Artificial intelligence-based optimization system and method

Abstract

The invention discloses an artificial intelligence-based optimization system and method, and belongs to the field of finance and tax process management. The method comprises the following steps: collecting data and preprocessing to obtain preprocessed data; extracting relevant features of finance and taxation data from the preprocessed data, constructing a feature matrix for training and optimizing an initial LSTM time sequence prediction model to obtain a final LSTM time sequence prediction model, and substituting finance and taxation data features of enterprises needing to predict enterprise taxation trends into the final LSTM time sequence prediction model to obtain a future taxation feature set; establishing objective functions, and distributing a weight for each objective function; setting a lowest tax compliance score as a constraint condition; using multi-target particle swarm optimization to jointly optimize the targets; making a decision according to an optimization result; through the artificial intelligence technology, optimal management of the finance and taxation process is realized, and the tax compliance, the cost effectiveness and the fund mobility of an enterprise are improved, so that the overall operation efficiency and benefits of the enterprise are improved.

Description

Technical Field

[0001] The present invention belongs to the field of financial and tax process management, and in particular, relates to an optimization system and method based on artificial intelligence. Background Art

[0002] Existing tax process management systems usually rely on manual operations and traditional algorithms, lacking the ability to respond to the latest tax policies in real time, resulting in inefficient decision-making and prone to compliance risks. In addition, traditional optimization methods are difficult to balance multiple goals such as tax compliance, cost-effectiveness and liquidity. Summary of the invention

[0003] 1. Technical issues to be resolved In response to the problems in the related technologies, the present invention provides an artificial intelligence-based optimization system and method, which integrates deep learning, multi-objective optimization and other technologies to overcome the above-mentioned technical problems existing in the existing related technologies and achieve global optimization of the enterprise's financial and tax management processes.

[0004] (II) Technical solution In order to solve the above technical problems, the present invention is achieved through the following technical solutions: S1. Data preprocessing module: preprocess historical tax data, financial data, market data and fiscal and taxation policy data to obtain preprocessed data; S2, model training module, extracts relevant features of financial and tax data from the cleaned data, constructs feature matrix and label matrix for training and testing the initial LSTM time series prediction model, obtains the final LSTM time series prediction model after training and optimization, brings the financial and tax data features of the enterprise whose tax trend needs to be predicted into the final LSTM time series prediction model, obtains the future tax feature set, and derives the tax trend of the enterprise based on the feature set; S3, Multi-objective Optimization Module: Set up three objective functions, namely, maximizing tax compliance, maximizing capital liquidity, and maximizing cost-effectiveness; assign a weight to each objective function; set a minimum tax compliance score as a constraint; use multi-objective particle swarm optimization to jointly optimize tax compliance, cost-effectiveness, and capital liquidity; S4, Decision support module: Provide decision support based on optimization results to help enterprises formulate financial and tax strategies.

[0005] Preferably, the S1 comprises the following steps: S11. Obtain historical tax data, financial data, market data and fiscal and taxation policy data; S12. Identify and process missing data points in historical tax data, financial data, market data, and fiscal and taxation policy data, and fill them with mean, median, and mode; delete duplicate data records, unify data formats, and obtain preprocessed data; By following these steps, you can ensure that the data used is accurate, complete, and consistent.

[0006] Preferably, S2 comprises the following steps: S21. Select a training data set, a test data set, and their corresponding label sets; construct a training matrix, a test matrix, and their corresponding label sets; collect and process the past financial and tax data of the enterprise whose tax trends need to be predicted to obtain predicted financial and tax data; extract time series and financial features from the predicted financial and tax data to obtain a prediction feature set; S22, establish an initial LSTM time series prediction model, use the training matrix and the first label matrix to train the initial LSTM time series prediction model; after the training is completed, obtain the trained LSTM time series prediction model, then input the test matrix and the second label matrix into the trained LSTM time series prediction model for testing, optimize the trained LSTM time series prediction model according to the test results, and obtain the final LSTM time series prediction model; substitute the predicted feature set into the final LSTM time series prediction model to obtain the future feature set; Through the above steps, the LSTM model can be used to train financial and tax data, learn long-term dependencies in the data, predict future tax burdens and trends, and adjust financial and tax strategies in a timely manner, thereby improving the efficiency and effectiveness of financial and tax management.

