Hierarchical-oriented intelligent cost accounting method

Abstract

The invention relates to a hierarchical-oriented intelligent cost accounting method, and belongs to the field of intelligent cost accounting. The method comprises five main steps. Acquiring cost data, preprocessing the cost data, and dividing the cost data into a training set and a test set; constructing a weighted time series addition decomposition model, and performing data extraction on cost data in the training set through the model to obtain a trend term, a season term and a residual term; processing the trend item, the season item and the residual item to obtain a new training sample set; a cost prediction model is constructed, the cost prediction model comprises a CNN residual module and a bidirectional LSTM module, and the new training sample set is input into the cost prediction model to obtain a predicted cost total amount; and evaluating the prediction performance of the model by performing error evaluation on the total cost of the test set and the total predicted cost. According to the invention, the accuracy of cost intelligent accounting and the capability of capturing complex cost change features are improved.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent cost accounting, and in particular relates to a hierarchical intelligent cost accounting method. Background Art

[0002] As the market becomes more and more perfect, competition becomes increasingly fierce, and the production environment of enterprises continues to change, traditional cost accounting methods often face problems such as heavy workload, low efficiency, and incorrect cost information at various levels. It is difficult to adapt to the refined cost management needs of enterprises. The rapid development of technologies such as the Internet of Things, cloud computing, and cloud storage has made it possible for intelligent cost accounting, enabling enterprises to achieve comprehensive, accurate, and efficient cost accounting. It also helps enterprises to detect cost anomalies in a timely manner and predict cost decisions. However, the existing intelligent cost accounting methods still face challenges in refined and hierarchical management. Therefore, studying an intelligent accounting method for multi-level costs to achieve refined intelligent management of cost accounting has become the key to improving the competitiveness of enterprises. Summary of the invention

[0003] In order to solve the above problems, the present invention provides a hierarchical cost intelligent accounting method.

[0004] In order to achieve the above object, the present invention is implemented through the following technical solutions: A hierarchical cost intelligent accounting method includes the following steps: S1. Obtain cost data, preprocess the cost data, and divide the cost data into a training set and a test set.

[0005] S2. construct a weighted time series additive decomposition model, and extract the cost data in the training set through the model to obtain trend items, seasonal items and residual items; S3. Process the trend term, seasonal term and residual term to obtain a new training sample set; S4. construct a cost prediction model, the cost prediction model includes a CNN residual module and a bidirectional LSTM module, and the new training sample set is input into the cost prediction model to obtain the total predicted cost; S5. Use the total cost in the test set and the predicted total cost to perform error assessment to evaluate the prediction performance of the model.

[0006] Further, step S1 includes: S11. Data collection: daily cost data is obtained from the internal financial system of the e-commerce management enterprise, and the cost data includes production raw material costs, fixed labor costs, fixed asset depreciation costs, office rental costs, management costs, server maintenance costs, product insurance costs, water and electricity costs, other production costs, advertising costs, and temporary labor costs; the cost data is divided into four levels, including direct costs, indirect costs, fixed costs, and other costs; the direct cost is the sum of production raw material costs and fixed labor costs, the indirect cost is the sum of fixed asset depreciation costs, server maintenance costs, and product insurance costs, the fixed cost is the sum of office rental costs, management costs, advertising costs, and water and electricity costs, and the other costs are the sum of other production costs and temporary labor costs; the total cost is the total of the four levels of direct costs, indirect costs, fixed costs, and other costs; S12. Data standardization: Standardize the cost data at each level. Specific operations include: unifying the data format, unit and data naming, filling the missing values ​​in the data at each level with the median value, and normalizing the cost data at each level; S13. Data storage and division: All data is divided into daily units, and the data characteristics of each day Including: date, total cost of the day, total direct cost, total indirect cost, total fixed cost and total other cost. All data are stored in Excel format to obtain a data set. , and the dataset Divide into training set and test set .

