WO2020211245A1 - Development trend data acquisition method and device - Google Patents

Abstract

A development trend data acquisition method and device. The method comprises: determining an object category to which a prediction object belongs; determining at least one factor corresponding to the object category; taking at least two objects belonging to the object category as sample objects, and separately obtaining historical development data of each sample object; separately extracting first historical factor data corresponding to each factor from within the historical development data of each sample object; by means of each piece of the extracted first historical factor data, training a machine learning model corresponding to the object category; extracting second historical factor data corresponding to each factor from within the historical development data of the prediction object; inputting each piece of second historical factor data into the machine learning model to obtain first prediction data; and according to the first prediction data, determining development trend data used for representing the development trend of the prediction object. The costs paid by analysis personnel for predicting the revenue of publicly listed companies may thus be reduced.

Description

Method and device for acquiring development trend data Technical field

This application relates to the field of data processing technology, and in particular to a method and device for acquiring development trend data.

Background technique

Listed companies are obliged to forecast their revenues and make public announcements. Therefore, analysts of listed companies need to make regular revenue forecasts for their listed companies. Under normal circumstances, analysts forecast the revenue of listed companies on a quarterly basis.

At present, analysts mainly use the following methods to predict the revenue of listed companies: determine multiple factors that have a direct impact on the revenue of the listed company according to the industry to which the listed company belongs, and use the real-time data of the determined factors to perform linear simulation Or polynomial fitting, and then predict the revenue of listed companies based on the fitted linear function or polynomial function.

According to the current method of forecasting the revenue of listed companies, the real-time data of each factor is used for linear fitting or polynomial fitting. Therefore, the timeliness of the data corresponding to each factor has a great influence on the forecast results. In order to ensure the revenue forecast Accuracy, analysts need to collect real-time data corresponding to each factor in a timely manner, resulting in higher costs for analysts to pay for the revenue forecast of listed companies.

Summary of the invention

This application provides a method and device for acquiring development trend data, the main purpose of which is to use the historical development data of multiple sample objects to train a machine learning model, and to obtain the machine by inputting historical development data exclusive for prediction into the trained machine learning model The first prediction data output by the learning model is further used to determine the development trend data of the prediction object according to the first prediction data. When applying this development trend data acquisition method to the revenue forecast of listed companies, it is not necessary for analysts to collect real-time data corresponding to the listed company in time, thereby reducing the cost of analysts’ revenue forecasting of listed companies.

In the first aspect, an embodiment of the present application provides a method for obtaining development trend data, including:

Determine the object category to which the predicted object belongs;

Determine at least one factor corresponding to the object category, wherein different factors correspond to different data statistics rules;

Taking at least two objects belonging to the object category as sample objects, and obtaining historical development data of each of the sample objects respectively;

Extracting the first historical factor data corresponding to each of the factors from the historical development data of each of the sample objects;

Training a machine learning model corresponding to the object category through each of the extracted first historical factor data;

Extracting second historical factor data corresponding to each of the factors from the historical development data of the prediction object;

Input each of the second historical factor data into the machine learning model to obtain the first prediction data output by the machine learning model;

The development trend data used to characterize the development trend of the prediction object is determined according to the first prediction data.

Optionally,

Before determining the development trend data used to characterize the development trend of the prediction object according to the first prediction data, the method further includes:

Fitting a polynomial function with historical development data of at least two of the sample objects, wherein the historical development data of each sample object satisfies the polynomial function;

Input the historical development data of the prediction object into the polynomial function to obtain second prediction data output by the polynomial function;

The determining development trend data used to characterize the development trend of the prediction object according to the first prediction data includes:

The development trend data of the prediction object is determined according to the first prediction data and the second prediction data.

Optionally,

Before the determining the development trend data of the prediction object according to the first prediction data and the second prediction data, the method further includes:

Fitting a time series model with the historical development data of at least two of the sample objects, wherein the change law of the historical development data of each of the sample objects over time conforms to the time series model;

Input the historical development data of the prediction object into the time series model to obtain the third prediction data output by the time series model;

The determining the development trend data of the prediction object according to the first prediction data and the second prediction data includes:

The development trend data of the prediction object is determined according to the first prediction data, the second prediction data, and the third prediction data.

Optionally,

The fitting a polynomial function using historical development data of at least two of the sample objects includes:

Determine the forecast period for forecasting the development trend of the forecast object;

Extracting first historical development data corresponding to each statistical period from the historical development data of each sample object according to the prediction period, wherein the statistical period corresponds to the prediction period in a time span;

According to each of the first historical development data corresponding to each of the statistical periods, the following polynomial function is fitted, where each of the first historical development data corresponding to each of the sample objects satisfies the polynomial function;

Wherein, the M represents the second prediction data relative to the current time; the k _i represents the weight coefficient fitted by machine learning; the x represents the previous statistical period relative to the current time Corresponding to the first historical development data; the x _i represents the first historical development data corresponding to the last i+1 statistical periods relative to the current time; the t+1 represents the current The number of statistical periods before the time.

Optionally,

The utilizing the historical development data of at least two of the sample objects to fit a time series model includes:

Determine the forecast period for forecasting the development trend of the forecast object;

Extracting the second historical development data corresponding to each statistical period from the historical development data of each sample object according to the prediction period, wherein the statistical period corresponds to the prediction period in a time span;

A time series model is fitted according to each of the second historical development data corresponding to each of the statistical periods, wherein the change rule over time of each of the second historical development data corresponding to each of the sample objects satisfies the time series model;

The form of the time series model is as follows:

(ΔM _t ) ² =K+k ₁ (ΔM _t-1 ) ² -k ₂ (ΔM _t-2 ) ² +ε _t -k ₃ ε _t-1

Wherein, the ΔM _t characterizes the difference between the third prediction data relative to the current time and the second historical development data corresponding to the last statistical period of the current time; the ΔM _t-1 characterizes the difference The difference between the second historical development data corresponding to the last statistical period of the current time and the second historical development data corresponding to the second statistical period before the current time; the ΔM _{t- 2} characterizing the difference between the second historical development data corresponding to the second statistical period before the current time and the second historical development data corresponding to the third statistical period before the current time; The ε _t characterizes the third prediction data relative to the current time; the ε _t-1 characterizes the second historical development data corresponding to the last statistical period of the current time; the K , The k ₁ , the k ₂ and the k ₃ are all weight coefficients fitted by machine learning.

Optionally, the fitting a time series model according to each second historical development data corresponding to each statistical period includes:

Performing a second difference on each of the second historical development data corresponding to each of the statistical periods to obtain a corresponding difference sequence;

According to the difference sequence, a list method is used to define the target equation corresponding to the model;

Solving the target equation to obtain an estimation result of the model;

Detecting the fitting effect of the model based on the goodness of fit;

After determining that the fitting effect of the model reaches a preset target, detecting the residual of the model;

When it is determined that the residual fluctuation of the model is within a preset fluctuation range, the model is determined as the time series model.

Optionally, the determining the development trend data of the prediction object according to the first prediction data, the second prediction data, and the third prediction data includes:

Perform a weighting operation on the first prediction data, the second prediction data, and the third prediction data to obtain the development trend data of the prediction object.

Optionally, the training of the machine learning model corresponding to the object category through each of the extracted first historical factor data includes:

For each of the factors, obtain at least one factor data corresponding to the factor in each of the past at least two years from the first historical factor data corresponding to the factor;

Using the factor data corresponding to each of the factors as a sample to train the factor coefficients corresponding to each of the factors;

Use each of the acquired factor coefficients to construct the following formula for calculating the first prediction data;

Wherein, the M′ represents the first prediction data; the n represents the number of the factors; the m represents the number of historical years covered by the first historical factor data; the x _{(i, 1)} Characterizing the factor data corresponding to the i-th factor of the predicted object in the previous year; the x _{(i, 2)} characterizing the factor corresponding to the i-th factor in the previous year of the predicted object Data; the k _i characterizes the factor coefficient corresponding to the i-th factor at the current time; the x _{(i, j)} characterizes the factor data corresponding to the i-th factor in the previous j-th year of the prediction object;

The machine learning model including the formula is constructed.

