University financial data management method and system

Abstract

The present application is applicable to the technical field of financial data management, and in particular to a method and system for financial data management in colleges and universities, the method comprising: obtaining a current financial data sequence, and determining a current financial identification sequence based on the current financial data sequence; predicting each financial identification of the next cycle of the current financial identification sequence based on the current financial identification sequence to obtain a predicted financial identification sequence; obtaining an actual financial data sequence, and determining an actual financial identification sequence based on the actual financial data sequence; comparing the actual financial identification sequence with the predicted financial identification sequence, and constructing an actual capital flow diagram based on the actual financial identification sequence, judging whether there is abnormal financial data in the actual financial data sequence, and obtaining a judgment result. The method can quickly discover abnormal financial data, while reducing manual intervention, and improving the intelligence level of financial management and data processing efficiency.

Description

Technical Field

[0001] The present application belongs to the technical field of financial data management, and in particular to a method and system for managing financial data of universities. Background Art

[0002] University financial data management involves the systematic recording, analysis, and oversight of a university's revenues, expenditures, and assets, facilitating efficient and transparent fund management. This management encompasses various revenue sources, including tuition, government grants, research funding, and social donations, as well as expenditure items such as teaching, research, infrastructure, and student services. Financial statements, such as balance sheets and income and expenditure statements, comprehensively reflect the university's financial status, supporting budgeting, cost control, and performance evaluation, and providing crucial decision-making support for sustainable development.

[0003] Existing technologies involve university financial data across multiple dimensions (such as revenue, expenditure, and budget execution), resulting in diversity and complexity. Due to the sheer volume of financial data, manual reconciliation and exception review processes are inefficient and prone to errors, making it difficult to comprehensively and promptly identify anomalies. For example, issues like irregular expenditures, abnormal fund flows, and duplicate reimbursements can lead to financial losses or even financial risks if not promptly identified.

[0004] To sum up, when managing the financial data of universities, there is a problem that it is difficult to detect abnormal data in a timely manner due to the huge amount of financial data and the low efficiency of manual abnormality review. Summary of the Invention

[0005] The embodiments of the present application provide a method and system for managing financial data of universities, which can solve the problem in related technologies of managing financial data of universities, that is, the huge amount of financial data and the low efficiency of manual abnormality review make it difficult to detect abnormal data in a timely manner.

[0006] In a first aspect, an embodiment of the present application provides a method for managing financial data of a university, including:

[0007] Obtaining a current financial data sequence, and determining a current financial identification sequence based on the current financial data sequence; wherein the current financial data sequence is obtained by arranging the financial data of colleges and universities within a certain period in chronological order, and the current financial identification sequence is used to represent the financial identification corresponding to each financial data in the current financial data sequence;

[0008] According to the current financial identification sequence, each financial identification of the next period of the current financial identification sequence is predicted to obtain a predicted financial identification sequence;

[0009] Acquire an actual financial data sequence, and determine an actual financial identification sequence based on the actual financial data sequence; wherein the actual financial data sequence is the financial data actually collected in the next period of the current financial identification sequence;

[0010] Comparing the actual financial identifier sequence with the predicted financial identifier sequence, and constructing an actual capital flow diagram based on the actual financial identifier sequence, determining whether there is abnormal financial data in the actual financial data sequence, and obtaining a determination result;

[0011] If the judgment result indicates that abnormal financial data exists in the actual financial data sequence, abnormality management is performed on the abnormal financial data.

