A water conservancy budget data summary analysis method and management system

Abstract

The present invention relates to a water conservancy budget data aggregation and analysis method and management system, belonging to the technical field of water conservancy budget data analysis and management. The method comprises the following steps: obtaining water conservancy budget data, performing data aggregation operations through a budget data collection module to obtain a water conservancy budget text data sample set; inputting the water conservancy budget text data sample set into a budget data preprocessing module for processing to obtain a simplified water conservancy budget text data sample set; performing hidden data conversion, feature optimization, and scale reconciliation operations through a budget data privacy protection module to obtain a standardized water conservancy budget text data set; and processing the standardized water conservancy budget text data set through a budget text data intelligent classification module to obtain a final decision output. The present invention can ensure data privacy while improving work efficiency and accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of water conservancy budget data analysis and management, and in particular relates to a water conservancy budget data summary analysis method and management system. Background Art

[0002] In recent years, the rapid development of artificial intelligence (AI) technology has enabled it to be integrated into various industries, holding enormous potential for application in data analysis. In modern business and government management, AI, through machine learning and deep learning algorithms, can quickly and accurately process and analyze vast amounts of data across industries. The water conservancy industry involves a vast amount of data related to manual budgeting, material procurement, and cost accounting. Traditional manual budget compilation and data analysis methods are inefficient and prone to errors. However, the automated and intelligent collection, aggregation, and analysis of water conservancy budget data using AI technology can significantly improve efficiency and accuracy. Furthermore, AI-powered management systems can dynamically analyze budget data, monitor its execution in real time, and provide a scientific basis for future budgetary needs. This approach not only optimizes budget management processes but also provides more accurate financial budget analysis and risk warning mechanisms for the water conservancy industry, significantly enhancing management efficiency and financial transparency.

[0003] In the field of modern water conservancy budget data management systems, although with the development of science and technology, more and more work content is processed by computers, the method still relies on manual budget data forecasting and analysis, which has problems such as low efficiency and insufficient accuracy. Therefore, the present invention proposes a water conservancy budget data aggregation and analysis method and management system, which can use deep learning algorithms to predict water conservancy budget data, optimize resource allocation, improve the scientificity and rationality of overall budget decision-making, and make budget management more efficient and intelligent. Summary of the Invention

[0004] In order to solve the above problems, the present invention provides a water conservancy budget data summary and analysis method and management system.

[0005] To achieve the above-mentioned purpose, the present invention is implemented through the following technical solutions:

[0006] The present invention provides a water conservancy budget data summary and analysis method, comprising the following steps:

[0007] S1. Obtain water conservancy budget data. The water conservancy budget data is aggregated through the budget data collection module to obtain a water conservancy budget text data sample set. ;

[0008] S2. Collect water conservancy budget text data samples Input into the budget data preprocessing module for disassembly operation, benchmark text difference fitting operation and simplest semantic representation operation to obtain a simplified water conservancy budget text data sample set ;

[0009] S3. Simplified water conservancy budget text data sample collection After the budget data privacy protection module performs hidden data operations, feature optimization and scale coordination operations, a standardized water conservancy budget text data set is obtained. ;

[0010] S4. Standardized water budget text data set After being processed by the budget text data intelligent classification module, the final decision output is obtained ; The budget text data intelligent classification module includes a text data feature encoder, an adaptive window attention text capture module and a hierarchical feature fusion output module.

[0011] Furthermore, step S1 specifically includes:

[0012] The water conservancy budget data includes manual real-time information input , local historical budget data and collection of publicly available industry information , the water conservancy budget data is aggregated through the budget data collection module to obtain a water conservancy budget text data sample set , the formula is as follows:

[0013] .

[0014] Furthermore, step S2 specifically includes:

[0015] S21. Water conservancy budget text data sample collection Input into the budget data preprocessing module, and after the semantic unit disassembly operation, the disassembled data sample set is obtained. , the formula is as follows:

[0016] ,

[0017] in, Represents the disassembled data sample set, , Indicates the After disassembly, there are a total of item; Represents the semantic unit disassembly function, Represents a collection of text data Break it down into semantic units according to the space separator;

[0018] S22. The disassembled data sample set Perform benchmark text difference fitting operation to obtain benchmark text set ,in, is the benchmark data item, indicating the The base text of each data item; get the base text set Then calculate each semantic unit The corresponding benchmark data items The difference value , if the difference value Less than a predetermined threshold , then keep this difference value On the contrary, the semantic unit Treated as a new benchmark data item, and finally a new benchmark text is obtained , the benchmark data has a total of The formula for the above process is as follows:

[0019] ,

[0020] ,

[0021] ,

[0022] in, Represents computational semantic units and candidate benchmarks The similarity between represents a candidate benchmark item, Representing semantic units and benchmark data items The character length of

[0023] S23. Collect the new benchmark text After the simplest semantic representation operation, a simplified water conservancy budget text data sample set is obtained , which uses a hash function to generate each benchmark data item Generate a unique identifier , when simplifying, if it already exists in the database Then for the corresponding Do not store the data to remove duplicates from the budget data text. The formula is as follows:

[0024] ,

[0025] ,

[0026] in, represents the hash algorithm, The dataset benchmark data has a total of item.

[0027] Furthermore, step S3 specifically includes:

[0028] S31. Add noise to the benchmark data item. The formula is as follows:

[0029] ,

[0030] in, represents the covariance matrix after adding noise, The mean is 0 and the variance is The Gaussian noise matrix, represents the mean of all benchmark data items, is the total number of benchmark data;

[0031] S32. Select the covariance matrix after adding noise Before The eigenvectors corresponding to the largest eigenvalues ​​constitute the dimensionality reduction matrix , and then the corresponding hidden features are obtained through calculation , the formula is as follows:

[0032] ,

[0033] in, is the dimension reduction matrix The transposed matrix of ; finally the standardized water conservancy budget text data set is obtained , ,in, express The total number of samples included, express feature dimension.

[0034] Furthermore, step S4 specifically includes:

[0035] The text data feature encoder is a bidirectional long short-term memory network Bi-LSTM; the adaptive window attention text capture module uses a fixed window mask matrix to limit the focus range, specifically:

[0036] S41. Standardized water budget text data set After processing by the text data feature encoder, the embedded features are obtained ;

[0037] S42. Embedding features in the adaptive window attention text capture module Mapped into query vector through the fully connected layer , key vector Sum value vector , perform dot product attention calculation on the query vector and the key vector to obtain the dot product attention , the formula is as follows:

[0038] ,

[0039] ,

[0040] in, 、 and Represent the query vector , key vector Sum value vector The trainable weight matrix, 、 and Represent the query vector , key vector Sum value vector The trainable bias of Represents the key vector The transposed vector of ; fixed window mask matrix The formula is as follows:

[0041] ,

[0042] in, Indicates the calculation scope of attention, specifically the meaning of vocabulary Only distance vocabulary The maximum range is Attention calculation, Indicates the vocabulary index currently being calculated for attention, Indicates the index of the word being focused on, Represents the maximum context window radius; context radius vector of the adaptive window attention text capture module The calculation process is:

[0043] ,

[0044] in, express Activation function used to limit the context radius vector The value of , represents the first weight parameter, represents the first offset, represents the second weight parameter, Represents the second bias. The four parameters are obtained through back propagation and optimization during model training. The range of the fixed window mask matrix and dot-product attention use Activation function calculates local attention score , the public notice is as follows:

[0045] ,

[0046] in, Expressive vocabulary distance vocabulary The maximum range is Attention calculation, Indicates the sub-feature corresponding to the vocabulary index currently being calculated for attention Sub-features corresponding to the index of the word being focused The local attention score between ; Represents sub-features The dot product attention calculation result of Represents the activation function; the sub-feature The local attention score and its value vector Multiply to get the adaptive window attention text feature , and finally obtain the feature set ;

[0047] S43. Set the features Input to the hierarchical feature fusion output module, perform weighted fusion on the features under different window sizes, and output the final data classification results , the formula is as follows:

[0048] ,

[0049] in, Represents sub-features The weight parameter, represents the weight matrix, Indicates the offset.

[0050] Furthermore, the threshold value in step S22 Manually adjust the threshold according to its impact on model performance. The initial range is .

[0051] The present invention also provides a water conservancy budget data management system, which executes the water conservancy budget data summary and analysis method, including:

[0052] Budget data collection module: used to obtain water conservancy budget data. The water conservancy budget data is aggregated by the budget data collection module to obtain a water conservancy budget text data sample set;

[0053] Budget data preprocessing module: used to input the water conservancy budget text data sample set into the budget data preprocessing module for disassembly operation, benchmark text difference fitting operation and simplest semantic representation operation, so as to obtain a simplified water conservancy budget text data sample set;

[0054] Budget data privacy protection module: used to input the simplified water conservancy budget text data sample set into the budget data privacy protection module for hidden data operation, feature optimization and scale reconciliation operation, and obtain a standardized water conservancy budget text data set;

[0055] Budget data prediction module: used to input the standardized water conservancy budget text data set into the budget text data intelligent classification module for processing to obtain the final decision output; the budget text data intelligent classification module includes a text data feature encoder, an adaptive window attention text capture module and a hierarchical feature fusion output module.

