Big data budget index accounting control method

Abstract

The invention discloses an accounting control method based on a big data budget index, and relates to the field of budget control, and the method comprises the steps: analyzing the information contribution degree of multi-source data in a standard data set through employing an entropy method, calculating the weight, carrying out the weighted integration, generating a composite data set, training a random forest model through employing the composite data set, and carrying out the calculation of a random forest model. Predicting a budget index, constructing a correlation graph through statistical analysis and a graph neural network, generating a budget index predicted value and the correlation graph, tracing the influence of the business motivation on the financial deviation by using the graph neural network according to the budget index predicted value and the correlation graph, and performing calculation through a long and short-term memory network. Generating a root cause analysis report and a budget interval; according to the method, through combined use of multiple algorithms, the accuracy and adaptability of the budget interval are improved, and a more reliable decision basis is provided for enterprise budget accounting control.

Description

Technical Field

[0001] This invention relates to the field of budget control, and in particular to an accounting control method based on big data budget indicators. Background Technology

[0002] In recent years, with the rapid development of big data technology, budget indicator accounting and control methods based on multi-source data analysis have been widely used in corporate financial management. Traditional budget accounting mainly relies on structured data from financial systems, such as transaction details and sales records, to predict budget indicators such as revenue, costs, and cash flow through statistical models or rule-driven methods. In recent years, technological advancements have driven the introduction of unstructured data, such as customer review text data from social media platforms. Natural language processing techniques are used to extract customer sentiment scores, enhancing the external perspective of budget forecasting. Simultaneously, the application of machine learning algorithms and graph neural networks enables enterprises to construct complex correlation graphs based on multi-source data, analyzing the dynamic relationship between business drivers and financial indicators.

[0003] Although traditional budget accounting methods have made great progress, there are still areas for improvement. First, existing budget forecasting models are mostly based on a single algorithm and fail to effectively combine time series analysis to capture the dynamic fluctuation patterns of budget indicators, thus limiting the accuracy of budget intervals. In addition, when processing multi-source data, there is often a lack of systematic evaluation of the contribution of data information, resulting in low efficiency in integrating structured and unstructured data and difficulty in accurately reflecting the influence weight of each data source on budget indicators. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an accounting control method based on big data budget indicators to solve the problem that existing budget forecasting models are mostly based on a single algorithm and fail to effectively combine time series analysis to capture the dynamic fluctuation patterns of budget indicators, thus limiting the accuracy of budget intervals.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides an accounting control method based on big data budget indicators, which includes,

[0008] Collect and preprocess multi-source data to generate a standard dataset;

[0009] The entropy method is used to analyze the information contribution of multi-source data in the standard dataset, calculate the weights and perform weighted integration to generate a composite dataset.

[0010] A random forest model is trained using a composite dataset to predict budget indicators. A correlation graph is constructed using statistical analysis and graph neural networks to generate predicted budget indicator values ​​and correlation graphs.

[0011] Based on the budget target forecasts and correlation graphs, a graph neural network is used to trace the impact of business drivers on financial deviations, and a long short-term memory network is used to calculate and generate budget ranges and root cause analysis reports.

[0012] Based on the root cause analysis report and budget range, the system automatically generates accounting entries and records audit logs by comparing real-time data with the budget range, generating monitoring alerts, accounting entries, and audit logs to complete the accounting control of budget indicators.

[0013] As a preferred embodiment of the accounting control method based on big data budget indicators described in this invention, the generation of the standard dataset specifically involves:

[0014] Collect internal and external data from the enterprise, and categorize internal and external data into structured and unstructured data;

[0015] The system performs field alignment, missing value detection, and outlier identification on structured data. After standardizing unstructured data, it performs sentiment analysis, extracts customer sentiment scores, and generates a standard dataset.

[0016] As a preferred embodiment of the accounting control method based on big data budget indicators described in this invention, wherein:

[0017] The entropy method is used to analyze the information contribution of multi-source data in a standard dataset. Weights are calculated and weighted integration is performed to generate a composite dataset.

[0018] Statistical analysis of multi-source data in a standard dataset was performed using Python-SciPy to generate information uncertainty values. Based on these information uncertainty values, the information contribution of the multi-source data was determined.

[0019] A weight allocation rule is established based on the information contribution. Multi-source data with higher information contribution are assigned larger weight values ​​and integrated into a weight vector.