[0012] The above technical solutions in the embodiments of the present application have at least the following technical effects:

[0013] The present application provides a university financial data management method. The method first obtains a current financial data sequence (based on the university financial data within a certain period arranged in chronological order) and determines a current financial identifier sequence based on the current financial data sequence. Then, based on the current financial identifier sequence, each financial identifier of the next period of the current financial identifier sequence is predicted to obtain a predicted financial identifier sequence. Then, an actual financial data sequence is obtained and, based on the actual financial data sequence (the financial data actually collected in the next period of the current financial identifier sequence), an actual financial identifier sequence is determined. Then, the actual financial identifier sequence is compared with the predicted financial identifier sequence, and an actual capital flow diagram is constructed based on the actual financial identifier sequence to determine whether there is abnormal financial data in the actual financial data sequence and obtain a judgment result. Finally, if the judgment result indicates that there is abnormal financial data in the actual financial data sequence, abnormal management is performed on the abnormal financial data. By comparing the actual financial identifier sequence with the predicted financial identifier sequence, the method can quickly discover abnormal financial data, improve financial risk management capabilities, and clearly display the flow, source, and use of funds through the actual capital flow diagram, thereby improving the transparency of fund management and helping university financial managers to intuitively analyze fund flows. This method can automatically extract, analyze and predict financial indicators, reduce manual intervention, reduce human errors, and improve the intelligence level of financial management and data processing efficiency.

[0014] In a second aspect, the present application provides a university financial data management system, including:

[0015] A current financial identification sequence determining unit is configured to obtain a current financial data sequence and determine a current financial identification sequence based on the current financial data sequence; wherein the current financial data sequence is obtained by chronologically arranging the financial data of colleges and universities within a certain period, and the current financial identification sequence is used to represent the financial identification corresponding to each financial data in the current financial data sequence;

[0016] A prediction unit, configured to predict each financial identifier of a next period of the current financial identifier sequence based on the current financial identifier sequence to obtain a predicted financial identifier sequence;

[0017] An actual financial identification sequence determining unit, configured to obtain an actual financial data sequence and determine an actual financial identification sequence based on the actual financial data sequence; wherein the actual financial data sequence is the financial data actually collected in the next period of the current financial identification sequence;

[0018] an abnormality judgment unit, configured to compare the actual financial identification sequence with the predicted financial identification sequence, construct an actual capital flow diagram based on the actual financial identification sequence, judge whether there is abnormal financial data in the actual financial data sequence, and obtain a judgment result;

[0019] The abnormality management execution unit is configured to execute abnormality management on the abnormal financial data if the judgment result indicates that abnormal financial data exists in the actual financial data sequence.

[0020] In a third aspect, an embodiment of the present application provides a university financial data management device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in any one of the embodiments of the first aspect when executing the computer program.

[0021] It can be understood that the beneficial effects of the second to third aspects mentioned above can be found in the relevant description of the first aspect mentioned above, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0023] Figure 1 This is a flow chart of a method for managing university financial data provided by an embodiment of the present application;

[0024] Figure 2It is a schematic diagram of the implementation process of predicting the five-level classification and amount of each financial identifier in the next cycle of the current financial identifier sequence in the university financial data management method provided by the embodiment of the present application. DETAILED DESCRIPTION

[0025] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.

[0026] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0027] It will also be understood that the term "and / or" used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0028] As used in this specification and the appended claims, the term "if" can be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]," depending on the context.

[0029] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0030] References to "one embodiment" or "some embodiments" in this specification mean that a particular feature, structure, or characteristic described in conjunction with that embodiment is included in one or more embodiments of the present application. Thus, phrases such as "in one embodiment," "in some embodiments," "in other embodiments," and "in other embodiments" appearing in various places in this specification do not necessarily refer to the same embodiment, but rather mean "one or more but not all embodiments," unless otherwise specifically emphasized. The terms "including," "comprising," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0031] In related technologies, university financial data encompasses multiple dimensions (such as revenue, expenditure, and budget execution), and is diverse and complex. Due to the sheer volume of financial data, manual reconciliation and exception review are inefficient and prone to errors, making it difficult to comprehensively and promptly detect anomalies. For example, issues such as irregular expenditures, abnormal fund flows, and duplicate reimbursements can lead to financial losses or even financial risks if not promptly detected.