In the second aspect, an embodiment of the present application also provides a development trend data acquisition device, including:

Category recognition module, factor recognition module, data acquisition module, first data extraction module, model training module, second data extraction module, model processing module, and data processing module;

The category recognition module is used to determine the object category to which the predicted object belongs;

The factor identification module is configured to determine at least one factor corresponding to the object category determined by the category identification module, wherein different factors correspond to different data statistics rules;

The data acquisition module is configured to use at least two objects belonging to the object category determined by the category recognition module as sample objects, and obtain historical development data of each sample object respectively;

The first data extraction module is configured to extract each of the factor corresponding to each of the factors determined by the factor identification module from the historical development data of each of the sample objects and acquired by the data acquisition module. The first historical factor data;

The model training module is configured to train a machine learning model corresponding to the object category through each of the first historical factor data extracted by the first data extraction module;

The second data extraction module is configured to extract second historical factor data corresponding to each of the factors by the factor identification module from the historical development data of the prediction object;

The model processing module is configured to input each of the second historical factor data extracted by the second data extraction module into the machine learning model trained by the model training module to obtain the output of the machine learning model First forecast data;

The data processing module is configured to determine development trend data used to characterize the development trend of the prediction object according to the first prediction data acquired by the model processing module.

In a third aspect, an embodiment of the present application also provides a computer device, including a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, it implements any of the foregoing The development trend data acquisition method.

In a fourth aspect, embodiments of the present application also provide a non-volatile computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the computer program described in any of the above-mentioned first aspects is implemented. Development trend data acquisition method.

The development trend data acquisition method, device, computer equipment, and non-volatile computer-readable storage medium provided by the embodiments of the present application determine the object category to which the predicted object belongs, and then determine one or more factors corresponding to the object category. Obtain the historical development data of at least two sample objects belonging to the object category, extract the first historical factor data corresponding to each factor from the historical development data of each sample object, and extract each historical development data from the predicted object One factor corresponds to the second historical factor data, and then each first historical factor data is used to train the machine learning model corresponding to the object category to which the predicted object belongs. After each second historical factor data is input into the trained machine learning model, the first The forecast data can then determine the development trend data of the forecast object according to the first forecast data. It can be seen that when listed companies are used as forecast objects and development trend data are used as revenue forecast data, historical revenue data of multiple sample companies belonging to the same industry category as listed companies are used to predict the revenue of listed companies , The timeliness requirements of the first historical factor data corresponding to each factor are low, so there is no need for analysts to collect real-time data corresponding to each factor in time, which can reduce the cost of analysts for revenue forecasting of listed companies.

Description of the drawings

In order to more clearly describe the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the drawings in the following description are For some of the embodiments of the application, for those of ordinary skill in the art, other drawings may be obtained based on these drawings without creative work.

FIG. 1 is a flowchart of a method for forecasting the revenue of a listed company according to an embodiment of the present application;

2 is a flowchart of a polynomial function fitting method provided by an embodiment of the present application;

FIG. 3 is a flowchart of a time series model fitting method provided by an embodiment of the present application;

4 is a flowchart of another method for forecasting the revenue of a listed company according to an embodiment of the present application;

Fig. 5 is a schematic diagram of a revenue forecasting device for listed companies provided by an embodiment of the present application.

detailed description

In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments These are part of the embodiments of this application, not all of them. Based on the examples of this application, all other embodiments obtained by those of ordinary skill in the art without creative work are protected by this application. range. It should be understood that the specific embodiments described here are only used to explain the application, and are not used to limit the application.

In the following, the forecast object is a listed company and the development trend data is the revenue forecast data of the listed company as an example, and the method and device for acquiring the development trend data provided by the embodiments of the present application will be described in detail. Specifically, the method corresponding to the method for acquiring development trend data is a method for forecasting the revenue of listed companies, and the method corresponding to the method for acquiring development trend data is a method for forecasting the revenue of listed companies.

As shown in Figure 1, an embodiment of the present application provides a method for forecasting the revenue of a listed company, including:

Step 101: Determine the industry category of the listed company that needs to perform revenue forecasting;

Step 102: Determine at least one factor corresponding to the industry category, where different factors correspond to different data statistics rules;

Step 103: Take at least two companies belonging to the industry category as sample companies, and obtain historical revenue data of each sample company respectively;

Step 104: Extract the first historical factor data corresponding to each factor from the historical revenue data of each sample company;

Step 105: Train a machine learning model corresponding to the industry category through each extracted first historical factor data;

Step 106: Extract the second historical factor data corresponding to each factor from the historical revenue data of the listed company;

Step 107: Input each second historical factor data into the machine learning model, and obtain the first revenue prediction result output by the machine learning model;

Step 108: Determine the predicted revenue data of the listed company according to the first revenue prediction result.

According to the method for forecasting the revenue of listed companies provided by the embodiments of the application, after determining the industry category of the listed company that needs to perform revenue forecasting, one or more factors corresponding to the industry category are determined, and then at least two of the industry categories are obtained Sample company’s historical revenue data, and extract the first historical factor data corresponding to each factor from the historical revenue data of each sample company, and extract the second historical factor corresponding to each factor from the historical revenue data of listed companies Historical factor data, and then use each first historical factor data to train the machine learning model corresponding to the industry category of the listed company. After each second historical factor data is input into the trained machine learning model, the first revenue forecast result is obtained, and then The forecasted revenue data of listed companies can be determined according to the first revenue forecast result. It can be seen that using the historical revenue data of multiple sample companies belonging to the same industry category as the listed company to predict the revenue of the listed company, the timeliness of the first historical factor data corresponding to each factor is relatively low, so There is no need for analysts to collect real-time data corresponding to each factor in time, which can reduce the cost of analysts’ revenue forecasts for listed companies.

In the embodiments of the present application, when determining at least one factor corresponding to the industry category of the listed company, it needs to be performed according to the cycle of revenue forecasting of the industry category of the listed company. For example, the industry category of a listed company usually collects revenue data on a quarterly basis. When forecasting the revenue of a listed company, the revenue data of the listed company in the next quarter is usually used. Correspondingly, last quarter's revenue, last quarter's total assets, last year's same quarter's revenue, and last year's total assets were identified as the four factors corresponding to the industry category of the listed company.

For example, after determining the last quarter’s revenue, last quarter’s total assets, last year’s same quarter revenue, and last year’s total assets as four factors, 3,000 companies belonging to the industry category of listed companies are determined as sample companies and obtained The historical revenue data of each sample company in the past ten years, and then the historical revenue data obtained from each sample company’s quarterly revenue and quarterly total assets in each of the past ten years are obtained as the first Historical factor data. After that, the machine learning model is trained through the extracted 240,000 (3000*10*4*2) first historical factor data, and the machine learning model corresponding to the industry category of the listed company is obtained. After that, the last quarter's revenue, total assets of the last quarter, revenue of the same quarter last year, and total assets of the same quarter last year of the listed company to be forecasted are input into the machine learning model, and the next quarter forecasted revenue of the listed company output from the machine learning model is obtained The data is used as the first revenue forecast result.