[0007] Further, step S2 includes: S21. Set the cost data of the trend item in days, with a period of dr=7; set consecutive dr days as a window, and use the sliding window average method to calculate the training set The trend item of daily cost data is expressed as follows: , in, Represents the estimated value of the cost at time node t, that is, the trend term, Indicates the original cost data value at the time node t+i days; S22. Set the cost data of seasonal items in months and calculate the training set The average value of cost data for each month is expressed as follows: , in, Indicates that the training set has Year, represents the cost data on the dth day of the mth month of the yth year, and Day represents the day of the mth month of the yth year. represents the average cost data of the mth month in all years, According to the trend item of the daily cost data in step S21, the trend item of the monthly cost data is obtained by cubic spline interpolation ; Calculate the monthly average after detrending , according to the number of holidays in each month Divide by the total number of days in the month Get the weight, perform weighted processing on the detrended monthly average, and get the weighted monthly average. The formula is as follows: , , , in, represents the monthly average after detrending, represents the weighted monthly average; expanding the weighted monthly average to each day yields , forming time series data in days; S23. Through the formula , get the time point The residual term , the final weighted time series additive decomposition model constructed It is expressed as: , in, Represents the training set Mid-time point data, Represent the trend term, seasonal term, and residual term respectively. Further, the operation of the cubic spline interpolation method is: S221. Constructing cubic polynomial segments To represent two adjacent time cost trend values, the formula is as follows: , When two adjacent time cost trend items are given and , is designed to pass through these points and provide smooth transitions between these points; Respectively represent the first coefficient, second coefficient, third coefficient, and fourth coefficient to be solved; t represents the variable at any time point, and its value range is [ ], is a specific time point, indicating the time corresponding to the i-th time cost trend value; S222. Constructing and solving systems of equations , the equation group is constructed using interpolation conditions and continuity conditions, and the formula is as follows: , , , , in, and Respectively represent the i-th and i+1-th cubic polynomials in The first derivative of the polynomial over the time segment, and Respectively represent the i-th and i+1-th cubic polynomials in Second derivatives of polynomials over time periods; S223. Find all After that, we get the spline interpolation function , select the 15th of each month as the node and enter After that, we can get the trend items of monthly cost data. .

[0008] Further, step S3 includes: S31. The result obtained in step S2 Convert it into a vector with months as the unit, and accumulate the cost data of each level for each month to get a new trend item , New Seasonal Items , the new residual term ; S32. In the new trend , New Seasonal Items , the new residual term A month is randomly selected as the starting point, and starting from the selected starting point, 12 consecutive months of data are extracted as a training sample, where the data of the first 11 months are used as input features and the data of the last month are used as prediction labels; the above operation is repeated N times to generate N independent training sample sets to obtain the final training sample set, which includes the trend item training sample set , seasonal item training sample set And the residual training sample set .