[0032] To address the above-mentioned issues, embodiments of the present application provide a university financial data management method and system. This method first obtains a current financial data sequence (based on the chronological arrangement of university financial data within a certain period), and based on the current financial data sequence, determines a current financial identifier sequence. Then, based on the current financial identifier sequence, each financial identifier for the next period of the current financial identifier sequence is predicted to obtain a predicted financial identifier sequence. An actual financial data sequence is then obtained, and based on the actual financial data sequence (the actual financial data collected in the next period of the current financial identifier sequence), an actual financial identifier sequence is determined. The actual financial identifier sequence is then compared with the predicted financial identifier sequence, and an actual funds flow diagram is constructed based on the actual financial identifier sequence to determine whether there is abnormal financial data in the actual financial data sequence, obtaining a judgment result. Finally, if the judgment result indicates that there is abnormal financial data in the actual financial data sequence, abnormality management is performed on the abnormal financial data. By comparing the actual financial identifier sequence with the predicted financial identifier sequence, this method can quickly identify abnormal financial data, improving financial risk management capabilities. Furthermore, the actual funds flow diagram can clearly display the flow, source, and use of funds, improving the transparency of fund management and helping university financial managers intuitively analyze fund flows. This method can automatically extract, analyze and predict financial indicators, reduce manual intervention, reduce human errors, and improve the intelligence level of financial management and data processing efficiency.

[0033] The university financial data management method provided in the embodiment of the present application can be applied to the university financial data management device. At this time, the university financial data management device is the executor of the university financial data management method provided in the embodiment of the present application. The embodiment of the present application does not impose any restrictions on the specific type of the university financial data management device.

[0034] For example, the university financial data management device can be a mobile phone, a tablet computer, a laptop computer, an ultra-mobile personal computer (UMPC), a netbook, a personal digital assistant (PDA), a desktop computer, a handheld device with wireless communication capabilities, a computing device or other processing device connected to a wireless modem, a computer, a laptop computer, a handheld computing device, customer premise equipment (CPE) and / or other devices for communicating on wireless systems and next-generation communication systems, such as mobile terminals in 5G networks or mobile terminals in future evolved public land mobile networks (PLMN), etc.

[0035] In order to better understand the university financial data management method provided by the embodiment of the present application, the specific implementation process of the university financial data management method provided by the embodiment of the present application is exemplarily introduced below.

[0036] Figure 1 A schematic flow chart of a university financial data management method provided by an embodiment of the present application is shown. The university financial data management method includes:

[0037] S100: Obtain a current financial data sequence and determine a current financial identification sequence based on the current financial data sequence. The current financial data sequence is obtained by chronologically arranging the financial data of universities within a certain period, and the current financial identification sequence is used to represent the financial identification corresponding to each financial data in the current financial data sequence.

[0038] It can be understood that the cycle can be a short cycle, such as day, week, or month (applicable to cash flow management); it can be a medium cycle, such as quarter (applicable to budget execution analysis); or it can be a long cycle, such as year (applicable to financial settlement and financial health assessment).

[0039] For example, the financial data of a university within a certain period can be obtained by querying the university's financial management system (such as Oracle EBS, SAP, UFIDA, Kingdee, etc.) through SQL; it can be obtained from the university's financial statements, such as income statements, expenditure statements, budget execution statements, etc.; the financial data of various departments of the university can be extracted from the ERP system; if the university's financial data is stored in Excel / CSV files, it can be read using tools such as Python.

[0040] Data cleaning and preprocessing of current financial data include: removing duplicate records, such as duplicate income or expenditure data; filling or deleting missing data, such as using data from the previous period to fill in or delete outliers; and standardizing all data fields, such as date format, monetary unit, and account classification. After data cleaning, the current financial data is arranged chronologically to form a current financial data sequence for subsequent trend analysis.

[0041] Financial identifiers can include the time, type, and amount of financial data. Financial data types can be categorized into income categories, such as tuition, accommodation fees, research funding, social donations, and other income (such as venue rental fees and investment returns); and expenditure categories, such as teaching expenses (teacher salaries, equipment purchases), research expenses (laboratory materials, paper publication fees), logistics expenses (property management, utilities), administrative expenses (office supplies, marketing), student activities expenses (scholarships, grants), and infrastructure expenses (new school projects and renovation costs).