Optionally, on the basis of the revenue forecasting method for listed companies shown in Figure 1, since each first historical revenue data reflects the previous revenue of each sample company, the corresponding first historical revenue data can be determined to correspond to The factor coefficients of different factors use the determined factor coefficients to reflect the revenue trend of each sample company over the years, and then the determined factor coefficients can be used to construct a machine learning model. The two historical factor data are processed to predict the revenue of the listed company and obtain the first revenue prediction result. The specific method of constructing a machine learning model may include the following steps:

S1: For each factor, obtain at least one factor data corresponding to the factor in each of the past at least two years from the first historical factor data corresponding to the factor;

S2: Use the acquired factor data as a sample to train the factor coefficients corresponding to each factor;

S3: Use the obtained factor coefficients to construct the following company for calculating the first revenue forecast result;

Among them, M′ represents the first revenue forecast result; n represents the number of factors; m represents the number of historical years covered by the first historical factor data; x _{(i, 1)} represents the first year of the listed company corresponding to the first factor data of the i factor; x _{(i, 2)} represents the factor data corresponding to the i-th factor in the second year of the listed company; k _i represents the factor coefficient corresponding to the i-th factor at the current time; x _{(i, j )} Characterize the factor data corresponding to the i-th factor in the j-th year of the listed company;

S4: Build a machine learning model including the above formula.

For example, the four factors identified for listed companies are the revenue of the previous quarter, the total assets of the previous quarter, the revenue of the same quarter last year, and the total assets of the same quarter last year. The first historical factor data obtained is the data of 3000 companies in the past 10 years. In terms of revenue data, a total of 60,000 factor data of 10*3000*2 can be determined based on the revenue factor of the previous quarter. Accordingly, the total assets of the previous quarter, the revenue of the same quarter last year and the total assets of the same quarter last year can be determined accordingly. 6 factor data are obtained. After that, the 240,000 factor data is used as sample data for machine learning, and 4 factor coefficients corresponding to the above 4 factors are fitted. After substituting the fitted four factor coefficients into the above formula, and substituting the historical revenue data corresponding to the above four factors of the previous years of the company into the above formula, the first revenue forecast for listed companies can be calculated. Revenue forecast results.

In the embodiment of the present application, the historical revenue data of each sample company is used to fit the factor coefficients corresponding to each factor, and the fitted factor coefficients are then used to construct a machine learning model corresponding to the industry category of the listed company. The machine learning model Reflects the revenue change trend of the industry category of the listed company, and then can use the constructed machine learning model to predict the revenue of the listed company. Because of the reference to the revenue change trend of other companies in the same industry category and the listed company to be predicted Historical revenue situation, which can more accurately predict the revenue of listed companies.

Optionally, by training the machine learning model, and inputting the second historical factor data extracted from the historical revenue data of the listed company into the machine learning model to obtain the first revenue prediction result, then the first revenue prediction can be directly The result is used as the forecasted revenue data of the listed company, and the forecasted revenue data of the listed company can also be determined by combining the revenue forecast results obtained through other forecasting methods. Regarding the method of determining forecasted revenue data by combining the revenue forecast results obtained by other forecasting methods, the following two methods can be used to determine the forecasted revenue data of listed companies:

Method 1: Combine the first revenue prediction result obtained through the machine learning model with the second revenue prediction result obtained through a polynomial function to determine the predicted revenue data of the listed company;

Method 2: Combine the first revenue forecast result obtained through the machine learning model, the second revenue forecast result obtained through the polynomial function, and the third revenue forecast result obtained through the time series model to determine the forecast of the listed company Revenue data.

The above two methods of determining forecasted revenue data by combining multiple revenue forecasting results are described below.

For method one:

On the basis of the listed company revenue forecasting method shown in Figure 1, after obtaining the first revenue prediction result through the machine learning model in step 107, and in step 108, determine the listed company’s predicted revenue data according to the first revenue prediction result Previously, the historical revenue data of each sample company can be used to fit a polynomial function, so that the historical revenue data of each sample company meets the polynomial function obtained by fitting, and then the historical revenue data of listed companies can be input to fit a polynomial function. The polynomial function of to obtain the second revenue forecast result output by the polynomial function. Correspondingly, in step 108, when determining the predicted revenue data of the listed company according to the first revenue forecast result, the predicted revenue data of the listed company may be determined according to the first revenue forecast result and the second revenue forecast result, which can be specifically calculated The weighted average of the first revenue forecast result and the second revenue forecast result, and the calculated weighted average is used as the forecast revenue data of the listed company.

Fit the polynomial function according to the historical revenue data of each sample company, and then input the historical revenue data of the listed company into the polynomial function to obtain the second revenue forecast result, and then according to the first revenue forecast result and the second revenue forecast result To determine the forecasted revenue data of listed companies, due to the combination of machine learning model and polynomial function two revenue forecasting methods to predict the revenue of listed companies, it can improve the accuracy of forecasting the revenue of listed companies.

In the embodiment of the present application, the historical revenue data of each sample company is used to fit the polynomial function. As shown in FIG. 2, the process of fitting the polynomial function can be realized by the following steps:

Step 201: Determine the forecast period for the listed company's revenue forecast;

Step 202: Extract the first historical revenue data corresponding to each statistical period from the historical revenue data of each sample company according to the determined prediction period, where the statistical period corresponds to the prediction period in a time span;

Step 203: Fit the following polynomial function according to each first historical revenue data corresponding to each statistical period, so that each first historical revenue data corresponding to each sample company satisfies the polynomial function;

Among them, M represents the revenue prediction result relative to the current time; k _i represents the weight coefficient fitted by machine learning; x represents the first historical revenue data corresponding to the previous statistical period relative to the current time; x _i represents The first historical revenue data corresponding to the last i+1 statistical period relative to the current time; t+1 represents the number of statistical periods before the current time.

For example, now it is necessary to forecast the revenue of listed companies in the next quarter, the forecast period for the revenue forecast of listed companies is quarter. Correspondingly, the statistical period for the historical revenue data of each sample company is also quarterly. . For each sample company, the historical revenue of the sample company is obtained from the historical revenue data of the sample company as the first historical revenue data. By fitting each first historical revenue data corresponding to each sample company, a polynomial function that can satisfy each first historical revenue data corresponding to each sample company is obtained.

After fitting the polynomial function, the listed company’s data corresponding to each statistical period is obtained from the listed company’s historical revenue data, and then the obtained data corresponding to each statistical period is input into the polynomial function to obtain the output of the polynomial function. 2. Revenue forecast results. For example, after determining that the forecast period for the listed company’s revenue forecast is quarterly, the listed company’s historical revenue data is obtained from the listed company’s historical revenue data for each quarter, and then the obtained listed company’s historical revenue data Input the polynomial function fitted to the revenue data of a quarter to obtain the second revenue forecast result corresponding to the listed company.

After the second revenue prediction result is obtained through the fitted polynomial function, the first revenue prediction result and the second revenue prediction result are calculated according to the weight values set in advance for the first revenue prediction result and the second revenue prediction result. The revenue forecast results are weighted and the calculation results are used as the forecast revenue data of the listed company. Specifically, the weight values for the first revenue prediction result and the second revenue prediction result can be set to 0.5, that is, the weighted average of the first revenue prediction result and the second revenue prediction result is taken as the listed company’s Forecast revenue data.

Targeting method two:

On the basis of the method for determining forecast revenue data provided in the above method, after obtaining the first revenue forecast result through the machine learning model and the second revenue forecast result through the polynomial function, the historical revenue of each sample company can be used The data is fitted to the time series model, so that the historical revenue data of each sample company conforms to the fitted time series model, and then the historical revenue data of the listed company is input into the fitted time series model. Obtain the third revenue forecast result output by the time series model. Correspondingly, the predicted revenue data of the listed company can be determined according to the first revenue prediction result, the second revenue prediction result, and the third revenue prediction result.