[0009] Further, step S4 includes: The trend item training sample set , seasonal item training sample set And the residual training sample set The three independent models are input into the constructed cost prediction model for training, namely the trend item prediction model , Seasonal Term Forecast Model , residual term prediction model ; From new trending items Extract the trend item data for the last 11 months As The trend item g(1) for the first month of the next year is predicted, and the new trend item The trend item data of the last 10 months in g(1) form a new set of 11 months of data ,Will As The trend item g(2) for the second month of the next year is predicted. The above operation is iterated, and each time the newly predicted trend item is combined with the trend item data of the previous 10 months to form the new trend item data of 11 months, until the trend items of the 12 months of the next year are predicted, and finally the trend items of 12 months are obtained. Similarly, we can get the seasonal items of the next 12 months. , the residual item for the next 12 months ;Will Add them together to get the total predicted cost , the formula is as follows: , Furthermore, the specific operations in the cost prediction model are: The CNN residual module includes a convolutional layer and a residual layer; the residual layer includes a first residual block, a second residual block, a third residual block and a fourth residual block; each residual block in the residual layer includes two convolutional layers, a batch normalization layer and an activation layer; Input the data in the final training sample set into the convolution layer to obtain the local features in the time series data The local features The first output feature is obtained through the first residual block , the local features and the first output feature Perform residual connection to obtain the first residual feature , the first residual feature After the ReLU activation function, the first deep feature is obtained ; The first deep feature The second output feature is obtained through the second residual block , the first deep feature and the second output feature Perform residual connection on the second residual feature , the second residual feature After the ReLU activation function, the second deep feature is obtained ; The second deep feature The third output feature is obtained through the third residual block , the second deep feature and the third output feature Perform residual connection to obtain the third residual feature , the third residual feature After the ReLU activation function, the third deep feature is obtained The third deep feature The fourth output feature is obtained through the fourth residual block , the third deep feature and the fourth output feature Perform residual connection to obtain the fourth residual feature , the fourth residual feature After the ReLU activation function, the fourth deep feature is obtained ; The bidirectional LSTM module includes two cascaded bidirectional LSTM layers; The fourth deep feature After the first bidirectional LSTM layer, the time series features are extracted to obtain preliminary time series features. , preliminary timing characteristics The high-level time series features are extracted through the second bidirectional LSTM layer to obtain high-level time series features. ; The high-level timing characteristics After the fully connected layer, the prediction result is obtained.

[0010] Further, step S5 includes: By correlation coefficient The prediction performance of the cost prediction model is evaluated, and the formula is expressed as follows: , in, Represents the total cost predicted by the model The average value in the kth month, The actual total cost The average value in the kth month.

[0011] The advantages of the present invention are: The present invention uses an improved time series decomposition model to perform hierarchical trend decomposition on historical cost data, which can more accurately capture the changing trends of various types of costs. Even if errors occur in the prediction of a certain type of cost, it will not have a significant impact on the entire cost forecast; by constructing a weighted time series additive decomposition model, trend items, seasonal items and residual items in cost data can be more effectively extracted, and the weighting mechanism more accurately reflects the contribution of different components in cost changes, improving the accuracy and reliability of cost decomposition; the combination of CNN and LSTM layers enables the model to more comprehensively understand the complexity and nonlinear characteristics of complex long-term cost data, thereby improving the accuracy and generalization of predictions. The present invention uses a deep learning intelligent model to analyze future cost trends, study the cost occurrence laws of enterprises, construct an intelligent accounting model, and verify and evaluate its effectiveness, which can provide a reference for future allocation of cost budgets and refined management. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] 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.

[0013] Figure 1 is a flow chart of the method of the present invention; Figure 2 The cost data collection and preprocessing flow chart of the method of the present invention; Figure 3 A case flow chart of the method of the present invention; Figure 4 is a sample curve chart of direct costs of the method of the present invention; Figure 5 is the trend term of the method of the present invention; Figure 6 is the seasonal term of the method of the present invention; Figure 7 is the residual term of the method of the present invention. DETAILED DESCRIPTION

[0014] 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.

[0015] Example 1 In this embodiment, Figure 1 As shown, the present invention provides a hierarchical cost intelligent accounting method, the specific steps of which include: S1. Obtain cost data, preprocess the cost data, and divide the cost data into a training set and a test set.

[0016] Specifically, Figure 2 As shown, S11. Data collection: daily cost data is obtained from the internal financial system of the e-commerce management enterprise, and the cost data includes production raw material costs, fixed labor costs, fixed asset depreciation costs, office rental costs, management costs, server maintenance costs, product insurance costs, water and electricity costs, other production costs, advertising costs, and temporary labor costs; the cost data is divided into 4 levels, including direct costs, indirect costs, fixed costs, and other costs; the direct cost is the sum of the production raw material cost and the fixed labor cost, and the indirect cost is the fixed asset depreciation cost, server maintenance cost, and product insurance cost. The fixed cost is the sum of office rental cost, management fee, advertising fee and water and electricity fee, and the other cost is the sum of other production cost and temporary labor cost; the total cost is the total of direct cost, indirect cost, fixed cost and other cost; S12. Data standardization: standardize the cost data of each level, including: unify data format, unit and data naming, fill missing values ​​in each level data according to median value, and normalize the cost data of each level; S13. Data storage and division: store all data in units of days, and the data characteristics of each day are Including: date, total cost of the day, total direct cost, total indirect cost, total fixed cost and total other cost, store all data in Excel format to get a data set , and the dataset Divide into training set and test set .

