Basic learner dynamic combination method and device based on bionic algorithm, and medium

Abstract

The invention provides a base learner dynamic combination method and device based on a bionic algorithm and a medium, and belongs to the technical field of classification prediction. According to the technical scheme, the method comprises the following steps: dividing a data set, determining a to-be-selected learner set, and training learners by using the data set; initializing a population, wherein individuals in the population are lists composed of trained to-be-selected learners; calculating the fitness value of each individual in the current population; generating a new individual through evolution operation, and combining the new individual with the population to obtain a new population; and iterating the steps until a termination condition is met, stopping iteration, and selecting an optimal individual as a final to-be-selected learner combination according to a final fitness value. According to the method, the potential base learner combination is automatically searched through the bionic algorithm, the search efficiency is improved, and the burden of manual selection is relieved.

Description

Technical Field

[0001] The invention relates to a base learner dynamic combination method, device and medium based on a bionic algorithm, belonging to the technical field of classification prediction. Background Art

[0002] Multi-learner fusion learning algorithms, such as the Stacking fusion learning algorithm, are a meta-learning method that uses the prediction results of multiple base learners as input and then uses a meta-learner to complete the final prediction. Therefore, the selection and combination of base learners have an important impact on the performance of the algorithm. However, in traditional learner fusion learning algorithms, the selection of base learners has the following two defects: First, it relies too much on manual experience. This method of selection is often based on subjective judgment and is often limited by personal experience, making it difficult to ensure that the selected base learners are optimal. Second, the combination of learners is fixed. Traditional multi-learner fusion learning algorithms generally select a set of base learners during the training process and fix them. However, in practical problems, as data changes and the complexity of the problem increases, the fixed combination of base learners may not be able to adapt to new data and problem requirements, resulting in a decrease in algorithm performance.

[0003] The bionic algorithm is an optimization algorithm inspired by biological evolution theory, which searches for the optimal solution by simulating the biological evolution process. It changes and optimizes the solution through evolutionary operations (selection, crossover, mutation) to find a better solution. The bionic algorithm is combined with the multi-learner fusion learning algorithm to use the advantages of the bionic algorithm to search for the optimal combination of base learners. The bionic algorithm can flexibly adapt to different tasks according to the characteristics of the problem and the characteristics of the data. By optimizing the combination of base learners, a more effective model combination method can be found to improve the overall prediction performance. At the same time, the bionic algorithm can also handle base learners of different types and attributes, making full use of the advantages of each base learner to make the integrated prediction more robust and generalizable. At present, there is a lack of a method to improve the existing base learner selection and combination method using the bionic algorithm. Summary of the invention

[0004] The purpose of the present invention is to provide a method, device and medium for dynamically combining base learners based on a bionic algorithm, so as to improve the accuracy of model classification problems and the generalization ability of the model.

[0005] In order to achieve the above object, the present invention is implemented through the following technical solutions: Step 1: Divide the data set, determine the set of candidate learners, and use the data set to train the learners; Step 2: Initialize the population, where the individuals in the population are a list of candidate learners that have completed training; Step 3: Calculate the fitness value of each individual in the current population; Step 4: Generate new individuals through evolutionary operations, and merge the new individuals with the population to obtain a new population; Step 5: Iterate step 2-step 4 until the termination condition is met and stop the iteration. According to the final fitness value, select the best individual as the final candidate learner combination.

[0006] Preferably, the data set division includes: dividing the data set into a training set, a validation set and a test set according to a ratio; and dividing the data set in a rolling manner.

[0007] Preferably, the set of candidate learners is a Bagging parallel integration algorithm and a Boosting serial integration algorithm, including a random forest classifier, a gradient boosted tree classifier, XGBoost, CatBoost, a lightweight gradient boosting machine, a logistic regression, a decision tree, a support vector machine, a k-nearest neighbor, and a naive Bayes.

[0008] Preferably, the fitness function is as follows: , in, represents the weight parameter that weighs learning performance and diversity, represents the learning performance score, represents the learning performance metric of the learner, Represents the diversity score of the learner.