[0056] The advantages of the present invention are:

[0057] The present invention proposes a water conservancy budget data summary and analysis method, which is used to solve the problem that data analysis in the field of water conservancy budget data management systems requires manual intervention. Traditional manual analysis methods have the problems of low efficiency and insufficient accuracy and are prone to privacy leakage. Using the water conservancy budget data summary and analysis method proposed by the present invention, the data can be uniformly processed and converted into data required by the data analysis model through the budget data preprocessing module. The budget data privacy protection module of this method can ensure the privacy of the data. The water conservancy budget data can be analyzed through the budget text data intelligent classification module, which greatly improves work efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] 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.

[0059] Figure 1 Flow chart of the steps of the method of the present invention. DETAILED DESCRIPTION

[0060] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments derived by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0061] Example 1

[0062] In this embodiment, Figure 1 As shown, the present invention provides a method for summarizing and analyzing water conservancy budget data, which specifically includes the following steps:

[0063] S1. Obtain water conservancy budget data. The water conservancy budget data is aggregated through the budget data collection module to obtain a water conservancy budget text data sample set. ;

[0064] Specifically, the water conservancy budget data includes manual real-time information input , local historical budget data and collection of publicly available industry information , the water conservancy budget data is aggregated through the budget data collection module to obtain a water conservancy budget text data sample set , the formula is as follows:

[0065] .

[0066] S2. Collect water conservancy budget text data samples Input into the budget data preprocessing module for disassembly operation, benchmark text difference fitting operation and simplest semantic representation operation to obtain a simplified water conservancy budget text data sample set ;

[0067] Specifically, S21. Water conservancy budget text data sample set Input into the budget data preprocessing module, and after the semantic unit disassembly operation, the disassembled data sample set is obtained. , the formula is as follows:

[0068] ,

[0069] in, Represents the disassembled data sample set, , Indicates the After disassembly, there are a total of item; Represents the semantic unit disassembly function, Represents a collection of text data Break it down into semantic units according to the space separator;

[0070] S22. The disassembled data sample set Perform benchmark text difference fitting operation to obtain benchmark text set ,in, is the benchmark data item, indicating the The base text of each data item; get the base text set Then calculate each semantic unit The corresponding benchmark data items The difference value , if the difference value Less than a predetermined threshold , then keep this difference value On the contrary, the semantic unit Treated as a new benchmark data item, and finally a new benchmark text is obtained , the benchmark data has a total of The formula for the above process is as follows:

[0071] ,

[0072] ,

[0073] ,

[0074] in, Represents computational semantic units and candidate benchmarks The similarity between represents a candidate benchmark item, Representing semantic units and benchmark data items The character length of

[0075] S23. Collect the new benchmark text After the simplest semantic representation operation, a simplified water conservancy budget text data sample set is obtained , which uses a hash function to generate each benchmark data item Generate a unique identifier , when simplifying, if it already exists in the database Then for the corresponding Do not store the data to remove duplicates from the budget data text. The formula is as follows:

[0076] ,

[0077] ,

[0078] in, represents the hash algorithm, The dataset benchmark data has a total of item.

[0079] Specifically, the threshold value in step S22 Manual adjustment is performed based on the impact of the threshold on model performance. In text classification research, the threshold The initial range is .

[0080] S3. Simplified water conservancy budget text data sample collection After the budget data privacy protection module performs hidden data operations, feature optimization and scale coordination operations, a standardized water conservancy budget text data set is obtained. ;

[0081] Specifically, S31. adds noise to the benchmark data item, and the formula is expressed as follows:

[0082] ,

[0083] in, represents the covariance matrix after adding noise, The mean is 0 and the variance is The Gaussian noise matrix, represents the mean of all benchmark data items, is the total number of benchmark data;

[0084] S32. Select the covariance matrix after adding noise Before The eigenvectors corresponding to the largest eigenvalues ​​constitute the dimensionality reduction matrix , and then the corresponding hidden features are obtained through calculation , the formula is as follows:

[0085] ,

[0086] in, is the dimension reduction matrix The transposed matrix of ; finally the standardized water conservancy budget text data set is obtained , ,in, express The total number of samples included, express feature dimension.