According to the historical revenue data of each sample company, the machine learning model, polynomial function and time series model are obtained, and then the historical revenue data of listed companies are input into the machine learning model, polynomial function and time series model, respectively, to obtain the corresponding The first revenue forecast result, the second revenue forecast result, and the third revenue forecast result, and then the forecasted revenue of the listed company is determined based on the first revenue forecast result, the second revenue forecast result, and the third revenue forecast result. Receive data. Because of the combination of machine learning model, polynomial function and time series model three revenue forecasting methods to predict the revenue of listed companies, the accuracy of forecasting the revenue of listed companies can be further improved.

In the embodiment of the present application, the historical revenue data of each sample company can be used to fit the time series model. As shown in FIG. 3, the process of fitting the time series model can be realized by the following steps:

Step 301: Determine the forecast period for the listed company's revenue forecast;

Step 302: Extract the second historical revenue data corresponding to each statistical period from the historical revenue data of each sample company according to the prediction period, where the statistical period corresponds to the prediction period in a time span;

Step 303: Fit a time series model according to each second historical revenue data corresponding to each statistical period, so that the time series model of each second historical revenue data corresponding to each sample company meets the time series model;

The form of the time series model is as follows:

(ΔM _t ) ² =K+k ₁ (ΔM _t-1 ) ² -k ₂ (ΔM _t-2 ) ² +ε _t -k ₃ ε _t-1

Among them, ΔM _t represents the difference between the revenue forecast results relative to the current time and the second historical revenue data corresponding to the previous statistical period of the current time; ΔM _t-1 represents the second historical data corresponding to the previous statistical period of the current time The difference between the revenue data and the second historical revenue data corresponding to the second statistical period before the current time; ΔM _t-2 represents the second historical revenue data corresponding to the second statistical period before the current time and the current time before The difference between the second historical revenue data corresponding to the third statistical period; ε _t represents the revenue forecast result relative to the current time; ε _t-1 represents the second historical revenue data corresponding to the previous statistical period at the current time ; K, k ₁ , k ₂ and k ₃ are weight coefficients fitted by machine learning.

For example, now it is necessary to forecast the revenue of listed companies in the next quarter, the forecast period for the revenue forecast of listed companies is quarter. Correspondingly, the statistical period for the historical revenue data of each sample company is also quarterly. . For each sample company, the historical revenue of the sample company is obtained from the historical revenue data of the sample company as the second historical revenue data. By fitting the second historical revenue data corresponding to each sample company, a time series model that can satisfy the change rule of each second historical revenue data of each sample company over time is obtained.

After fitting the time series model, obtain the listed company's data corresponding to each statistical period from the listed company's historical revenue data, and then input the obtained data corresponding to each statistical period into the time series model to obtain the time series model Output the third revenue forecast result. For example, after determining that the forecast period for the listed company’s revenue forecast is quarterly, the listed company’s historical revenue data is obtained from the listed company’s historical revenue data for each quarter, and then the obtained listed company’s historical revenue data The revenue data of a quarter is entered into the fitted time series model, and the third revenue forecast result corresponding to the listed company is obtained.

It should be noted that, in the actual business realization process, the first historical revenue data for the user fitting a polynomial function and the second historical revenue data for fitting the time series model may be the same data.

Optionally, on the basis of the method for determining the predicted revenue data of listed companies provided in the second method, after obtaining the first revenue prediction result, the second revenue prediction result, and the third revenue prediction result, the The first revenue prediction result, the second revenue prediction result, and the third revenue prediction result are weighted, and the result of the weighted calculation is used as the predicted revenue data of the listed company.

Specifically, corresponding weighting coefficients can be set for the first revenue forecast result, the second revenue forecast result, and the third revenue forecast result according to actual needs to control the weight of each revenue forecast result to the final forecast revenue data . For example, the weighting coefficient corresponding to the first revenue prediction result can be set to 0.4, the weighting coefficient corresponding to the second revenue prediction result is set to 0.3, and the weighting coefficient corresponding to the third revenue prediction result is set to 0.3.

Set the corresponding weighting coefficients for the first revenue forecast result, the second revenue forecast result and the third revenue forecast result respectively, and the first revenue forecast result and the second revenue forecast result according to the preset weighting coefficient The result and the third revenue forecast result are weighted, and the result of the weighted calculation is used as the forecast revenue data of the listed company. By setting up a weighting system, on the one hand, it is possible to balance the impact of the three revenue forecast results on the calculation results, weigh the pros and cons of the three revenue forecast methods, and make the final forecasted revenue data more accurate. Adjust the weighting coefficient according to the needs, so as to meet the individual needs of different users, and improve the applicability of the listed company's revenue forecasting method.

Optionally, on the basis of the method of fitting the time series model shown in FIG. 3, step 303 fits the time series model according to each second historical revenue data corresponding to each statistical period. Specifically, the ARIMA time series model can be fitted. The fitting process of the ARIMA time series model can be implemented in the following ways:

Perform the second difference of each second historical revenue data corresponding to each statistical period to obtain the corresponding difference sequence;

According to the difference sequence, use the tabulation method to define the target equation corresponding to the model;

Solve the target equation to obtain the estimated result of the model;

Detect the fit effect of the model based on the goodness of fit;

After confirming that the fitting effect of the model reaches the pre-set target, the residual error of the model is detected;

When it is determined that the residual fluctuation of the model is within the preset fluctuation range, the model is determined as a time series model.

In the embodiment of the present application, in the process of fitting the time series model, the fitting effect of the model is first detected based on the goodness of fit, and the linear residual of the model is detected after the fitting effect reaches the preset target After the linear residual of the model is within the preset fluctuation range, the model is determined as a time series model. By detecting the fitting effect and linear residual of the model, after the fitting effect and linear residual meet the preset conditions, the model is determined as a time series model to ensure the accuracy of the generated time series model, and then The accuracy of the third revenue forecast result obtained through the time series model can be guaranteed.

The following takes the method for obtaining forecasted revenue data provided in the second method as an example to further describe the method for forecasting the revenue of a listed company provided in the embodiment of the present application. As shown in FIG. 4, the method may include the following steps:

Step 401: Determine the industry category of the listed company that needs to perform revenue forecasting.

In the embodiment of the application, when it is necessary to predict the revenue of a listed company, the industry category to which the listed company belongs must first be determined.

For example, if it is now necessary to predict the revenue of listed company A in the next quarter, first determine the industry category A to which listed company A belongs.

Step 402: Obtain historical revenue data of multiple sample companies belonging to the determined industry category.

In this embodiment of the application, after determining the industry category of the listed company that needs to perform revenue forecasting, at least two companies are randomly selected as sample companies from each company belonging to the determined industry category, and then each sample is obtained The company’s historical revenue data.

For example, select 3000 companies from companies belonging to industry category A as the sample companies, and then obtain the historical revenue data of each sample company in the 3000 sample companies in the past 10 years. Among them, if the sample company has been established for less than 10 years, all historical data of the sample company will be obtained.

Step 403: Train a machine learning model based on historical revenue data of each sample company.