[0009] Preferably, the diversity score calculation formula of the learner is as follows: , , in, represents the number of samples correctly predicted by both learners, and represents the number of samples that a learner correctly predicts and a learner incorrectly predicts. represents the number of samples that are incorrectly predicted by both learners, Represents the number of learners.

[0010] Preferably, the specific calculation formula of the learning performance metric of the learner is as follows: , in, is the weight factor, Indicates the importance of time complexity, Indicates the base time, represents the training time of the learner, represents the test time of the learner.

[0011] Preferably, The value is 0.8. The value is 0.6. The value is 0.0237.

[0012] Preferably, the evolution operation includes a selection operation, a crossover operation and a mutation operation; The selection operation is as follows: Calculate the probability of each individual being selected according to the individual's fitness value , the specific formula is as follows: , in, Represents the normalized fitness value; according to Arrange the values ​​from high to low to construct a roulette wheel, generate a random number, and accumulate the probability of selecting individuals in the roulette wheel in a cycle; When the accumulated selection probability is greater than the random number, the loop stops and the current individual is selected; The crossover operation is as follows: Select two individuals as parents through selection operation, randomly generate crossover points, cut the learners of these two parent individuals at the crossover points, splice the first half of the first parent individual with the second half of the second parent individual to generate the first new individual, splice the first half of the second parent individual with the second half of the first parent individual to generate the second new individual, compare the original individual and the newly generated individual, select the individual with the higher fitness value as the offspring, and replace the original individual; The mutation operation is as follows: Select an individual from the population for mutation operation, randomly select a mutation point position in the learner of the individual and randomly select a new base learner to replace the base learner at the mutation point, generate a new individual, compare the original individual with the newly generated individual, select the individual with a high fitness value as the offspring, and replace the original individual.

[0013] The advantages of the present invention are: the present invention automatically searches for potential base learner combinations through a bionic algorithm, thereby improving the efficiency of the search and reducing the burden of manual selection. Through random operations and diversity mechanisms, a global search is performed in the entire search space, and the possibility of different learner combinations can be explored to find a better solution. Through the evolutionary iterative process, the combination of base learners can be dynamically added, deleted or adjusted to adapt to changes in the problem and changes in the performance of the learner; this flexibility can cope with the needs of adaptive combinations of learners in real-time or dynamic environments, and continuously optimize the performance of the combined learner. Therefore, a method for selecting base learners in a fusion learning algorithm using a bionic algorithm is proposed, which is an effective method to solve the defects of traditional base learner selection. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0015] Figure 1 This is a flow chart of the dynamic combination method of base learners based on the bionic algorithm; Figure 2 is the AUC value of the candidate learner in the training set; Figure 3 is the EWA score curve of each learner; Figure 4 It is the MCC coefficient heat map; Figure 5 For rolling experiments of existing methods; Figure 6 This is a rolling experiment of the method proposed in the present invention. DETAILED DESCRIPTION

[0016] 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 only 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.

[0017] Example 1 like Figure 1 As shown in the figure, a dynamic combination method of base learners based on bionic algorithm is implemented by the following technical solutions: S1: Divide the data set and determine the set of learners to be selected, use the data set to train the learners, and use the learner combination with good performance on the test set as the individuals of the population; the data set division includes: dividing the training set, the validation set and the test set according to the proportion; and dividing the data set in a rolling manner.

[0018] The candidate learner set is a Bagging parallel ensemble algorithm and a Boosting serial ensemble algorithm, including a random forest classifier, a gradient boosting tree classifier, XGBoost, CatBoost, a lightweight gradient boosting machine (LightGBM), a logistic regression, a decision tree, a support vector machine, a k-nearest neighbor, and a naive Bayes.

[0019] S2: Initialize the population, where the individuals in the population are a list of candidate learners that have completed training.

[0020] In this embodiment, a population is initialized and the population size population_size=10 is specified. , ,in Indicates the number of candidate learners, 0 means not selected, 1 means selected, specify . Then the initialized individual is , the population after initialization is .

[0021] S3: Calculate the fitness value of each individual in the current population.

