preprocessing is performed on the variables;

[0151] Construct a dynamic factor model and determine the dynamic relationship and time series between variables in the dynamic factor model;

[0152] Perform initial random assignment of model parameters for the dynamic factor model.

[0153] It should be noted that variable selection: predictive factors and benchmark indicators are selected as model variables from the current modeling factor library.

[0154] Model building: Construct a dynamic factor model to determine the dynamic relationship and time series characteristics between variables.

[0155] Parameter initialization: Assign initial values ​​to model parameters in preparation for algorithm iteration. These values ​​can be obtained through statistical analysis of historical data or by random assignment.

[0156] Suppose an automobile manufacturer wants to predict sales for the next quarter. In the previous quarter, they have built a modeling population factor library containing multiple predictors. When entering a new forecast period, they perform the following steps:

[0157] Updated data: Collect the latest economic indicators, consumer confidence survey results, industry reports, and more.

[0158] Re-evaluate: Using the latest data, calculate the correlation of each factor with the expected sales volume for the next quarter through regression analysis.

[0159] Screening factors: Based on the results of the correlation analysis, select the factors that are most correlated with the expected sales volume.

[0160] Build the current factor library: The screened factors are combined into the current modeling factor library for building new prediction models.

[0161] Through this process, automakers are able to ensure that their predictive models are based on the latest and most relevant data, thereby improving the accuracy and reliability of predictions.

[0162] S105, iteratively converge the dynamic factor model through an expectation maximization algorithm to obtain model parameters of the converged dynamic factor model;

[0163] Furthermore, the dynamic factor model is iteratively converged through the expectation maximization algorithm, and the model parameters of the dynamic factor model after convergence are obtained as follows:

[0164] Calculate the first conditional expectation of latent factors of the dynamic factor model by Kalman filtering method;

[0165] Calculate the second conditional expectation of the preset error term of the dynamic factor model;

[0166] According to the first conditional expectation and the second conditional expectation, the model parameters of the dynamic factor model are updated by maximizing the log-likelihood function of the observed data;

[0167] Based on the preset optimization algorithm, optimize and adjust the model parameters of the dynamic factor model;

[0168] If the model parameters converge or the current iteration round reaches the preset number of iterations, the model parameters of the converged dynamic factor model are obtained, otherwise the first conditional expectation of the latent factors of the dynamic factor model is returned to recalculate.

[0169] It should be noted that Kalman filtering: Use the Kalman filtering method to estimate the conditional expectation of latent factors and error terms.

[0170] Parameter update: Update the model parameters by maximizing the log-likelihood function using conditional expectation.

[0171] Convergence judgment: Check whether the model parameters have converged. If not, return to step S105 to continue iteration.

[0172] The expectation maximization (EM) algorithm is an iterative algorithm used in statistics to estimate the parameters of probability models with hidden variables. In the context of dynamic factor models, the EM algorithm is particularly useful because it allows us to estimate unobservable hidden factors in the model (such as underlying economic trends or market conditions) and optimize model parameters. The following is a specific explanation of the application of the EM algorithm in dynamic factor models:

[0173] Step 1: Initialize model parameters

[0174] Parameter assignment: Assign initial values ​​to the parameters of the dynamic factor model (such as factor loadings, variance of error terms, etc.). These initial values ​​can be based on previous research, expert knowledge, or randomly selected.

[0175] Step 2: E (Expectation Step)

[0176] Conditional expectation of latent factors: Use Kalman filtering or other state-space methods to estimate the conditional expectation of latent factors. Kalman filtering is a recursive filter that estimates the state of a dynamic system based on a series of observations, even if the data is noisy.

[0177] Conditional expectation of error terms: Calculate the conditional expectation of the error terms given in the model. This usually involves estimating the statistical distribution parameters of the error terms, such as the mean and variance.

[0178] Step 3: M-step (Maximization Step)

[0179] Parameter update: Using the conditional expectation of the latent factors and error terms obtained in step E, the model parameters are updated by maximizing the log-likelihood function of the observed data. The purpose of this step is to find the model parameters that best explain the observed data.