[0087] S4. Standardized water budget text data set After being processed by the budget text data intelligent classification module, the final decision output is obtained ; The budget text data intelligent classification module includes a text data feature encoder, an adaptive window attention text capture module and a hierarchical feature fusion output module.

[0088] Specifically, the text data feature encoder is a bidirectional long short-term memory network Bi-LSTM; the adaptive window attention text capture module uses a fixed window mask matrix to limit the focus range, specifically:

[0089] S41. Standardized water budget text data set After processing by the text data feature encoder, the embedded features are obtained ;

[0090] S42. Embedding features in the adaptive window attention text capture module Mapped into query vector through the fully connected layer , key vector Sum value vector , perform dot product attention calculation on the query vector and the key vector to obtain the dot product attention , the formula is as follows:

[0091] ,

[0092] ,

[0093] in, 、 and Represent the query vector , key vector Sum value vector The trainable weight matrix, 、 and Represent the query vector , key vector Sum value vector The trainable bias of Represents the key vector The transposed vector of ; fixed window mask matrix The formula is as follows:

[0094] ,

[0095] in, Indicates the calculation scope of attention, specifically the meaning of vocabulary Only distance vocabulary The maximum range is Attention calculation, Indicates the vocabulary index currently being calculated for attention, Indicates the index of the word being focused on, Represents the maximum context window radius; context radius vector of the adaptive window attention text capture module The calculation process is:

[0096] ,

[0097] in, express Activation function used to limit the context radius vector The value of , represents the first weight parameter, represents the first offset, represents the second weight parameter, Represents the second bias. The four parameters are obtained through back propagation and optimization during model training. The range of the fixed window mask matrix and dot-product attention use Activation function calculates local attention score , the public notice is as follows:

[0098] ,

[0099] in, Expressive vocabulary distance vocabulary The maximum range is Attention calculation, Indicates the sub-feature corresponding to the vocabulary index currently being calculated for attention Sub-features corresponding to the index of the word being focused The local attention score between ; Represents sub-features The dot product attention calculation result of Represents the activation function; the sub-feature The local attention score and its value vector Multiply to get the adaptive window attention text feature , and finally obtain the feature set ;

[0100] S43. Set the features Input to the hierarchical feature fusion output module, perform weighted fusion on the features under different window sizes, and output the final data classification results , which is the required budget data inferred by the proposed method based on historical information, real-time input data and industry public information . , the formula is as follows:

[0101] ,

[0102] in, Represents sub-features The weight parameter, represents the weight matrix, Indicates the offset.

[0103] Example 2

[0104] This embodiment provides a water conservancy budget data management system, which executes the water conservancy budget data aggregation and analysis method described in Example 1, including:

[0105] Budget data collection module: used to obtain water conservancy budget data. The water conservancy budget data is aggregated by the budget data collection module to obtain a water conservancy budget text data sample set;

[0106] Budget data preprocessing module: used to input the water conservancy budget text data sample set into the budget data preprocessing module for disassembly operation, benchmark text difference fitting operation and simplified minimalist semantic representation operation, so as to obtain a simplified water conservancy budget text data sample set;

[0107] Budget data privacy protection module: used to input the water conservancy budget text data sample set into the budget data privacy protection module to perform hidden data operations, feature optimization and scale harmonization operations, and obtain a standardized water conservancy budget text data set;

[0108] Budget data prediction module: used to input the standardized water conservancy budget text data set into the budget text data intelligent classification module for processing to obtain the final decision output; the budget text data intelligent classification module includes a text data feature encoder, an adaptive window attention text capture module and a hierarchical feature fusion output module.

[0109] Example 3

[0110] In order to better demonstrate the technical effect of the present invention, the method of the present invention was used to conduct experiments, as follows:

[0111] The data is collected from quarterly budget data obtained from a company's official channels. To ensure the authority and accuracy of the data, the data is collected and verified through the company's annual reports, quarterly reports, and public data. The data covers nearly ten years of quarterly data (budget data from 2015 to 2024) to ensure that the data has time series characteristics. The dataset includes the specific time of each quarter, such as 2024Q1 and 2024Q2, as well as the budget amount for each quarter (such as labor costs, travel expenses, and the total budget execution status, etc.). The FNBD dataset is constructed based on the above data. The FNBD dataset is an .xlsx file. The specific details of the file are as follows: 1. Table structure: The budget data for each quarter are arranged in rows, and each row represents the budget information for a quarter; 2. Column names in the table: (1) Time: Indicates the specific quarterly time, in the format of YYYYQX, where YYYY represents the year and QX represents the quarter (such as 2024Q1 represents the first quarter of 2024); (2) Human resource cost: The human resource cost budget amount for the quarter, in monetary units (such as RMB, US dollars, etc.); (3) Travel expenses: The travel expense budget amount for the quarter, in monetary units; (4) Total budget execution: The total amount of budget execution for the quarter, in monetary units; (5) Other budget items: If there are other budget items (such as equipment procurement, marketing promotion, etc.), list them in this column, in monetary units; (6) Remarks: Used to record special circumstances or explanations of the budget data for the quarter (such as budget adjustments, abnormal circumstances, etc.). The dataset is divided into training and test sets in chronological order. In the experiment, the training set accounts for 80% and the test set accounts for 20%. Specifically, the data covers the period from 2015 to 2024, with the training set allocated to 2015 to 2022 and the test set to 2023 and 2024.

[0112] The experiment compared the proposed water conservancy budget data aggregation and analysis method with other existing budget models under the same environmental configuration conditions. The three comparison systems used were: an LSTM time series forecasting system, an ARIMA budget forecasting model, and a Transformer model. The error between the predicted results and the true values ​​was analyzed using three experimental metrics: absolute error (AE), mean absolute error (MAE), and mean absolute percentage error (MAPE).

[0113] Table 1 Comparison results between the budget forecasted by the method of the present invention and the actual expenditure amount

[0114]

[0115] Experimental results (Table 1) show that the proposed forecasting model achieved a mean absolute error (MAE) of 16.5, which is relatively small compared to the actual expenditure figures (hundreds of units). The mean absolute percentage error (MAPE) was only 3.05%, indicating that the model's forecasted expenditures are close to actual expenditures, with low deviations, meeting the accuracy requirements for fiscal budget forecasts. Observing the data, we find that the trends in predicted and actual expenditures are largely consistent, demonstrating that the model effectively captures the changing trends in fiscal expenditures.

[0116] Table 2 Comparison results between the method of the present invention and other prediction models

[0117]

[0118] From the experimental results in Table 2, it can be seen that the proposed method model performs best in budget expenditure forecasting, and is significantly better than the LSTM, ARIMA, and Transformer models in terms of both absolute error (MAE) and relative error (MAPE).

[0119] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are 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 will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A water conservancy budget data summary and analysis method, characterized in that: The following steps are involved: S1. Obtain water conservancy budget data. The water conservancy budget data is aggregated through the budget data collection module to obtain a water conservancy budget text data sample set. ; S2. Collect water conservancy budget text data samples Input into the budget data preprocessing module for disassembly operation, benchmark text difference fitting operation and simplest semantic representation operation to obtain a simplified water conservancy budget text data sample set ; S3. Simplified water conservancy budget text data sample collection After the budget data privacy protection module performs hidden data operations, feature optimization and scale coordination operations, a standardized water conservancy budget text data set is obtained. ; S31. Add noise to the benchmark data item. The formula is as follows: ， in, represents the covariance matrix after adding noise, represents the benchmark data item, The mean is 0 and the variance is The Gaussian noise matrix, represents the mean of all benchmark data items, is the total number of benchmark data; S32. Select the covariance matrix after adding noise Before The eigenvectors corresponding to the largest eigenvalues ​​constitute the dimensionality reduction matrix , and then the corresponding hidden features are obtained through calculation , the formula is as follows: ， in, is the dimension reduction matrix The transposed matrix of the water conservancy budget text data set is finally obtained , ,in, express The total number of samples included, express characteristic dimensions; S4. Standardized water budget text data set After being processed by the budget text data intelligent classification module, the final decision output is obtained ; The budget text data intelligent classification module includes a text data feature encoder, an adaptive window attention text capture module and a hierarchical feature fusion output module.