[0022] The fitness function is as follows: , in, represents the weight parameter that weighs learning performance and diversity, represents the learning performance score, represents the learning performance metric of the learner, Represents the diversity score of the learner.

[0023] The diversity score of the learner has the following meaning: Assuming learners, for each learner , sum it with the MCC (Matthews correlation coefficient) values ​​of all other learners and divide it by Get the learner The average MCC value is calculated as follows: , , It should be noted that: for the two-classification problem, and For two different learners, based on a test data set , Representation classifier and The number of samples that are correctly predicted; for and The number of samples that are incorrectly predicted; for Correctly predicting The number of samples that were incorrectly predicted; for Correctly predicting The number of samples that were incorrectly predicted, Represents the number of learners.

[0024] The specific calculation formula of the learning performance metric of the learner is as follows: , in, is the weight factor, indicating the importance of the AUC value. Indicates the importance of time complexity. When , EWA degenerates into AUC; when When , EWA degenerates into the time complexity, Indicates the base time, represents the training time of the learner, represents the test time of the learner, The higher the value, the lower the time complexity and the better the model performance. The higher the value of EWA, the better the algorithm performs in terms of time and predictive ability.

[0025] In this embodiment, The value is 0.8. The value is 0.6. The value is 0.0237.

[0026] It should be noted that: S4: Generate new individuals through evolutionary operations, and merge the new individuals with the population to obtain a new population.

[0027] The evolutionary operation includes a selection operation, a crossover operation and a mutation operation; The selection operation is as follows: Calculate the probability of each individual being selected according to the individual's fitness value , the specific formula is as follows: , in, Represents the normalized fitness value; according to Arrange the values ​​from high to low to construct a roulette wheel, generate a random number, and accumulate the probability of selecting individuals in the roulette wheel in a cycle; When the accumulated selection probability is greater than the random number, the loop stops and the current individual is selected; The crossover operation is as follows: Select two individuals as parents through selection operation, randomly generate crossover points, cut the learners of these two parent individuals at the crossover points, splice the first half of the first parent individual with the second half of the second parent individual to generate the first new individual, splice the first half of the second parent individual with the second half of the first parent individual to generate the second new individual, compare the original individual and the newly generated individual, select the individual with the higher fitness value as the offspring, and replace the original individual; The mutation operation is as follows: Select an individual from the population for mutation operation, randomly select a mutation point position in the learner of the individual and randomly select a new base learner to replace the base learner at the mutation point, generate a new individual, compare the original individual with the newly generated individual, select the individual with a high fitness value as the offspring, and replace the original individual.

[0028] S5: Iterate step 2-step 4 until the termination condition is met and stop the iteration. According to the final fitness value, select the best individual as the final candidate learner combination. Specifically, select the individual with the highest fitness value as the best individual. Here, the maximum number of iterations is set to 100.

[0029] The embodiment of the present disclosure also provides a base learner dynamic combination device based on a bionic algorithm, including a processor and a memory. Optionally, the device may also include a communication interface and a bus. The processor, the communication interface, and the memory may communicate with each other through the bus. The communication interface may be used for information transmission. The processor may call the logic instructions in the memory to execute the base learner dynamic combination method based on the bionic algorithm of the above embodiment.

[0030] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product.

[0031] The memory, as a computer-readable storage medium, can be used to store software programs and computer executable programs, such as program instructions / modules corresponding to the method in the embodiment of the present disclosure. The processor executes the function application and data processing by running the program instructions / modules stored in the memory, that is, the base learner dynamic combination method based on the bionic algorithm in the above embodiment is implemented.

[0032] The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required for at least one function; the data storage area may store data created according to the use of the terminal device, etc. In addition, the memory may include a high-speed random access memory and may also include a non-volatile memory.

[0033] An embodiment of the present disclosure provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured to execute the above-mentioned base learner dynamic combination method based on the bionic algorithm.

[0034] The computer-readable storage medium mentioned above may be a transient computer-readable storage medium or a non-transitory computer-readable storage medium.