[0007] Preferably, S21 comprises the following steps: S211, determine that the year interval of the training data set and the test data set is α; the corresponding year interval of the label data set is β, where α and β are continuous and uninterrupted, and the year interval of α is before β; S212. Ensure that the data in the selected year range can represent the normal operation of the company and avoid including data from abnormal years, which include but are not limited to those caused by one-off events (such as mergers and acquisitions, legal proceedings, and natural disasters); S213. Ensure that the time series of the data is continuous without interruptions or missing data, because the LSTM model relies on the continuity of time series data to learn long-term dependencies. S214. Ensure that the amount of data in each time period is equal. If the amount of data in some time periods in the data set is larger than that in other time periods, the model will be biased towards the characteristics of these time periods. In this case, consider resampling the data to make the amount of data in each time period equal. S215. Selecting a training data set from preprocessed dataA ={ A 1, A 2,..., A i,..., A N} , select the first label dataset A′ ={ A′ 1, A′ 2,..., A′ i,..., A′ N} ;in, A i Indicates the training data set i The financial and tax data of a company in the year interval α, A′ i Indicates i The financial and tax situation of a company in the year interval β, N Represents the total number of companies; Selected test data set B ={ B 1, B 2,..., B i,..., B N} , select the second label dataset B′ ={ B′ 1, B′ 2,..., B′ i,..., B′ N} ;in, B i Indicates the training data set i The financial and tax data of a company in the year interval α, represents the fiscal and tax situation of the i-th company in the year interval β; S216. From the training data set A ={ A 1, A 2,..., A i,..., A N} , test data set B ={ B1, B 2,..., B i,..., B N} And the first label dataset A′ ={ A′ 1, A′ 2,..., A′ i,..., A′ N} , the second label dataset B′ ={ B′ 1, B′ 2,..., B′ i,..., B′ N} Extract time series features, such as year, month, and day, to capture the impact of time on fiscal and tax data; extract financial features, such as income, expenditure, and tax rate, to capture the impact of financial features on fiscal and tax data; construct a training feature matrix , the first label matrix ;in A′′ in Indicates i The company is in the first n Features, A′′′ in Indicates i The company is in the year interval β n Features; build test feature matrix , the second label matrix ;in B′′ in Indicates i The company is in the first n Features, B′′′ in Indicates i The company is in the year interval β n Features S217. Collect and process the past financial and tax data of the enterprise whose tax trend needs to be predicted to obtain the predicted financial and tax data; extract the time series and financial features from the predicted financial and tax data to obtain the predicted feature set. C ={ C 1, C 2,..., C i,..., CN} ;in C i The company’s i Features The above steps can ensure the quality of the data and avoid the influence of individual data on the training results of the model, so as to make more accurate predictions on future fiscal and tax data. The LSTM time series prediction model can be trained more effectively, making the final trained model more universal and accurate in predicting fiscal and tax trends.

[0008] Preferably, the S22 comprises the following steps: S221, construct an initial LSTM time series prediction model, and set the number of iterations of the initial LSTM time series prediction model training to a′ , the batch size is b′ , the optimizer is adam optimizer and the initial learning rate is c′ ; S222, set the current number of training iterations to d And the training error threshold is d′ The training error of the initial LSTM time series prediction model is calculated using the missing function, and the function expression is as follows: ; in, n is the sample size; y i It is i The true label of each sample, which takes the value 0 or 1; It is i The predicted probability of a sample indicates the probability that the model predicts that the sample belongs to category 1; according to the batch size b′ , adam optimizer and initial learning rate c′ The training matrix B and the first label matrix B′ Input into the initial LSTM time series prediction model for training. When the training error of the initial LSTM time series prediction model is < d′ when or d ≥ a' When , the training is stopped to obtain the trained LSTM time series prediction model; S223. Setting accuracy threshold d′′ , the test matrix B′′ and the second label matrix B′′′ Input into the trained LSTM time series prediction model for testing to obtain the test accuracy d′′′ ;when d′′′ ≥ d′′When , the trained LSTM time series prediction model is used as the final LSTM time series prediction model; when d′′′ < d′′ , return to S222 to continue training the trained LSTM time series prediction model until d′′′ ≥ d′ Until the time, the final LSTM time series prediction model is obtained; S224, predicting feature set C Substitute into the final LSTM time series prediction model to obtain the future feature set C′ ={ C′ 1, C′ 2,..., C′ i,..., C′ n} ,in , C′ i The company’s future i Features By setting the relevant parameters of the LSTM time series prediction model, and then inputting the training matrix into the LSTM time series prediction model for training, a trained LSTM time series prediction model is obtained. The test matrix is ​​used to test the trained LSTM time series prediction model. According to the test results, the trained LSTM time series prediction model is optimized to make the LSTM time series prediction model more accurate.