[0042] For example, each piece of financial data can be assigned to a corresponding type based on attributes such as the data's source, amount, and purpose. Data can be automatically categorized using the financial system's pre-set accounting classification system. Data can also be automatically categorized by identifying keywords in the data (e.g., salary, scholarship). Historical financial records can be compared and categorized based on historical similarities. Financial personnel can manually confirm special cases that don't meet the automatic classification rules. For example, if a source of income is tuition payments, its type is tuition income; if an expenditure is incurred for laboratory equipment purchases, its type is scientific research expenditure.

[0043] After the financial data is categorized, the time and amount in the financial data are extracted and the financial data identifiers are arranged in chronological order to generate a financial identifier sequence. For example, the financial identifier is 2025.03.01 - Tuition Income - 500,000; the financial identifier is 2025.03.03 - Teaching Expenses - 20,000.

[0044] This step facilitates the structuring and standardization of financial data, laying the foundation for subsequent financial forecasting, cash flow analysis, and anomaly detection.

[0045] In one possible implementation, S100, determining a current financial identification sequence based on a current financial data sequence, includes:

[0046] S110: Extract key information for each financial data item from the current financial data sequence based on an identification template. The identification template includes the teaching cycle, five-level classification, funding attributes, and amount. The key information for each financial data item includes the teaching cycle, five-level classification, funding attributes, and amount. The teaching cycle is obtained by converting the time of the financial data into the academic year number, semester phase, and teaching week. The five-level classification divides the financial data into five levels: major category, medium category, minor category, special item, and sub-item according to the standards of the Ministry of Education. Funding attributes include funding source type and funding destination type.

[0047] It can be understood that each financial data in the current financial data sequence can include a timestamp (such as 2025.03.10), transaction amount, fund category (such as tuition income, purchase of precision instruments, etc.), relevant departments (such as teaching units, scientific research units, administrative departments), funding sources (such as government grants, self-raised funds (tuition, accommodation fees), social funds) and fund purposes (such as wages, equipment purchases, and scholarship issuance).

[0048] The semester stage can be divided into the preparation period for the start of school (the first two weeks of the teaching week), the teaching operation period (the 3rd to 16th weeks of the teaching week), the examination week (the last two weeks of the teaching week), and the holidays (winter vacation / summer vacation).

[0049] The five-level classification is constructed based on the "Guidelines for Financial Accounting of Colleges and Universities" of the Ministry of Education. The major category represents the highest-level classification of fund use, the middle category represents the second-level classification of funds, the minor category represents specific expenditure or income categories, the special item represents specific projects that are further subdivided, and the sub-item represents the specific flow of individual funds.

[0050] For example, the timestamp of each financial data item can be converted into an academic year, semester, and teaching week. For example, if the timestamp of a financial data item is 2025.03.10, then the teaching period for this financial data item is the 2024-2025 academic year, the semester is the teaching period, and the teaching week is week 5. Converting natural time (timestamps) into a university-specific teaching calendar system (teaching cycle) can solve the problem that universal time cannot reflect the cyclical fluctuations of university finances.

[0051] Based on the funding category of each financial data item, a five-level classification tree can be constructed: major category - medium category - minor category - special item - sub-item. For example, if the funding category of a financial data item is precision instrument purchase, the five-level classification of this financial data could be scientific research expenditure - vertical project - equipment expenses - national key R&D program - precision instrument purchase. Structuring funding categories into a five-level classification tree allows for more detailed tracking of funding flows than traditional three-level classification.

[0052] You can extract the fund attribute label for each financial data item based on its source and purpose. For example, if the source of funds for a financial data item is government grants and its purpose is student aid, then the fund attribute for the financial data item is government grants -> student aid.

[0053] S120 , filling key information of each financial data into an identification template to obtain a financial identification corresponding to each financial data.