[0035] The technical solution of the embodiment of the present disclosure can be embodied in the form of a software product, which is stored in a storage medium and includes one or more instructions for enabling a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the embodiment of the present disclosure. The aforementioned storage medium may be a non-transient storage medium, including: a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a disk or an optical disk, and other media that can store program codes, or a transient storage medium.

[0036] Example 2 Figure 1 The process steps of the method for dynamically selecting base learners based on the bionic algorithm are demonstrated. The data set of this embodiment is Financial_Fraud.csv, which is real data, and there are 18699 positive and negative samples in the data set. Each data contains 72 columns of features such as "equity turnover rate B", "capital expenditure and depreciation and amortization ratio", and "non-current asset turnover rate B_3y". Based on this data set, the bionic algorithm is used to select the optimal learner combination for the fusion learning algorithm, and the output of the method is a list of learner combinations. In order to verify the effectiveness of the method, it is necessary to use the Stacking fusion learning algorithm and meta-learning k-neighbors to verify the performance and generalization ability of the Stacking model using the method proposed by the invention.

[0037] The specific steps can be divided into the following steps: 1. Data source and preprocessing Data Source The selection of non-fraud samples is mainly considered from two aspects: one is the representativeness of fraud samples and non-fraud samples, and the other is the industry representativeness. After excluding the sample data of the financial industry, companies that have been punished for corporate financial fraud are selected as fraud samples. Considering that "inflated profits", "fictitious assets" and "false records (misleading statements)" are the most likely purposes of corporate financial statement fraud, and considering that these violations have a greater impact on investors' investment activities and the operation of the capital market. Therefore, "inflated profits", "fictitious assets" and "false records (misleading statements)" are selected from the total table of violation handling as fraud samples, and the fraud sample label is set to 1. In terms of the non-fraud sample set, in order to avoid data bias of companies where potential fraud has not yet been discovered, the present invention eliminates data samples that have any violation records and sets the non-fraud sample label to 0. The number of fraud incidents reached a peak in 2018, with a total of 353 fraud incidents that year, but the number of fraud samples that year accounted for No more than 2%. Total percentage of fraud samples It is 16.0756%, so it can be seen that there is a data category imbalance problem in the financial fraud sample data set.

[0038] Data preprocessing The missing values ​​of the original data in the database are partly caused by the corresponding company's financial report not disclosing relevant data, or the company may not have any relevant business on this item. For this part of the missing data, the database is partially processed as 0 and partly as NA missing values, so the missing values ​​are filled with 0. If the missing ratio exceeds 80% of certain features, delete them directly. Otherwise, if it is a continuous feature, the missing value is filled with 0; if it is a categorical feature, the weight of evidence (WOE) encoding is adopted. The Isolation Forest (IF) model is used to learn on the training set, and samples with anomaly scores exceeding 0.7 percentile are removed from the training set; then the Local Outlier Factor (LOF) model is used to learn on the training set, and samples with anomaly scores exceeding 0.7 percentile are removed from the training set, and 11,288 samples are finally determined. After information value (IV) screening, binning and discretization, population stability index (PSI) screening, correlation screening, and stepwise regression operations, 72 columns of features were reduced to 45 columns, and finally 45 features were determined to be involved in financial fraud prediction.

[0039] 2. Model training (1) Experimental environment Anaconda3 is used as the programming tool. The detailed hardware and software configuration of the experiment is shown in Table 1.

[0040] Table 1 Software and hardware configuration table (2) Hyperparameter setting The number of iterations is set to 50, and the population size is set to 10. The candidate learner set is logistic regression (LR), decision tree (DT), support vector machine (SVM), KNN, naive Bayes (NB), RandomForestClassifier (RF), GradientBoostingClassifier (GB), XGBoost, catBoost and LightGBM (LGBM). The number of learners in the base learner combination is set to 6. Roulette is selected for the selection operation; single-point crossover is selected for the crossover operation; and the mutation probability is 0.1. =0.0237, w=0.9.

[0041] (3) Dataset division The training set and validation set select data from 2000 to 2018. The ratio of training set and validation set data is 7:3. The data from 2019 to 2020 are out-of-time sample data.