In this embodiment of the application, after determining the industry category of the listed company that needs to perform revenue forecasting, at least one factor corresponding to the industry category is determined. Then, for each sample company, the first historical factor data corresponding to each factor is extracted from the historical revenue data of the sample company. Then, the extracted first historical factor data is used to train a machine learning model relative to the determined industry category. Among them, the trained machine learning model can predict future revenue data based on historical revenue data. Specifically, the machine learning model may include the following formula:

Among them, M′ represents the first revenue forecast result; n represents the number of factors; m represents the number of historical years covered by the first historical factor data; x _{(i, 1)} represents the first year of the listed company corresponding to the first factor data of the i factor; x _{(i, 2)} represents the factor data corresponding to the i-th factor in the second year of the listed company; k _i represents the factor coefficient corresponding to the i-th factor at the current time; x _{(i, j )} Characterize the factor data corresponding to the i-th factor in the previous j-th year of the listed company

For example, the last quarter's revenue, last quarter's total assets, last year's same quarter revenue, and last year's same quarter total assets are identified as the four factors corresponding to industry category A. Then, for each of the 3000 sample companies, the sample company’s quarterly revenue and quarterly total assets for each year and quarter in the past 10 years are extracted from the historical revenue data of the sample company in the past 10 years as the first A historical factor data, which can extract a total of 240,000 first historical factor data of 3000*10*4*2, and then use these 240,000 first historical factor data to train machine learning model A.

Step 404: Use the machine learning model to process the historical revenue data of the listed company to obtain the first revenue prediction result.

In this embodiment of the application, the second historical factor data corresponding to each factor is extracted from the historical revenue data of the listed company, and then each extracted second historical factor data is input into the trained machine learning model, and the machine learning The model predicts the revenue of the listed company based on each second historical factor data, and obtains the first revenue prediction result output by the machine learning model.

For example, obtain the historical revenue data of listed company A in the past 10 years, and then extract the quarterly revenue and quarterly total assets of listed company A in each of the past 10 years from the obtained historical revenue data as the second Historical factor data, so you can extract a total of 80 second historical factor data of 10*4*2. Then input the 80 second historical factor data into machine learning model A, and obtain the first revenue prediction result output by machine learning model A. Specifically, the obtained 80 second historical factor data can be substituted into the above formula to calculate the first revenue forecast result.

Step 405: Fit a polynomial function through historical revenue data of each sample company.

In the embodiment of this application, according to the forecast period of the listed company’s revenue forecast, the first historical revenue data of each statistical period is extracted from the historical revenue data of each sample company, and then the extracted first historical revenue data is used. A polynomial function is fitted to historical revenue data, so that each first historical revenue data corresponding to each sample company satisfies the fitted polynomial function. Among them, the statistical period and the forecast period correspond in time span.

The form of the fitted polynomial function is as follows:

Among them, M represents the revenue prediction result relative to the current time; k _i represents the weight coefficient fitted by machine learning; x represents the first historical revenue data corresponding to the previous statistical period relative to the current time; x _i represents The first historical revenue data corresponding to the last i+1 statistical period relative to the current time; t+1 represents the number of statistical periods before the current time.

For example, since the forecast period for the revenue forecast of listed company A is quarterly, for each sample company in the 3000 sample companies, extract the historical revenue data of the sample company from the past 10 years. The quarterly revenue data is used as the first historical revenue data, which can extract a total of 120,000 first historical revenue data of 3000*10*4. Then, the 120,000 first historical revenue data are used to fit a polynomial function, so that each first historical revenue data corresponding to each sample company meets the polynomial function.

Step 406: Use a polynomial function to process historical revenue data of the listed company to obtain a second revenue forecast result.

In this embodiment of the application, the historical revenue data of listed companies is extracted according to the statistical cycle, the historical revenue data corresponding to each statistical period is obtained, and then the extracted historical revenue data corresponding to each statistical period is input The polynomial function is used to predict the revenue of the listed company based on the input historical revenue data, and obtain the second revenue prediction result output by the polynomial function.

For example, obtain the historical revenue data of listed company A in the past 10 years, and then extract the quarterly revenue of listed company A from the obtained historical revenue data for each and every quarter of each of the past 10 years, so as to extract 10* 4A total of 40 quarterly revenues. Then input the 40 quarterly revenue into the following polynomial function A to obtain the second revenue forecast result output by the polynomial function A;

Among them, in the polynomial function A, M represents the relative second revenue forecast result; k _i represents the weight coefficient fitted by machine learning; x represents the quarterly revenue of listed company A in the previous quarter; x _i represents the listing Company A’s quarterly revenue for i+1 quarters.

Step 407: Fit a time series model through historical revenue data of each sample company.

In this embodiment of the application, according to the forecast period of the listed company’s revenue forecast, the second historical revenue data of each statistical period is extracted from the historical revenue data of each sample company, and then the extracted second historical revenue data is used. 2. The historical revenue data is fitted to the time series model, so that the change rule of each second historical revenue data corresponding to each sample company over time meets the time series model. Among them, the statistical period and the forecast period correspond in time span.

The form of the fitted time series model is as follows:

(ΔM _t ) ² =K+k ₁ (ΔM _t-1 ) ² -k ₂ (ΔM _t-2 ) ² +ε _t -k ₃ ε _t-1

Among them, ΔM _t represents the difference between the revenue forecast results relative to the current time and the second historical revenue data corresponding to the previous statistical period of the current time; ΔM _t-1 represents the second historical data corresponding to the previous statistical period of the current time The difference between the revenue data and the second historical revenue data corresponding to the second statistical period before the current time; ΔM _t-2 represents the second historical revenue data corresponding to the second statistical period before the current time and the current time before The difference between the second historical revenue data corresponding to the third statistical period; ε _t represents the revenue forecast result relative to the current time; ε _t-1 represents the second historical revenue data corresponding to the previous statistical period at the current time ; K, k ₁ , k ₂ and k ₃ are weight coefficients fitted by machine learning.

For example, since the forecast period for the revenue forecast of listed company A is quarterly, for each sample company in the 3000 sample companies, extract the historical revenue data of the sample company from the past 10 years. The quarterly revenue data is used as the second historical revenue data, which can be extracted to 3000*10*4, a total of 120,000 second historical revenue data. Then, the 120,000 second historical revenue data is used to fit the time series model, so that each second historical revenue data corresponding to each sample company meets the multiple time series model.

Step 408: Use the time series model to process the historical revenue data of the listed company to obtain the third revenue forecast result.

In this embodiment of the application, the historical revenue data of listed companies is extracted according to the statistical cycle, the historical revenue data corresponding to each statistical period is obtained, and then the extracted historical revenue data corresponding to each statistical period is input The time series model uses the time series model to predict the revenue of the listed company according to the input historical revenue data, and obtains the third revenue prediction result output by the polynomial function.

For example, obtain the historical revenue data of listed company A in the past 10 years, and then extract the quarterly revenue of listed company A from the obtained historical revenue data for each and every quarter of each of the past 10 years, so as to extract 10* 4A total of 40 quarterly revenues. Then input the 40 quarterly revenue into the following time series model A to obtain the third revenue forecast result output by the time series model A;

(ΔM ₄₀ ) ² =K+k ₁ (ΔM ₃₉ ) ² -k ₂ (ΔM ₃₈ ) ² +ε ₄₀ -k ₃ ε ₃₉

Among them, ΔM ₄₀ represents the difference between the third revenue forecast and the quarterly revenue of listed company A in the previous quarter; ΔM ₃₉ represents the quarterly revenue of listed company A in the previous quarter and the quarterly revenue of listed company A in the previous second quarter. The difference in revenue; ΔM ₃₈ represents the difference between the quarterly revenue of listed company A in the previous second quarter and the quarterly revenue of listed company A in the previous third quarter; ε ₄₀ represents the third revenue forecast result of listed company A; ε ₃₉ represents the quarterly revenue of listed company A in the previous quarter; K, k ₁ , k ₂ and k ₃ are all weight coefficients fitted by machine learning.

Step 409: Determine the predicted revenue data of the listed company according to the first revenue prediction result, the second revenue prediction result, and the third revenue prediction result.