[0054] For example, after extracting the teaching period, five-level classification, funding attributes, and amount for each financial data item, they are entered into an identification template to form a standardized financial identification: {Academic Year Number}-{Semester Period}-{Teaching Week} / {Five-Level Classification} / {Fund Source Type}->{Fund Destination Type} / {Amount}. For example, a financial identification might be: 2024-2025 Academic Year - Teaching Period - Week 5 / Scientific Research Expenditure - Longitudinal Project - Equipment Expenses - National Key R&D Program - Precision Instrument Purchase / Self-Raised Funds -> Equipment Purchase / 20,000.

[0055] S130 , arranging the financial identifier corresponding to each financial data according to the arrangement order of each financial data in the current financial data sequence to obtain a current financial identifier sequence.

[0056] For example, each generated financial identifier is arranged according to the original order of the current financial data sequence to form a current financial identifier sequence. The current financial identifier sequence can be used for financial analysis, forecasting, fund flow tracking, etc.

[0057] The above steps are conducive to the standardization and traceability of university financial data, providing a solid foundation for subsequent financial analysis, forecasting and anomaly detection.

[0058] S200 , predicting each financial identifier of the next period of the current financial identifier sequence based on the current financial identifier sequence to obtain a predicted financial identifier sequence.

[0059] It is understandable that since university financial data has cyclical patterns (e.g. tuition income is concentrated at the beginning of the semester, research funding expenditures have a quarterly allocation rhythm, etc.), the financial identifier sequence of the next cycle can be predicted by analyzing the patterns of historical financial data.

[0060] For example, linear regression, sliding average method or exponential smoothing can be used to identify the long-term growth or decline trend of the historical financial indicator sequence; Fourier transform or autocorrelation analysis can be used to identify whether the historical financial indicator sequence has a pattern that repeats at fixed time intervals (periodicity); seasonal decomposition of time series (STL) can be used to analyze whether the historical financial indicator sequence is affected by seasonal factors, such as semester changes, fiscal year cycles, etc.; the relationship between different financial indicators can be analyzed through the Pearson correlation coefficient and Granger causality test to determine whether there is a dependency relationship between certain expenditures or income.

[0061] You can select an appropriate forecasting model based on the trends, cyclical characteristics, seasonality, and correlations of your financial data. For example, if the data trend is stable, you can choose ARIMA or exponential smoothing; if the data exhibits long-term dependencies, you can choose an LSTM neural network. Once you've selected an appropriate forecasting model, you can train the model using historical financial data and use a subset of historical data (such as data from the most recent year) as a test set to evaluate the model's accuracy and optimize model parameters, such as adjusting the ARIMA order or the number of LSTM layers. After model training is complete, input the current financial indicator data, and the model will output a sequence of predicted financial indicators for the next period.

[0062] The forecast results can be adjusted based on the experience of university financial managers. For example, if new policies are expected to be introduced in the next cycle, the budget ratio can be adjusted manually; if a certain expenditure is significantly higher than the historical trend, the forecast data can be reassessed.

[0063] This step can predict the financial situation of colleges and universities and provide a scientific basis for identifying abnormal financial data.

[0064] In one possible implementation, see Figure 2 S200 predicts each financial identifier of the next cycle of the current financial identifier sequence based on the current financial identifier sequence to obtain a predicted financial identifier sequence, including:

[0065] S210 , encoding the teaching cycle of each financial identifier in the current financial identifier sequence to obtain a semester phase feature vector and a teaching week feature vector for each financial identifier.

[0066] For example, one-hot encoding can be used to convert the semester stage to which each financial identifier belongs into a vector representation, such as the feature vector of the school preparation period is (1, 0, 0, 0), the feature vector of the teaching operation period is (0, 1, 0, 0), the feature vector of the exam week is (0, 0, 1, 0), and the feature vector of the holiday is (0, 0, 0, 1).

[0067] Because teaching weeks are continuous variables, directly using integer representations may result in a poor model understanding of the teaching week feature vector. Therefore, sinusoidal positional encoding can be used. Sinusoidal positional encoding preserves periodic information and is particularly suitable for time series forecasting tasks. It maps teaching weeks into a high-dimensional vector space, making the distances between adjacent weeks more regular.