[0042] (4) Model training Based on the above unbalanced data set, each candidate learner model is trained using the processed data. After the training, the average AUC value of each learner in multiple experiments on the validation set is ( Figure 2 As shown in the figure) and EWA score, the EWA score curve is as follows Figure 3 According to the calculation formula of MCC, the MCC correlation coefficient between the base learners is calculated, as shown in Figure 4 shown.

[0043] 3. Model Evaluation Using recall ( ) and AUC value are used to evaluate the performance of this method. The calculation formula for recall rate is: , where TP represents the number of cases where the true value is positive and the prediction is also positive; FN represents the number of cases where the true value is positive but the prediction is negative. The AUC value (Area Under the Curve) is widely used as a prediction model evaluation indicator because it is not easily affected by sample imbalance. The AUC value compares the prediction effects of different prediction models by the area under the ROC curve. The ROC curve reflects the relationship between the recall rate and the false positive rate (FPR = FP / (FP + TN)) of the model at different thresholds. Generally speaking, the higher the AUC value, the better.

[0044] In order to verify the effectiveness of the method of the present invention, the Stacking algorithm is used as the fusion learning algorithm, and the KNN algorithm is selected as the meta-learner.

[0045] (1) Model performance evaluation Here, one of the base learners is selected by the present invention through the bionic algorithm, and the other is selected by the traditional manual method based on the size of the AUC value. Figure 2 The AUC values ​​of each learner are shown in Table 5. The parameters of the same learner model are set the same. The base learner combination obtained by the method of the present invention is: {NB, GB, DT, LGBM, KNN, CatBoost}; the base learner combination obtained by manual selection is: {catBoost, SVM, LR, LGBM, XGBoost, GB}. After multiple experiments, the average evaluation index values ​​of the out-of-time sample test results are shown in Table 1. Table 2 Model prediction evaluation indicators From the perspective of the model's evaluation indicators, the base learner combination method selected by the bionic algorithm proposed in the present invention has a higher recall value and AUC value for model prediction than the traditional manual selection method of the base learner. The experimental results show that the AUC values ​​of the two base learner selection methods are similar, about 3% higher, but the Recall value of the present invention is much higher than the traditional method, about 11%, which shows that the base learner selection method proposed in the present invention has a stronger ability to discriminate fraud samples and capture financial fraud samples. In other words, when the data set is an unbalanced data set, the method proposed in the present invention is better than the previous method in predicting classification problems, and the performance has been greatly improved.

[0046] (2) Model generalization ability evaluation In the real world, we can only know the information of the whole year of the previous year, and we cannot predict the fraud information and financial information in the future. Therefore, this part uses a rolling method to test the robustness and generalization ability of the improved model. Here, the rolling is carried out in units of five years. 2010-2014 is a group as the training set and test set, a group of base learners is selected, and 2015 is used as the out-of-time sample; 2011-2015 is a group as the training set, a group of base learners is selected, and 2016 is used as the out-of-time sample; then every year, a fixed time of 5 years is used as a window for rolling, and so on.

[0047] Here, one of the base learners is dynamically selected by the present invention through the bionic algorithm, and the other is selected based on the size of the AUC value by the traditional manual method, and the parameters of the same learner model are set the same. The base learner combination obtained by the method in the present invention is 6 groups: {DT, GB, LR, LGBM, KNN, CatBoost}, {NB, GB, DT, LGBM, RF, XGBoost}; according to the size of the ACU value in the figure, the first 6 base learner combinations obtained by manual selection are: {catBoost, SVM, LR, LGBM, XGBoost, GB}. The fusion learning algorithm uses the Stacking algorithm, and the meta-learner uses the KNN algorithm.

[0048] According to the real world, the window rolling method is used for prediction. The ROC curve of the traditional base learner rolling experiment is as follows: Figure 5 As shown in the figure, it can be seen that when the base learner is a fixed learner, the AUC value fluctuates dramatically, indicating that the data characteristics have changed over time, and the fixed base learner combination cannot adapt to the changes in the problem. Figure 6 From the results, we can see that although the data features have changed, the AUC value has not fluctuated much, which also shows that the dynamic combination base learner enables the classification prediction model to have better generalization ability.