[0020] By fusing multi-source data from the standard dataset with the weight values ​​in the weight vector, a weighted data value is generated, forming a composite dataset.

[0021] As a preferred embodiment of the big data-based budget indicator-based accounting control method of the present invention, wherein: the step of using a composite dataset to train a random forest model to predict budget indicators specifically involves...

[0022] Load the weighted data values ​​from the composite dataset as input features, and load the historical budget execution results as training labels;

[0023] Initialize the random forest model, set the number of decision trees and the maximum depth, and build multiple decision trees based on the composite dataset through multiple iterations. After learning the mapping relationship between weighted data values ​​and historical budget execution results, the overall mapping relationship is formed by aggregation, and the trained random forest model is obtained. Input the weighted data values ​​of the current period to generate the budget indicator prediction values ​​for the next period.

[0024] As a preferred embodiment of the big data-based budget indicator accounting control method of the present invention, the step of constructing a correlation graph through statistical analysis and graph neural networks to generate predicted values ​​of budget indicators and a correlation graph specifically involves:

[0025] Load the weighted data values ​​from the composite dataset, generate statistical correlation values ​​between the weighted data values ​​based on the Pearson correlation coefficient, and filter out the business data values ​​whose correlation with the predicted values ​​of the budget indicators is higher than the correlation threshold based on the statistical correlation values.

[0026] The selected business data values ​​and budget indicator prediction values ​​are defined as graph nodes. PyTorch-Geometric is used to load the graph nodes and statistical correlation values. The dynamic causal relationship between nodes is analyzed through graph neural network, graph edges are established, and a correlation graph between business drivers and financial indicators is generated.

[0027] As a preferred embodiment of the big data-based budget indicator-based accounting control method of the present invention, the step of using a graph neural network to trace the impact of business drivers on financial deviations based on the predicted values ​​of budget indicators and correlation graphs specifically involves...

[0028] The budget target forecasts are compared with the actual execution data of the enterprise to calculate the budget target deviation value of each budget target;

[0029] Load the correlation graph between business drivers and financial indicators and the deviation values ​​of budget indicators. Use a graph neural network to analyze the connection relationships between graph nodes and the deviation values ​​of budget indicators, generate the impact weight of business drivers on financial deviations, and obtain a root cause analysis report.

[0030] As a preferred embodiment of the accounting control method based on big data budget indicators described in this invention, the step of generating the budget interval through a long short-term memory network specifically involves:

[0031] Load the predicted values ​​of budget indicators, the deviation values ​​of budget indicators, and the industry price index data values ​​in the standard dataset, and organize them into input features according to the time series.

[0032] Initialize a long short-term memory network, and learn the fluctuation patterns of input features through forget gate, input gate, and output gate to predict the fluctuation range of revenue, cost, and cash flow in the next cycle;

[0033] Extract the upper and lower limits of each budget indicator from the fluctuation range, integrate them into JSON format, and form a budget range.

[0034] As a preferred embodiment of the big data-based budget indicator-based accounting control method of the present invention, the method of automatically generating accounting entries and recording audit logs based on root cause analysis reports and budget intervals by comparing budget intervals with real-time data specifically involves:

[0035] Using the root cause analysis report and the upper and lower limits of the budget indicators for the budget range, import them into the budget monitoring center, obtain the actual execution data of the budget indicators from the enterprise resource planning center, and update them to the budget monitoring center at a preset frequency.

[0036] Compare the actual execution data of budget targets with the upper and lower limits of budget targets to generate budget deviation monitoring alerts;

[0037] Budget deviation monitoring alerts, root cause analysis reports, budget ranges, and operator information are integrated into an audit log and stored using blockchain technology.

[0038] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the accounting control method based on big data budget indicators as described in the first aspect of the present invention.

[0039] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the accounting control method based on big data budget indicators as described in the first aspect of the present invention.

[0040] The beneficial effects of this invention are as follows: By analyzing the information contribution of multi-source data using the entropy method based on a standard dataset and generating a composite dataset, and by combining long short-term memory networks with predicted budget indicator values, budget indicator deviation values, and industry price index data from market platforms to predict the reasonable fluctuation range of budget intervals, this invention achieves systematic weight assessment of multi-source data and accurate capture of the dynamic fluctuation patterns of budget indicators. This overcomes the shortcomings of low integration efficiency caused by the lack of data contribution assessment, ensures the effective integration of structured and unstructured data, accurately reflects the influence weight of each data source on budget indicators, and transcends the limitations of single-algorithm prediction. Through time-series analysis using long short-term memory networks, it captures the fluctuation patterns of revenue, costs, and cash flow, improving the accuracy and adaptability of budget intervals and providing a more reliable decision-making basis for enterprise budget accounting and control. Attached Figure Description

[0041] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a flowchart of an accounting control method based on big data budget indicators.

