WO2020124977A1 - Method and apparatus for processing production data, computer device, and storage medium - Google Patents

Abstract

A method for processing production data comprises: receiving an input prediction time, and obtaining, according to the prediction time, corresponding historical production data from a server; inputting the prediction time and the historical production data into a model classifier to obtain predicted production data corresponding to the prediction time, wherein the predicted production data is obtained by means of a single-model prediction result of an optimal prediction single model, the optimal prediction single model being selected by the model classifier according to the prediction time and the historical production data; or the predicted production data is calculated and obtained by means of a corresponding weight and a corresponding single-model prediction result of each of multiple prediction single models, the prediction single models being generated according to the prediction time and the historical production data; and obtaining an investment file corresponding to the predicted production data, and sending the investment file to an investment terminal.

Description

Production data processing method, device, computer equipment and storage medium

Cross-reference of related applications

This application requires the priority of the Chinese patent application filed on December 19, 2018, with the application number 2018115552084, and the application name is "Model Classifier Establishment Method, Device, Computer Equipment, and Storage Media", all of which are approved by The reference is incorporated in this application.

Technical field

The present application relates to a production data processing method, device, computer equipment and storage medium.

Background technique

In the financial industry, operating income refers to the various incomes obtained by the sale of products or the provision of labor services in the production and operation activities of the enterprise. It is related to the survival and development of the enterprise and is of great significance to the operation of the enterprise. Therefore, accurately predict the enterprise Revenue is an important part of investment analysis.

However, the inventor realized that the current production data, that is, the operating income data is processed according to a single model, that is, the historical production data is input into the single model so that the predicted production data corresponding to the historical production data can be obtained, but This processing method is only based on a single model. Such a predicted single model is easy to ignore certain special factors or lead to extreme predictions. Therefore, the accuracy of the model is not enough, resulting in inaccurate production data processing results.

Summary of the invention

According to various embodiments disclosed in the present application, a production data processing method, apparatus, computer equipment, and storage medium are provided.

A production data processing method, including:

Receiving the input prediction time, and obtaining corresponding historical production data from the server according to the prediction time;

Input the predicted time and historical production data to the model classifier to obtain the predicted production data corresponding to the predicted time, the predicted production data is the optimal selected by the model classifier according to the predicted time and historical production data Obtained by predicting the single model prediction result of the single model, or the predicted production data is calculated according to the weight corresponding to each of the predicted single models and the corresponding single model prediction result generated according to the prediction time and historical production data of;

Obtain the investment file corresponding to the predicted production data, and send the investment file to the investment terminal.

A production data processing device, including:

The receiving module is used to receive the input predicted time, and obtain corresponding historical production data from the server according to the predicted time;

A processing module, configured to input the predicted time and historical production data to a model classifier to obtain predicted production data corresponding to the predicted time, the predicted production data is the model classifier based on the predicted time and historical production Obtained by the single model prediction result of the optimal prediction single model selected by the data, or the predicted production data is the weight corresponding to each of the predicted single models generated according to the prediction time and historical production data and the corresponding single Calculated by the model prediction results;

The sending module is used to obtain the investment file corresponding to the predicted production data, and send the investment file to the investment terminal.

A computer device includes a memory and one or more processors. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the one or more processors are executed The following steps:

Receiving the input prediction time, and obtaining corresponding historical production data from the server according to the prediction time;

Input the predicted time and historical production data to the model classifier to obtain the predicted production data corresponding to the predicted time, the predicted production data is the optimal selected by the model classifier according to the predicted time and historical production data Obtained by predicting the single model prediction result of the single model, or the predicted production data is calculated according to the weight corresponding to each of the predicted single models and the corresponding single model prediction result generated according to the prediction time and historical production data of;

Obtain the investment file corresponding to the predicted production data, and send the investment file to the investment terminal.

One or more non-volatile computer-readable storage media storing computer-readable instructions. When the computer-readable instructions are executed by one or more processors, the one or more processors perform the following steps:

Receiving the input prediction time, and obtaining corresponding historical production data from the server according to the prediction time;

Input the predicted time and historical production data to the model classifier to obtain the predicted production data corresponding to the predicted time, the predicted production data is the optimal selected by the model classifier according to the predicted time and historical production data Obtained by predicting the single model prediction result of the single model, or the predicted production data is calculated according to the weight corresponding to each of the predicted single models and the corresponding single model prediction result generated according to the prediction time and historical production data of;

Obtain the investment file corresponding to the predicted production data, and send the investment file to the investment terminal.

The details of one or more embodiments of the application are set forth in the drawings and description below. Other features and advantages of this application will become apparent from the description, drawings, and claims.

BRIEF DESCRIPTION

In order to more clearly explain the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. Those of ordinary skill in the art can obtain other drawings based on these drawings without creative work.

FIG. 1 is an application scenario diagram of a production data processing method according to one or more embodiments.

FIG. 2 is a schematic flowchart of a production data processing method according to one or more embodiments.

3 is a block diagram of a production data processing device according to one or more embodiments.

Figure 4 is a block diagram of a computer device in accordance with one or more embodiments.

detailed description

In order to make the technical solutions and advantages of the present application clearer, the following describes the present application in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

The production data processing method provided by this application can be applied to the application environment shown in FIG. 1. The terminal 102 communicates with the server 104 via the network. The terminal 102 can obtain historical production data from the server 104, so that it can be trained according to the production data to obtain multiple predicted single models, and then trained according to the multiple predicted single models and historical production data to obtain a model classifier, which can be based on the model The classifier obtains the most accurate prediction single model or integrates the prediction results of multiple prediction single models to obtain the most accurate prediction result, which improves the accuracy of the prediction result. Specifically, the terminal 102 can obtain the training prediction time, and input the training prediction time and historical production data into the prediction single model to obtain a single model prediction result, so that the first model selection feature can be obtained according to the single model prediction result and historical production data , The first model selection feature fully takes into account the prediction error of each prediction single model, and in order to fully take into account the characteristics of historical production data, the terminal 102 can also obtain the second model selection feature based only on the historical production data, so that the terminal 102 can The first model selection feature, the second model selection feature, and the prediction single model are trained to obtain a model classifier, so that the trained model classifier can be used to predict the production data at the prediction time, that is, the most accurate prediction is obtained according to the model classifier The prediction single model or the prediction results of multiple prediction single models are combined to obtain the most accurate prediction result, which improves the accuracy of the prediction result. The terminal 102 may be, but not limited to, various personal computers, notebook computers, smart phones, tablet computers, and portable wearable devices. The server 104 may be implemented by an independent server or a server cluster composed of multiple servers.