[0049] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for dynamic combination of base learners based on bionic algorithm, characterized in that: The steps include: Step 1: Divide the data set, determine the set of candidate learners, and use the data set to train the learners; Step 2: Initialize the population, where the individuals in the population are a list of candidate learners that have completed training; Step 3: Calculate the fitness value of each individual in the current population; Step 4: Generate new individuals through evolutionary operations, and merge the new individuals with the population to obtain a new population; Step 5: Iterate step 2-step 4 until the termination condition is met and stop the iteration. According to the final fitness value, select the best individual as the final candidate learner combination.

2. The method for dynamically combining base learners based on a bionic algorithm according to claim 1, characterized in that: The data set division includes: dividing the training set, the validation set and the test set according to a ratio; and dividing the data set in a rolling manner.

3. The method for dynamically combining base learners based on a bionic algorithm according to claim 1, characterized in that: The candidate learner set is a Bagging parallel integration algorithm and a Boosting serial integration algorithm, including a random forest classifier, a gradient boosted tree classifier, XGBoost, CatBoost, a lightweight gradient boosting machine, a logistic regression, a decision tree, a support vector machine, a k-nearest neighbor, and a naive Bayes.

4. The method for dynamically combining base learners based on a bionic algorithm according to claim 1, characterized in that: The fitness function is as follows: ， in, represents the weight parameter that weighs learning performance and diversity, represents the learning performance score, represents the learning performance metric of the learner, Represents the diversity score of the learner.

5. The method for dynamically combining base learners based on a bionic algorithm according to claim 4, characterized in that: The diversity score calculation formula of the learner is as follows: ， ， in, represents the number of samples correctly predicted by both learners, and represents the number of samples that a learner correctly predicts and a learner incorrectly predicts. represents the number of samples that are incorrectly predicted by both learners, Represents the number of learners.

6. The method for dynamically combining base learners based on a bionic algorithm according to claim 5, characterized in that: The specific calculation formula of the learning performance metric of the learner is as follows: ， in, is the weight factor, Indicates the importance of time complexity, Indicates the base time, represents the training time of the learner, represents the test time of the learner.

7. The method for dynamically combining base learners based on a bionic algorithm according to claim 6, characterized in that: The value is 0.

8. The value is 0.

6. The value is 0.0237.

8. The method for dynamically combining base learners based on a bionic algorithm according to claim 3, characterized in that: The evolutionary operation includes a selection operation, a crossover operation and a mutation operation; The selection operation is as follows: Calculate the probability of each individual being selected according to the individual's fitness value , the specific formula is as follows: ， in, Represents the normalized fitness value; according to Arrange the values ​​from high to low to construct a roulette wheel, generate a random number, and accumulate the probability of selecting individuals in the roulette wheel in a cycle; When the accumulated selection probability is greater than the random number, the loop stops and the current individual is selected; The crossover operation is as follows: Select two individuals as parents through selection operation, randomly generate crossover points, cut the learners of these two parent individuals at the crossover points, splice the first half of the first parent individual with the second half of the second parent individual to generate the first new individual, splice the first half of the second parent individual with the second half of the first parent individual to generate the second new individual, compare the original individual and the newly generated individual, select the individual with the higher fitness value as the offspring, and replace the original individual; The mutation operation is as follows: Select an individual from the population for mutation operation, randomly select a mutation point position in the learner of the individual and randomly select a new base learner to replace the base learner at the mutation point, generate a new individual, compare the original individual with the newly generated individual, select the individual with a high fitness value as the offspring, and replace the original individual.

9. A base learner dynamic combination device based on a bionic algorithm, comprising a processor and a memory storing program instructions, characterized in that: The processor is configured to execute the base learner dynamic combination method based on the bionic algorithm as described in any one of claims 1-8 when running the program instructions.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the program is executed by a processor, the method described in any one of claims 1 to 8 is implemented.

Images