[0043] Figure 2 Flowchart for generating standard datasets for data preprocessing.

[0044] Figure 3 Build a flowchart for the association graph.

[0045] Figure 4 This is a flowchart for budget monitoring and accounting control. Detailed Implementation

[0046] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0047] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0048] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0049] Reference Figures 1-4 This is one embodiment of the present invention, which provides an accounting control method based on big data budget indicators, including the following steps:

[0050] S1. Collect multi-source data and preprocess it to generate a standard dataset;

[0051] The specific steps are as follows:

[0052] S1.1: Collect internal and external data of the enterprise, and classify internal and external data into structured data and unstructured data;

[0053] Internal enterprise data includes transaction details from the financial system, sales records and logistics information from the operations system, and equipment operation logs from the production system; external data includes industry price index data from market platforms, macroeconomic indicator data, and customer comment text data from social media platforms.

[0054] Transaction details, sales records, logistics information, equipment operation logs, industry price indexes, and macroeconomic indicators are categorized as structured data, while customer review text data is categorized as unstructured data.

[0055] S1.2: Perform field alignment, missing value detection, and outlier identification on structured data. Perform sentiment analysis on unstructured data after standardization and extract customer sentiment scores. Specifically: (1) Extract structured data and perform field alignment using Python-Pandas by mapping fields from different sources (e.g., "income" and "sales amount") to a unified naming convention to ensure consistent field names. (2) Use Python-Pandas to examine each field of the structured data, identify missing values, and fill them with the average of the corresponding fields. For example, fill the missing income field with the average income in the financial data. (3) Use Python-Pandas to analyze the structured data, identify outliers, and maintain data integrity by removing records that deviate significantly from the normal range (e.g., cost values ​​exceeding three times the standard deviation). For unstructured data, perform standardization using Apache-NiFi to convert the text to a unified encoding format and remove irrelevant characters. Then, use spaCy to perform sentiment analysis on the standardized unstructured data, analyze the sentiment of the text through a pre-trained language model, extract customer sentiment scores, and generate scores between 0 and 1, where 1 represents highly positive sentiment.

[0056] The pre-trained language model is a natural language processing model based on neural networks, specifically designed for tasks such as sentiment analysis. The training process is as follows: (1) Collect large-scale public text datasets, such as online comments, social media posts, and product reviews, ensuring that they contain diverse sentiment expressions, such as positive, negative, and neutral comments. (2) Preprocess these text data, unify the encoding format, remove irrelevant characters, and label each text with a sentiment tag, such as "positive" corresponding to 1, "negative" corresponding to 0, and "neutral" corresponding to 0.5. (3) Use a neural network architecture for initial training, input the pre-processed text data and sentiment tags, and repeatedly adjust the parameters of the neural network architecture to enable the neural network architecture to learn the correspondence between text features (such as vocabulary and syntax) and sentiment tags. During the training process, the neural network architecture gradually improves the accuracy of predicting sentiment through multiple rounds of iterative optimization, for example, it can identify "good service" as representing positive sentiment and "slow logistics" as representing negative sentiment. After training is completed, the neural network architecture is saved as a pre-trained language model and integrated into the spaCy tool. spaCy loads a pre-trained language model and inputs standardized unstructured data into it one by one. The pre-trained language model processes the text content of each comment through word segmentation, word vector embedding, and semantic analysis. For example, it identifies "good service" as positive sentiment and "slow logistics" as negative sentiment, generating a customer sentiment score between 0 and 1. For example, "good service" scores 0.9.

[0057] Structured and unstructured data are uniformly adjusted to the range of 0 to 1 and then integrated into a JSON file with a unified format to form a standard dataset.

[0058] S2. Use the entropy method to analyze the information contribution of multi-source data in the standard dataset, calculate the weights and perform weighted integration to generate a composite dataset.