[0017] S2. construct a weighted time series additive decomposition model, and extract the cost data in the training set through the model to obtain trend items, seasonal items and residual items; Specifically, S21. Set the cost data of the trend item in days, with a period of dr=7; set consecutive dr days as a window, and use the sliding window average method to calculate the training set The trend item of daily cost data is expressed as follows: , in, Represents the estimated value of the cost at time node t, that is, the trend term, Indicates the original cost data value at the time node t+i days; S22. Set the cost data of seasonal items in months and calculate the training set The average value of cost data for each month is expressed as follows: , in, Indicates that the training set has Year, represents the cost data on the dth day of the mth month of the yth year, and Day represents the day of the mth month of the yth year. represents the average cost data of the mth month in all years, According to the trend item of the daily cost data in step S21, the trend item of the monthly cost data is obtained by cubic spline interpolation ; Calculate the monthly average after detrending , according to the number of holidays in each month Divide by the total number of days in the month Get the weight, perform weighted processing on the detrended monthly average, and get the weighted monthly average. The formula is as follows: , , , in, represents the monthly average after detrending, represents the weighted monthly average; expanding the weighted monthly average to each day yields , forming time series data in days; S23. Through the formula , get the time point The residual term , the final weighted time series additive decomposition model constructed It is expressed as: , in, Represents the training set Mid-time point data, They represent trend term, seasonal term and residual term respectively.

[0018] Specifically, step S22 includes: S221. Constructing cubic polynomial segments To represent two adjacent time cost trend values, the formula is as follows: , When two adjacent time cost trend items are given and , is designed to pass through these points and provide smooth transitions between these points; Respectively represent the first coefficient, second coefficient, third coefficient, and fourth coefficient to be solved; t represents the variable at any time point, and its value range is [ ], is a specific time point, indicating the time corresponding to the i-th time cost trend value; S222. Constructing and solving systems of equations , the equation group is constructed using interpolation conditions and continuity conditions, and the formula is as follows: , , , , in, and Respectively represent the i-th and i+1-th cubic polynomials in The first derivative of the polynomial over the time segment, and Respectively represent the i-th and i+1-th cubic polynomials in Second derivatives of polynomials over time periods; S223. Find all After that, we get the spline interpolation function , select the 15th of each month as the node and enter After that, we can get the trend items of monthly cost data. .

[0019] S3. Process the trend term, seasonal term and residual term to obtain a new training sample set; Specifically, S31. The obtained in step S2 Convert it into a vector with months as the unit, and accumulate the cost data of each level for each month to get a new trend item , New Seasonal Items , the new residual term ; S32. In the new trend , New Seasonal Items , the new residual term A month is randomly selected as the starting point, and starting from the selected starting point, 12 consecutive months of data are extracted as a training sample, where the data of the first 11 months are used as input features and the data of the last month are used as prediction labels; the above operation is repeated N times to generate N independent training sample sets to obtain the final training sample set, which includes the trend item training sample set , seasonal item training sample set And the residual training sample set .