[0180] Optimization algorithm: Use preset optimization algorithms (such as gradient descent, Newton's method, etc.) to adjust model parameters to improve the model's fit to the data.

[0181] Step 4: Iteration and convergence judgment

[0182] Iteration process: Repeat the E and M steps until the model parameters converge or the preset number of iterations is reached. Each iteration aims to improve the estimation accuracy of the model parameters.

[0183] Convergence judgment: After each iteration, the changes in model parameters are evaluated. If the change is less than a certain threshold or reaches the preset number of iterations, the model parameters can be considered to have converged.

[0184] Suppose we are using a dynamic factor model to predict car sales, and the model contains some unobservable latent factors, such as potential market demand and economic cycles. We first assign initial values ​​to the model parameters, and then use the Kalman filter method to estimate the conditional expectations of the latent factors in the E step. Then, in the M step, we use these expected values ​​to update the model parameters to maximize the log-likelihood function of the observed data. We repeat this process until the model parameters converge. Finally, we use these parameters to predict future sales and provide decision support for automobile manufacturers.

[0185] In this way, the EM algorithm allows us to handle complex models with latent variables and provide more accurate prediction results in applications such as car sales forecasting.

[0186] S106. Calculate the predicted values ​​of the prediction factors and benchmark indicators contained in the current modeling factor library at the prediction time based on the model parameters.

[0187] It should be noted that once the model parameters converge, they will be used to predict future benchmark indicator values, such as car sales. The final model parameters are used for predictive analysis to support decision making.

[0188] Furthermore, after obtaining the predicted values ​​of each prediction factor and the benchmark index contained in the current modeling factor library at the prediction time according to the model parameters, it also includes:

[0189] The predicted values ​​of the prediction factors and benchmark indicators are subjected to frequency conversion processing by a preset frequency conversion method to obtain the predicted values ​​corresponding to the target frequency;

[0190] The predicted values ​​of the prediction factors and benchmark indicators are seasonally restored to obtain the corresponding predicted values ​​under the target seasonal pattern.

[0191] It should be noted that in time series analysis and forecasting, it is often necessary to convert data to different time frequencies or adjust the data to reflect specific seasonal patterns. The following is a specific explanation of these two steps:

[0192] In time series analysis and forecasting, it is often necessary to transform data to a different time frequency or adjust the data to reflect a specific seasonal pattern. The following is a detailed explanation of these two steps:

[0193] Step 1: Frequency conversion

[0194] Frequency conversion refers to the process of converting time series data from one frequency to another. In car sales forecasting, this may involve converting monthly forecasts to quarterly or annual forecasts, or vice versa. The following are the specific steps of frequency conversion:

[0195] Determine the target frequency: First, identify the target frequency you need to transform the data to. For example, if the model outputs monthly forecasts, but you need quarterly data, then the target frequency is quarterly.

[0196] Select frequency conversion method: Select the appropriate frequency conversion method according to the characteristics and requirements of the data. Common methods include summation, averaging, interpolation, etc.

[0197] Apply frequency conversion algorithm: Apply frequency conversion algorithm to the forecast values ​​of the predictor and benchmark indicators to get the forecast values ​​corresponding to the target frequency. For example, if you need data from monthly to quarterly, you may need to sum or average the forecast values ​​for three months.

[0198] Verify results: Check whether the data after frequency conversion processing is reasonable and ensure that no errors or deviations are introduced during the conversion process.

[0199] Step 2: Seasonal Restore

[0200] Seasonal reduction refers to removing seasonal effects from time series data in order to more accurately analyze and predict non-seasonal trends and patterns. Here are the specific steps for seasonal reduction:

[0201] Identify seasonal patterns: Analyze historical data to identify and understand seasonal patterns in the data. This may involve calculating and comparing data for different time periods (such as the same period of the year).

[0202] Seasonal adjustment: Use statistical methods (such as seasonal indices) to adjust forecasts to reflect historical seasonal patterns. This usually involves removing seasonal effects from the forecasts in order to produce seasonally neutral data.

