Enterprise asset estimation method and device, computer disclosure book and storage medium

Abstract

The invention discloses an enterprise asset estimation method and device, a computer disclosure book and a storage medium. The method comprises the steps of obtaining to-be-predicted enterprise related data; preprocessing the to-be-predicted enterprise related data to obtain a preprocessing result; inputting the preprocessing result into a multi-target asset prediction model for asset prediction to obtain a prediction result; performing constraint adjustment according to the prediction result to obtain an estimation result; and outputting the estimation result. By implementing the method provided by the invention, the prediction precision and multi-target synchronous prediction can be improved, the interpretability and transparency of the model are improved, the stability and reliability of prediction are ensured, the consumption of computing resources is reduced, real-time prediction is supported, and the application requirements of high efficiency and real-time performance are met.

Description

Technical Field

[0001] The present invention relates to big data analysis, and more specifically to an enterprise asset estimation method, device, computer briefing document and storage medium. Background Art

[0002] In finance and corporate management, asset estimation is an important task in evaluating the value of corporate assets, which involves making reasonable estimates of the total assets, fixed assets, and current assets of the company. Traditional asset estimation methods mainly include financial analysis, statistical regression analysis, and traditional machine learning methods, each with its own advantages and disadvantages.

[0003] Financial analysis is to estimate the total assets and fixed assets of an enterprise by analyzing its financial statements and financial ratios such as debt-to-asset ratio and current ratio. Financial analysis relies on the financial data provided by the company and can reflect the asset status more intuitively. However, it is highly dependent on data quality and is only applicable to enterprises with complete and transparent financial statements. It also lacks consideration of non-financial factors such as market environment and management capabilities. Statistical regression analysis uses statistical techniques such as linear regression and multivariate regression to establish a mathematical model between assets and related features through historical data to estimate assets. Regression analysis can reveal the relationship between assets and some variables, but it has strict assumptions about data distribution and may produce large errors when dealing with complex nonlinear relationships and high-dimensional data. Traditional machine learning methods such as decision trees and random forests can handle some nonlinear relationships and perform well in structured data prediction tasks. These methods have high requirements for feature selection and data preprocessing, and are prone to overfitting problems, especially when multiple prediction targets are involved. It is difficult to ensure that the prediction results of the model have consistent business logic between different targets. The deep learning method uses deep learning algorithms such as neural networks such as LSTM to estimate assets. It is particularly good at processing unstructured data such as text and images, and can capture more complex nonlinear relationships. However, the deep learning method has high computational overhead, high data volume requirements, and lacks sufficient interpretability. In addition, it is difficult to embed business logic constraints into deep learning models, which may make the model output contradict actual business needs.

[0004] In summary, the existing asset estimation methods generally have the following problems: First, there is a lack of effective processing of multi-target predictions, and it is impossible to directly model the relationship between targets; second, there is a lack of sufficient business logic support, which easily leads to results that do not meet actual business needs; third, in the case of high errors, they fail to effectively deal with the risks brought by errors; fourth, they are highly data-dependent and lack interpretability, especially the black box characteristics of deep learning models are more obvious; fifth, in application scenarios such as real-time prediction, the computing resources consumption is large, and it is difficult to meet real-time requirements.

[0005] Therefore, it is necessary to design a new method to improve prediction accuracy and multi-target synchronous prediction, improve the interpretability and transparency of the model, ensure the stability and reliability of the prediction, reduce the consumption of computing resources, support real-time prediction, and meet the application requirements of high efficiency and real-time performance. Summary of the invention

[0006] The purpose of the present invention is to overcome the defects of the prior art and provide an enterprise asset estimation method, device, computer briefing document and storage medium.

[0007] To achieve the above purpose, the present invention adopts the following technical solution: a method for estimating enterprise assets, comprising:

[0008] Obtain relevant data of the enterprise to be predicted;

[0009] Preprocessing the data related to the enterprise to be predicted to obtain a preprocessing result;

[0010] Inputting the preprocessing results into a multi-objective asset prediction model to perform asset prediction to obtain a prediction result;

[0011] Perform constraint adjustments based on the prediction results to obtain an estimation result;

[0012] The estimation result is outputted.

[0013] A further technical solution is: preprocessing the data related to the enterprise to be predicted to obtain a preprocessing result, including:

[0014] The data related to the enterprise to be predicted is processed for null values, abnormal values, and repeated values, and feature correlation analysis is performed to obtain preprocessing results.

[0015] Its further technical solution is: the multi-objective asset prediction model is a model obtained by performing multi-objective training on a sample set formed by preprocessing the company's internal public data and industry and province numerical data using the XGBoost algorithm combined with a custom loss function and an evaluation function.

[0016] The further technical solution is: the multi-objective asset prediction model is a model obtained by performing multi-objective training on a sample set formed by preprocessing the internal public data of the enterprise and the numerical data of the industry and province using the XGBoost algorithm combined with a custom loss function and an evaluation function, including:

[0017] Obtaining public internal data of enterprises and numerical data of industries and provinces;

[0018] Preprocessing the enterprise's internal public data and industry and province numerical data to obtain a sample set;

[0019] Use the XGBoost algorithm to build a multi-objective model for simultaneously predicting total assets and fixed assets;

[0020] Customize loss function and evaluation function;

[0021] Using the sample set to train the multi-objective model, wherein the model performance is improved through parameter tuning and early stopping mechanism;

[0022] The loss function and the evaluation function are used to evaluate the trained multi-objective model to obtain a multi-objective asset prediction model.

[0023] Its further technical solution is: the loss function includes the total asset error and relative error percentage, the fixed asset error and relative error percentage, the penalty factor of the total assets, the penalty factor of the fixed assets and the constraint penalty, and the multi-objective model parameters are adjusted by calculating the gradient and Hessian matrix of the loss function relative to the trained multi-objective model parameters.

[0024] Its further technical solution is: the evaluation function includes the absolute error percentage of total assets, the absolute error percentage of fixed assets, the constraint violation rate and the average error percentage; wherein the constraint violation rate refers to the proportion of the fixed asset forecast value exceeding the total asset forecast value; the average error percentage refers to the average value of the absolute error percentage of total assets, the absolute error percentage of fixed assets and the constraint violation rate.