[0009] Preferably, S3 comprises the following steps: S31. Collect data related to the decision variables from the preprocessed data to obtain a decision variable data set, including but not limited to cost data, tax rate data, cash flow, and total assets; S32, establishing an objective function and constructing constraint conditions based on the decision variable data set; S33. Use multi-objective particle swarm optimization algorithm to optimize the maximum tax compliance, maximum capital liquidity and maximum cost-effectiveness; obtain the Pareto optimal solution set through the multi-objective particle swarm optimization algorithm; Through these steps, the comprehensive effects of tax compliance, capital liquidity and cost-effectiveness are taken into consideration, which helps enterprises better balance the relationship between tax compliance, capital liquidity and cost-effectiveness.

[0010] Preferably, the S32 comprises the following steps: S321, use a score w To represent the tax compliance of the enterprise, the objective function is to maximize the score w, that is, to maximize tax compliance, this score w It is a comprehensive assessment based on historical compliance records, internal control system assessments, and compliance audit results; weights are set for historical compliance records, internal control system assessments, and compliance audit results w 1、 w 2、 w 3 ; For historical compliance records, internal control systems, and compliance audits, assign a score based on their performance y 1、 y 2、 y 3 , which results in a tax compliance score w = w 1× y 1＋ w 2× y 2＋ w 3× y 3 , where the weight w 1、 w 2、 w 3 is a real number between 0 and 1, the sum of all weights equals 1, and the fraction y 1、 y 2、 y 3 It is a value between 0 and 1, where 1 indicates full compliance and 0 indicates full non-compliance; S322. Liquidity is expressed as the ratio of cash flow to total assets. x = x 1÷ x 2 ; The objective function is to maximize x , that is, maximize the liquidity of funds; x 1 represents cash flow, x 2 represents total assets; S323. Use profit margin to express cost-effectiveness, that is, the ratio of profit divided by cost. z =( z 1 - z 2 - z 3 )÷ z2 ; The objective function is to maximize z , that is, maximize the liquidity of funds; z 1 Indicates income, z 2 Indicates cost, z 3 Indicates the amount of tax; S324. Assign a weight to each objective function based on the importance and priority of the objective. Tax compliance w The weight of w′ , liquidity x The weight of x′ , cost-effective z The weight of z′ , w′ + x' + z' = 1; For each objective function, calculate its score on the current solution w′′ , x′′ , z′′ ; Use the weighted summation method to multiply the score of each target by its weight and add them together to get the comprehensive fitness value s= w′ × w′′ + x' × x'' + z' × z'' ; S325. Set a minimum tax compliance score w′′′ As constraints, ensure that all solutions satisfy the constraints; Through these steps, we ensure that enterprises can maximize their liquidity and cost-effectiveness while complying with regulations, thereby supporting their long-term sustainable development and competitiveness.

[0011] Preferably, the S33 comprises the following steps: S331. Create a particle swarm based on the range of decision variables, where each particle represents a solution. The position of each particle is a vector containing the values ​​of all decision variables. Initialize the parameters of the particle swarm optimization algorithm based on the decision data, randomly initialize the speed and position of each particle, and define individual extreme values. P best and global extremum G best ; S332, calculate the fitness function value of each particle, and compare the fitness function value of each particle with the individual extreme value P best In contrast, if the fitness function value of the particle is higher than the individual extreme value P best , then assign the fitness function value of the particle to the individual extreme value P best ; S333. Compare the individual extreme values ​​of each particle P best , select the individual extreme value with the highest fitness function value P best and global extreme value G best For comparison, if the individual extreme value P best Above the global extreme G best , then update the global extreme value G best The value of S334, according to individual extreme values P best and global extremum G best Update inertia weight and learning factor; S335, updating the inertia weight and learning factor; S336, defining the maximum number of Pareto optimal solutions as δ, and increasing the Pareto optimal solution count each time a Pareto optimal solution is obtained; defining the maximum number of iterations ζ, and increasing the iteration count after each iteration; S337, repeat steps S331-S336, when the number of Pareto optimal solutions reaches δ or the number of runs reaches the maximum number of iterations ζ, end the algorithm and finally output the Pareto optimal solution set; Through the above steps, the multi-objective particle swarm optimization algorithm finds the balance point among tax compliance, capital liquidity, and cost-effectiveness, and provides a global optimal solution set, thereby supporting enterprises to make more comprehensive and wise decisions between tax compliance, capital liquidity, and cost-effectiveness.