[0203] Apply seasonality reduction: seasonality reduction is applied to the forecast values ​​of the predictor and the benchmark indicator to obtain the corresponding forecast values ​​under the target seasonal pattern. This may involve dividing the forecast values ​​by the corresponding seasonal index.

[0204] Validation and interpretation: Check whether the seasonally restored data follows the expected seasonal pattern and ensure the interpretability of the results.

[0205] Suppose an automobile manufacturer uses a dynamic factor model to forecast sales for the next few months. The output of the model is monthly sales forecasts. For quarterly planning, the manufacturer needs to convert these monthly forecasts into quarterly forecasts. They might sum the sales forecasts for three months to get a forecast for total quarterly sales.

[0206] In addition, if the sales data shows significant seasonal fluctuations (such as a holiday sales peak), the manufacturer may also need to seasonally restore the forecast values ​​to more accurately predict sales in future seasons. They may use historical seasonality indices to adjust the forecast values ​​to obtain a forecast result after eliminating the seasonal effects.

[0207] By following these steps, manufacturers can obtain more accurate and useful forecast information to support their production planning, inventory management and marketing strategies.

[0208] See also Figure 2 , Figure 2 Schematic diagram of the structure of the indicator data prediction device in the embodiment of the present application. Figure 2 As shown, specifically:

[0209] A factor tree construction unit 201 is used to construct a prediction factor tree, wherein the prediction factor tree includes a plurality of prediction factors, and the prediction factors are connected according to the influence mode and the conduction chain relationship;

[0210] A first matching degree calculation unit 202 is used to determine a benchmark indicator, calculate a first matching degree between the factor data of each prediction factor and the benchmark indicator, and screen and generate a modeling overall factor library;

[0211] The second matching degree calculation unit 203 is used to calculate the second matching degree between the factor data of each prediction factor in the overall modeling factor library and the benchmark index within the prediction period, and screen and generate the current modeling factor library;

[0212] A model building unit 204 is used to build a dynamic factor model using the prediction factors and benchmark indicators included in the current modeling factor library as variables;

[0213] The model iteration unit 205 is used to iteratively converge the dynamic factor model through an expectation maximization algorithm to obtain model parameters of the dynamic factor model after convergence;

[0214] The prediction unit 206 is used to calculate the prediction values ​​of each prediction factor and the benchmark index contained in the current modeling factor library at the prediction time according to the model parameters.

[0215] Another embodiment of the present invention provides an indicator data prediction device, such as Figure 3 As shown, the device 10 includes:

[0216] One or more processors 110 and memory 120, Figure 3 A processor 110 is used as an example for description. The processor 110 and the memory 120 may be connected via a bus or other means. Figure 3 The example of connecting through bus is taken in the following.

[0217] The processor 110 is used to complete various control logics of the device 10, and it can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), a single-chip microcomputer, an ARM (Acorn RISCMachine) or other programmable logic device, a discrete gate or transistor logic, a discrete hardware component, or any combination of these components. In addition, the processor 110 can also be any traditional processor, microprocessor or state machine. The processor 110 can also be implemented as a combination of computing devices, for example, a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors combined with a DSP and / or any other such configuration.

[0218] The memory 120 is a non-volatile computer-readable storage medium that can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions corresponding to the method for constructing a multilingual phoneme representation model in the embodiment of the present invention. The processor 110 executes various functional applications and data processing of the device 10 by running the non-volatile software programs, instructions and units stored in the memory 120, that is, implements the method for constructing a multilingual phoneme representation model in the above method embodiment.