In the embodiment of the present application, after obtaining the first revenue prediction result, the second revenue prediction result, and the third revenue prediction result, the first revenue prediction result, the second revenue prediction result, and the third revenue prediction result are calculated. The revenue forecast result is weighted and the result of the weighted calculation is used as the forecast revenue data of the listed company.

Specifically, when the first revenue prediction result, the second revenue prediction result, and the third revenue prediction result are weighted, the first revenue prediction result, the second revenue prediction result, and the third revenue prediction result can be separately calculated in advance. The corresponding weighting coefficient is set for the revenue forecast result. The three weighting coefficients corresponding to the first revenue prediction result, the second revenue prediction result, and the third revenue prediction result can be equal, and the three weighting factors are all 1/3. In addition, different weighting coefficients can be set for the first revenue prediction result, the second revenue prediction result, and the third revenue prediction result. Specifically, it can be used to obtain the first revenue prediction result, the second revenue prediction result and The third method of revenue prediction results is used to predict the revenue of the listed company in the previous historical year, and then the weighting coefficient is determined based on the prediction results of the three to predict the true revenue of the historical year. For example, use the method of obtaining the first revenue forecast to predict the last quarter's revenue as X1, use the method of obtaining the second revenue forecast to predict the last quarter's revenue as X2, and use the method to obtain the third revenue forecast. The method predicts that the revenue of the previous quarter is X3, and the revenue of the previous quarter is actually X4, and then the absolute value of the difference between X1, X2, and X3 and X4 can be used to determine the corresponding first revenue forecast result , The weighting coefficient of the second revenue prediction result and the third revenue prediction result, the larger the absolute value of the difference, the smaller the corresponding weighting coefficient.

For example, use the following formula to calculate the forecasted revenue data of a listed company:

(The first revenue forecast result + the second revenue forecast result + the third revenue forecast result).

As shown in FIG. 5, an embodiment of the present application provides a revenue forecasting device for listed companies, including: a category recognition module 501, a factor recognition module 502, a data acquisition module 503, a first data extraction module 504, and a model training module 505 , The second data extraction module 506, the model processing module 507 and the data processing module 508;

The category identification module 501 is used to determine the industry category of the listed company that needs to perform revenue forecasting;

The factor identification module 502 is configured to determine at least one factor corresponding to the industry category determined by the category identification module 501, wherein different factors correspond to different data statistics rules;

The data acquisition module 503 is configured to use at least two companies belonging to the industry category determined by the category identification module 501 as sample companies, and obtain historical revenue data of each sample company respectively;

The first data extraction module 504 is configured to extract the first historical factor data corresponding to each factor determined by the factor identification module 502 from the historical revenue data of each sample company and acquired by the data acquisition module 503;

The model training module 505 is configured to train a machine learning model corresponding to the industry category through each first historical factor data extracted by the first data extraction module 504;

The second data extraction module 506 is used to extract the factor identification module 502 from the historical revenue data of the listed company to determine the second historical factor data corresponding to each factor;

The model processing module 507 is configured to input each second historical factor data extracted by the second data extraction module 506 into the machine learning model trained by the model training module 505 to obtain the first revenue prediction result output by the machine learning model;

The data processing module 508 is configured to determine the predicted revenue data of the listed company according to the first revenue prediction result obtained by the model processing module 507.

In the embodiment of the present application, the category identification module 501 can be used to perform step 101 in the above method embodiment, the factor identification module 502 can be used to perform step 102 in the above method embodiment, and the data acquisition module 503 can be used to perform the above method embodiment. In step 103, the first data extraction model 504 can be used to perform step 104 in the above method embodiment, the model training module 505 can be used to perform step 105 in the above method embodiment, and the second data extraction module 506 can be used to perform the above method. In step 106 in the embodiment, the model processing module 507 can be used to execute step 107 in the above method embodiment, and the data processing module 508 can be used to execute step 108 in the above method embodiment.

It should be noted that the information interaction and execution process among the various modules included in the device embodiment are based on the same inventive concept as the above method embodiment, and the specific content can be referred to the description in the above method embodiment. No longer. In addition, this device embodiment may also include other modules for executing each step in the foregoing method embodiment.

The embodiments of the present application also provide a computer device, including a memory and a processor, and a computer program is stored on the memory. When the processor executes the computer program stored on the memory, the listed company operation provided by the foregoing embodiments can be implemented. Method of income forecasting.

The embodiments of the present application also provide a non-volatile computer-readable storage medium with a computer program stored on the non-volatile computer-readable storage medium, and the above-mentioned various implementations can be implemented when the stored computer storage is executed. Examples of listed companies’ revenue forecasting methods.

In summary, the listed company revenue forecasting methods, devices, computer equipment, and non-volatile computer-readable storage medium provided by the various embodiments of this application are determined after determining the industry category of the listed company that needs to perform revenue forecasting One or more factors corresponding to the industry category, then obtain the historical revenue data of at least two sample companies belonging to the industry category, and extract the first history corresponding to each factor from the historical revenue data of each sample company Factor data, and extract the second historical factor data corresponding to each factor from the historical revenue data of listed companies, and then use each first historical factor data to train the machine learning model corresponding to the industry category of the listed company. After the two historical factor data is input into the trained machine learning model, the first revenue forecast result is obtained, and then the forecast revenue data of the listed company can be determined according to the first revenue forecast result. It can be seen that using the historical revenue data of multiple sample companies belonging to the same industry category as the listed company to predict the revenue of the listed company, the timeliness of the first historical factor data corresponding to each factor is relatively low, so There is no need for analysts to collect real-time data corresponding to each factor in time, which can reduce the cost of analysts’ revenue forecasts for listed companies.

It should be noted that in this article, the terms "including", "including" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements not only includes those elements, It also includes other elements that are not explicitly listed, or elements inherent to the process, device, article, or method. If there are no more restrictions, the element defined by the sentence "including a..." does not exclude the existence of other identical elements in the process, device, article or method that includes the element.

The serial numbers of the foregoing embodiments of the present application are for description only, and do not represent the superiority of the embodiments. Through the description of the above embodiments, those skilled in the art can clearly understand that the method of the above embodiments can be implemented by means of software plus the necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases the former is better.的实施方式。 Based on this understanding, the technical solution of this application essentially or the part that contributes to the existing technology can be embodied in the form of a software product, and the computer software product is stored in a storage medium (such as ROM/RAM) as described above. , Magnetic disk, optical disk), including several instructions to make a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) execute the method described in each embodiment of the present application.

The above are only preferred embodiments of this application, and do not limit the scope of this application. Any equivalent structure or equivalent process transformation made using the content of the description and drawings of this application, or directly or indirectly used in other related technical fields , The same reason is included in the scope of patent protection of this application.