[0012] Preferably, S4 comprises the following steps: S41. Through the future feature set C′ ={ C′ 1, C′ 2,..., C′ i,..., C′ n} Determine the predicted corporate tax trends; policymakers make adjustments based on the predicted corporate tax trends; S42. Decision makers make the most suitable decision for the current development of the enterprise based on the focus of the enterprise's current development and the Pareto optimal solution set obtained by the multi-objective particle swarm optimization algorithm; The above steps will help companies better understand and manage their tax risks while optimizing their financial position and cash flow, thereby supporting their long-term development and strategic goals.

[0013] An artificial intelligence-based optimization system, characterized by comprising a data preprocessing module, a model training module, a multi-objective optimization module and a decision support module; The data preprocessing module is used to process the collected historical tax data, financial data, market data and fiscal and taxation policy data to maintain the accuracy and consistency of the data; The model training module is used to extract the features of financial and tax data, construct a feature matrix, train, test and optimize the model to obtain the final model, substitute the financial and tax data features of the enterprise whose tax trend needs to be predicted into the final model to obtain the tax trend feature set of the enterprise; The multi-objective optimization module is used to jointly optimize tax compliance, cost-effectiveness and capital liquidity to obtain a Pareto optimal solution set; The decision support module is used to adjust the tax trend of the enterprise according to the tax trend feature set and obtain the Pareto optimal solution set.

[0014] (III) Beneficial effects The present invention has the following beneficial effects: Through the data preprocessing module, the system of the present invention can ensure the accuracy and consistency of the collected historical tax data, financial data, market data and fiscal and tax policy data, thereby improving data quality; through the model training module, the system can extract the characteristics of fiscal and tax data, construct a feature matrix, train and optimize the model, so as to predict the tax trend of the enterprise; through the multi-objective optimization module, the system can jointly optimize tax compliance, cost-effectiveness and capital liquidity, and obtain the Pareto optimal solution set, thereby helping the enterprise to formulate the optimal tax strategy; through the decision support module, the system can adjust the fiscal and tax strategy of the enterprise according to the tax trend feature set and the Pareto optimal solution set, thereby providing decision support for the enterprise.

[0015] At the same time, the system can automate and optimize financial and tax processes through artificial intelligence technology to improve the company's operational efficiency and benefits; through the multi-objective optimization module, the system can jointly optimize tax compliance and help companies reduce tax risks; through the multi-objective optimization module, the system can jointly optimize capital liquidity and improve the company's capital turnover efficiency.

[0016] In summary, through artificial intelligence technology, the system can achieve optimized management of financial and tax processes, improve the company's tax compliance, cost-effectiveness and capital liquidity, thereby improving the company's overall operational efficiency and benefits.

[0017] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the invention, the drawings required for describing the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For ordinary technicians in this field, they can also obtain drawings based on these drawings without paying creative work.

[0019] Figure 1 The figure is a schematic diagram of the electronic manufacturer name identification process based on the bill of materials of the present invention. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the invention to clearly and completely describe the technical solutions in the embodiments of the invention. Obviously, the described embodiments are only part of the embodiments of the invention, not all of the embodiments. Based on the embodiments in the invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the invention.

[0021] In the description of the present invention, it is necessary to understand that the terms "opening", "upper", "lower", "top", "middle", "inside" and the like indicating orientation or positional relationship are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the components or elements referred to must have a specific orientation, be constructed and operate in a specific orientation, and therefore cannot be understood as limiting the invention.