[0025] A further technical solution is: using the sample set to train the multi-objective model, wherein the model performance is improved through parameter tuning and early stopping mechanism, including:

[0026] The multi-objective model is trained using the sample set by setting major hyperparameters, hyperparameters are adjusted using grid search, and early stopping mechanism is adopted for tuning.

[0027] The present invention also provides an enterprise asset estimation device, comprising:

[0028] A data acquisition unit, used to acquire relevant data of the enterprise to be predicted;

[0029] A preprocessing unit, used for preprocessing the data related to the enterprise to be predicted to obtain a preprocessing result;

[0030] An asset prediction unit, used for inputting the preprocessing result into a multi-objective asset prediction model to perform asset prediction to obtain a prediction result;

[0031] A constraint adjustment unit, used to perform constraint adjustment according to the prediction result to obtain an estimation result;

[0032] An output unit is used to output the estimation result.

[0033] The present invention further provides a computer device, which includes a memory and a processor. The memory stores a computer program, and the processor implements the above method when executing the computer program.

[0034] The present invention also provides a storage medium, wherein the storage medium stores a computer program, and the computer program implements the above method when executed by a processor.

[0035] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention ensures the quality of input data, reduces the influence of noise, and improves the accuracy of model prediction by cleaning and standardizing the relevant data of the enterprise to be predicted; designs a customized multi-objective loss function, optimizes multiple prediction targets, realizes synchronous prediction between targets, and enhances the adaptability and accuracy of the model; adjusts business logic constraints based on the prediction results to ensure that the results meet the actual financial requirements, while improving the reliability and interpretability of the prediction; avoids overfitting and reduces unnecessary consumption of computing resources by introducing an early stopping mechanism and dynamically adjusting the training process, while ensuring the stability of the model; supports rapid feedback and adjustment through real-time evaluation and training of the early stopping mechanism, meets the application requirements of high efficiency and real-time, and improves the transparency and comprehensibility of the prediction results.

[0036] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0038] Figure 1 A schematic diagram of an application scenario of the enterprise asset estimation method provided by an embodiment of the present invention;

[0039] Figure 2 A schematic diagram of a process flow of an enterprise asset estimation method provided by an embodiment of the present invention;

[0040] Figure 3 A schematic block diagram of an enterprise asset estimation device provided by an embodiment of the present invention;

[0041] Figure 4 A schematic block diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0044] It should also be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0045] It should be further understood that the term "and / or" used in the present description and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0046] See also Figure 1 and Figure 2 , Figure 1 A schematic diagram of an application scenario of the enterprise asset estimation method provided in an embodiment of the present invention. Figure 2A schematic flow chart of an enterprise asset estimation method provided by an embodiment of the present invention. The enterprise asset estimation method is applied to a server. The server interacts with the terminal for data, and by processing null values, outliers and duplicate values, and performing feature correlation analysis, the data quality is effectively improved, ensuring that the data input into the model is more accurate and meaningful, thereby improving the prediction accuracy. The XGBoost algorithm is combined with a custom loss function and an evaluation function to construct a multi-objective model for simultaneously predicting total assets and fixed assets, ensuring the synchronous prediction of the two asset targets and enhancing the comprehensiveness of the model. By introducing the total asset error, fixed asset error and constraint penalty factor into the loss function, the model training process is optimized, the accuracy and stability of the prediction results are improved, and the occurrence of unreasonable predictions is avoided. An evaluation function including the absolute error percentage, the constraint violation rate and the average error percentage is designed to ensure a comprehensive evaluation of the prediction results, thereby improving the interpretability and transparency of the model. The hyperparameter configuration of the model is optimized through grid search and early stopping mechanism, overfitting is reduced, the stability and reliability of the model are improved, and unnecessary consumption of computing resources is avoided. By using the tuned multi-objective model, the consumption of computing resources is reduced, the computing efficiency is improved, real-time prediction is supported, and the efficiency requirements in practical applications are met. By optimizing model performance and computing efficiency, this method can support real-time asset prediction, ensure the timeliness of prediction and the efficiency of application, and meet the real-time requirements of actual business.

[0047] Figure 2 FIG. 1 is a flow chart of the enterprise asset estimation method provided by an embodiment of the present invention. Figure 2 As shown, the method includes the following steps S110 to S150.

[0048] S110, obtaining relevant data of the enterprise to be predicted.

[0049] In this embodiment, the enterprise-related data to be predicted refers to the enterprise's internal public data as well as industry and province numerical data.

[0050] Before conducting an asset estimation, you first need to collect various data related to the target company; select the required company information from the company's internal database, including but not limited to the company's profile (non-sensitive information), such as the tax number and name after hash encoding, to protect the company's privacy.

[0051] Industry and province numerical data: Focus on the primary industries of manufacturing and construction, because enterprises in these two industries usually have significant fixed assets, which is of great value for multi-objective learning, that is, predicting total assets and fixed assets at the same time. The data types collected cover industry statistics, such as the maximum value of total assets, quantiles (such as current assets, actual tax paid, prepaid payments, etc.).

[0052] S120: preprocessing the data related to the enterprise to be predicted to obtain a preprocessing result.

[0053] In this embodiment, the preprocessing result refers to the result obtained after data cleaning and other processing are performed on the relevant data of the enterprise to be predicted.

[0054] Specifically, the data related to the enterprise to be predicted is processed for null values, abnormal values, and repeated values, and feature correlation analysis is performed to obtain preprocessing results.

[0055] In this embodiment, in order to ensure the quality of the data, a series of cleaning and preprocessing operations are required before formal use, including:

[0056] Null value processing: Evaluate data columns with missing values ​​and delete features with a missing ratio of more than 50% to ensure the overall integrity and reliability of the data set.

[0057] Outlier detection and processing: Use statistical methods or machine learning algorithms to identify and process outliers and reduce the impact of noise on the model.

[0058] Duplicate value removal: Check and remove possible redundant records in the data set to prevent duplicate data from affecting model training results.

[0059] Feature selection optimization: Analyze the correlation between features based on statistical indicators such as the Pearson correlation coefficient, remove redundant features with high correlation (such as correlation greater than 0.8), and improve model efficiency.

[0060] Of course, after preprocessing, you can also perform feature engineering on the data to extract information that is helpful to the model. The tasks at this stage include but are not limited to:

[0061] Create new feature variables based on business logic and domain knowledge, such as calculating certain financial ratios, to provide the model with more dimensional information. For non-numeric features, such as provinces or industry categories, use appropriate encoding methods to convert them into numerical form for easier model processing. Standardize or normalize numerical features so that data of different magnitudes can be compared on the same scale, which helps improve model performance.