[0059] The specific steps are as follows:

[0060] S2.1: Perform statistical analysis on the numerical distribution of each data category in the standard dataset, and calculate the information uncertainty of each data category. Specifically, use Python-SciPy to perform statistical analysis on the numerical distribution of each feature. By evaluating the distribution characteristics of feature values ​​in the standard dataset, generate indicators reflecting the uniformity of distribution, such as whether the numerical distribution of the income field is concentrated or dispersed. Python-SciPy generates an information uncertainty value for each feature based on the distribution characteristics, reflecting the degree of uncertainty in the information expression of financial data, operational data, production data, and external data. The expression is:

[0061]

[0062] Among them, H jLet represent the information uncertainty value, or entropy value, of the j-th data type in the standard dataset; n represents the total number of data values ​​of the j-th data type in the standard dataset; i represents the index of the i-th data value of the j-th data type in the standard dataset; and p represents the index of the i-th data value of the j-th data type in the standard dataset. ij represents the normalized probability of the i-th data value of the j-th data type in the standard dataset, and log represents the natural logarithm operation.

[0063] S2.2: Determine the information contribution of each data type based on the information uncertainty value. The lower the information uncertainty, the higher the information contribution. Python-SciPy analyzes the information uncertainty value of each feature and evaluates the information expression ability in the standard dataset by comparing the size of the uncertainty value. For example, the income field with lower uncertainty indicates that the information expression is more stable.

[0064] Python-SciPy sorts various data types according to their information expressive power based on the relative magnitude of their information uncertainty values, and assigns information contribution levels accordingly. Data with lower information uncertainty values ​​receive higher information contribution levels; for example, the income field, with the lowest uncertainty, receives the highest information contribution level.

[0065] S2.3: Establish weight allocation rules based on the information contribution of each type of data in the standard dataset: the higher the information contribution, the greater the weight allocated, and finally integrate them to form a weight vector.

[0066] All data in the standard dataset are fused with the corresponding weight values ​​in the weight vector to obtain the weighted data result, i.e., the weighted data value. The weighted data result is then integrated according to a unified field structure to form a composite dataset containing the weighted data value.

[0067] S3. Use a composite dataset to train a random forest model to predict budget indicators and construct a correlation graph through statistical analysis and graph neural networks to generate predicted budget indicator values ​​and correlation graphs.

[0068] The specific steps are as follows:

[0069] S3.1: Use a composite dataset as feature input to train the random forest model. During the training process, use historical budget execution results as training labels to enable the random forest model to learn the mapping relationship between each data point in the composite dataset and the budget indicators.

[0070] Specifically, the process involves loading a composite dataset as feature input and collecting historical budget execution results (including actual revenue, cost, and cash flow values ​​for the past 12 months) as training labels. Using Python-scikit-learn, these data values ​​and labels are loaded, and the random forest model is initialized, setting the number of decision trees (e.g., 100) and maximum depth (e.g., 10 layers). Through multiple iterations, the random forest model randomly samples subsets of data values ​​and features from the composite dataset to construct multiple decision trees. Each tree learns a local mapping relationship between weighted data values ​​and revenue, cost, and cash flow. Specifically, a subset of data values ​​(e.g., 70% of the data) is randomly sampled from the composite dataset using sampling with replacement, generating different training subsets for each decision tree. At each node of each decision tree, a subset is randomly sampled from all features of the weighted data values ​​(e.g., the square root of the total number of features). For example, a tree might only consider weighted sales order volume and weighted industry price index, selecting the most... The optimal split point is used to recursively split and generate child nodes until the termination condition of a maximum depth of 10 layers or fewer than 2 samples is met. This allows each decision tree to learn the local mapping relationship between weighted data values ​​and actual values ​​of revenue, cost, and cash flow. For example, a tree may learn the pattern that revenue increases when the weighted sales order volume is high. Finally, the random forest model aggregates the prediction results of all decision trees to form an overall mapping relationship and saves it as a trained random forest model. The termination condition of a maximum depth of 10 layers or fewer than 2 samples is based on the hyperparameters set by the random forest model when constructing decision trees to balance the complexity and prediction performance of the random forest model.

[0071] After training is complete, the weighted data values ​​of the current period are input into the trained random forest model. The trained random forest model generates budget indicator predictions for the next period, including revenue predictions, cost predictions, and cash flow predictions.