Specifically, multiple forecast single models are generated in advance in this application, and the production forecast data can be predicted through the multiple forecast single models, for example, revenue data; however, the single model prediction results of the forecast single model may be biased. , Resulting in the final determination of which single-model prediction result of the single-model prediction is accurate, so in order to avoid this situation, this application has generated multiple targets based on historical production data and the single-model prediction results of each predicted single model in advance Model selection features. The target model selection features may include features for measuring the accuracy of prediction results of individual models and features of historical production data's own attributes. Therefore, the present application can select features and predict single models according to the multiple target models for training to obtain the best prediction results of production data, for example, select features according to the target model to select the best predictive single model, or select according to the target model Feature to obtain the weight of each prediction single model, so that the prediction result of production data can be obtained according to the weight and the single model prediction result of each prediction single model.

In one of the embodiments, as shown in FIG. 2, a production data processing method is provided. The method is applied to the terminal in FIG. 1 as an example for description, including the following steps:

S202: Obtain historical production data.

Specifically, the historical production data is data generated by an enterprise in past transactions or production and life, for example, it may be historical revenue data of various enterprises. The terminal may first send a historical revenue data acquisition instruction to the financial server corresponding to each enterprise, so that after receiving the historical revenue data acquisition instruction, the financial service of each enterprise sends the historical revenue data to the corresponding terminal, and the terminal is receiving After reaching the historical revenue data, store the historical revenue data, for example, in a secure database or server. In addition, the financial servers of each enterprise can also periodically submit historical revenue data of each enterprise to the terminal in the most recent period to ensure the timeliness of the historical revenue data stored on the terminal side.

And optionally, after the terminal obtains the historical production data, the historical production data can also be pre-processed to improve the accuracy of model training. For example, the terminal may delete abnormal data or process historical production data to obtain historical production data needed for model training. Deleting the abnormal data may include the terminal determining whether there is incomplete data, zero or empty data in the historical production data, and if there is, deleting the abnormal data. The terminal processes the historical production data to obtain the historical production data required for model training. The historical production data can be differentiated. For example, after deleting the abnormal data, the terminal can judge that the obtained historical production data needs to be differentiated. For example, when the obtained history When the production data is half a year of historical production data, the terminal can subtract the historical production data of the first quarter from the historical production data of the half year, so that the historical production data of the second quarter can be obtained.

S204: Obtain the training prediction time, and input the training prediction time and historical production data into the prediction single model to obtain a single model prediction result. The prediction single model is obtained by pre-training.

Specifically, the prediction single model may include multiple categories, such as a time series model, a trend fitting model, a time series fitting comprehensive model, and a multi-factor model. The time series model can be the arima model, which is a model built using the time series data itself. The historical production data mainly includes two fields, one is the time field, and the other is the revenue data field, so that the two Field to build the arima model; the trend-fitting model can be a polyfit model, which builds discrete points based on known revenue data, that is, time is on the x-axis, and historical revenue data is on the y-axis, and then based on discrete points The combined construction function is used as the trend-fitting model, that is, the obtained fitting function is the trend-fitting model; the time-series fitting comprehensive model may be a prophet model, which generates a segmented model based on the continuous growth of revenue data in segments , Where the revenue data can be segmented according to different growth rates, and then the function corresponding to each segment can be obtained, and all functions can be combined to obtain a single prediction model; the multi-factor model can be the xgboost model, which is based on the classifier The idea is to build a forecasting single model, that is, classify historical revenue data according to time and revenue data, and then learn the classified historical revenue data to obtain a forecasting single model.

Specifically, the training prediction time is also the historical time, that is, the time that has passed. When training the model, the historical production data needs to be divided into sample data and verification data according to the training prediction time, where the sample data is used for prediction, verification The data is used to adjust the prediction model or to check whether the prediction result is accurate.

The terminal can use the historical production data corresponding to the training prediction time as verification data; and the historical production data before the training prediction time as sample data; the terminal inputs the sample data and training prediction time into the prediction single model to obtain each prediction The single model prediction result output by the single model. For example, the training prediction time is the third quarter of 2018, you can obtain historical production data before 2018 as sample data, and enter the sample data into the prediction single model to obtain the single model prediction result corresponding to the training prediction time. For example, the prediction results of the arima model, the prediction results of the polyfit model, the prediction results of the prophet model, and the prediction results of the xgboost model can be obtained.

S206: The comprehensive single-model prediction results and the historical production data's own attributes are constructed to characterize the error between the single-model prediction results and the historical production data and the target model selection characteristics that characterize the historical production data's essential characteristics.

Specifically, the target model selection feature may include a feature for measuring the accuracy of each single model prediction result and the feature of historical production data's own attributes, for example, the target model selection feature may include a first model selection feature and a second model selection feature , Where the first model selection feature fully takes into account the error between the prediction results of the single model and the actual results in the historical production data, and the second model selection feature is a feature used to identify the essence of the historical production data.

Specifically, the first model selection feature fully considers the error between the prediction results of the single model and the real results in the historical production data, so that the prediction accuracy of each prediction single model can be considered. The single-model prediction result is the prediction result, and the historical production data is a representation of the real result, so the first model selection feature can be constructed by the prediction result and the real result, for example, the corresponding training in the single-model prediction result and the historical production data can be selected The ratio of the real result of the prediction time is used as the first model selection feature. Optionally, the first model selection feature may be the prediction error of each prediction single model from the previous cycle and/or the prediction error of each prediction single model from the previous year.

Specifically, the second model selection feature is a feature used to identify the essence of historical production data. The second model selection feature may be calculated based on historical production data before the training prediction time, which may include periodic strength indicators and/or Trend strength indicator. The cyclical strength indicator is used to represent the periodicity in the historical production data, and the trend strength indicator is used to represent the trend of the historical production data, such as growth or decline.

S208: Train the target model selection feature and the prediction single model to obtain a model classifier. The model classifier is used to select the optimal prediction single model from the prediction single model according to the target model selection feature or to select the feature based on the target model for each A corresponding weight of the prediction single model is established, and a prediction model is obtained according to the weight and the prediction result of the corresponding single model.

Specifically, the training process of the model classifier may be to use the predicted single model as the Y value, and then use the constructed first model selection feature and the second model selection feature as the X value, and the corresponding relationship between the Y value and the X value Learn to get the model classifier. The terminal first determines the training prediction time, and then determines the Y value corresponding to the training prediction time based on the single model prediction value and the historical production data corresponding to the training prediction time, so that the correspondence between the Y value and the X value can be established, and all the Y values can be established. Mark the X value in the coordinate system to obtain multiple discrete points, and then fit the multiple discrete points to obtain the model classifier.