[0062] S130, inputting the preprocessing result into a multi-objective asset prediction model to perform asset prediction to obtain a prediction result.

[0063] In this embodiment, the prediction results refer to the predicted value of fixed assets and the predicted value of total assets.

[0064] The multi-objective asset prediction model is a model obtained by performing multi-objective training on a sample set formed by preprocessing the company's internal public data and industry and province numerical data using the XGBoost algorithm combined with a custom loss function and an evaluation function.

[0065] In one embodiment, the above-mentioned step S130 may include steps S131 - S136 .

[0066] S131. Obtain the company’s internal public data as well as industry and province numerical data.

[0067] In this embodiment, the internal public data of the enterprise includes the enterprise portrait (sensitive data is removed, and the enterprise tax number and name are replaced by hash code), various relevant numerical data of the industry and province of the enterprise (mainly statistical information of the industry, such as the maximum total assets of this industry, the total assets percentile, the current assets percentile, the actual tax paid percentile, the prepayment percentile, etc.), and only the enterprises with the first-level industry of manufacturing and construction are taken, because enterprises in these two industries generally have fixed assets. This modeling is mainly multi-objective learning of fixed assets and total assets.

[0068] By combining internal enterprise data with external data from industries and provinces, we can gain a more comprehensive understanding of the financial performance of enterprises and the market environment they are in. Rich data sources help improve the accuracy and stability of forecasts, because multi-dimensional data can reveal more potential relationships and influencing factors.

[0069] S132. Preprocess the enterprise's internal public data and industry and province numerical data to obtain a sample set.

[0070] In this embodiment, the sample set refers to the result obtained after preprocessing the internal public data of the enterprise and the numerical data of the industry and province. It should be noted that the internal public data of the enterprise and the numerical data of the industry and province here are historical data and already have real fixed assets and total assets data.

[0071] Data preprocessing ensures the quality of data input to the model, improves training efficiency and model prediction accuracy. Processing missing values, outliers, and duplicate data can prevent the model from being affected by noise and reduce false predictions. Feature selection helps remove redundant and irrelevant features, avoid overfitting, and improve the generalization ability of the model.

[0072] In addition, the sample set can be divided into training set, validation set and test set in the ratio of 7:1.5:1.5. Because the model uses XGBoost, the model itself can handle non-numerical features and null values ​​very well. The algorithm can automatically find the best direction according to the distribution of the data and place the missing values ​​on the optimal leaf node, reducing the steps of data preprocessing.

[0073] S133. Use the XGBoost algorithm to build a multi-objective model for simultaneously predicting total assets and fixed assets.

[0074] In this example, XGBoost is an efficient gradient boosting decision tree algorithm with good generalization ability, which is particularly suitable for processing complex data relationships. It supports multi-objective learning and can optimize multiple outputs at the same time, meeting the need to simultaneously predict total assets and fixed assets in this case.

[0075] The XGBoost (Extreme Gradient Boosting) algorithm is used, which is an integrated learning method based on gradient boosting and is suitable for the prediction of large-scale data sets and complex problems.

[0076] In this step, XGBoost is used to build a multi-objective model to predict total assets and fixed assets at the same time. XGBoost has become a widely used machine learning algorithm in the industry due to its efficient computing power and excellent prediction performance. The multi-objective model allows multiple related targets (such as total assets and fixed assets) to be predicted simultaneously, rather than just a single target, thereby improving the accuracy and efficiency of the overall prediction. XGBoost is able to handle large-scale data sets and reduce overfitting, so it is very suitable for big data environments in actual business scenarios.

[0077] S134. Customize loss function and evaluation function.

[0078] In this embodiment, the loss function includes the total asset error and relative error percentage, the fixed asset error and relative error percentage, the penalty factor of the total assets, the penalty factor of the fixed assets and the constraint penalty, and the multi-objective model parameters are adjusted by calculating the gradient and Hessian matrix of the loss function relative to the trained multi-objective model parameters.

[0079] Specifically, the design goal of the custom loss function is to adjust according to the error of total assets and fixed assets, and to perform dynamic penalties in combination with business logic. The goal of business requirements is to have as many fixed assets and total assets with a relative error rate within 0.5 as possible, and fixed assets should be smaller than total assets. The main components of the loss function can be summarized as the total asset error and relative error percentage, the fixed asset error and relative error percentage, the penalty factor of total assets, the penalty factor of fixed assets, and the constraint penalty.

[0080] Total assets error:

[0081] Fixed asset error:

[0082] in, is the predicted value of total assets, is the predicted value of fixed assets, y total is the true value of total assets, y fixe d is the actual value of fixed assets.

[0083] Total Assets Relative Error Percentage: This section calculates the relative error between the predicted and actual total assets. It is defined as:

[0084] Fixed Assets Relative Error Percentage: Similarly, calculate the relative error of fixed assets. Defined as:

[0085]

[0086] Here ò is a small value used to avoid division by zero errors.

[0087] The loss function uses a penalty factor to control the impact of large errors, especially when the prediction error exceeds a set threshold, the loss function will increase the corresponding penalty. Specifically, the model's sensitivity to error is adjusted, especially when the prediction error exceeds a certain threshold, the penalty for large errors is increased by increasing the loss value, thereby improving the model's prediction accuracy and ensuring that the prediction results are consistent with business logic.

[0088] Set the error threshold δ = 0.5, and dynamically calculate the penalty factor based on the error percentage. The penalty factor is related to the error size and hyperparameters. There are several specific cases:

[0089] Penalty factor for total assets:

[0090]

[0091] Penalty factor for fixed assets:

[0092]

[0093] Where M=8, M is a hyperparameter in the loss function, obtained through grid search, and is also a maximum penalty threshold set to prevent the penalty from growing infinitely.

[0094] Different penalties are imposed on different business logics, and dynamic penalties are imposed based on the size of the error and the hyperparameter M. When the relative error percentage is greater than the set threshold, the penalty is multiplied by M. When the relative error percentage is less than the negative threshold, the penalty is multiplied by Since the data of this application are basically data of small and medium-sized enterprises, generally speaking, the values ​​of fixed assets and total assets will not be too large, so when the predicted relative error percentage is greater than the set threshold, this embodiment will increase the penalty factor.