[0020] S4. construct a cost prediction model, the cost prediction model includes a CNN residual module and a bidirectional LSTM module, and the new training sample set is input into the cost prediction model to obtain the total predicted cost; The trend item training sample set , seasonal item training sample set And the residual training sample set The three independent models are input into the constructed cost prediction model for training, namely the trend item prediction model , Seasonal Term Forecast Model , residual term prediction model ; From new trending items Extract the trend item data for the last 11 months As The trend item g(1) for the first month of the next year is predicted, and the new trend item The trend item data of the last 10 months in g(1) form a new set of 11 months of data ,Will As The trend item g(2) for the second month of the next year is predicted. The above operation is iterated, and each time the newly predicted trend item is combined with the trend item data of the previous 10 months to form the new trend item data of 11 months, until the trend items of the 12 months of the next year are predicted, and finally the trend items of 12 months are obtained. Similarly, we can get the seasonal items of the next 12 months. , the residual item for the next 12 months ;Will Add them together to get the total predicted cost , the formula is as follows: .

[0021] Specifically, the operations in the cost prediction model are: The CNN residual module includes a convolutional layer and a residual layer; the residual layer includes a first residual block, a second residual block, a third residual block and a fourth residual block; each residual block in the residual layer includes two convolutional layers, a batch normalization layer and an activation layer; Input the data in the final training sample set into the convolution layer to obtain the local features in the time series data The local features The first output feature is obtained through the first residual block , the local features and the first output feature Perform residual connection to obtain the first residual feature , the first residual feature After the ReLU activation function, the first deep feature is obtained ; The first deep feature The second output feature is obtained through the second residual block , the first deep feature and the second output feature Perform residual connection on the second residual feature , the second residual feature After the ReLU activation function, the second deep feature is obtained ; The second deep feature The third output feature is obtained through the third residual block , the second deep feature and the third output feature Perform residual connection to obtain the third residual feature , the third residual feature After the ReLU activation function, the third deep feature is obtained The third deep feature The fourth output feature is obtained through the fourth residual block , the third deep feature and the fourth output feature Perform residual connection to obtain the fourth residual feature , the fourth residual feature After the ReLU activation function, the fourth deep feature is obtained ; The bidirectional LSTM module includes two cascaded bidirectional LSTM layers; The fourth deep feature After the first bidirectional LSTM layer, the time series features are extracted to obtain preliminary time series features. , preliminary timing characteristics The high-level time series features are extracted through the second bidirectional LSTM layer to obtain high-level time series features. ; The high-level timing characteristics After the fully connected layer, the prediction result is obtained.

[0022] S5. Use the total cost in the test set and the predicted total cost to perform error assessment to evaluate the prediction performance of the model.

[0023] Specifically, through the correlation coefficient The prediction performance of the cost prediction model is evaluated, and the formula is expressed as follows: , in, Represents the total cost predicted by the model The average value in the kth month, The actual total cost The average value in the kth month.

[0024] Example 2 In order to illustrate the effectiveness of the model in this paper, we use the data collected by the method of the present invention to conduct a comparative experiment between the method of the present invention and the traditional STL model and LSTNet model. The STL model is a method for time series decomposition, and the LSTNet model is a deep learning framework that combines convolutional neural networks and recurrent neural networks to process long-term and short-term time series patterns. From the results, it can be seen that the correlation coefficient CC between the predicted value and the true value of the model of the method of the present invention reaches 0.8536, which is higher than the correlation coefficient between the predicted value and the true value of the STL model and the LSTNet model, proving the advantages of the hierarchical cost intelligent accounting method proposed in this invention.

[0025] Table 1 Comparison of experimental data between the proposed method and the traditional STL and LSTNet models Secondly, we use the data from the first 11 months of 2023 as historical data and the data from the last month as labels to illustrate the cost prediction process with specific cases. Figure 4 ) as an example, the case flow is as follows Figure 3 As shown, firstly, the improved time series additive model is applied to decompose it into trend terms (see Figure 5 ), seasonal items (see Figure 6 ) and the residual term (see Figure 7), and then input the three decomposed components into the cost forecast model respectively, and obtain the three predicted trend items, seasonal items and residual items. The three items are superimposed to obtain the forecast value for December. Then we compare the original direct cost label with the predicted cost label and calculate the correlation coefficient value. The method of the present invention can accurately perform cost accounting.