2. A water conservancy budget data summary and analysis method according to claim 1, characterized in that: Step S1 specifically includes: The water conservancy budget data includes manual real-time information input , local historical budget data and collection of publicly available industry information , the water conservancy budget data is aggregated through the budget data collection module to obtain a water conservancy budget text data sample set , the formula is as follows: 。 3. A water conservancy budget data summary and analysis method according to claim 2, characterized in that: Step S2 specifically includes: S21. Water conservancy budget text data sample collection Input into the budget data preprocessing module, and after the semantic unit disassembly operation, the disassembled data sample set is obtained. , the formula is as follows: ， in, Represents the disassembled data sample set, , Indicates the After the disassembly, there are a total of item; Represents the semantic unit disassembly function, Represents a collection of text data Break it down into semantic units according to the space separator; S22. The disassembled data sample set Perform benchmark text difference fitting operation to obtain benchmark text set ,in, is the benchmark data item, indicating the The base text of each data item; get the base text set Then calculate each semantic unit The corresponding benchmark data items The difference value , if the difference value Less than a predetermined threshold , then keep this difference value On the contrary, the semantic unit Treated as a new benchmark data item, and finally a new benchmark text is obtained , the benchmark data has a total of The formula for the above process is as follows: ， ， ， in, Represents computational semantic units and candidate benchmarks The similarity between represents a candidate benchmark item, Representing semantic units and benchmark data items The character length of S23. Collect the new benchmark text After the simplest semantic representation operation, a simplified water conservancy budget text data sample set is obtained , which uses a hash function to generate each benchmark data item Generate a unique identifier , when simplifying, if it already exists in the database Then for the corresponding Do not store the data to remove duplicates from the budget data text. The formula is as follows: ， ， in, represents the hash algorithm, The dataset benchmark data has a total of item.

4. A water conservancy budget data summary and analysis method according to claim 3, characterized in that: Step S4 specifically includes: The text data feature encoder is a bidirectional long short-term memory network Bi-LSTM; the adaptive window attention text capture module uses a fixed window mask matrix to limit the focus range, specifically: S41. Standardized water budget text data set After processing by the text data feature encoder, the embedded features are obtained ; S42. Embedding features in the adaptive window attention text capture module Mapped into query vector through the fully connected layer , key vector Sum value vector , perform dot product attention calculation on the query vector and the key vector to obtain the dot product attention , the formula is as follows: ， ， in, 、 and Represent the query vector , key vector Sum value vector The trainable weight matrix, 、 and Represent the query vector , key vector Sum value vector The trainable bias of Represents the key vector The transposed vector of ; fixed window mask matrix The formula is as follows: ， in, Indicates the calculation scope of attention, specifically the meaning of vocabulary Only distance vocabulary The maximum range is Attention calculation, Indicates the vocabulary index currently being calculated for attention, Indicates the index of the word being focused on, Represents the maximum context window radius; context radius vector of the adaptive window attention text capture module The calculation process is: ， in, express Activation function used to limit the context radius vector The value of , represents the first weight parameter, represents the first offset, represents the second weight parameter, Represents the second bias. The four parameters are obtained through back propagation and optimization during model training. The range of the fixed window mask matrix and dot-product attention use Activation function calculates local attention score , the public notice is as follows: ， in, Expressive vocabulary distance vocabulary The maximum range is Attention calculation, Indicates the sub-feature corresponding to the vocabulary index currently being calculated for attention Sub-features corresponding to the index of the word being focused The local attention score between ; Represents sub-features The dot product attention calculation result of Represents the activation function; the sub-feature The local attention score and its value vector Multiply to get the adaptive window attention text feature , and finally obtain the feature set ; S43. Set the features Input to the hierarchical feature fusion output module, perform weighted fusion on the features under different window sizes, and output the final data classification results , the formula is as follows: ， in, Represents sub-features The weight parameter, represents the weight matrix, Indicates the offset.

5. A water conservancy budget data summary and analysis method according to claim 4, characterized in that: The threshold value in step S22 Manually adjust the threshold according to its impact on model performance. The initial range is .

6. A water conservancy budget data management system, which executes the water conservancy budget data summary and analysis method according to claim 1, characterized in that: include: Budget data collection module: used to obtain water conservancy budget data. The water conservancy budget data is aggregated by the budget data collection module to obtain a water conservancy budget text data sample set; Budget data preprocessing module: used to input the water conservancy budget text data sample set into the budget data preprocessing module for disassembly operation, benchmark text difference fitting operation and simplest semantic representation operation, so as to obtain a simplified water conservancy budget text data sample set; Budget data privacy protection module: used to input the simplified water conservancy budget text data sample set into the budget data privacy protection module for hidden data operation, feature optimization and scale reconciliation operation, and obtain a standardized water conservancy budget text data set; Budget data prediction module: used to input the standardized water conservancy budget text data set into the budget text data intelligent classification module for processing to obtain the final decision output; the budget text data intelligent classification module includes a text data feature encoder, an adaptive window attention text capture module and a hierarchical feature fusion output module.

Images