[0095] Business logic requires that fixed assets should be less than total assets. If the fixed assets predicted by the model exceed the total assets, an additional penalty needs to be imposed. This part of the constraint can be expressed as:

[0096]

[0097] The optimization process is guided by calculating the gradient (first-order derivative) and Hessian matrix (second-order derivative) of the loss function with respect to the model parameters:

[0098] Gradient (first derivative): used to calculate the partial derivative of the loss function with respect to each model parameter, indicating how to update the parameters. In each iteration, the value of the gradient is used to measure the degree to which the predicted value deviates from the true value and guide how the model is updated.

[0099] Hessian matrix (second derivative): used to calculate the second-order derivative of the loss function, helping to adjust the step size during the optimization process so that the model can converge more stably; describes the curvature of the loss curve. Hessian is used in XGBoost to adjust the update step size of each step to make the optimization process more stable and efficient.

[0100] Specifically, the calculation process of the above gradient and Hessian matrix is ​​as follows:

[0101] Gradient of total assets:

[0102] Gradient of fixed assets:

[0103] The Hessian matrix values ​​correspond to the second-order derivatives of each objective, adjusted based on the penalty factor.

[0104] The Hessian matrix of total assets:

[0105] The Hessian matrix of fixed assets:

[0106]

[0107] Among them, α is the adjustment parameter of the second-order derivative, which is used to update the step size. It is also set as a hyperparameter of the loss function here, α = 6.5, which is also obtained by grid search.

[0108] The final gradient matrix:

[0109] The final Hessian matrix:

[0110] The evaluation function includes the absolute error percentage of total assets, the absolute error percentage of fixed assets, the constraint violation rate and the average error percentage; wherein the constraint violation rate refers to the proportion of the fixed asset forecast value exceeding the total asset forecast value; the average error percentage refers to the average value of the absolute error percentage of total assets, the absolute error percentage of fixed assets and the constraint violation rate.

[0111] In this example, the custom evaluation function custom_eval is designed to monitor the prediction error in real time during the training process to ensure that the model is not only optimized in terms of accuracy, but also complies with actual business rules, especially the constraints of total assets and fixed assets. According to the description you provided, this evaluation function mainly includes the following aspects:

[0112] The absolute error percentage for total assets calculates the percentage of absolute error between the predicted and actual values ​​of total assets and measures whether the error exceeds a set threshold. The absolute error percentage for fixed assets is similar to the error for total assets, where the error for fixed assets is also measured by calculating the percentage between the predicted and actual values ​​to measure the severity of the error. The percentage by which the fixed asset forecast exceeds the total asset forecast is used to assess whether the fixed asset forecast violates the business rule (i.e., the fixed asset forecast cannot exceed the total asset forecast). Constraint violations are measured by calculating the percentage by which the fixed asset forecast exceeds the total asset forecast. The average values ​​of these three indicators are evaluated to provide an overall error evaluation standard. The average error percentage combines the total asset error, fixed asset error, and constraint violations to measure the overall performance of the model.

[0113] Specifically, in addition to the custom loss function, a custom evaluation function custom_eval is also designed to monitor the prediction error during training. The specific implementation method is as follows:

[0114] Error percentage calculation: Calculate the absolute error percentage for total assets and fixed assets, and set a threshold to measure the severity of the error.

[0115] Constraint violation rate: Evaluates the ratio of the fixed asset forecast value exceeding the total asset forecast value to ensure that the model prediction results are within the range of actual business rules.

[0116] Absolute error percentage of total assets:

[0117] Absolute error percentage for fixed assets:

[0118] Here ò is a small value used to avoid division by zero errors.

[0119] The error threshold δ is set to 0.5 to determine whether the error exceeds a certain ratio.

[0120] The proportion of total assets whose error exceeds the threshold:

[0121]

[0122] The proportion of fixed assets with errors exceeding the threshold:

[0123]

[0124] Where n is the number of samples, 1(·) is an indicator function that returns 1 if the error percentage is greater than the threshold δ, and 0 otherwise.

[0125] The percentage of the forecast value of fixed assets exceeding the forecast value of total assets: The indicator function 1(·) here means that if the forecast value of fixed assets is greater than the forecast value of total assets, it returns 1, otherwise it returns 0.

[0126] Finally, the mean error percentage is defined as the average of the three ratios above:

[0127]

[0128] By customizing the loss function, you can tailor the model more precisely to your business needs, such as emphasizing the importance of relative error rate and ensuring that fixed asset forecasts do not exceed total assets.

[0129] Custom evaluation functions help developers understand the performance of the model in different aspects, such as error percentage, constraint violation rate, etc., making it easier to adjust model parameters to achieve better results.

[0130] S135. Training the multi-objective model using the sample set, wherein the model performance is improved through parameter tuning and early stopping mechanism.

[0131] In this embodiment, the multi-objective model is trained using the sample set by setting major hyperparameters, and hyperparameters are adjusted using grid search, and an early stopping mechanism is used for tuning.

[0132] Specifically, parameter tuning (e.g., through grid search) can help find the optimal hyperparameter combination and further improve the performance of the model. The early stopping mechanism prevents the model from being overtrained, that is, the training is terminated early when the performance on the validation set no longer improves, which not only saves computing resources but also avoids overfitting problems.

[0133] Specifically, Grid Search is a common hyperparameter optimization method that searches for the optimal combination by traversing all possible hyperparameter combinations.

[0134] In this embodiment, different hyperparameters are selected for trial through grid search, such as learning rate, tree depth, minimum number of sample splits, subsample ratio, etc., and the performance of each combination on the validation set is calculated, and finally the best performing combination is selected.

[0135] Grid search traverses all preset hyperparameter spaces to ensure that the best combination of each hyperparameter is found, thereby improving the performance of the model. By adjusting important hyperparameters, grid search can help find the optimal parameter configuration, so that the model can perform better on both the training set and the validation set, improving the accuracy and stability of the prediction.

[0136] The early stopping mechanism is designed to prevent the model from overfitting. During the training process, the model is continuously optimized, but too many training times may cause the model to overfit, especially when the error of the validation set starts to rise.

[0137] When using the early stopping mechanism, if the performance (such as loss or accuracy) on the validation set no longer improves after a number of rounds of training, the training process will stop early. This strategy can significantly save computing resources and help avoid overfitting of the model.