[0219] The memory 120 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and applications required for at least one function; the data storage area may store data created according to the use of the device 10, etc. In addition, the memory 120 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 120 may optionally include a memory remotely arranged relative to the processor 110, and these remote memories may be connected to the device 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0220] One or more units are stored in the memory 120, and when executed by one or more processors 110, the following steps are implemented:

[0221] Construct a prediction factor tree, which contains several prediction factors, and the prediction factors are connected according to the influence mode and the transmission chain relationship;

[0222] Determine the benchmark indicator, calculate the first matching degree between the factor data of each predictor and the benchmark indicator, and screen and generate the overall factor library for modeling;

[0223] During the forecast period, the second matching degree between the factor data of each forecast factor in the overall modeling factor library and the benchmark index is calculated, and the current modeling factor library is screened and generated;

[0224] The dynamic factor model is constructed by taking the predictive factors and benchmark indicators contained in the current modeling factor library as variables;

[0225] The dynamic factor model is iteratively converged through the expectation maximization algorithm to obtain the model parameters of the dynamic factor model after convergence;

[0226] The predicted values ​​of each prediction factor and benchmark indicator contained in the current modeling factor library at the prediction time are calculated based on the model parameters.

[0227] An embodiment of the present invention provides a non-volatile computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are executed by one or more processors, any implementation of an indicator data prediction method described in the above embodiments is implemented.

[0228] As an example, non-volatile storage media can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) as external cache memory. By way of illustration and not limitation, RAM can be obtained in many forms such as synchronous RAM (SRAM), dynamic RAM, (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), and direct Rambus RAM (DRRAM). The disclosed memory components or memories of the operating environment described herein are intended to include one or more of these and / or any other suitable types of memory.

[0229] In the embodiments of the present application, a method, apparatus, device and readable storage medium for predicting indicator data are provided. By building a tree-type multi-factor system, a dynamic factor model is constructed and trained based on a quantitative screening mechanism of factor data and benchmark indicators. By iterating the dynamic factor model, the predicted value of the benchmark indicator is obtained within the prediction period. The rich prediction factors can be matched and transformed according to the needs of different benchmark indicators, and can cope with indicator prediction in various scenarios. At the same time, the quantitative screening mechanism avoids the subjectivity of human selection and reduces dependence on manual labor. It solves the technical problems of the existing historical sales trend continuation method ignoring external factors that may affect sales, and the micro-expectation summary method having difficulty in information collection and labor consumption, which are easy to lead to deviations in the prediction results.

[0230] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0231] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein, for example. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0232] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0233] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0234] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0235] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0236] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (full name in English: Read-Only Memory, English abbreviation: ROM), random access memory (full name in English: Random Access Memory, English abbreviation: RAM), disk or optical disk and other media that can store program codes.

[0237] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

[0238] It should be noted that if software tools or components other than those of our company appear in the embodiments of the present application, they are only used for illustrative purposes and do not represent actual use.

Claims

1. A method for predicting indicator data, characterized in that: include: Constructing a prediction factor tree, wherein the prediction factor tree includes a plurality of prediction factors, and the prediction factors are connected according to influence modes and conduction chain relationships; Determine a benchmark indicator, calculate a first matching degree between the factor data of each of the prediction factors and the benchmark indicator, and screen and generate a total factor library for modeling; During the forecast period, the second matching degree between the factor data of each of the forecast factors in the overall modeling factor library and the benchmark indicator is calculated, and the current modeling factor library is screened and generated; Constructing a dynamic factor model using the prediction factors and the benchmark indicators contained in the current modeling factor library as variables; Iteratively converge the dynamic factor model through an expectation maximization algorithm to obtain model parameters of the dynamic factor model after convergence; The predicted values ​​of each of the prediction factors and the benchmark indicators contained in the current modeling factor library at the prediction time are calculated based on the model parameters.

2. The indicator data prediction method according to claim 1, characterized in that: The constructing of a prediction factor tree, wherein the prediction factor tree includes a plurality of prediction factors, wherein the prediction factors are connected according to the influence mode and the conduction chain relationship, further includes: Counting the data update frequency and the amount of historical data of the factor data of each of the prediction factors in the prediction factor tree; The prediction factors used for predicting indicator data are screened and determined according to the data update frequency and the amount of historical data.