Claims (20)

1. A method for obtaining development trend data, including:

   Determine the object category to which the predicted object belongs;

   Determine at least one factor corresponding to the object category, wherein different factors correspond to different data statistics rules;

   Taking at least two objects belonging to the object category as sample objects, and obtaining historical development data of each of the sample objects respectively;

   Extracting the first historical factor data corresponding to each of the factors from the historical development data of each of the sample objects;

   Training a machine learning model corresponding to the object category through each of the extracted first historical factor data;

   Extracting second historical factor data corresponding to each of the factors from the historical development data of the prediction object;

   Input each of the second historical factor data into the machine learning model to obtain the first prediction data output by the machine learning model;

   The development trend data used to characterize the development trend of the prediction object is determined according to the first prediction data.
2. The method according to claim 1,

   Before determining the development trend data used to characterize the development trend of the prediction object according to the first prediction data, the method further includes:

   Fitting a polynomial function with historical development data of at least two of the sample objects, wherein the historical development data of each sample object satisfies the polynomial function;

   Input the historical development data of the prediction object into the polynomial function to obtain second prediction data output by the polynomial function;

   The determining development trend data used to characterize the development trend of the prediction object according to the first prediction data includes:

   The development trend data of the prediction object is determined according to the first prediction data and the second prediction data.
3. According to the method of claim 2,

   Before the determining the development trend data of the prediction object according to the first prediction data and the second prediction data, the method further includes:

   Fitting a time series model with the historical development data of at least two of the sample objects, wherein the change law of the historical development data of each of the sample objects over time conforms to the time series model;

   Input the historical development data of the prediction object into the time series model to obtain the third prediction data output by the time series model;

   The determining the development trend data of the prediction object according to the first prediction data and the second prediction data includes:

   The development trend data of the prediction object is determined according to the first prediction data, the second prediction data, and the third prediction data.
4. The method according to claim 2 or 3, wherein the fitting a polynomial function using historical development data of at least two of the sample objects comprises:

   Determine the forecast period for forecasting the development trend of the forecast object;

   Extracting first historical development data corresponding to each statistical period from the historical development data of each sample object according to the prediction period, wherein the statistical period corresponds to the prediction period in a time span;

   According to each of the first historical development data corresponding to each of the statistical periods, the following polynomial function is fitted, where each of the first historical development data corresponding to each of the sample objects satisfies the polynomial function;

   

   Wherein, the M represents the second prediction data relative to the current time; the k _i represents the weight coefficient fitted by machine learning; the x represents the previous statistical period relative to the current time Corresponding to the first historical development data; the x _i represents the first historical development data corresponding to the last i+1 statistical periods relative to the current time; the t+1 represents the current The number of statistical periods before the time.
5. The method according to claim 3, wherein the fitting a time series model using the historical development data of at least two of the sample objects comprises:

   Determine the forecast period for forecasting the development trend of the forecast object;

   Extracting the second historical development data corresponding to each statistical period from the historical development data of each sample object according to the prediction period, wherein the statistical period corresponds to the prediction period in a time span;

   A time series model is fitted according to each of the second historical development data corresponding to each of the statistical periods, wherein the change rule over time of each of the second historical development data corresponding to each of the sample objects satisfies the time series model;

   The form of the time series model is as follows:

   (ΔM _t ) ² =K+k ₁ (ΔM _t-1 ) ² -k ₂ (ΔM _t-2 ) ² +ε _t -k ₃ ε _t-1

   Wherein, the ΔM _t characterizes the difference between the third prediction data relative to the current time and the second historical development data corresponding to the last statistical period of the current time; the ΔM _t-1 characterizes the difference The difference between the second historical development data corresponding to the last statistical period of the current time and the second historical development data corresponding to the second statistical period before the current time; the ΔM _{t- 2} characterizing the difference between the second historical development data corresponding to the second statistical period before the current time and the second historical development data corresponding to the third statistical period before the current time; The ε _t characterizes the third prediction data relative to the current time; the ε _t-1 characterizes the second historical development data corresponding to the last statistical period of the current time; the K , The k ₁ , the k ₂ and the k ₃ are all weight coefficients fitted by machine learning.
6. The method according to claim 5, wherein the fitting a time series model according to each of the second historical development data corresponding to each of the statistical periods comprises:

   Performing a second difference on each of the second historical development data corresponding to each of the statistical periods to obtain a corresponding difference sequence;

   According to the difference sequence, a list method is used to define the target equation corresponding to the model;

   Solving the target equation to obtain an estimation result of the model;

   Detecting the fitting effect of the model based on the goodness of fit;

   After determining that the fitting effect of the model reaches a preset target, detecting the residual of the model;

   When it is determined that the residual fluctuation of the model is within the preset fluctuation range, the model is determined as the time series model.
7. According to the method of claim 3, 5 or 6,

   The determining the development trend data of the prediction object according to the first prediction data, the second prediction data, and the third prediction data includes:

   Performing a weighted operation on the first prediction data, the second prediction data, and the third prediction data to obtain the development trend data of the prediction object;

   and / or,

   The training of the machine learning model corresponding to the object category through each of the extracted first historical factor data includes:

   For each of the factors, obtain at least one factor data corresponding to the factor in each of the past at least two years from the first historical factor data corresponding to the factor;

   Using the factor data corresponding to each of the factors as a sample to train the factor coefficients corresponding to each of the factors;

   Use each of the acquired factor coefficients to construct the following formula for calculating the first prediction data;

   

   Wherein, the M′ represents the first prediction data; the n represents the number of the factors; the m represents the number of historical years covered by the first historical factor data; the x _{(i, 1)} Characterizing the factor data corresponding to the i-th factor of the predicted object in the previous year; the x _{(i, 2)} characterizing the factor corresponding to the i-th factor in the previous year of the predicted object Data; the k _i characterizes the factor coefficient corresponding to the i-th factor at the current time; the x _{(i, j)} characterizes the factor data corresponding to the i-th factor in the previous j-th year of the prediction object;

   The machine learning model including the formula is constructed.
8. A development trend data acquisition device, including: a category identification module, a factor identification module, a data acquisition module, a first data extraction module, a model training module, a second data extraction module, a model processing module, and a data processing module;

   The category recognition module is used to determine the object category to which the predicted object belongs;

   The factor identification module is configured to determine at least one factor corresponding to the object category determined by the category identification module, wherein different factors correspond to different data statistics rules;

   The data acquisition module is configured to use at least two objects belonging to the object category determined by the category recognition module as sample objects, and obtain historical development data of each sample object respectively;

   The first data extraction module is configured to extract each of the factor corresponding to each of the factors determined by the factor identification module from the historical development data of each of the sample objects and acquired by the data acquisition module. The first historical factor data;

   The model training module is configured to train a machine learning model corresponding to the object category through each of the first historical factor data extracted by the first data extraction module;

   The second data extraction module is configured to extract second historical factor data corresponding to each of the factors by the factor identification module from the historical development data of the prediction object;

   The model processing module is configured to input each of the second historical factor data extracted by the second data extraction module into the machine learning model trained by the model training module to obtain the output of the machine learning model First forecast data;

   The data processing module is configured to determine development trend data used to characterize the development trend of the prediction object according to the first prediction data acquired by the model processing module.
9. The device according to claim 8, further configured to: before determining the development trend data used to characterize the development trend of the prediction object according to the first prediction data, further:

   Fitting a polynomial function with historical development data of at least two of the sample objects, wherein the historical development data of each sample object satisfies the polynomial function;

   Input the historical development data of the prediction object into the polynomial function to obtain second prediction data output by the polynomial function;

   The determining development trend data used to characterize the development trend of the prediction object according to the first prediction data includes:

   The development trend data of the prediction object is determined according to the first prediction data and the second prediction data.
10. The device according to claim 9, further used for:

    Before the determining the development trend data of the prediction object according to the first prediction data and the second prediction data, further:

    Fitting a time series model with the historical development data of at least two of the sample objects, wherein the change law of the historical development data of each of the sample objects over time conforms to the time series model;

    Input the historical development data of the prediction object into the time series model to obtain the third prediction data output by the time series model;

    The determining the development trend data of the prediction object according to the first prediction data and the second prediction data includes:

    The development trend data of the prediction object is determined according to the first prediction data, the second prediction data, and the third prediction data.
11. The device according to claim 9 or 10, wherein the device uses historical development data of at least two of the sample objects to fit a polynomial function, further comprising:

    Determine the forecast period for forecasting the development trend of the forecast object;

    Extracting first historical development data corresponding to each statistical period from the historical development data of each sample object according to the prediction period, wherein the statistical period corresponds to the prediction period in a time span;