[0138] The early stopping mechanism can effectively prevent the model from "remembering" the noise of the training data during training, thereby avoiding overfitting. This makes the model more generalizable and performs better when faced with new data. By stopping training when the performance of the validation set no longer improves, a lot of computing time and resources can be saved, avoiding unnecessary computing costs. The early stopping mechanism makes the training process more efficient, shortens the training time, and ensures the training quality of the model.

[0139] Combination of parameter tuning and early stopping mechanism: During the training process, grid search finds the optimal hyperparameter combination for the model to ensure that the model is trained in the best state; while the early stopping mechanism ensures that there is no overfitting during the training process and effectively saves resources. This comprehensive method can effectively improve the overall performance of the model, ensuring accuracy and preventing overfitting, and ultimately obtaining a multi-objective prediction model with excellent performance.

[0140] In addition, the main hyperparameters include:

[0141] eta(learning rate) = 0.03, the learning rate controls the weight adjustment of each tree. A lower learning rate usually requires more iterations to converge.

[0142] max_depth (tree depth) = 6, which refers to the maximum depth of the tree. This value controls the complexity of the tree. A tree that is too deep is prone to overfitting, while a tree that is too shallow may lead to underfitting.

[0143] colsample_bytree (column sampling ratio) = 0.85, the proportion of features randomly sampled when each tree is built. Usually set to 0.7 to 0.9, which helps reduce overfitting.

[0144] subsample (row sampling ratio) = 0.8, the ratio of training set samples for each tree, randomly selecting samples for training. The smaller the value, the more the model can reduce overfitting, but if it is too small, it may lead to underfitting.

[0145] lambda (L2 regularization term) = 1, L2 regularization term, used to control the complexity of the model. The larger the value, the stronger the regularization, which can avoid overfitting of the model.

[0146] Gamma (minimum splitting loss) = 0.2, used to control the splitting of the tree. The larger the value, the higher the model's requirements for each split and the more difficult it is to split. A higher gamma helps reduce overfitting.

[0147] num_rounds (number of training rounds) = 1000; the total number of training rounds. Usually, the training process will be stopped early according to the early stopping mechanism.

[0148] early_stopping_rounds = 80; if the model does not improve within 80 rounds, training will automatically stop to avoid overfitting.

[0149] S136. Use the loss function and the evaluation function to evaluate the trained multi-objective model to obtain a multi-objective asset prediction model.

[0150] In this embodiment, a rigorous evaluation process is used to confirm whether the model has sufficient prediction accuracy and stability, thereby enhancing confidence in the model. Based on the evaluation results, possible deficiencies in the model can be identified, providing direction for future improvements.

[0151] XGBoost supports custom loss functions by defining a function to control the optimization target during training. The target of this embodiment is multi-task regression or classification problem, and the loss function can be defined according to specific needs. Similarly, XGBoost also allows custom evaluation functions. During the training process, we use early_stopping_rounds to implement the early stopping mechanism. The training process will automatically stop if the performance of the validation set does not improve within the specified number of rounds. XGBoost will output the loss value and evaluation results of each round during the training process by default. In order to facilitate subsequent parameter adjustments, these results can be recorded in a log file.

[0152] Through the above settings and processes, you can configure the hyperparameters of the XGBoost model, including learning rate, tree depth, column sampling ratio, etc. Use custom loss functions and evaluation functions to ensure that the training process meets the requirements of multi-objective optimization. Use the early stopping mechanism to avoid overfitting and improve training efficiency. Record training logs for subsequent review and adjustment of the training process.

[0153] Summary of the settings: eta=0.03; max_depth=6; colsample_bytree=0.85; subsample=0.8; lambda=1; gamma=0.2; early_stopping_rounds=80; num_rounds=1000.

[0154] Based on these parameter configurations, an optimized XGBoost model can be effectively trained and the parameters can be adjusted as needed to improve performance.

[0155] S140: performing constraint adjustment according to the prediction result to obtain an estimation result.

[0156] In this embodiment, in the multi-objective asset prediction model, although the model can predict multiple target variables (such as total assets and fixed assets) at the same time, since the targets of the multi-objective regression model are not always related, it sometimes outputs illogical results. For example, the predicted value of fixed assets may be greater than the predicted value of total assets, which is unacceptable in practice because fixed assets are part of total assets and usually fixed assets will not be greater than total assets.

[0157] After the model completes the forecast, apply constraint adjustments to the forecast values ​​of fixed assets, using the following rules:

[0158] Fixed assets = min (fixed assets forecast value, total assets forecast value);

[0159] Specifically, if the model predicts a fixed asset value that is greater than the corresponding total asset value, the forecast value of the fixed asset is adjusted to the total asset value. This adjustment ensures that the forecast value of the fixed asset will never exceed the forecast value of the total asset.

[0160] This adjustment is based on the actual financial situation of the business, because total assets include all assets (current and non-current assets), and fixed assets usually only account for a portion of them. Obviously, the total value of fixed assets cannot exceed the total value of the total assets of the business, so this adjustment step ensures that the forecast results are in line with common sense.

[0161] In this way, the output of the model not only meets the requirements in a statistical sense, but also meets the needs of actual business, improving the business interpretability and applicability of the model.

[0162] This constraint adjustment not only solves the problem that the model may output unreasonable predictions, but also enhances the credibility of the model's prediction results. In practical applications, companies can rely on these predictions to make decisions, such as capital allocation, fixed asset investment, etc. By ensuring the logical consistency that fixed assets do not exceed total assets, the prediction results can better reflect the company's true financial situation.

[0163] In short, this constraint adjustment mechanism is an effective way to combine financial logic and machine learning models, ensuring the rationality and feasibility of the model prediction results in practical applications.

[0164] S150: Output the estimation result.

[0165] Output the estimation results to the terminal display.

[0166] The method of this embodiment embeds business logic into the model through a custom loss function and penalty mechanism, and designs a dynamic evaluation and training callback mechanism. The early stopping mechanism can be regarded as a callback mechanism, which provides a new idea for the application of XGBoost in multi-objective prediction. It can optimize multiple prediction targets (such as total assets and fixed assets) at the same time, effectively control errors, and ensure that the prediction results are consistent with business logic, thereby improving the accuracy, applicability and training efficiency of the model.