[0022] Embodiment 1: See also Figure 1 , the present invention is an optimization method based on artificial intelligence, comprising the following steps: S1. Preprocess historical tax data, financial data, market data and fiscal and taxation policy data to obtain preprocessed data; The S1 comprises the following steps: S11. Obtain historical tax data, financial data, market data and fiscal and taxation policy data; S12. Identify and process missing data points in historical tax data, financial data, market data, and fiscal and taxation policy data, and fill them with mean, median, and mode; delete duplicate data records, unify data formats, and obtain preprocessed data; S2. Extract relevant features of financial and tax data from preprocessed data, construct feature matrix and label matrix for training and testing initial LSTM time series prediction model, and obtain final LSTM time series prediction model after training and optimization; bring financial and tax data features of enterprises whose tax trends need to be predicted into final LSTM time series prediction model to obtain future tax feature set; The S2 comprises the following steps: S21. Select a training data set, a test data set, and their corresponding label sets; construct a training matrix, a test matrix, and their corresponding label sets; collect and process the past financial and tax data of the enterprise whose tax trends need to be predicted to obtain predicted financial and tax data; extract time series and financial features from the predicted financial and tax data to obtain a prediction feature set; The S21 comprises the following steps: S211, determine that the year interval of the training data set and the test data set is α; the corresponding year interval of the label data set is β, where α and β are continuous and uninterrupted, and the year interval of α is before β; S212. Ensure that the data within the selected year range can represent the normal operation of the company; S213. Ensure that the time series of data is continuous without interruption or missing; S214, ensuring that the amount of data in each time period is equal; S215. Selecting a training data set from preprocessed data A ={ A 1, A 2,..., A i,..., A N} , select the first label dataset A′ ={ A′ 1, A′ 2,..., A′ i,..., A′ N} ;in, A i Indicates the training data set i The financial and tax data of a company in the year interval α, A′ i Indicates i The financial and tax situation of a company in the year interval β, N Represents the total number of companies; Selected test data set B ={ B 1, B 2,..., B i,..., B N} , select the second label dataset B′ ={ B′ 1, B′2,..., B′ i,..., B′ N} ;in, B i Indicates the training data set i The financial and tax data of a company in the year interval α, represents the fiscal and tax situation of the i-th company in the year interval β; S216. From the training data set A ={ A 1, A 2,..., A i,..., A N} , test data set B ={ B 1, B 2,..., B i,..., B N} And its corresponding first label dataset A′ ={ A′ 1, A′ 2,..., A′ i,..., A′ N} , the second label dataset B′ ={ B′ 1, B′ 2,..., B′ i,..., B′ N} Extract time series features, such as year, month, and day, to capture the impact of time on fiscal and tax data; extract financial features, such as income, expenditure, and tax rate, to capture the impact of financial features on fiscal and tax data; construct a training feature matrix , the first label matrix ;in A′′ in Indicates i The company is in the first n Features, A′′′ in Indicates i The company is in the first n Features; build test feature matrix , the second label matrix ;in B′′ in Indicates i The company is in the first n feature, B′′′in represents the i The company is in the year interval β n Features S217. Collect and process the past financial and tax data of the enterprise whose tax trend needs to be predicted to obtain the predicted financial and tax data; extract the time series and financial features from the predicted financial and tax data to obtain the predicted feature set. C ={ C 1, C 2,..., C i,..., C N} ;in C i The company’s i Features S22, establish an initial LSTM time series prediction model, use the training matrix and the first label matrix to train the initial LSTM time series prediction model; after the training is completed, obtain the trained LSTM time series prediction model, then input the test matrix and the second label matrix into the trained LSTM time series prediction model for testing, optimize the trained LSTM time series prediction model according to the test results, and obtain the final LSTM time series prediction model; substitute the predicted feature set into the final LSTM time series prediction model to obtain the future feature set; The S22 comprises the following steps: S221, construct an initial LSTM time series prediction model, and set the number of iterations of the initial LSTM time series prediction model training to a′ , the batch size is b′ , the optimizer is adam optimizer and the initial learning rate is c′ ; S222, set the current number of training iterations to c′ And the training error threshold is d′ The training error of the initial LSTM time series prediction model is calculated using the missing function, and the function expression is as follows: ; in, n is the sample size; y i It is i The true label of each sample, which takes the value 0 or 1; It is i The predicted probability of a sample indicates the probability that the model predicts that the sample belongs to category 1; according to the batch size b′ , adam optimizer and initial learning rate c′ The training matrix B and the first label matrix B′ Input into the initial LSTM time series prediction model for training. When the training error of the initial LSTM time series