S210: The model classifier obtained by training predicts the production data at the prediction time.

After obtaining the model classifier, the terminal can input the prediction time, so that the terminal can obtain the corresponding historical production data from the server according to the prediction time, and then input the prediction time and the historical production data into the model classifier to obtain the prediction production corresponding to the prediction time Data, and after obtaining the predicted production data corresponding to the predicted time, the server can obtain the corresponding investment file according to the predicted production data, and send the investment file to the investment terminal, so that the terminal can determine the corresponding investment plan according to the investment file . Where the forecast production data is obtained by the single model prediction result of the optimal forecast single model selected by the model classifier according to the forecast time and historical production data, or the forecast production data is generated according to the forecast time and historical production data The weight corresponding to the model and the corresponding prediction result of the single model are calculated.

The above production data processing method fully considers the single model prediction results and historical production data of multiple single models when building the model classifier, and constructs the first model selection feature based on the single model prediction results and historical production data. The second model selection feature is constructed based on historical production data, and the model classifier is obtained by training based on the first model selection feature and the second model selection feature, so that the model classifier fully considers the characteristics of each model and the characteristics of historical production data , Which can improve the accuracy of model prediction.

In one of the embodiments, training target model selection features and predicting a single model to obtain a model classifier may include: extracting real results from historical production data; calculating the difference between the real results and the single model prediction results to obtain the difference The smallest predictive single model is used as the optimal predictive single model; the optimal predictive single model and the target model are trained to select features to obtain the model classifier.

In one of the embodiments, training target model selection features and predicting a single model to obtain a model classifier may include: extracting real results from historical production data; calculating the ratio of the single model prediction results to the real results, and obtaining predictions based on the ratios Single model weights; train the weights of the predicted single model and the target model selection features to obtain the model classifier.

Specifically, the model classifier can be divided into two types, which are different according to the Y value, where the Y value can predict a single model, or the weight of each predicted single model, where the Y value is a predicted single model, Then, the optimal single prediction model is selected through the model classifier, and the Y value is the weight of each prediction single model. The weight of each prediction single model is obtained through the model classifier, so that each prediction single model is fully considered Forecast results.

Specifically, when the Y value is a single prediction model, the terminal can extract the real results from the historical production data, that is, the real results corresponding to the training prediction time, and then compare the real results with the single model prediction results to obtain the real results The closest single model prediction result, and use the predicted single model corresponding to the single model prediction result as the Y value, and then learn the Y value and the corresponding X value, such as fitting the Y value and the X value to get each The weight corresponding to the X value, so that a model selector can be established.

When the Y value is the weight of each prediction single model, the terminal can extract the real results from the historical production data, that is, the real results corresponding to the training prediction time, and then compare the real results with the single model prediction results, through each The ratio of the prediction result of the single model to the true result can obtain the probability of each prediction single model as the optimal single model, that is, the weight, and optionally, the terminal can classify the possibility of a prediction single model as the optimal single model. One normalization can obtain the weights corresponding to each prediction single model, so that the terminal can learn these weights and the corresponding X value to obtain the model classifier. In this way, when the first model selection feature and the second model selection feature are input into the model selector, the model classifier outputs the weight of each predicted single model, for example, the first predicted single model weight is 0.25, and the second predicted single model weight It is 0.15, the weight of the third prediction model is 0.2, and the weight of the fourth model is 0.4. Therefore, the terminal can calculate the final prediction result based on the single-model prediction results and weights of the prediction single models, that is, the revenue data corresponding to the prediction time.

In the above embodiment, the Y value can predict a single model, or the weight of each predicted single model; when the Y value is a predicted single model, it can choose an optimal single model; when the Y value is each prediction The weight of the single model fully considers the prediction result of each prediction single model, and can guarantee the accuracy of the prediction result obtained by the model classifier.

In one of the embodiments, inputting the training prediction time and historical production data into the prediction single model to obtain a single model prediction result may include: obtaining the feature period corresponding to the training prediction time; calculating the corresponding period corresponding to the feature period through the prediction single model Predicted value; constructing the first model selection feature based on the single model prediction result and historical production data, which may include: extracting the true value corresponding to the feature period from the historical production data; calculating the first model selection feature based on the predicted value and the true value. The characteristic period can be the previous period or the previous period. When the previous period is the first period, the first model selection feature is the forecast error. When the first period is the last period, the first model selection feature is the previous period. cycle.

In one of the embodiments, constructing a second model selection feature based on historical production data may include: obtaining a preset cycle length and interval, segmenting the historical production data according to the preset cycle length; obtaining the corresponding score in each interval The historical production data of the segment, and sort the acquired historical production data, and mark the sequence value of the sorted historical production data; calculate the deviation value of the sequence value of the segmented historical production data corresponding to each interval; calculate the deviation The average of the values gives the periodic strength indicator as the second model selection feature.

In one of the embodiments, constructing a second model selection feature based on historical production data may include: obtaining a preset period length, segmenting the historical production data according to the preset period length; according to the historical production data of each segment The historical production data of the previous section is calculated to obtain the increase and decrease range; according to the increase and decrease range, the increase and decrease range mark value of each section of the historical production data is calculated; the average value of the increase and decrease range mark value is calculated to obtain the trend strength indicator as The second model selects features.

Specifically, the first model selection feature and the second model selection feature are involved in the above model classifier, and for convenience, the two features are hereinafter referred to as model selection features for description. That is, the model selection characteristics may include the prediction error of the previous cycle prediction result, the prediction error of the previous cycle prediction result, the periodic strength index, and the trend strength index.

Among them, the period in the forecast error of the previous cycle’s forecast results can refer to a quarter or a year, and the forecast error of the previous cycle’s forecast results is for each forecast model and training forecast time. Each forecast unit The models all correspond to a prediction error of the prediction result of the previous period. The terminal can obtain the training prediction time, and then obtain the previous period of the previous period, and calculate the first prediction value corresponding to the previous period of the previous period by calculating the single model of the prediction, and extract the first prediction value corresponding to the previous period of the previous period from the historical production data For the real value, the terminal then obtains the prediction error of the prediction result of the previous period based on the ratio of the first predicted value and the first true value, that is, the first model selection feature described above.