[0072] S3.2 performs correlation analysis on the weighted data values ​​in the composite dataset, calculating the statistical correlation between each weighted data value; based on the statistical correlation results, business data values ​​with a correlation higher than the correlation threshold with the budget indicator prediction values ​​are selected, and the selected business data values ​​and budget indicator prediction values ​​are defined as graph nodes. Graph edges are established based on the statistical correlation between nodes and the graph neural network training results, forming a correlation graph between business drivers and financial indicators, specifically:

[0073] S3.2.1: Load the weighted data values ​​from the composite dataset using Python-NetworkX, perform correlation analysis on these weighted data values ​​using the Pearson correlation coefficient, and generate statistical correlation values ​​by evaluating the strength of the statistical association between each pair of weighted data values. For example, to analyze the correlation strength between weighted sales order volume and weighted cost, the expression is:

[0074]

[0075] Where, ρ ab The statistical correlation value between the a-th weighted data value and the b-th weighted data value in the composite dataset ranges from [-1, 1], where t represents the total number of records for either the a-th or b-th weighted data value in the composite dataset, and V ma This represents the m-th record of the a-th weighted data value in a composite dataset, for example, the m-th value of the weighted sales order quantity, V. mb This represents the m-th record of the b-th weighted data value in the composite dataset. This represents the average of the a-th weighted data value in the composite dataset. This represents the average of the b-th weighted data value in the composite dataset, and m represents the index of the m-th record of the a-th weighted data value in the composite dataset.

[0076] S3.2.2: Filter out business data values ​​that are more correlated with the predicted values ​​of budget indicators than the correlation threshold (e.g., 0.7) based on statistical correlation values. The correlation threshold is set based on historical correlation analysis.

[0077] S3.2.3: Define the selected business data values ​​and budget indicator forecasts as graph nodes. Use PyTorch-Geometric to load the defined graph nodes and statistical correlation values. Analyze the dynamic causal relationships between graph nodes using the graph neural network training results, and establish graph edges, such as the edge connecting weighted sales order volume and revenue forecast. Finally, integrate the graph nodes and edges to generate a correlation graph between business drivers and financial indicators.

[0078] The training process of a graph neural network is as follows:

[0079] (1) Collect historical composite datasets (including weighted revenue, cost, sales order volume, transportation time, equipment uptime, price index, consumer price index and customer sentiment score for the past 12 months) and corresponding historical budget execution results, including actual values ​​of revenue, cost and cash flow.

[0080] (2) Use Python-NetworkX to generate an initial graph structure, defining historical weighted data values ​​and budget execution results as nodes, and establishing edges based on statistical correlation values ​​(e.g., Pearson correlation coefficient). Use PyTorch-Geometric to load the initial graph structure, initialize the graph neural network, and set graph convolutional layers (e.g., 2 layers) and hidden units (e.g., 64). Through multiple iterations, the graph neural network updates node representations through graph convolutional layers, learning dynamic causal relationships between nodes, such as the potential impact of weighted sales order volume on revenue. After training, save the graph neural network model for analyzing nodes and statistical correlation values ​​in the current period, generating a correlation graph between business drivers and financial indicators.

[0081] S4. Based on the predicted values ​​of budget indicators and the correlation graph, use graph neural networks to trace the impact of business drivers on financial deviations, and use long short-term memory networks to calculate and generate budget ranges and root cause analysis reports.

[0082] The specific steps are as follows:

[0083] S4.1: Compare the predicted values ​​of budget indicators with the actual execution data to calculate the budget indicator deviation values ​​for each budget indicator. Specifically, based on the generated predicted values ​​of budget indicators and the actual execution data obtained from the Enterprise Resource Planning Center, the budget indicator deviation values ​​are generated by comparing the differences between the predicted and actual revenue values, the predicted and actual cost values, and the predicted and actual cash flow values. For example, if the predicted revenue value is higher than the actual value, a positive deviation value is generated.

[0084] S4.2: The correlation graph between business drivers and financial indicators is used as input to the graph neural network. Based on the connections between nodes in the correlation graph and the budget indicator deviation values, the graph neural network analyzes the strength of the impact of each business driver on the budget deviation and outputs the weights of the impact of business drivers on the financial deviation. Specifically,

[0085] The system loads a graph of the relationship between business drivers and financial indicators, along with budget indicator deviations. A graph neural network analyzes the connections between nodes (e.g., the edge between weighted sales order volume and revenue forecast) and budget indicator deviations. The graph neural network updates the representation of each node through graph convolutional layers, incorporating budget indicator deviations into the node update process. It assesses the strength of the impact of business drivers (e.g., weighted sales order volume, weighted shipping time) on revenue deviations, cost deviations, and cash flow deviations. Based on the node update results and connections, it generates the weights of the business drivers' impact on financial deviations. For example, assigning a weight of 0.6 to weighted sales order volume indicates a strong impact on revenue deviations, while assigning a weight of 0.3 to weighted shipping time indicates a weaker impact on cost deviations.