[0068] The sinusoidal position encoding is defined as , ,in, Indicates the number of teaching weeks (18 weeks), represents the encoding dimension (set to 4 or 8 to control computational complexity), and The coding values for the even and odd dimensions are respectively corresponding. 10000 is the scaling factor used to make the values of different weeks distributed in a more appropriate range. For example, the teaching week of a certain financial symbol is 5, the encoding dimension is 4, and the sinusoidal position encoding is calculated: , , , , the characteristic vector of teaching week is [0.4999, 0.8660, 0.0099, 0.9999].

[0069] This step can make full use of the teaching cycle information, capture the characteristics of the semester stage, and retain the time series characteristics of the teaching weeks, thereby improving the accuracy of financial data forecasting.

[0070] S220 , classify each financial identifier in the current financial identifier sequence into five categories to obtain a five-category feature vector for each financial identifier. The five-category feature vector includes a major category feature vector, a medium category feature vector, a minor category feature vector, a special feature vector, and a sub-item feature vector.

[0071] Exemplarily, one-hot encoding can be used to encode major categories, medium categories, and minor categories to obtain major category feature vectors, medium category feature vectors, and minor category feature vectors; label encoding can be used to encode special items and sub-items to obtain special item feature vectors and sub-item feature vectors.

[0072] S230 , converting the capital attribute in each financial identifier in the current financial identifier sequence into a capital attribute feature vector through a type mapping matrix.

[0073] For example, a table of number-source type and number-destination type can be constructed. Based on these tables and the probability of each source type flowing to each destination type, a type mapping matrix can be constructed. The source and destination of funds for each financial identifier are converted into one-hot vectors, namely, a fund source vector and a fund destination vector. Based on the type mapping matrix, the fund source vector and the fund destination vector are mapped and transformed to obtain a fund attribute feature vector.

[0074] The assumption number-source type table is as follows:

[0075]

[0076] Table 1

[0077] The number-destination type table is as follows:

[0078]

[0079] Table 2

[0080] Here we can build a 5×5 type mapping matrix, which represents the relationship weight of funds from source type (row index) to destination type (column index). for , each element Represents the source type of funds To destination type The value range is 0~1 (1 means high correlation, 0 means no correlation). For example, It indicates that government funding is mainly used for teachers’ salaries, and the correlation is relatively high.

[0081] The source of funds for a certain financial identifier is tuition income, and the destination of funds is equipment procurement. They are converted into one-hot vectors through one-hot encoding, that is, the source vector is S=[0,1,0,0,0], and the destination vector is D=[0,0,1,0,0]. Calculate the product of the source vector and the type mapping matrix, and the product of the transposed determinant of the destination vector and the type mapping matrix. Add the two products to obtain the fund attribute feature vector, that is, , , the capital attribute characteristic vector is .

[0082] This step is conducive to enhancing the accuracy of subsequent fund flow modeling and can be effectively used in university financial analysis and forecasting.

[0083] S240 , normalizing the amount in each financial identifier in the current financial identifier sequence to obtain an amount feature vector for each financial identifier.

[0084] For example, the Min-Max normalization method can be used to map the amount in each financial identifier in the current financial identifier sequence to between 0 and 1 or -1 and 1, which helps improve the stability of the amount prediction. That is, the difference between the amount of each financial identifier and the minimum amount in the current financial identifier sequence, as well as the difference between the maximum amount and the minimum amount in the current financial identifier sequence, are calculated. The ratio of the two (the difference between the amount of each financial identifier and the minimum amount in the current financial identifier sequence / the difference between the maximum amount and the minimum amount in the current financial identifier sequence) is the amount feature vector for each financial identifier.

[0085] S250, based on the semester stage feature vector and the five-class classification feature vector of each financial identifier, use the classification model to predict the five-class classification of each financial identifier in the next period of the current financial identifier sequence to obtain a class prediction result.