[0167] The traditional XGBoost model is usually applied to single-target prediction. However, this embodiment designs a customized multi-target loss function so that the model can optimize multiple prediction targets at the same time and dynamically adjust the penalty intensity according to the error of each target. The innovation of the loss function is reflected in two aspects:

[0168] When the prediction error exceeds the preset threshold, the model will automatically increase the penalty and prioritize the optimization of targets with large errors, thereby improving the accuracy of the prediction.

[0169] Differentiate between positive and negative errors and apply differentiated penalties based on business needs. For example, different penalties are imposed on overestimation or underestimation errors to meet specific financial forecasting requirements.

[0170] This custom loss function enhances the performance of XGBoost in multi-objective prediction, enabling it to not only accurately predict total assets and fixed assets, but also ensure that the error is controlled within a reasonable range.

[0171] When predicting corporate financial data, fixed assets should always be less than or equal to total assets, which is basic business logic. This embodiment embeds this business logic constraint in the loss function to ensure that when the predicted value of fixed assets exceeds total assets, the model automatically increases losses to avoid unreasonable prediction results. In this way, while optimizing the error, the model can also ensure that the prediction meets the actual financial requirements, improve the business interpretability and applicability of the model, and is particularly suitable for financial prediction tasks that require multi-objective constraints.

[0172] This embodiment also designs a custom training callback function to monitor the error changes of the model in real time and output the evaluation results regularly. Combined with the early stopping mechanism, the model can adjust the training process according to the real-time evaluation results and automatically stop in advance at the iteration with the best performance, thereby avoiding overfitting and improving training efficiency and stability.

[0173] This embodiment is not only applicable to asset prediction, but can also be extended to multi-objective scenarios such as bank credit assessment and portfolio management. By flexibly configuring the loss function, evaluation mechanism and early stopping adjustment, the model is highly adaptable and versatile, and can meet the multi-objective prediction needs in various financial fields.

[0174] By properly designing the loss function and evaluation mechanism, the prediction results of the model are highly consistent with actual business needs. The built-in penalty mechanism not only controls the error, but also provides users with a clear explanation to help them understand the source of the prediction error and the direction of model optimization. At the same time, the real-time output of the evaluation results allows users to clearly understand the error status in each training iteration, helping business analysis and decision-making, and improving the comprehensibility and trust of the model in practical applications.

[0175] The core advantage of this embodiment is that it can effectively embed complex business logic into the model, improve prediction accuracy and business adaptability, while ensuring the interpretability and efficiency of the model, and is suitable for a wide range of multi-objective financial forecasting tasks.

[0176] Specifically, the above method can accurately predict the total assets and fixed assets of an enterprise at the same time by introducing a custom loss function and an evaluation function. The custom loss function fully considers the errors of multiple target variables and imposes a penalty when the error exceeds the preset threshold, thereby enhancing the model's sensitivity to large errors. This design ensures that the model can maintain a high prediction accuracy on multiple targets, meeting the actual needs of enterprises in multi-target simultaneous prediction.

[0177] This model adds business constraints to the loss function, such as the fixed asset forecast value cannot exceed the total asset forecast value. If the fixed asset forecast value is greater than the total asset forecast value, the model will adjust the loss function by adding a penalty factor, thereby prompting the model to optimize the forecast results. This mechanism ensures that the forecast results are consistent with financial common sense, avoids unreasonable forecast values, and improves the practical application value of the model.

[0178] The introduced adaptive penalty factor is dynamically adjusted according to the size of the prediction error, and can impose higher penalties on larger errors. This makes the model more sensitive to outliers and extreme situations, thereby effectively reducing the risk of large errors in predictions and enhancing the robustness and stability of the model in practical applications.

[0179] The model based on the XGBoost algorithm has excellent training performance and can efficiently process large-scale data sets. Through the advantages of parallel computing and incremental learning, the model can complete training in a relatively short time. At the same time, the use of a customized early stopping mechanism and training callback function can automatically terminate training when the model performance stagnates, saving computing resources and preventing overfitting, thereby improving training efficiency.

[0180] This embodiment designs a custom loss function and evaluation function, which has high flexibility. The structure of the loss function and the penalty strategy can be flexibly adjusted according to different business needs and target variables. This flexibility makes the model not only suitable for enterprise asset forecasting, but also can be extended to other multi-objective forecasting tasks, with good adaptability and scalability.

[0181] The XGBoost model automatically handles missing values ​​and outliers, reducing the complexity of data preprocessing. The model is insensitive to feature scaling, so it has lower requirements in feature engineering, simplifies data preparation, and improves development efficiency.

[0182] Through accurate total asset and fixed asset forecasts, enterprises can effectively predict financial risks and optimize the decision-making process. This helps enterprises to plan asset allocation and allocate financial resources more reasonably, thereby reducing financial management risks and enhancing the scientificity and rationality of decision-making.

[0183] It can be seen that the method of this embodiment can meet the needs of daily financial analysis, management and risk control of enterprises.

[0184] The above-mentioned enterprise asset estimation method ensures the quality of input data, reduces the impact of noise, and improves the accuracy of model prediction by cleaning and standardizing the relevant data of the enterprise to be predicted; designs a customized multi-objective loss function, optimizes multiple prediction targets, and realizes synchronous prediction between targets, thereby enhancing the adaptability and accuracy of the model; adjusts business logic constraints based on the prediction results to ensure that the results meet the actual financial requirements, while improving the reliability and interpretability of the prediction; avoids overfitting and reduces unnecessary consumption of computing resources while ensuring model stability by introducing an early stopping mechanism and dynamically adjusting the training process; supports rapid feedback and adjustment through real-time evaluation and training early stopping mechanisms, meets the application requirements of high efficiency and real-time performance, and improves the transparency and comprehensibility of the prediction results.

[0185] Figure 3 is a schematic block diagram of an enterprise asset estimation device 300 provided in an embodiment of the present invention. Figure 3 As shown, corresponding to the above enterprise asset estimation method, the present invention also provides an enterprise asset estimation device 300. The enterprise asset estimation device 300 includes a unit for executing the above enterprise asset estimation method, and the device can be configured in a server. Figure 3 The enterprise asset estimation device 300 includes a data acquisition unit 301, a preprocessing unit 302, an asset prediction unit 303, a constraint adjustment unit 304 and an output unit 305.