Specifically, a training prediction time of one prediction single model is used for explanation. Assuming that the training prediction time is the fourth quarter of 2018, the terminal first obtains the first prediction value corresponding to the prediction single model in the third quarter of 2018, Then the terminal extracts the first true value of the forecasted single model from the historical production data in the third quarter of 2018, and uses the ratio of the first predicted value and the first true value as the prediction error of the previous cycle's prediction result.

Among them, the period of the forecast error of the forecast result of the previous period of the previous year can refer to a quarter or a year. The forecast error of the forecast result of the previous period of the previous period is for each forecast single model and training prediction time. The models all correspond to a forecast error compared to the forecast result of the previous period. The terminal can obtain the training prediction time, and then obtain the previous period of the previous year, and calculate the second prediction value corresponding to the previous period of the previous year by calculating the single model of the prediction, and extract the second prediction value corresponding to the previous period of the previous year from the historical production data. For the true value, the terminal then obtains the prediction error of the prediction result of the previous period based on the ratio of the second predicted value and the second true value, that is, the above-mentioned first model selection feature.

Specifically, a training prediction time of one of the prediction single models is used for explanation. Assuming that the training prediction time is the fourth quarter of 2018, the terminal first obtains the second prediction value corresponding to the prediction single model in the fourth quarter of 2017, Then the terminal extracts the second true value of the forecasted single model from the historical production data in the fourth quarter of 2017, and uses the ratio of the second predicted value and the second true value as the forecast error of the previous year’s forecast results.

Among them, the periodic strength index is used to characterize the periodic characteristics in historical production data. After acquiring the historical production data, the terminal obtains a preset cycle length and interval, for example, the preset cycle length is quarter, and the interval may be a year, that is, an interval may include multiple cycles. The terminal can segment the historical production data according to the cycle length, and then obtain and sort the historical production data in the corresponding segment in each interval, and mark the sorted historical production data to obtain the sequence value, and then calculate each The deviation value of the sequence value of the historical production data corresponding to the segment in an interval, and calculating the average value of the deviation values to obtain the periodic strength index, that is, the above-mentioned second model selection feature.

Specifically, taking the cycle length as the quarter and the interval length as the year as an example for explanation, it is assumed that the historical production data includes revenue data for 2015, 2016, and 2017, so the terminal first divides the revenue data according to the cycle length The segment is divided into the first quarter, second quarter, third quarter and fourth quarter of 2015, the first quarter, second quarter, third quarter and fourth quarter of 2016 and the first quarter and second quarter of 2017 , The third quarter and the fourth quarter, and then sort the corresponding segments in each interval, and mark the sequence value of the sorted segments, as shown in Table 1 below:

Table 1 Sorting results

Ordinal value	the first season	Second quarter	the third quater	Fourth quarter
2015	1	2	3	2
2016	2	1	2	3
2017	3	3	1	1

As shown in the table above, 2015 is the first interval, 2016 is the second interval, and 2017 is the third interval. The corresponding segment in each interval is the first quarter of 2015, the first quarter of 2016, and In the first quarter of 2017, the terminal can sort the historical production data of the three segments, that is, sort the historical revenue data, and mark the sequence value of each segment after sorting, for example, the maximum value is 1, followed by For 2, mark it in turn. In this way, the corresponding segments of each interval are sorted and completed.

After the sorting is completed, the terminal obtains the sequence value in each interval, and calculates the deviation value according to the sequence value. For example, the sequence value of the first interval corresponding to 2015 includes 1, 2, 3, and 2, and the terminal calculates 1. The deviation values of 2, 3, and 2, for example, the standard deviations of 1, 2, 3, and 2 can be calculated. Similarly, the terminal can also calculate the standard deviations of other intervals, and finally use the average value of all standard deviations as the periodic strength indicator .

Among them, the trend strength index is used to characterize the trend characteristics of historical production data. After acquiring historical production data, the terminal acquires a preset period length, for example, the preset period length is quarter. The terminal can segment the historical production data according to the length of the cycle, and then calculate the increase and decrease range according to the historical production data of the current segment and the historical production data of the previous segment, and can determine the increase and decrease of each segment according to the increase and decrease The historical production data is marked to obtain the increase-decrease amplitude mark value, and the terminal finally calculates the average value of the increase-decrease amplitude mark value to obtain the trend strength index, that is, the above-mentioned second model selection feature.

Specifically, taking the cycle length as a quarter for example, it is assumed that the historical production data includes revenue data for 2015, 2016, and 2017, so the terminal first segments the revenue data according to the cycle length, divided into 2015 Q1, Q2, Q3 and Q4 of 2016, Q1, Q2, Q3 and Q4 of 2016 and Q1, Q2, Q3 and Q1 of 2017 The fourth quarter, and then calculate the increase and decrease value for each segment, for example, the slope of the connection between the first quarter of 2015 revenue data and the second quarter of 2015 revenue data can be used as the second quarter of 2015 The increase and decrease amplitude values can also be calculated for other segments, that is, the increase and decrease amplitude values for other periods. After calculating the increase and decrease amplitude value, the terminal calculates the increase and decrease amplitude mark value according to the increase and decrease amplitude value, for example, when the increase and decrease amplitude value is greater than 0, the increase and decrease amplitude value is +1, when the increase and decrease amplitude value is less than When 0, the increase and decrease amplitude tag value is -1, when the increase and decrease amplitude is equal to 0, the increase and decrease amplitude tag value is 0. After acquiring all the increase and decrease amplitude mark values, the terminal may calculate the average value of the mark amplitude mark values, and these average values are the trend strength index, that is, the above-mentioned second model selection feature.

In the above embodiment, based on historical production data and each forecasting single model, the prediction error of the previous-period forecast results, the forecast error of the previous-period forecast results, the periodic strength index, and the trend strength index can be calculated. Then the model classifier is trained to ensure the accuracy of the model classifier's prediction.

Specifically, the training process of the model trainer will be described in detail below in conjunction with the above model selection features:

Use the above-mentioned forecast error of the previous cycle forecast results, the forecast error of the previous year’s forecast results, cyclic strength indicators, and trend strength indicators as model selection features; use historical data as revenue from 2015 to 2017 Data as an example, for example, when the obtained historical production revenue is half a year of historical revenue data, the terminal can subtract the historical revenue data of the first quarter from the historical revenue data of the half year, so that the second quarter of the Historical revenue data, so that the terminal can calculate the above periodic strength index A and trend strength index B based on the standardized historical revenue data, and then calculate the chain-on-cycle previous period forecast corresponding to each forecasting single model according to the training forecast time The forecast error of the results and the forecast error of the forecast results of the previous period compared with the previous period. Assuming that there are 4 forecast single models, there are 4 forecast errors C, D, E, F and 4 of the previous period. The prediction errors M, N, P, Q of the period prediction result, so the model classifier Y = a1*A+b1*B+c1*C+d1*D+e1*E+f1*F+m1*M+n1 *N+p1*P+q1*Q, where a1, b1, c1, d1, e1, f1, m1, n1, p1, and q1 are the weights for selecting features for each model, and a1, b1, c1 can be obtained by fitting training , D1, e1, f1, m1, n1, p1, and q1, after the values of a1, b1, c1, d1, e1, f1, m1, n1, p1, and q1 are determined, the model classifier can be obtained, and After obtaining the model classifier, when the user enters the predicted time, the terminal can obtain the historical revenue data corresponding to the predicted time, and then input the predicted time and historical revenue data into the model classifier to obtain the corresponding to the predicted time Forecast revenue data.