prediction model is < d' when or d ≥ a' When , the training is stopped to obtain the trained LSTM time series prediction model; S223. Setting accuracy threshold d′′ , the test matrix B′′ and the second label matrix B′′′ Input into the trained LSTM time series prediction model for testing to obtain the test accuracy d′′′ ;when d′′′ ≥ d′′ When , the trained LSTM time series prediction model is used as the final LSTM time series prediction model; when d′′′ < d′′ , return to S222 to continue training the trained LSTM time series prediction model until d′′′ ≥ d′ Until the time, the final LSTM time series prediction model is obtained; S224, predicting feature set C Substitute into the final LSTM time series prediction model to obtain the future feature set C′ ={ C′ 1, C′ 2,..., C′ i,..., C′ n} ,in , C′ i The company’s future i Features S3. Establish three objective functions, namely, maximizing tax compliance, maximizing capital liquidity, and maximizing cost-effectiveness; assign a weight to each objective function; set a minimum tax compliance score as a constraint; use multi-objective particle swarm optimization to jointly optimize tax compliance, cost-effectiveness, and capital liquidity; The S3 comprises the following steps: S31, collecting data related to the decision variables from the preprocessed data to obtain a decision variable data set; S32, establishing an objective function and constructing constraint conditions based on the decision variable data set; The S32 comprises the following steps: S321, use a score w To represent the tax compliance of the enterprise, the objective function is to maximize this score w , which maximizes tax compliance; this score w It is a comprehensive assessment based on historical compliance records, internal control system assessments, and compliance audit results; weights are set for historical compliance records, internal control system assessments, and compliance audit results w 1、 w 2、 w 3 ; For historical compliance records, internal control systems, and compliance audits, assign a score based on their performance y 1、 y 2、 y 3 , which results in a tax compliance score w = w 1× y 1＋ w 2× y 2＋ w 3× y 3 , where the weight w 1、 w 2、 w 3 is a real number between 0 and 1, the sum of all weights equals 1, and the fraction y 1、 y 2、 y 3 It is a value between 0 and 1, where 1 indicates full compliance and 0 indicates full non-compliance; S322. Liquidity is expressed as the ratio of cash flow to total assets. x = x 1÷ x 2 ; The objective function is to maximize x , that is, to maximize liquidity, where x 1 represents cash flow, x 2 represents total assets; S323. Use profit margin to express cost-effectiveness, that is, the ratio of profit divided by cost. z =( z 1 - z 2 - z 3 )÷ z 2 ; The objective function is to maximize z , that is, maximize the liquidity of funds; z 1 Indicates income, z 2 Indicates cost, z 3 Indicates the amount of tax; S324. Assign a weight to each objective function based on the importance and priority of the objective. Tax compliance w The weight of w′ , liquidity x The weight of x′ , cost-effective z The weight of z′ , w′ + x' + z' = 1; For each objective function, calculate its score on the current solution w′′ , x′′ , z′′ ; Use the weighted summation method to multiply the score of each target by its weight and add them together to get the comprehensive fitness value s= w′ × w′′ + x' × x'' + z' × z'' ; S325. Set a minimum tax compliance score w′′′ As constraints, all solutions need to satisfy the constraints; S33. Use multi-objective particle swarm optimization algorithm to optimize the maximum tax compliance, maximum capital liquidity and maximum cost-effectiveness; obtain the Pareto optimal solution set through the multi-objective particle swarm optimization algorithm; The S33 comprises the following steps: S331. Create a particle swarm based on the range of decision variables, where each particle represents a solution. The position of each particle is a vector containing the values ​​of all decision variables. Initialize the parameters of the particle swarm optimization algorithm based on the decision data, randomly initialize the speed and position of each particle, and define individual extreme values. P best and global extremum G best ; S332, calculate the fitness function value of each particle, and compare the fitness function value of each particle with the individual extreme value P best In contrast, if the fitness function value of the particle is higher than the individual extreme value P best , then assign the fitness function value of the particle to the individual extreme value P best ; S333. Compare the individual extreme values ​​of each particle P best , select the individual extreme value with the highest fitness function value P best and global extreme value G best For comparison, if the individual extreme value P best Above the global extreme G best , then update the global extreme value G best The value of S334, according to individual extreme values P best and global extremum G best Update inertia weight and learning factor; S335, updating the inertia weight and learning factor; S336, defining the maximum number of Pareto optimal solutions as δ, and increasing the Pareto optimal solution count each time a Pareto optimal solution is obtained; defining the maximum number of iterations ζ, and increasing the iteration count after each iteration; S337, repeat steps S331-S336, when the number of Pareto optimal solutions reaches δ or the number of runs reaches the maximum number of iterations ζ, end the algorithm and finally output the Pareto optimal solution set; S4. Provide decision support based on optimization results to help enterprises formulate financial and taxation strategies.