[0186] The data acquisition unit 301 is used to acquire the relevant data of the enterprise to be predicted; the preprocessing unit 302 is used to preprocess the relevant data of the enterprise to be predicted to obtain the preprocessing result; the asset prediction unit 303 is used to input the preprocessing result into the multi-objective asset prediction model to perform asset prediction to obtain the prediction result; the constraint adjustment unit 304 is used to adjust the constraints according to the prediction result to obtain the estimation result; the output unit 305 is used to output the estimation result.

[0187] In one embodiment, the preprocessing unit 302 is used to process null values, abnormal values, and repeated values ​​of the enterprise-related data to be predicted, and perform feature correlation analysis to obtain a preprocessing result.

[0188] In one embodiment, the device also includes a model training unit, which is used to obtain a model by performing multi-objective training on a sample set formed by preprocessing internal public data of the enterprise and numerical data of industries and provinces using an XGBoost algorithm combined with a custom loss function and an evaluation function.

[0189] In one embodiment, the model training unit is further used to:

[0190] Obtain the enterprise's internal public data and industry and province numerical data; pre-process the enterprise's internal public data and industry and province numerical data to obtain a sample set; use the XGBoost algorithm to build a multi-objective model for simultaneously predicting total assets and fixed assets; customize the loss function and evaluation function; use the sample set to train the multi-objective model, wherein the model performance is improved through parameter tuning and early stopping mechanism; specifically, train the multi-objective model with the sample set by setting the main hyperparameters, use grid search to adjust the hyperparameters, and use the early stopping mechanism for tuning. Use the loss function and the evaluation function to evaluate the trained multi-objective model to obtain a multi-objective asset prediction model.

[0191] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the above-mentioned enterprise asset estimation device 300 and each unit can refer to the corresponding description in the aforementioned method embodiment, and for the convenience and brevity of description, it will not be repeated here.

[0192] The enterprise asset estimation device 300 can be implemented in the form of a computer program. The computer program can be implemented in Figure 4 Runs on the computer device shown.

[0193] See also Figure 4 , Figure 4 5 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device 500 may be a server, wherein the server may be an independent server or a server cluster composed of multiple servers.

[0194] See also Figure 4 The computer device 500 includes a processor 502 , a memory and a network interface 505 connected via a system bus 501 , wherein the memory may include a non-volatile storage medium 503 and an internal memory 504 .

[0195] The non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions, and when the program instructions are executed, the processor 502 can execute an enterprise asset valuation method.

[0196] The processor 502 is used to provide computing and control capabilities to support the operation of the entire computer device 500 .

[0197] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute an enterprise asset valuation method.

[0198] The network interface 505 is used to communicate with other devices over the network. Figure 4 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present application, and does not constitute a limitation on the computer device 500 to which the solution of the present application is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0199] The processor 502 is used to run the computer program 5032 stored in the memory to implement the following steps:

[0200] Acquire relevant data of the enterprise to be predicted; preprocess the relevant data of the enterprise to be predicted to obtain preprocessing results; input the preprocessing results into a multi-objective asset prediction model to perform asset prediction to obtain prediction results; adjust constraints according to the prediction results to obtain estimation results; output the estimation results.

[0201] Among them, the multi-objective asset prediction model is a model obtained by performing multi-objective training on a sample set formed by preprocessing the company's internal public data and industry and province numerical data using the XGBoost algorithm combined with a custom loss function and an evaluation function.

[0202] In one embodiment, when the processor 502 implements the step of preprocessing the data related to the enterprise to be predicted to obtain the preprocessing result, the processor 502 specifically implements the following steps:

[0203] The data related to the enterprise to be predicted is processed for null values, abnormal values, and repeated values, and feature correlation analysis is performed to obtain preprocessing results.

[0204] In one embodiment, when the processor 502 implements the step of obtaining a model by performing multi-objective training on a sample set formed by preprocessing the enterprise's internal public data and industry and province numerical data using the XGBoost algorithm combined with a custom loss function and an evaluation function, the multi-objective asset prediction model specifically implements the following steps:

[0205] Obtain the enterprise's internal public data as well as industry and province numerical data; preprocess the enterprise's internal public data as well as industry and province numerical data to obtain a sample set; use the XGBoost algorithm to build a multi-objective model for simultaneously predicting total assets and fixed assets; customize the loss function and the evaluation function; use the sample set to train the multi-objective model, wherein the model performance is improved through parameter tuning and early stopping mechanism; use the loss function and the evaluation function to evaluate the trained multi-objective model to obtain a multi-objective asset prediction model.

[0206] Among them, the loss function includes the total asset error and relative error percentage, the fixed asset error and relative error percentage, the penalty factor of total assets, the penalty factor of fixed assets and the constraint penalty, and the multi-objective model parameters are adjusted by calculating the gradient and Hessian matrix of the loss function relative to the trained multi-objective model parameters.

[0207] The evaluation function includes the absolute error percentage of total assets, the absolute error percentage of fixed assets, the constraint violation rate and the average error percentage; wherein the constraint violation rate refers to the proportion of the fixed asset forecast value exceeding the total asset forecast value; the average error percentage refers to the average value of the absolute error percentage of total assets, the absolute error percentage of fixed assets and the constraint violation rate.

[0208] In one embodiment, the processor 502 implements the following steps when implementing the step of training the multi-objective model using the sample set, wherein the model performance is improved by parameter tuning and early stopping mechanism:

[0209] The multi-objective model is trained using the sample set by setting major hyperparameters, hyperparameters are adjusted using grid search, and early stopping mechanism is adopted for tuning.

[0210] It should be understood that in the embodiment of the present application, the processor 502 may be a central processing unit (CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0211] It can be understood by those skilled in the art that all or part of the processes in the method for implementing the above embodiment can be completed by instructing the relevant hardware through a computer program. The computer program includes program instructions, and the computer program can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiment of the above method.

[0212] Therefore, the present invention also provides a storage medium. The storage medium may be a computer-readable storage medium. The storage medium stores a computer program, wherein when the computer program is executed by a processor, the processor executes the following steps:

[0213] Acquire relevant data of the enterprise to be predicted; preprocess the relevant data of the enterprise to be predicted to obtain preprocessing results; input the preprocessing results into a multi-objective asset prediction model to perform asset prediction to obtain prediction results; adjust constraints according to the prediction results to obtain estimation results; output the estimation results.