[0086] For example, the five-level classification can be regarded as a hierarchical structure. Based on the Random Forest, each level category is predicted step by step. Five independent random forest classifiers can be constructed and trained to predict the major category (major category classification model), medium category (medium category classification model), small category (small category classification model), special category (special category classification model), and sub-item (sub-item classification model) respectively.

[0087] Training process: Financial data from universities over the past 3-5 years is collected and converted into financial identifiers. A collection of financial identifier sequences from various periods is constructed. This collection of financial identifier sequences from various periods is divided into a training set (80%) and a validation set (20%). The input data in the training and validation sets includes the semester feature vectors and five-class classifications for the financial identifier sequence from the previous period. The output data includes the five-class classification of each financial identifier from the next period corresponding to the financial identifier sequence from the previous period.

[0088] You can set the number of decision trees (e.g., 100-500 trees), maximum depth (e.g., 10-20), minimum number of sample splits (e.g., 5-10), and minimum number of sample leaves (e.g., 2-5). Bootstrap sampling is used to generate multiple sub-datasets from the training set. Each decision tree is trained using a different sub-dataset. The optimal feature split point is selected using information gain or the Gini coefficient, and all decision trees vote to determine the final five-class classification result. Model optimization can use K-Fold cross-validation (K=5 or 10) to avoid overfitting. Accuracy, F1-score, and recall are calculated. Grid search or Bayesian optimization is used to adjust the number of trees for improved stability. The maximum depth (max_depth) is adjusted to control model complexity and prevent overfitting.

[0089] After the model training is completed, the semester stage feature vector and five-class classification feature vector of each financial identifier are input into the model, and the model outputs the five-class classification prediction results.

[0090] Optionally, see Figure 2 S250, based on the semester stage feature vector and the five-class classification feature vector of each financial identifier, uses the classification model to predict the five-class classification of each financial identifier in the next cycle of the current financial identifier sequence, and obtains the category prediction results, including:

[0091] S251: For each financial identifier, perform the following operations: Based on the hierarchical attention mechanism, calculate the weight of each eigenvector in the five-level classification feature vector, perform a weighted sum of the weight of each eigenvector and each eigenvector to obtain a classification comprehensive feature vector; and select the classification weight corresponding to the semester stage eigenvector from the classification weight matrix. The classification weight matrix is pre-set and includes classification weights corresponding to eigenvectors of different semester stages.

[0092] For example, the feature vector of the major category can be set as , the middle class feature vector is set as , the small class feature vector is set as , the special feature vector is set as , the sub-item eigenvector is set to , we can use splicing, weighted summation, dimensionality reduction projection and other methods to 、 、 、 、 Combine to form the total eigenvector ,like , the contribution (weight) of each level to the final classification is calculated through the hierarchical attention mechanism. The calculation formula is ,in, is a trainable hierarchical attention parameter. Based on the calculated hierarchical weight, each feature vector in the five-level classification feature vector is weighted and summed to obtain the classification comprehensive feature vector, that is, ,in, Represents the comprehensive feature vector of the classification. The hierarchical attention mechanism of the five-level classification can solve the problem that traditional single-layer classification cannot handle the complex system of universities.

[0093] The classification weight matrix can be set as - Preparation period for the start of school, -Teaching operation period, -Exam week, -Holiday. According to the semester phase feature vector of the financial identifier, select the corresponding classification weight The setting of the classification weight matrix can overcome the prediction bias of the general model at different stages of the semester.

[0094] The product of the corresponding classification weight and the classification comprehensive feature vector can be used as the input feature vector of the classification model, that is, .

[0095] S252 , based on the classification weight and classification comprehensive feature vector of each financial identifier, a classification model is used to predict the five-level classification of each financial identifier in the next cycle of the current financial identifier sequence to obtain a category prediction result. The classification model is a machine learning model.

[0096] For example, the training set and validation set in step S250 may be used to convert the five-class classification of each financial identifier in the training set into an input feature vector through the method of step S251 .