[0026] 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 hierarchical cost intelligent accounting method, characterized in that: The following steps are involved: S1. Obtain cost data, preprocess the cost data, and divide the cost data into a training set and a test set; S2. construct a weighted time series additive decomposition model, and extract the cost data in the training set through the model to obtain trend items, seasonal items and residual items; S3. Process the trend term, seasonal term and residual term to obtain a new training sample set; S4. construct a cost prediction model, the cost prediction model includes a CNN residual module and a bidirectional LSTM module, and the new training sample set is input into the cost prediction model to obtain the total predicted cost; S5. Use the total cost in the test set and the predicted total cost to perform error assessment to evaluate the prediction performance of the model.

2. A hierarchical cost intelligent accounting method according to claim 1, characterized in that: Step S1 includes: S11. Data collection: daily cost data is obtained from the internal financial system of the e-commerce management enterprise, and the cost data includes production raw material costs, fixed labor costs, fixed asset depreciation costs, office rental costs, management costs, server maintenance costs, product insurance costs, water and electricity costs, other production costs, advertising costs, and temporary labor costs; the cost data is divided into four levels, including direct costs, indirect costs, fixed costs, and other costs; the direct cost is the sum of production raw material costs and fixed labor costs, the indirect cost is the sum of fixed asset depreciation costs, server maintenance costs, and product insurance costs, the fixed cost is the sum of office rental costs, management costs, advertising costs, and water and electricity costs, and the other costs are the sum of other production costs and temporary labor costs; the total cost is the total of the four levels of direct costs, indirect costs, fixed costs, and other costs; S12. Data standardization: Standardize the cost data at each level. Specific operations include: unifying the data format, unit and data naming, filling the missing values ​​in the data at each level with the median value, and normalizing the cost data at each level; S13. Data storage and division: All data is divided into daily units, and the data characteristics of each day Including: date, total cost of the day, total direct cost, total indirect cost, total fixed cost and total other cost. All data are stored in Excel format to obtain a data set. , and the dataset Divide into training set and test set .

3. A hierarchical cost intelligent accounting method according to claim 2, characterized in that: Step S2 includes: S21. Set the cost data of the trend item in days, with a period of dr=7; set consecutive dr days as a window, and use the sliding window average method to calculate the training set The trend item of daily cost data is expressed as follows: ， in, Represents the estimated value of the cost at time node t, that is, the trend term, Indicates the original cost data value at the time node t+i days; S22. Set the cost data of seasonal items in months and calculate the training set The average value of cost data for each month is expressed as follows: ， in, Indicates that the training set has Year, represents the cost data on the dth day of the mth month of the yth year, and Day represents the day of the mth month of the yth year. represents the average cost data of the mth month in all years, According to the trend item of the daily cost data in step S21, the trend item of the monthly cost data is obtained by cubic spline interpolation ; Calculate the monthly average after detrending , according to the number of holidays in each month Divide by the total number of days in the month Get the weight, perform weighted processing on the detrended monthly average, and get the weighted monthly average. The formula is as follows: ， ， ， in, represents the monthly average after detrending, represents the weighted monthly average; expanding the weighted monthly average to each day yields , forming time series data in days; S23. Through the formula , get the time point The residual term , the final weighted time series additive decomposition model constructed It is expressed as: ， in, Represents the training set Mid-time point data, They represent trend term, seasonal term and residual term respectively.

4. A hierarchical cost intelligent accounting method according to claim 3, characterized in that: The operation of cubic spline interpolation is: S221. Constructing cubic polynomial segments To represent two adjacent time cost trend values, the formula is as follows: ， When two adjacent time cost trend items are given and , is designed to pass through these points and provide smooth transitions between these points; Respectively represent the first coefficient, second coefficient, third coefficient, and fourth coefficient to be solved; t represents the variable at any time point, and its value range is [ ], is a specific time point, indicating the time corresponding to the i-th time cost trend value; S222. Constructing and solving systems of equations , the equation group is constructed using interpolation conditions and continuity conditions, and the formula is as follows: ， ， ， ， in, and Respectively represent the i-th and i+1-th cubic polynomials in The first derivative of the polynomial over the time segment, and Respectively represent the i-th and i+1-th cubic polynomials in Second derivatives of polynomials over time periods; S223. Find all After that, we get the spline interpolation function , select the 15th of each month as the node and enter After that, we can get the trend items of monthly cost data. .