    According to each of the first historical development data corresponding to each of the statistical periods, the following polynomial function is fitted, where each of the first historical development data corresponding to each of the sample objects satisfies the polynomial function;

    

    Wherein, the M represents the second prediction data relative to the current time; the k _i represents the weight coefficient fitted by machine learning; the x represents the previous statistical period relative to the current time Corresponding to the first historical development data; the x _i represents the first historical development data corresponding to the last i+1 statistical periods relative to the current time; the t+1 represents the current The number of statistical periods before the time.
12. The device according to claim 10, wherein the device uses historical development data of at least two of the sample objects to fit a time series model, further comprising:

    Determine the forecast period for forecasting the development trend of the forecast object;

    Extracting the second historical development data corresponding to each statistical period from the historical development data of each sample object according to the prediction period, wherein the statistical period corresponds to the prediction period in a time span;

    A time series model is fitted according to each of the second historical development data corresponding to each of the statistical periods, wherein the change rule over time of each of the second historical development data corresponding to each of the sample objects satisfies the time series model;

    The form of the time series model is as follows:

    (ΔM _t ) ² =K+k ₁ (ΔM _t-1 ) ² -k ₂ (ΔM _t-2 ) ² +ε _t -k ₃ ε _t-1

    Wherein, the ΔM _t characterizes the difference between the third prediction data relative to the current time and the second historical development data corresponding to the last statistical period of the current time; the ΔM _t-1 characterizes the difference The difference between the second historical development data corresponding to the last statistical period of the current time and the second historical development data corresponding to the second statistical period before the current time; the ΔM _{t- 2} characterizing the difference between the second historical development data corresponding to the second statistical period before the current time and the second historical development data corresponding to the third statistical period before the current time; The ε _t characterizes the third prediction data relative to the current time; the ε _t-1 characterizes the second historical development data corresponding to the last statistical period of the current time; the K , The k ₁ , the k ₂ and the k ₃ are all weight coefficients fitted by machine learning.
13. The device according to claim 12, said device fitting a time series model according to each of said second historical development data corresponding to each of said statistical periods, further comprising:

    Performing a second difference on each of the second historical development data corresponding to each of the statistical periods to obtain a corresponding difference sequence;

    According to the difference sequence, a list method is used to define the target equation corresponding to the model;

    Solving the target equation to obtain an estimation result of the model;

    Detecting the fitting effect of the model based on the goodness of fit;

    After determining that the fitting effect of the model reaches a preset target, detecting the residual of the model;

    When it is determined that the residual fluctuation of the model is within a preset fluctuation range, the model is determined as the time series model.
14. The device according to claim 10, 12 or 13,

    The device determining the development trend data of the prediction object according to the first prediction data, the second prediction data, and the third prediction data further includes:

    Performing a weighted operation on the first prediction data, the second prediction data, and the third prediction data to obtain the development trend data of the prediction object;

    and / or,

    The training of the machine learning model corresponding to the object category through each of the extracted first historical factor data includes:

    For each of the factors, obtain at least one factor data corresponding to the factor in each of the past at least two years from the first historical factor data corresponding to the factor;

    Using the factor data corresponding to each of the factors as a sample to train the factor coefficients corresponding to each of the factors;

    Use each of the acquired factor coefficients to construct the following formula for calculating the first prediction data;

    

    Wherein, the M′ represents the first prediction data; the n represents the number of the factors; the m represents the number of historical years covered by the first historical factor data; the x _{(i, 1)} Characterizing the factor data corresponding to the i-th factor of the predicted object in the previous year; the x _{(i, 2)} characterizing the factor corresponding to the i-th factor in the previous year of the predicted object Data; the k _i characterizes the factor coefficient corresponding to the i-th factor at the current time; the x _{(i, j)} characterizes the factor data corresponding to the i-th factor in the previous j-th year of the prediction object;

    The machine learning model including the formula is constructed.
15. A computer device includes a memory and a processor, the memory stores a computer program, and the steps of a method for acquiring development trend data when the processor executes the computer program include:

    Determine the object category to which the predicted object belongs;

    Determine at least one factor corresponding to the object category, wherein different factors correspond to different data statistics rules;

    Taking at least two objects belonging to the object category as sample objects, and obtaining historical development data of each of the sample objects respectively;

    Extracting the first historical factor data corresponding to each of the factors from the historical development data of each of the sample objects;

    Training a machine learning model corresponding to the object category through each of the extracted first historical factor data;

    Extracting second historical factor data corresponding to each of the factors from the historical development data of the prediction object;

    Input each of the second historical factor data into the machine learning model to obtain the first prediction data output by the machine learning model;

    The development trend data used to characterize the development trend of the prediction object is determined according to the first prediction data.
16. The computer device according to claim 1, before the determining the development trend data used to characterize the development trend of the prediction object according to the first prediction data, further comprising:

    Fitting a polynomial function with historical development data of at least two of the sample objects, wherein the historical development data of each sample object satisfies the polynomial function;

    Input the historical development data of the prediction object into the polynomial function to obtain second prediction data output by the polynomial function;

    The determining development trend data used to characterize the development trend of the prediction object according to the first prediction data includes:

    The development trend data of the prediction object is determined according to the first prediction data and the second prediction data.
17. The computer device according to claim 2, before said determining the development trend data of the prediction object according to the first prediction data and the second prediction data, further comprising:

    Fitting a time series model with the historical development data of at least two of the sample objects, wherein the change law of the historical development data of each of the sample objects over time conforms to the time series model;

    Input the historical development data of the prediction object into the time series model to obtain the third prediction data output by the time series model;

    The determining the development trend data of the prediction object according to the first prediction data and the second prediction data includes:

    The development trend data of the prediction object is determined according to the first prediction data, the second prediction data, and the third prediction data.
18. A non-volatile computer-readable storage medium having a computer program stored thereon, and the steps of implementing a method for acquiring development trend data when the computer program is executed by a processor include:

    Determine the object category to which the predicted object belongs;

    Determine at least one factor corresponding to the object category, wherein different factors correspond to different data statistics rules;

    Taking at least two objects belonging to the object category as sample objects, and obtaining historical development data of each of the sample objects respectively;

    Extracting the first historical factor data corresponding to each of the factors from the historical development data of each of the sample objects;

    Training a machine learning model corresponding to the object category through each of the extracted first historical factor data;

    Extracting second historical factor data corresponding to each of the factors from the historical development data of the prediction object;

    Input each of the second historical factor data into the machine learning model to obtain the first prediction data output by the machine learning model;

    The development trend data used to characterize the development trend of the prediction object is determined according to the first prediction data.
19. The storage medium according to claim 18, before the determining the development trend data used to characterize the development trend of the prediction object according to the first prediction data, further comprising:

    Fitting a polynomial function with historical development data of at least two of the sample objects, wherein the historical development data of each sample object satisfies the polynomial function;

    Input the historical development data of the prediction object into the polynomial function to obtain second prediction data output by the polynomial function;

    The determining development trend data used to characterize the development trend of the prediction object according to the first prediction data includes:

    The development trend data of the prediction object is determined according to the first prediction data and the second prediction data.
20. The storage medium according to claim 19, before said determining the development trend data of the prediction object according to the first prediction data and the second prediction data, further comprising:

    Fitting a time series model with the historical development data of at least two of the sample objects, wherein the change law of the historical development data of each of the sample objects over time conforms to the time series model;

    Input the historical development data of the prediction object into the time series model to obtain the third prediction data output by the time series model;

    The determining the development trend data of the prediction object according to the first prediction data and the second prediction data includes:

    The development trend data of the prediction object is determined according to the first prediction data, the second prediction data, and the third prediction data.

Images