[0086] S4.3: Based on the weighting of the impact of business drivers on financial deviations, and in conjunction with historical budget adjustment records, identify the business drivers that have a significant impact on revenue deviations, cost deviations, and cash flow deviations. For example, weighted sales order volume is identified as a major influencing factor due to its high weight.

[0087] This process extracts adjustment patterns related to high-weight business drivers from historical budget adjustment records. It then integrates these high-weight business drivers with the extracted budget adjustment patterns. Specifically, it loads the impact weights of business drivers on financial deviations and historical budget adjustment records. Python-Pandas analyzes the impact weights of business drivers on financial deviations, identifying high-weight business drivers, such as weighted sales order volume, which is identified as a major influencing factor due to its weight of 0.6. Python-Pandas filters records related to high-weight business drivers from historical budget adjustment records, such as extracting records where an increase in weighted sales order volume leads to an upward adjustment of the revenue budget, forming adjustment patterns, such as "a 10% increase in sales order volume leads to a 5% increase in the revenue budget." Python-Pandas then associates high-weight business drivers (e.g., weighted sales order volume) with their corresponding adjustment patterns (e.g., an upward adjustment of the revenue budget) according to a unified field structure, completing the integration and generating a root cause analysis report. The root cause analysis report includes the correspondence between business drivers, deviation types, and adjustment patterns.

[0088] S4.4: The predicted values ​​of budget indicators, budget deviation values, and industry price index data from the standard dataset are used as input features to the Long Short-Term Memory (LSTM) network. The LTM network learns the historical fluctuation patterns of budget indicators based on these input features and predicts the reasonable fluctuation range of budget indicators. Specifically:

[0089] The predicted values ​​of budget indicators, the deviation values ​​of budget indicators, and the industry price index data are organized into time series to form time series data, which serves as the input features of the Long Short-Term Memory Network. The Long Short-Term Memory Network analyzes the time series relationship between the predicted values ​​of budget indicators, the deviation values ​​of budget indicators, and the industry price index data layer by layer through forget gates, input gates, and output gates, and learns the fluctuation patterns of revenue, cost, and cash flow, such as the pattern of revenue prediction rising with the industry price index.

[0090] Based on the learned fluctuation patterns, the Long Short-Term Memory network predicts the reasonable fluctuation range of revenue, costs, and cash flow in the next cycle. For example, the fluctuation range of revenue is ±5%.

[0091] Extract the upper and lower limits of each budget indicator from the predicted fluctuation range. For example, the upper limit for revenue is the predicted value plus 5%, and the lower limit is the predicted value minus 5%. Finally, integrate the upper and lower limits of revenue, cost, and cash flow into JSON format to form the budget range.

[0092] S4.4.1: To predict the reasonable fluctuation range of budget indicators, the training process of the Long Short-Term Memory Network is as follows: Collect historical data from the past 24 months, including the predicted value of the budget indicator, the deviation value of the budget indicator, and the industry price index data value of the market platform in the standard dataset. Organize them into input features according to the time series, and use the actual fluctuation range of the historical budget indicator (e.g., the upper and lower limits of revenue, based on the upper and lower limits of actual execution data) as training labels.

[0093] Initialize a Long Short-Term Memory (LSTM) network with a two-layer structure of 128 units per layer. Use the Adam optimizer to adjust the learning rate (e.g., 0.001), and measure the difference between the predicted and actual fluctuation ranges using the mean squared error loss function. During training, the LSM network retains key historical information (e.g., long-term trends in industry price index data) through a forget gate, incorporates new data (e.g., recent deviations) through an input gate, and generates a prediction of the reasonable fluctuation range for budget indicators in the next period through an output gate. It learns the temporal mapping relationship between predicted budget indicator values, budget indicator deviations, and the industry price index data of the market platform and the fluctuation range; for example, it learns the pattern that an increase in industry price index data leads to an expansion of income fluctuation range.

[0094] Through 100 iterations of optimization, the Long Short-Term Memory (LSTM) network gradually improves its prediction accuracy. After training, the LTM network model is saved and used to process the current period's budget indicator predictions, budget indicator deviations, and industry price index data from the market platform to predict the budget range for the next period.