3. The indicator data prediction method according to claim 1, characterized in that: The constructing of a prediction factor tree, wherein the prediction factor tree includes a plurality of prediction factors, wherein the prediction factors are connected according to the influence mode and the conduction chain relationship, further includes: Performing data preprocessing on the factor data included in each of the prediction factors; The preprocessed factor data is standardized and converted into a target data format.

4. The indicator data prediction method according to claim 1, characterized in that: The step of calculating the first matching degree between the factor data of each of the prediction factors and the benchmark indicator and screening and generating the overall factor library for modeling is specifically as follows: For a plurality of preset evaluation dimensions, a first matching degree between the factor data of each of the prediction factors and the benchmark indicator is calculated respectively by using a plurality of preset models; Counting the statistical values ​​of the first matching degree of each of the prediction factors in all the preset evaluation dimensions; After sorting all the prediction factors in descending order according to the statistical value of the first matching degree, a target number of the prediction factors are selected in order to generate a modeling overall factor library.

5. The indicator data prediction method according to claim 1, characterized in that: The construction of the dynamic factor model using the prediction factors and the benchmark indicators contained in the current modeling factor library as variables is specifically as follows: Taking the prediction factors and the benchmark variables contained in the current modeling factor library as variables, performing data preprocessing on the variables; Constructing a dynamic factor model and determining the dynamic relationship and time series between the variables in the dynamic factor model; Perform initial random assignment of model parameters to the dynamic factor model.

6. The indicator data prediction method according to claim 5, characterized in that: The dynamic factor model is iteratively converged by the expectation maximization algorithm, and the model parameters of the dynamic factor model after convergence are specifically: Calculate the first conditional expectation of the latent factors of the dynamic factor model by Kalman filtering method; Calculating a second conditional expectation of a preset error term of the dynamic factor model; According to the first conditional expectation and the second conditional expectation, updating the model parameters of the dynamic factor model by maximizing the log-likelihood function of the observed data; Based on a preset optimization algorithm, optimizing and adjusting the model parameters of the dynamic factor model; If the model parameters converge or the current iteration round reaches the preset number of iterations, the converged model parameters of the dynamic factor model are obtained, otherwise the first conditional expectation of the latent factors of the dynamic factor model is returned to be recalculated.

7. The indicator data prediction method according to claim 1, characterized in that: After obtaining the predicted values ​​of the prediction factors and the benchmark indicators contained in the current modeling factor library at the prediction time by calculating the model parameters, the following further comprises: Performing frequency conversion processing on the prediction factor and the prediction value of the benchmark indicator by a preset frequency conversion method to obtain a prediction value corresponding to a target frequency; The predicted values ​​of the prediction factors and the benchmark indicators are seasonally restored to obtain corresponding predicted values ​​under the target seasonal pattern.

8. An indicator data prediction device, characterized in that: include: A factor tree construction unit, used to construct a prediction factor tree, wherein the prediction factor tree includes a plurality of prediction factors, and the prediction factors are connected according to influence modes and conduction chain relationships; A first matching degree calculation unit is used to determine a benchmark indicator, calculate a first matching degree between the factor data of each of the prediction factors and the benchmark indicator, and screen and generate a modeling overall factor library; A second matching degree calculation unit is used to calculate the second matching degree between the factor data of each of the prediction factors in the overall modeling factor library and the benchmark indicator within the prediction period, and screen and generate the current modeling factor library; A model building unit, used to build a dynamic factor model using the prediction factors and the benchmark indicators contained in the current modeling factor library as variables; A model iteration unit, used for iteratively converging the dynamic factor model through an expectation maximization algorithm to obtain model parameters of the dynamic factor model after convergence; The prediction unit is used to calculate the predicted values ​​of each prediction factor and the benchmark indicator contained in the current modeling factor library at the prediction time according to the model parameters.

9. An indicator data prediction device, characterized in that: The device comprises a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the indicator data prediction method described in any one of claims 1-7 according to the instructions in the program code.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store program code, and the program code is used to execute the indicator data prediction method according to any one of claims 1-7.

Images