[0214] Among them, the multi-objective asset prediction model is a model obtained by performing multi-objective training on a sample set formed by preprocessing the company's internal public data and industry and province numerical data using the XGBoost algorithm combined with a custom loss function and an evaluation function.

[0215] In one embodiment, when the processor executes the computer program to implement the step of preprocessing the data related to the enterprise to be predicted to obtain the preprocessing result, the processor specifically implements the following steps:

[0216] The data related to the enterprise to be predicted is processed for null values, abnormal values, and repeated values, and feature correlation analysis is performed to obtain preprocessing results.

[0217] In one embodiment, when the processor executes the computer program to implement the multi-objective asset prediction model, the model step is obtained by performing multi-objective training on a sample set formed by preprocessing the internal public data of the enterprise and the numerical data of the industry and province using the XGBoost algorithm combined with a custom loss function and an evaluation function, and specifically implements the following steps:

[0218] Obtain the enterprise's internal public data as well as industry and province numerical data; preprocess the enterprise's internal public data as well as industry and province numerical data to obtain a sample set; use the XGBoost algorithm to build a multi-objective model for simultaneously predicting total assets and fixed assets; customize the loss function and the evaluation function; use the sample set to train the multi-objective model, wherein the model performance is improved through parameter tuning and early stopping mechanism; use the loss function and the evaluation function to evaluate the trained multi-objective model to obtain a multi-objective asset prediction model.

[0219] Among them, the loss function includes the total asset error and relative error percentage, the fixed asset error and relative error percentage, the penalty factor of total assets, the penalty factor of fixed assets and the constraint penalty, and the multi-objective model parameters are adjusted by calculating the gradient and Hessian matrix of the loss function relative to the trained multi-objective model parameters.

[0220] The evaluation function includes the absolute error percentage of total assets, the absolute error percentage of fixed assets, the constraint violation rate and the average error percentage; wherein the constraint violation rate refers to the proportion of the fixed asset forecast value exceeding the total asset forecast value; the average error percentage refers to the average value of the absolute error percentage of total assets, the absolute error percentage of fixed assets and the constraint violation rate.

[0221] In one embodiment, the processor implements the step of training the multi-objective model using the sample set when executing the computer program, wherein the step of improving the model performance by parameter tuning and early stopping mechanism specifically implements the following steps:

[0222] The multi-objective model is trained using the sample set by setting major hyperparameters, hyperparameters are adjusted using grid search, and early stopping mechanism is adopted for tuning.

[0223] The storage medium may be a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk, etc., which are computer-readable storage media that can store program codes.

[0224] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0225] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of each unit is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0226] The steps in the method of the embodiment of the present invention can be adjusted in order, combined and deleted according to actual needs. The units in the device of the embodiment of the present invention can be combined, divided and deleted according to actual needs. In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0227] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, terminal, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention.

[0228] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for estimating enterprise assets, characterized in that: include: Obtain relevant data of the enterprise to be predicted; Preprocessing the data related to the enterprise to be predicted to obtain a preprocessing result; Inputting the preprocessing results into a multi-objective asset prediction model to perform asset prediction to obtain a prediction result; Perform constraint adjustments based on the prediction results to obtain an estimation result; The estimation result is outputted.

2. The enterprise asset estimation method according to claim 1, characterized in that: The preprocessing of the data related to the enterprise to be predicted to obtain a preprocessing result includes: The data related to the enterprise to be predicted is processed for null values, abnormal values, and repeated values, and feature correlation analysis is performed to obtain preprocessing results.

3. The enterprise asset estimation method according to claim 2, characterized in that: The multi-objective asset prediction model is a model obtained by performing multi-objective training on a sample set formed by preprocessing the company's internal public data and industry and province numerical data using the XGBoost algorithm combined with a custom loss function and an evaluation function.

4. The enterprise asset estimation method according to claim 3, characterized in that: The multi-objective asset prediction model is a model obtained by performing multi-objective training on a sample set formed by preprocessing the internal public data of the enterprise and the numerical data of the industry and province using the XGBoost algorithm combined with a custom loss function and an evaluation function, including: Obtaining public internal data of enterprises as well as numerical data of industries and provinces; Preprocessing the enterprise's internal public data and industry and province numerical data to obtain a sample set; Use the XGBoost algorithm to build a multi-objective model for simultaneously predicting total assets and fixed assets; Customize loss function and evaluation function; Using the sample set to train the multi-objective model, wherein the model performance is improved through parameter tuning and early stopping mechanism; The loss function and the evaluation function are used to evaluate the trained multi-objective model to obtain a multi-objective asset prediction model.

5. The enterprise asset estimation method according to claim 4, characterized in that: The loss function includes the total asset error and relative error percentage, the fixed asset error and relative error percentage, the penalty factor of the total assets, the penalty factor of the fixed assets and the constraint penalty, and the multi-objective model parameters are adjusted by calculating the gradient and Hessian matrix of the loss function relative to the trained multi-objective model parameters.

6. The enterprise asset estimation method according to claim 4, characterized in that: The evaluation function includes the absolute error percentage of total assets, the absolute error percentage of fixed assets, the constraint violation rate and the average error percentage; wherein the constraint violation rate refers to the proportion of the fixed asset forecast value exceeding the total asset forecast value; the average error percentage refers to the average value of the absolute error percentage of total assets, the absolute error percentage of fixed assets and the constraint violation rate.

7. The enterprise asset estimation method according to claim 4, characterized in that: The method of training the multi-objective model using the sample set, wherein the model performance is improved through parameter tuning and early stopping mechanism, comprises: The multi-objective model is trained using the sample set by setting major hyperparameters, hyperparameters are adjusted using grid search, and early stopping mechanism is adopted for tuning.

8. An enterprise asset estimation device, characterized in that: include: A data acquisition unit, used to acquire relevant data of the enterprise to be predicted; A preprocessing unit, used for preprocessing the data related to the enterprise to be predicted to obtain a preprocessing result; An asset prediction unit, used for inputting the preprocessing result into a multi-objective asset prediction model to perform asset prediction to obtain a prediction result; A constraint adjustment unit, used to perform constraint adjustment according to the prediction result to obtain an estimation result; An output unit is used to output the estimation result.

9. A computer device, characterized in that: The computer device comprises a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images