[0095] S5. Based on the root cause analysis report and budget range, the system automatically generates accounting entries and records audit logs by comparing real-time data with the budget range, generating monitoring alerts, accounting entries, and audit logs to complete the accounting control of budget indicators.

[0096] The specific steps are as follows:

[0097] To achieve accounting control of budget targets, generate monitoring alerts, accounting entries, and audit logs, perform the following operations:

[0098] Based on the root cause analysis report and budget range generated by S4, the upper and lower limits of the budget indicators in the root cause analysis report and budget range are loaded using Python-Pandas and imported into the budget monitoring center for real-time comparison. Actual execution data of the budget indicators (including actual revenue, actual cost, and actual cash flow for the current period, e.g., actual revenue of 9.5 million yuan) is obtained from the Enterprise Resource Planning Center and updated to the budget monitoring center via Apache-Kafka at a preset frequency (e.g., daily) to maintain data real-time performance.

[0099] Using Python-Pandas in the budget monitoring center, actual execution data for budget targets are compared with their upper and lower limits. For example, comparing actual revenue of 9.5 million yuan with the upper limit of 10 million yuan and the lower limit of 9 million yuan determines whether it exceeds the budget range. If the actual execution data exceeds the upper limit or falls below the lower limit, Python-Pandas generates a budget deviation monitoring alert. The alert includes deviation details, such as a revenue deviation of -200,000 yuan. This budget deviation monitoring alert is sent to financial management personnel via internal enterprise communication, including the deviation type and value.

[0100] Based on the weighting of the impact of business drivers on financial deviations in the root cause analysis report (e.g., a weighted average sales order volume weight of 0.6), identify the main responsible departments for budget deviations, such as the sales department, where a deviation in weighted average sales order volume leads to revenue deviations. Based on the correspondence between business drivers and financial deviations in the root cause analysis report (e.g., a decrease in weighted average sales order volume leading to revenue deviations), determine accounting adjustments, such as adjusting revenue account balances.

[0101] Automated workflow tools are used to generate journal entries based on accounting adjustments, such as debiting revenue adjustment accounts and crediting main revenue accounts. The generated journal entries are then synchronized to the Enterprise Resource Planning (ERP) center via API to update accounting records and complete the accounting process.

[0102] The audit log, compiled using Python-Pandas, integrates budget deviation monitoring alerts, root cause analysis reports, budget ranges, and the generation times and personnel information of accounting entries into a JSON format. This audit log is then stored on a blockchain using Hyperledger Fabric to ensure immutability and is used for subsequent financial audits and accountability.

[0103] This embodiment also provides a computer device applicable to the accounting control method based on big data budget indicators, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the accounting control method based on big data budget indicators proposed in the above embodiment.

[0104] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0105] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the accounting control method based on big data budget indicators as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0106] In summary, this invention achieves systematic weight assessment of multi-source data and accurate capture of dynamic fluctuation patterns in budget indicators by: analyzing the information contribution of multi-source data using the entropy method based on a standard dataset to generate a composite dataset; and predicting the reasonable fluctuation range of budget intervals by combining long short-term memory networks with budget indicator predictions, budget indicator deviations, and industry price index data from market platforms. This overcomes the shortcomings of low integration efficiency caused by the lack of data contribution assessment, ensures the effective fusion of structured and unstructured data, accurately reflects the influence weight of each data source on budget indicators, and transcends the limitations of single-algorithm prediction. By capturing the fluctuation patterns of revenue, costs, and cash flow through time-series analysis using long short-term memory networks, it improves the accuracy and adaptability of budget intervals, providing a more reliable decision-making basis for enterprise budget accounting and control.