5. A hierarchical cost intelligent accounting method according to claim 4, characterized in that: Step S3 includes: S31. The result obtained in step S2 Convert it into a vector with months as the unit, and accumulate the cost data of each level for each month to get a new trend item , New Seasonal Items , the new residual term ; S32. In the new trend , New Seasonal Items , the new residual term A month is randomly selected as the starting point, and starting from the selected starting point, 12 consecutive months of data are extracted as a training sample, where the data of the first 11 months are used as input features and the data of the last month are used as prediction labels; the above operation is repeated N times to generate N independent training sample sets to obtain the final training sample set, which includes the trend item training sample set , seasonal item training sample set And the residual training sample set .

6. A hierarchical cost intelligent accounting method according to claim 5, characterized in that: Step S4 includes: The trend item training sample set , seasonal item training sample set And the residual training sample set The three independent models are input into the constructed cost prediction model for training, namely the trend item prediction model , Seasonal Term Forecast Model , residual term prediction model ; From new trending items Extract the trend item data for the last 11 months As The trend item g(1) for the first month of the next year is predicted, and the new trend item The trend item data of the last 10 months in g(1) form a new set of 11 months of data ,Will As The trend item g(2) for the second month of the next year is predicted. The above operation is iterated, and each time the newly predicted trend item is combined with the trend item data of the previous 10 months to form the new trend item data of 11 months, until the trend items of the 12 months of the next year are predicted, and finally the trend items of 12 months are obtained. Similarly, we can get the seasonal items of the next 12 months. , the residual item for the next 12 months ;Will Add them together to get the total predicted cost , the formula is as follows: 。 7. A hierarchical cost intelligent accounting method according to claim 6, characterized in that: The specific operations in the cost prediction model are: The CNN residual module includes a convolutional layer and a residual layer; the residual layer includes a first residual block, a second residual block, a third residual block and a fourth residual block; each residual block in the residual layer includes two convolutional layers, a batch normalization layer and an activation layer; Input the data in the final training sample set into the convolution layer to obtain the local features in the time series data The local features The first output feature is obtained through the first residual block , the local features and the first output feature Perform residual connection to obtain the first residual feature , the first residual feature After the ReLU activation function, the first deep feature is obtained ; The first deep feature The second output feature is obtained through the second residual block , the first deep feature and the second output feature Perform residual connection on the second residual feature , the second residual feature After the ReLU activation function, the second deep feature is obtained ; The second deep feature The third output feature is obtained through the third residual block , the second deep feature and the third output feature Perform residual connection to obtain the third residual feature , the third residual feature After the ReLU activation function, the third deep feature is obtained The third deep feature The fourth output feature is obtained through the fourth residual block , the third deep feature and the fourth output feature Perform residual connection to obtain the fourth residual feature , the fourth residual feature After the ReLU activation function, the fourth deep feature is obtained ; The bidirectional LSTM module includes two cascaded bidirectional LSTM layers; The fourth deep feature After the first bidirectional LSTM layer, the time series features are extracted to obtain preliminary time series features. , preliminary timing characteristics The high-level time series features are extracted through the second bidirectional LSTM layer to obtain high-level time series features. ; The high-level timing characteristics After the fully connected layer, the prediction result is obtained.

8. The hierarchical cost intelligent accounting method according to claim 7, characterized in that: Step S5 includes: By correlation coefficient The prediction performance of the cost prediction model is evaluated, and the formula is expressed as follows: ， in, Represents the total cost predicted by the model The average value in the kth month, The actual total cost The average value in the kth month.

Images