In the above embodiment, when building the model classifier, the single-model prediction results and historical production data of multiple single models are fully considered, and the first model selection feature is constructed based on the single-model prediction results and historical production data. The historical production data is constructed to obtain the second model selection feature, so that the model classifier is obtained by training according to the first model selection feature and the second model selection feature, so that the model classifier fully considers the characteristics of each model and the characteristics of historical production data, Therefore, the accuracy of model prediction can be improved.

It should be understood that although the steps in the flowchart of FIG. 2 are displayed in order according to the arrows, the steps are not necessarily executed in the order indicated by the arrows. Unless clearly stated in this article, the execution of these steps is not strictly limited in order, and these steps may be executed in other orders. Moreover, at least a part of the steps in FIG. 2 may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed and completed at the same time, but may be executed at different times, the execution of these sub-steps or stages The order is not necessarily sequential, but may be performed in turn or alternately with at least a part of other steps or sub-steps or stages of other steps.

In one embodiment, as shown in FIG. 3, a production data processing device is provided, including: a receiving module 100, a processing module 200, and a sending module 300, wherein:

The receiving module 100 is configured to receive the input predicted time, and obtain corresponding historical production data from the server according to the predicted time.

The processing module 200 is configured to input the predicted time and historical production data to a model classifier to obtain predicted production data corresponding to the predicted time, the predicted production data is the model classifier according to the predicted time and history Obtained from the single-model prediction result of the optimal prediction single model selected by the production data, or the predicted production data is the weight corresponding to each of the predicted single models and the corresponding Single model prediction results are calculated.

The sending module 300 is configured to obtain the investment file corresponding to the predicted production data, and send the investment file to the investment terminal.

In one of the embodiments, the above production data processing device further includes:

The historical data acquisition module is used to acquire historical production data.

The single model prediction result obtaining module is used to obtain the training prediction time, and input the training prediction time and historical production data into the prediction single model to obtain the single model prediction result, and the prediction single model is obtained by pre-training.

The feature construction module is used to synthesize the single-model prediction results and the self-attributes of historical production data to construct the target model selection characteristics that characterize the errors between the single-model prediction results and the real results in the historical production data and the essential characteristics of the historical production data.

The training module is used to train the first model selection feature, the second model selection feature and the prediction single model to obtain a model classifier, and the model classifier is used to select the optimal prediction single model from the prediction single model according to the target model selection feature or It is used to establish corresponding weights for each of the prediction single models according to the target model selection characteristics, and obtain a prediction model according to the weights and the corresponding single model prediction results. In one of the embodiments, the training module may include:

The first extraction unit is used to extract real results from historical production data.

The first determining unit is used to calculate the difference between the real result and the prediction result of the single model, and obtain the prediction single model with the smallest difference as the optimal prediction single model.

The first training unit is used to train the optimal prediction single model and the target model selection features to obtain a model classifier.

In one of the embodiments, the training module may include:

The second extraction unit is used to extract real results from historical production data.

The second determining unit is used to calculate the ratio between the prediction result of the single model and the real result, and obtain the weight of the prediction single model according to the ratio.

The second training unit is used to train the weights of the predicted single model and the selected features of the target model to obtain a model classifier.

In one of the embodiments, the feature building module may include:

The first feature construction unit is used to construct a first model selection feature based on the single model prediction results and historical production data;

The second feature construction unit is used to construct a second model selection feature based on historical production data.

In one of the embodiments, the single-model prediction result acquisition module may include:

The first period obtaining unit is used to obtain a characteristic period corresponding to the training prediction time.

The first prediction value calculation unit is used to obtain the prediction value corresponding to the characteristic period through calculation of the prediction single model.

The first feature building unit may include:

The real value acquisition unit is used to extract the real value corresponding to the characteristic period from the historical production data;

The calculation unit is configured to calculate the first model selection feature according to the first predicted value and the true value.

In one of the embodiments, the second feature construction unit may include:

The first segmenting unit is used to obtain a preset period length and interval, and segment the historical production data according to the preset period length.

The first marking unit is used to obtain the corresponding segmented historical production data in each section, sort the acquired historical production data, and mark the sequence value of the sorted historical production data.

The deviation value calculation unit is used to calculate the deviation value of the sequence value of the historical production data of the segment corresponding to each section.

The periodic strength index calculation unit is used to calculate the average value of the deviation values to obtain the periodic strength index as the second model selection feature.

In one of the embodiments, the second feature construction unit may include:

The second segmenting unit is used to obtain a preset period length, and segment the historical production data according to the preset period length.

The second marking unit is used to calculate the increase and decrease range according to the historical production data of each segment and the historical production data of the previous segment.

The calculation unit of the increase/decrease amplitude value is used to obtain the increase/decrease amplitude mark value of the historical production data of each segment according to the increase/decrease amplitude.

The trend strength index calculation unit is used to calculate the average value of the increase and decrease amplitude marker values to obtain the trend strength index as the second model selection feature. For the specific limitation of the production data processing device, please refer to the above limitation on the method of establishing the model classifier, which will not be repeated here. Each module in the above model classifier building device may be implemented in whole or in part by software, hardware, and a combination thereof. The above modules may be embedded in the hardware form or independent of the processor in the computer device, or may be stored in the memory in the computer device in the form of software so that the processor can call and execute the operations corresponding to the above modules.

In one embodiment, a computer device is provided. The computer device may be a terminal, and an internal structure diagram thereof may be shown in FIG. 4. The computer equipment includes a processor, a memory, a network interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer-readable instructions. The internal memory provides an environment for the operation of the operating system and computer-readable instructions in the non-volatile storage medium. The network interface of the computer device is used to communicate with external terminals through a network connection. The computer readable instructions are executed by the processor to implement a model classifier building method. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covered on the display screen, or may be a button, a trackball, or a touchpad provided on the computer device housing , Can also be an external keyboard, touchpad or mouse.