[0107] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for accounting and control based on big data budget indicators, characterized in that: The method comprises the following steps: Collecting multi-source data and preprocessing to generate a standard data set; Using entropy method to analyze the information contribution of multi-source data in the standard data set, calculating the weight and integrating by weight to generate a composite data set; Using the composite data set to train the random forest model, predicting the budget index and constructing the correlation graph through statistical analysis and graph neural network to generate the budget index prediction value and the correlation graph; According to the budget index prediction value and the correlation graph, using the graph neural network to trace the influence of business causes on financial deviation, and using the long short-term memory network to calculate the budget interval and the root cause analysis report; Based on the root cause analysis report and the budget interval, comparing the budget interval with the real-time data to automatically generate the accounting entries and record the audit log, generating the monitoring alarm, the accounting entries and the audit log, and completing the accounting control of the budget index. 2.The big data budget index accounting control method according to claim 1, wherein: The standard data set is generated in detail as follows: Collecting enterprise internal data and external data, and dividing the enterprise internal data and external data into structured data and unstructured data; Aligning the fields of the structured data, detecting the missing values and identifying the abnormal values, and performing sentiment analysis on the standardized unstructured data to extract customer emotion scores to generate the standard data set. 3.The big data budget index accounting control method according to claim 2, wherein: The entropy method is used to analyze the information contribution of multi-source data in the standard data set, calculate the weight and integrate by weight to generate a composite data set, which is specifically as follows: Using Python-SciPy to perform statistical analysis on the multi-source data in the standard data set to generate information uncertainty values, and determining the information contribution of the multi-source data according to the information uncertainty values, According to the information contribution, a weight distribution rule is established, and the multi-source data with higher information contribution is allocated with larger weight value, and is integrated into a weight vector; The multi-source data in the standard data set is fused with the weight values in the weight vector to generate weighted data values to form a composite data set. 4.The big data budget index accounting control method according to claim 3, wherein: The composite data set is used to train the random forest model to predict the budget index, which is specifically as follows: Load the weighted data values in the composite data set as input features, and load the historical budget execution results as training labels; Initialize the random forest model, set the number of decision trees and the maximum depth, and through multiple iterations, build multiple decision trees based on the composite data set, learn the mapping relationship between the weighted data values and the historical budget execution results, and then form the overall mapping relationship through aggregation to obtain the trained random forest model, input the weighted data values of the current period to generate the budget index prediction value of the next period.

5. The big data budgetary index accounting control method of claim 4, wherein: The correlation graph is constructed by statistical analysis and graph neural network to generate the budget index prediction value and the correlation graph, which is specifically as follows: Load the weighted data values in the composite data set, generate the statistical correlation values between the weighted data values based on the Pearson correlation coefficient, and select the business data values with higher correlation than the correlation threshold value with the budget index prediction value according to the statistical correlation values; The selected business data values and the budget index prediction value are defined as graph nodes, the graph nodes and the statistical correlation values are loaded using PyTorch-Geometric, the dynamic causal relationship between the nodes is analyzed through the graph neural network, the graph edges are established, and the correlation graph of the business causes and the financial indicators is generated. 6.The big data budget index accounting control method based on big data according to claim 5, wherein: The influence of the business cause on the financial deviation is traced by using the graph neural network according to the budget index prediction value and the correlation graph, specifically, The budget index deviation value of each budget index is calculated by comparing the budget index prediction value with the actual execution data of the enterprise; The influence weight of the business cause on the financial deviation is generated by the graph neural network analyzing the connection relationship between the graph nodes and the budget index deviation value, and the root cause analysis report is obtained.

7. The big data budgetary index accounting control method of claim 6, wherein: The budget interval is generated by the long short-term memory network, specifically, The budget index prediction value, the budget index deviation value and the industry price index data value in the standard data set are loaded and arranged as input features in time series; The long short-term memory network is initialized, and the fluctuation law of the input features is learned through the forget gate, the input gate and the output gate to predict the fluctuation range of the next period income, cost and cash flow; The upper limit value and the lower limit value of each budget index are extracted from the fluctuation range and integrated into JSON format to form the budget interval. 8.The big data budget index accounting control method based on claim 7, characterized in that: Based on the root cause analysis report and the budget interval, the accounting entries are automatically generated and the audit log is recorded by comparing the budget interval with the real-time data, specifically, The upper limit value and the lower limit value of the budget index of the root cause analysis report and the budget interval are imported into the budget monitoring center, the actual execution data of the budget index is obtained from the enterprise resource planning center and updated to the budget monitoring center at a preset frequency; The budget deviation monitoring alarm is generated by comparing the actual execution data of the budget index with the upper limit value and the lower limit value of the budget index; The budget deviation monitoring alarm, the root cause analysis report, the budget interval and the operator information are integrated into the audit log, which is stored by the blockchain method. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The processor executes the computer program to realize the steps of the budget index accounting control method based on big data according to any one of claims 1-8.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the steps of the budget index accounting control method based on big data according to any one of claims 1-8.

Images