[0023] The S4 comprises the following steps: S41. Through the future feature set C′ ={ C′ 1, C′ 2,..., C′ i,..., C′ n} Determine the predicted corporate tax trends; policymakers make adjustments based on the predicted corporate tax trends; S42. Decision makers make the most suitable decision for the current development of the enterprise based on the focus of the current development of the enterprise and the Pareto optimal solution set obtained by the multi-objective particle swarm optimization algorithm.

[0024] Embodiment 2: An artificial intelligence-based optimization system, characterized by comprising a data preprocessing module, a model training module, a multi-objective optimization module and a decision support module; The data preprocessing module is used to process the collected historical tax data, financial data, market data and fiscal and taxation policy data to maintain the accuracy and consistency of the data; The model training module is used to extract the features of financial and tax data, construct a feature matrix, train, test and optimize the model to obtain the final model, substitute the financial and tax data features of the enterprise whose tax trend needs to be predicted into the final model to obtain the tax trend feature set of the enterprise; The multi-objective optimization module is used to jointly optimize tax compliance, cost-effectiveness and capital liquidity to obtain a Pareto optimal solution set; The decision support module is used to adjust the tax trend of the enterprise according to the tax trend feature set and obtain the Pareto optimal solution set.

[0025] In the description of this specification, the description with reference to the terms "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0026] The preferred embodiments of the invention disclosed above are only used to help explain the invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the invention, so that those skilled in the art can understand and use the invention well.

Claims

1. An optimization method based on artificial intelligence, characterized in that: The following steps are involved: S1. Preprocess historical tax data, financial data, market data and fiscal and taxation policy data to obtain preprocessed data; S2. Extract relevant features of financial and tax data from preprocessed data, construct feature matrix and label matrix for training and testing initial LSTM time series prediction model, obtain final LSTM time series prediction model after training and optimization, bring financial and tax data features of enterprises whose tax trends need to be predicted into final LSTM time series prediction model, and obtain future tax feature set; S3. Establish three objective functions, namely, maximizing tax compliance, maximizing capital liquidity, and maximizing cost-effectiveness; assign a weight to each objective function; and set a minimum tax compliance score as a constraint; Use multi-objective particle swarm optimization to jointly optimize tax compliance, cost-effectiveness, and capital liquidity; S4. Make decisions based on the optimization results.

2. The artificial intelligence optimization method according to claim 1, characterized in that: The S1 comprises the following steps: S11. Obtain historical tax data, financial data, market data and fiscal and taxation policy data; S12. Identify and process missing data points in historical tax data, financial data, market data, and fiscal and taxation policy data, and fill them with mean, median, and mode; delete duplicate data records and unify data formats; and obtain preprocessed data.

3. The artificial intelligence optimization method according to claim 1, characterized in that: The S2 comprises the following steps: S21. Select a training data set, a test data set, and their corresponding label sets; construct a training matrix, a test matrix, and their corresponding label sets; collect and process the past financial and tax data of the enterprise whose tax trends need to be predicted to obtain predicted financial and tax data; extract time series and financial features from the predicted financial and tax data to obtain a prediction feature set; S22. Establish an initial LSTM time series prediction model, and use the training matrix and the first label matrix to train the initial LSTM time series prediction model; after the training is completed, a trained LSTM time series prediction model is obtained, and then the test matrix and the second label matrix are input into the trained LSTM time series prediction model for testing, and the trained LSTM time series prediction model is optimized according to the test results to obtain a final LSTM time series prediction model; substitute the predicted feature set into the final LSTM time series prediction model to obtain a future feature set.

4. The artificial intelligence optimization method according to claim 3, characterized in that: The S21 comprises the following steps: S211, determine that the year interval of the training data set and the test data set is α; the corresponding year interval of the label data set is β, where α and β are continuous and uninterrupted, and the year interval of α is before β; S212. Ensure that the data within the selected year range can represent the normal operation of the company; S213. Ensure that the time series of data is continuous without interruption or missing; S214, ensuring that the amount of data in each time period is equal; S215, selecting a training data set and a first label data set from the preprocessed data; selecting a test data set and a second label data set; S216, extracting time series features and financial features from the training data set, the test data set, the first label data set, and the second label data set; constructing a training feature matrix, a first label matrix, a test feature matrix, and a second label matrix; S217. Collect and process the past financial and tax data of the enterprise whose tax trend needs to be predicted to obtain the predicted financial and tax data; extract the time series and financial features from the predicted financial and tax data to obtain the predicted feature set. C .

5. The artificial intelligence-based optimization method according to claim 3, characterized in that: The S22 comprises the following steps: S221, construct an initial LSTM time series prediction model, and set the number of iterations of the initial LSTM time series prediction model training to a′ The batch size is b′ , the optimizer is adam optimizer and the initial learning rate is c′ ; S222, set the current number of training iterations to d And the training error threshold is d′ The training error of the initial LSTM time series prediction model is calculated using the missing function, and the function expression is as follows: ； in, n is the sample size; y i It is i The true label of each sample, which takes the value 0 or 1; It is i The predicted probability of a sample indicates the probability that the model predicts that the sample belongs to category 1; according to the batch size b′ , adam optimizer and initial learning rate c′ The training matrix and the first label matrix are input into the initial LSTM time series prediction model for training. When the training error of the initial LSTM time series prediction model is less than d′ when or d≥a′ When , the training is stopped to obtain the trained LSTM time series prediction model; S223. Setting accuracy threshold d′′ , the test matrix and the second label matrix are input into the trained LSTM time series prediction model for testing to obtain the test accuracy d′′′ ;when d′′′ ≥ d′′ When , the trained LSTM time series prediction model is used as the final LSTM time series prediction model; when d′′′ < d′′ , return to S222 to continue training the trained LSTM time series prediction model until d′′′ ≥ d′′ Until the time, the final LSTM time series prediction model is obtained; S224, predicting feature set C Substitute into the final LSTM time series prediction model to obtain the future feature set C′ .