Those skilled in the art may understand that the structure shown in FIG. 4 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer equipment to which the solution of the present application is applied. The specific computer equipment may It includes more or fewer components than shown in the figure, or some components are combined, or have a different component arrangement.

A computer device includes a memory and one or more processors. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the one or more processors perform the following steps: receiving an input predicted time And obtain the corresponding historical production data from the server according to the predicted time; input the predicted time and historical production data into the model classifier to obtain the predicted production data corresponding to the predicted time. The predicted production data is the model classifier based on the predicted time and The single-model prediction result of the best-predicted single model selected by historical production data, or the predicted production data is calculated based on the prediction time and the weight corresponding to each predicted single model generated by the historical production data and the corresponding single-model prediction result . Obtain the investment file corresponding to the predicted production data, and send the investment file to the investment terminal. .

In one of the embodiments, the training method of the model classifier involved in the execution of the computer-readable instructions by the processor includes: obtaining historical production data; obtaining training prediction time, and inputting the training prediction time and historical production data into the prediction single model The single-model prediction results are obtained in. The single-model prediction is pre-trained. The comprehensive single-model prediction results and the self-attribute construction of historical production data characterize the error between the single-model prediction results and the real results in the historical production data and characterize the history. The target model selection feature of the essential characteristics of the production data; and training the target model selection feature and the prediction single model to obtain the model classifier. The model classifier is used to select the optimal prediction single model or the prediction single model from the prediction single model according to the target model selection feature. It is used to select a feature according to the target model to establish a corresponding weight for each of the prediction single models, and obtain a prediction model according to the weight and the corresponding single model prediction result.

In one embodiment, when the processor executes the computer-readable instructions, the target class model selection feature and the predicted single model are trained to obtain the model classifier, which may include: extracting the real results from the historical production data; calculating the real results and the single The difference between the prediction results of the model, the prediction single model with the smallest difference is obtained as the optimal prediction single model; and the model classifier is obtained by training the selection characteristics of the optimal prediction single model and the target model.

In one embodiment, when the processor executes the computer-readable instructions, the target class model selection feature and the prediction single model are trained to obtain the model classifier, which may include: extracting the real results from the historical production data; calculating the single model prediction results The ratio with the real result is used to obtain the weight of the predicted single model; and the weight of the predicted single model and the selection characteristics of the target model are trained to obtain the model classifier.

In one of the embodiments, when the processor executes the computer-readable instructions, the target model selection feature constructed based on the single model prediction result and historical production data may include: constructing the first model based on the single model prediction result and historical production data Model selection features; and constructing a second model selection feature based on historical production data.

In one embodiment, the input of the training prediction time and historical production data into the prediction single model obtained by the processor executing the computer-readable instructions to obtain the single-model prediction result may include: acquiring the characteristic period corresponding to the training prediction time; The prediction single model calculates the prediction value corresponding to the characteristic period; and the first model selection feature constructed based on the single model prediction result and historical production data when the processor executes the computer-readable instructions may include: extracting from historical production data and The true value corresponding to the feature period; the first model selection feature is calculated according to the predicted value and the true value.

In one embodiment, when the processor executes the computer-readable instructions, the selection feature of constructing the second model based on the historical production data may include: obtaining a preset cycle length and interval, and dividing the historical production data according to the preset cycle length Segment; obtain the historical production data of the corresponding segment in each interval, and sort the acquired historical production data, and mark the sequence value of the sorted historical production data; calculate the historical production of the segment corresponding to each interval The deviation value of the sequence value of the data; and calculating the average value of the deviation values to obtain the periodic strength index as the second model selection feature.

In one embodiment, the selection feature of the second model constructed from the historical production data implemented by the processor when executing the computer-readable instructions may include: obtaining a preset cycle length, and segmenting the historical production data according to the preset cycle length; Calculate the increase/decrease range based on the historical production data of each segment and the historical production data of the previous segment; obtain the increase/decrease amplitude tag value of each segment's historical production data based on the increase/decrease range; and calculate the increase/decrease amplitude tag The average of the values gives the trend strength indicator as the second model selection feature. One or more non-volatile computer-readable storage media storing computer-readable instructions, which when executed by one or more processors, cause the one or more processors to perform the following steps: receive input predictions Time, and obtain the corresponding historical production data from the server according to the predicted time; input the predicted time and historical production data into the model classifier to obtain the predicted production data corresponding to the predicted time, the predicted production data is the model classifier according to the predicted time The single-model prediction result of the optimal single-model model selected from the historical production data, or the predicted production data is calculated according to the weight of each predicted single model generated according to the prediction time and historical production data and the corresponding single-model prediction result of. Obtain the investment file corresponding to the predicted production data, and send the investment file to the investment terminal.

In one of the embodiments, the training method of the model classifier involved in the execution of the computer-readable instructions by the processor includes: acquiring historical production data; acquiring training prediction time, and inputting the training prediction time and historical production data to the prediction sheet The single-model prediction results are obtained from the model, and the predicted single-model prediction is obtained in advance; the comprehensive single-model prediction results and the self-attribute construction of the historical production data characterize the error between the single-model prediction results and the real results in the historical production data and characterize The target model selection feature of the essential characteristics of historical production data; and training the target model selection feature and the prediction single model to obtain a model classifier. The model classifier is used to select the optimal prediction single model from the prediction single model according to the target model selection feature or It is used to establish corresponding weights for each of the prediction single models according to the target model selection characteristics, and obtain a prediction model according to the weights and the corresponding single model prediction results.

In one embodiment, when the computer-readable instructions are executed by the processor, the target class model selection feature and the predicted single model are trained to obtain the model classifier, which may include: extracting the real results from the historical production data; calculating the real results and The difference between the prediction results of the single model, the prediction single model with the smallest difference is obtained as the optimal prediction single model; and the model classifier is obtained by training the selection features of the optimal prediction single model and the target model.

In one embodiment, when the computer readable instructions are executed by the processor, the target class model selection feature and the prediction single model are trained to obtain a model classifier, which may include: extracting real results from historical production data; calculating the single model prediction The ratio between the result and the real result is the weight of the predicted single model according to the ratio; and the model classifier is obtained by training the weight of the predicted single model and the selection characteristics of the target model.

In one of the embodiments, when the computer readable instructions are executed by the processor, the target model selection feature constructed based on the single model prediction result and historical production data may include: constructing the first model based on the single model prediction result and historical production data A model selection feature; and constructing a second model selection feature based on historical production data.

In one embodiment, when the computer-readable instructions are executed by the processor, inputting the training prediction time and historical production data into the prediction single model to obtain a single-model prediction result may include: acquiring a characteristic period corresponding to the training prediction time; The predicted value corresponding to the characteristic period is calculated by predicting the single model; and the first model selection feature constructed based on the single model prediction result and historical production data when the processor executes the computer-readable instructions may include: extracting from the historical production data The true value corresponding to the feature period; the first model selection feature is calculated based on the predicted value and the true value.

In one embodiment, when the computer-readable instructions are executed by the processor, the selection feature of constructing the second model based on historical production data may include: acquiring a preset cycle length and interval, and performing historical production data according to the preset cycle length Segment; obtain the historical production data of the corresponding segment in each interval, sort the acquired historical production data, and mark the sequence value of the sorted historical production data; calculate the history of the segment corresponding to each interval The deviation value of the sequence value of the production data; and calculating the average value of the deviation values to obtain the periodic strength index as the second model selection feature.

In one embodiment, when the computer-readable instructions are executed by the processor, the selection feature of constructing the second model based on historical production data may include: obtaining a preset period length, and segmenting the historical production data according to the preset period length ; Calculate the increase and decrease range according to the historical production data of each section and the historical production data of the previous section; obtain the increase and decrease range mark value of the historical production data of each section according to the increase and decrease range; and calculate the increase and decrease range The average value of the marker values gives a trend strength indicator as the second model selection feature. Those of ordinary skill in the art may understand that all or part of the process in the method of the foregoing embodiments may be completed by instructing relevant hardware through computer-readable instructions, and the computer-readable instructions may be stored in a non-volatile computer In the readable storage medium, when the computer-readable instructions are executed, they may include the processes of the foregoing method embodiments. Wherein, any reference to the memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and/or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

The technical features of the above embodiments can be arbitrarily combined. In order to simplify the description, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be It is considered as the scope described in this specification.

The above-mentioned embodiments only express several implementations of the present application, and their descriptions are more specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be noted that, for those of ordinary skill in the art, without departing from the concept of the present application, a number of modifications and improvements can also be made, which all fall within the protection scope of the present application. Therefore, the protection scope of the patent of this application shall be subject to the appended claims.

Claims (20)

1. A production data processing method, including:

   Receiving the input prediction time, and obtaining corresponding historical production data from the server according to the prediction time;

   Input the predicted time and historical production data to the model classifier to obtain the predicted production data corresponding to the predicted time, the predicted production data is the optimal selected by the model classifier according to the predicted time and historical production data Obtained by predicting the single model prediction result of the single model, or the predicted production data is calculated according to the weight corresponding to each of the predicted single models and the corresponding single model prediction result generated according to the prediction time and historical production data of;

   Obtain the investment file corresponding to the predicted production data, and send the investment file to the investment terminal.
2. The method according to claim 1, wherein the training method of the model classifier includes:

   Obtain historical production data;

   Obtain the training prediction time, and input the training prediction time and the historical production data into a prediction single model to obtain a single model prediction result, and the prediction single model is obtained by pre-training;

   Constructing a target model selection characteristic that characterizes the error between the single model prediction result and the real result in the historical production data and characterizes the essential characteristic of the historical production data by synthesizing the single model prediction result and the self-attributes of the historical production data;

   Training the target model selection feature and the prediction single model to obtain a model classifier, the model classifier is used to select the optimal prediction single model from the prediction single model according to the target model selection feature or use In order to select a feature according to the target model, a corresponding weight is established for each of the prediction single models, and a prediction model is obtained according to the weight and the corresponding prediction result of the single model.
3. The method according to claim 2, wherein the training of the target model selection feature and the prediction single model to obtain a model classifier includes:

   Extract real results from the historical production data;

   Calculating the difference between the real result and the prediction result of the single model, and obtaining the prediction single model with the smallest difference as the optimal prediction single model; and

   Train the optimal prediction single model and the target model selection feature to obtain a model classifier.
4. The method according to claim 2, wherein the training of the first model selection feature, the second model selection feature, and the predicted single model to obtain a model classifier includes:

   Extract real results from the historical production data;

   Calculating the ratio of the prediction result of the single model to the true result, and obtaining the weight of the prediction single model according to the ratio; and

   Train the weights of the predicted single model and the selected features of the target model to obtain a model classifier.
5. The method according to any one of claims 2 to 4, wherein the feature selection based on the prediction result of the single-model prediction and the historical production data includes:

   Construct a first model selection feature based on the single model prediction results and the historical production data; and

   A second model selection feature is constructed based on the historical production data.
6. The method according to claim 5, wherein the input of the training prediction time and the historical production data into the prediction single model to obtain a single model prediction result includes:

   Obtaining a characteristic period corresponding to the training prediction time; and

   Calculating the prediction value corresponding to the characteristic period through the prediction single model calculation;

   The constructing a first model selection feature based on the prediction result of the single model and the historical production data includes:

   Extract the true value corresponding to the characteristic period from the historical production data; and

   The first model selection feature is calculated according to the predicted value and the true value.
7. The method according to claim 5, characterized in that the selection feature constructed by constructing the second model based on the historical production data includes:

   Acquiring a preset period length and interval, and segmenting the historical production data according to the preset period length;

   Obtain the corresponding segmented historical production data in each interval, sort the acquired historical production data, and mark the sequence value of the sorted historical production data;

   Calculate the deviation of the sequence value of the historical production data corresponding to each section; and

   The average value of the deviation values is calculated to obtain a periodic strength index as a second model selection feature.
8. The method according to claim 5, characterized in that the selection feature constructed by constructing the second model based on the historical production data includes:

   Acquiring a preset period length, and segmenting the historical production data according to the preset period length;

   Calculate the increase or decrease according to the historical production data of each segment and the historical production data of the previous segment;

   Obtain the increase and decrease range marker value of the historical production data of each segment according to the increase and decrease range; and

   Calculate the average value of the increase and decrease amplitude marker values to obtain a trend strength indicator as the second model selection feature.
9. A production data processing device, including:

   The receiving module is used to receive the input predicted time, and obtain corresponding historical production data from the server according to the predicted time;

   A processing module, configured to input the predicted time and historical production data to a model classifier to obtain predicted production data corresponding to the predicted time, the predicted production data is the model classifier based on the predicted time and historical production Obtained by the single model prediction result of the optimal prediction single model selected by the data, or the predicted production data is the weight corresponding to each of the predicted single models generated according to the prediction time and historical production data and the corresponding single Calculated by the model prediction results;