6. The artificial intelligence optimization method according to claim 1, characterized in that: The S3 comprises the following steps: S31, collecting data related to the decision variables from the preprocessed data to obtain a decision variable data set; S32, establishing an objective function and constructing constraint conditions based on the decision variable data set; S33. Use the multi-objective particle swarm optimization algorithm to perform multi-objective optimization on maximizing tax compliance, maximizing capital liquidity and maximizing cost-effectiveness; obtain the Pareto optimal solution set through the multi-objective particle swarm optimization algorithm.

7. The artificial intelligence-based optimization method according to claim 6, characterized in that: The S32 comprises the following steps: S321, use a score w To represent the tax compliance of the enterprise, the objective function is to maximize this score w , i.e. maximizing tax compliance; S322. Liquidity is expressed as the ratio of cash flow to total assets. x=x 1÷ x 2; The objective function is to maximize x , that is, maximizing the liquidity of funds; S323. Use profit margin to express cost-effectiveness, that is, the ratio of profit divided by cost. z =( z 1- z 2- z 3)÷ z 2; The objective function is to maximize z , that is, maximizing the liquidity of funds; S324, assigning a weight to each objective function according to the importance and priority of the objective; S325. Set a minimum tax compliance score w′′′ As constraints, all solutions must satisfy the constraints.

8. The artificial intelligence-based optimization method according to claim 6, characterized in that: The S33 comprises the following steps: S331. Create a particle swarm based on the range of decision variables, where each particle represents a solution. The position of each particle is a vector containing the values ​​of all decision variables. Initialize the parameters of the particle swarm optimization algorithm based on the decision data, randomly initialize the speed and position of each particle, and define individual extreme values. P best and global extremum G best ; S332, calculate the fitness function value of each particle, and compare the fitness function value of each particle with the individual extreme value P best In contrast, if the fitness function value of the particle is higher than the individual extreme value P best , then assign the fitness function value of the particle to the individual extreme value P best ; S333. Compare the individual extreme values ​​of each particle P best , select the individual extreme value with the highest fitness function value P best and global extreme value G best For comparison, if the individual extreme value P best Above the global extreme G best , then update the global extreme value G best The value of S334, according to individual extreme values P best and global extremum G best Update inertia weight and learning factor; S335, updating the inertia weight and learning factor; S336, defining the maximum number of Pareto optimal solutions as δ, and increasing the Pareto optimal solution count each time a Pareto optimal solution is obtained; defining the maximum number of iterations ζ, and increasing the iteration count after each iteration; S337. Repeat steps S331-S336. When the number of Pareto optimal solutions reaches δ or the number of runs reaches the maximum number of iterations ζ, the algorithm ends and finally outputs the Pareto optimal solution set.

9. The artificial intelligence-based optimization method according to claim 1, characterized in that: The S4 comprises the following steps: S41. The predicted corporate tax trend is obtained through the future feature set; the decision maker makes adjustments based on the predicted corporate tax trend; S42. Decision makers make the most suitable decision for the current development of the enterprise based on the focus of the current development of the enterprise and the Pareto optimal solution set obtained by the multi-objective particle swarm optimization algorithm.

10. A system for implementing the artificial intelligence optimization method according to any one of claims 1 to 9, characterized in that: It includes data preprocessing module, model training module, multi-objective optimization module and decision support module; The data preprocessing module is used to process the collected historical tax data, financial data, market data and fiscal and taxation policy data to maintain the accuracy and consistency of the data; The model training module is used to extract the features of financial and tax data, construct a feature matrix, train, test and optimize the model to obtain the final model, substitute the financial and tax data features of the enterprise whose tax trend needs to be predicted into the final model to obtain the tax trend feature set of the enterprise; The multi-objective optimization module is used to jointly optimize tax compliance, cost-effectiveness and capital liquidity to obtain a Pareto optimal solution set; The decision support module is used to adjust the tax trend of the enterprise according to the tax trend feature set and obtain the Pareto optimal solution set.

Images