   The sending module is used to obtain the investment file corresponding to the predicted production data, and send the investment file to the investment terminal.
10. A computer device includes a memory and one or more processors. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the one or more processors, the one or more Each processor performs the following steps:

    Receiving the input prediction time, and obtaining corresponding historical production data from the server according to the prediction time;

    Input the predicted time and historical production data to the model classifier to obtain the predicted production data corresponding to the predicted time, the predicted production data is the optimal selected by the model classifier according to the predicted time and historical production data Obtained by predicting the single model prediction result of the single model, or the predicted production data is calculated according to the weight corresponding to each of the predicted single models and the corresponding single model prediction result generated according to the prediction time and historical production data of;

    Obtain the investment file corresponding to the predicted production data, and send the investment file to the investment terminal.
11. The computer device according to claim 10, wherein the generation method of the model classifier involved in the execution of the computer-readable instructions by the processor includes:

    Obtain historical production data;

    Obtain the training prediction time, and input the training prediction time and the historical production data into a prediction single model to obtain a single model prediction result, and the prediction single model is obtained by pre-training;

    Constructing a target model selection characteristic that characterizes the error between the single model prediction result and the real result in the historical production data and characterizes the essential characteristic of the historical production data by synthesizing the single model prediction result and the self-attributes of the historical production data; and

    Training the target model selection feature and the prediction single model to obtain a model classifier, the model classifier is used to select the optimal prediction single model from the prediction single model according to the target model selection feature or use In order to select a feature according to the target model, a corresponding weight is established for each of the prediction single models, and a prediction model is obtained according to the weight and the corresponding prediction result of the single model.
12. The computer device according to claim 11, characterized in that the training of the selected feature of the target model and the predicted single model is achieved by the processor when the processor executes the computer-readable instructions to obtain a model classifier ,include:

    Extract real results from the historical production data;

    Calculating the difference between the real result and the prediction result of the single model, and obtaining the prediction single model with the smallest difference as the optimal prediction single model; and

    Train the optimal prediction single model and the target model selection feature to obtain a model classifier.
13. The computer device of claim 11, wherein the pair of the first model selection feature, the second model selection feature, and the prediction list implemented when the processor executes the computer-readable instructions The model is trained to obtain a model classifier, including:

    Extract real results from the historical production data;

    Calculating the ratio of the prediction result of the single model to the true result, and obtaining the weight of the prediction single model according to the ratio; and

    Train the weights of the predicted single model and the selected features of the target model to obtain a model classifier.
14. The computer device according to any one of claims 11 to 13, characterized in that the prediction result based on the single-model prediction and the historical production data implemented when the processor executes the computer-readable instructions Target model selection features, including:

    Construct a first model selection feature based on the single model prediction results and the historical production data; and

    A second model selection feature is constructed based on the historical production data.
15. The computer device according to claim 14, characterized in that the input of the training prediction time and the historical production data to the prediction list model is realized when the processor executes the computer-readable instructions The single model prediction results are obtained in, including:

    Obtaining a characteristic period corresponding to the training prediction time; and

    Calculating the prediction value corresponding to the characteristic period through the prediction single model calculation;

    The first model selection feature constructed based on the single-model prediction result and the historical production data, implemented by the processor executing the computer-readable instructions, includes:

    Extract the true value corresponding to the characteristic period from the historical production data; and

    The first model selection feature is calculated according to the predicted value and the true value.
16. The computer device according to claim 14, characterized in that the selection feature constructed by the processor when the computer readable instructions are executed to construct the second model based on the historical production data includes:

    Acquiring a preset period length and interval, and segmenting the historical production data according to the preset period length;

    Obtain the corresponding segmented historical production data in each interval, sort the acquired historical production data, and mark the sequence value of the sorted historical production data;

    Calculate the deviation of the sequence value of the historical production data corresponding to each section; and

    The average value of the deviation values is calculated to obtain a periodic strength index as a second model selection feature.
17. The computer device according to claim 14, characterized in that the selection feature constructed by the processor when the computer readable instructions are executed to construct the second model based on the historical production data includes:

    Acquiring a preset period length, and segmenting the historical production data according to the preset period length;

    Calculate the increase or decrease according to the historical production data of each segment and the historical production data of the previous segment;

    Obtain the increase and decrease range marker value of the historical production data of each segment according to the increase and decrease range; and

    Calculate the average value of the increase and decrease amplitude marker values to obtain a trend strength indicator as the second model selection feature.
18. One or more non-volatile computer-readable storage media storing computer-readable instructions, which when executed by one or more processors, cause the one or more processors to perform the following steps:

    Receiving the input prediction time, and obtaining corresponding historical production data from the server according to the prediction time;

    Input the predicted time and historical production data to the model classifier to obtain the predicted production data corresponding to the predicted time, the predicted production data is the optimal selected by the model classifier according to the predicted time and historical production data Obtained by predicting the single model prediction result of the single model, or the predicted production data is calculated according to the weight corresponding to each of the predicted single models and the corresponding single model prediction result generated according to the prediction time and historical production data of;

    Obtain the investment file corresponding to the predicted production data, and send the investment file to the investment terminal.
19. The storage medium according to claim 18, wherein the generation method of the model classifier involved in the execution of the computer-readable instructions by the processor includes:

    Obtain historical production data;

    Obtain the training prediction time, and input the training prediction time and the historical production data into a prediction single model to obtain a single model prediction result, and the prediction single model is obtained by pre-training;

    Constructing a target model selection characteristic that characterizes the error between the single model prediction result and the real result in the historical production data and characterizes the essential characteristic of the historical production data by synthesizing the single model prediction result and the self-attributes of the historical production data;

    Training the target model selection feature and the prediction single model to obtain a model classifier, the model classifier is used to select the optimal prediction single model from the prediction single model according to the target model selection feature or use Selecting features according to the target model to establish a corresponding weight for each of the prediction single models, and obtaining a prediction model according to the weights and the corresponding prediction results of the single models; and

    The model classifier obtained by training predicts the production data at the prediction time.
20. The storage medium according to claim 19, wherein the computer-readable instructions are executed by the processor to obtain the model classification by training the target model selection feature and the prediction single model Devices, including:

    Extract real results from the historical production data;

    Calculating the difference between the real result and the prediction result of the single model, and obtaining the prediction single model with the smallest difference as the optimal prediction single model; and

    Train the optimal prediction single model and the target model selection feature to obtain a model classifier.

Images