[0097] The classification model can be a Random Forest, XGBoost, or a neural network (MLP / Transformer). A neural network is used here. Construct the neural network structure (input layer, hidden layer, output layer): The input layer is constructed using the dimensions of the input feature vector. The hidden layer is constructed using a multi-layer fully connected network, normalization layers, and dropout. The ReLU activation function can be used to construct a multi-layer fully connected network to enhance the model's nonlinear learning capabilities. Normalization layers prevent vanishing or exploding gradients, and dropout prevents overfitting. The output layer can use a Softmax activation to output a probability distribution for a five-class classification.

[0098] Training process: Use Xavier to initialize weights and set bias to zero. Calculate the output of the input feature vector after passing through the hidden layer. Obtain classification probabilities through the Softmax layer. Calculate gradients to update weights. Use the Adam optimizer for gradient updates. Use Mini-Batch SGD to improve training efficiency. The batch size can be set to 32–256. Adjust the number of hidden layers and neurons. Add Dropout to prevent overfitting. Use K-Fold cross-validation (K=5 or 10) to facilitate the model's generalization ability on different datasets. Calculate accuracy, F1, recall, and confusion matrix to verify the model's output.

[0099] After the model training is completed, the input feature vector corresponding to each financial identifier in the current financial identifier sequence is input into the model, and the model outputs the probability distribution of the five-level classification of each financial identifier in the next period.

[0100] These steps are combined with the hierarchical attention mechanism to improve the accuracy of classification prediction and consider semester stage information to make the prediction more reasonable. The neural network model can accurately predict the five-level classification of financial identification for the next cycle, providing scientific data support for university financial management.

[0101] S260, based on the teaching week feature vector, fund attribute feature vector and amount feature vector of each financial identifier, use the regression model to predict the amount corresponding to each five-level classification in the category prediction result to obtain an amount prediction result.

[0102] Exemplarily, the amount can also be predicted using random forests, with the fund attribute feature vector, teaching week feature vector, and amount feature vector of each financial identifier in the previous period in the training set and validation set in step S250 and the corresponding five-level classification of each financial identifier in the next period marked as input data, and the amount of each financial identifier in the next period marked as output data.

[0103] You can randomly select multiple data subsets from the training set and train multiple decision trees. Within each tree, find the optimal feature split point (e.g., five-class classification vs. instructional week). Select the optimal split feature using information gain or minimum variance to construct a regression tree. Adjust the number of decision trees and control the maximum tree depth to prevent overfitting. Set a minimum sample split and adjust the minimum number of leaf samples to reduce model complexity. Use bootstrap sampling to train multiple sub-models, calculate the predicted amount for each decision tree, and take the weighted average of all trees as the final predicted amount.

[0104] After the training is completed, the teaching week feature vector, fund attribute feature vector and amount feature vector of each financial identifier in the current financial identifier sequence and each five-level classification in the category prediction result are input, and the model outputs the amount corresponding to each five-level classification in the category prediction result.

[0105] Optionally, see Figure 2 S260, based on the teaching week feature vector, fund attribute feature vector and amount feature vector of each financial identifier, the amount corresponding to each five-level classification in the category prediction result is predicted using the regression model to obtain the amount prediction result, including:

[0106] S261: Based on the fund attributes and fund attribute feature vectors of each financial identifier, a current fund flow graph is constructed. Based on the current fund flow graph, a graph convolutional network is used to generate fund flow features. The nodes of the current fund flow graph represent each financial identifier, and the edges of the current fund flow graph represent the direction of fund flow.

[0107] Exemplarily, each financial identifier is used as a node in the current capital flow graph, and the node feature can be a capital attribute feature vector. The directed edges between nodes are determined according to the source and destination of funds of each financial identifier (such as funds flow from A to B, A->B). The weight of the directed edge can be calculated based on the amount of capital flow, that is, the ratio of the outflow amount of each edge to the total outflow amount.

[0108] Based on the weights of directed edges, a capital flow adjacency matrix is constructed. Based on the current capital flow graph and the capital flow adjacency matrix, a graph convolutional network is used to calculate the capital flow characteristics of each layer, propagate information between layers, and minimize the loss function, updating the weight matrix to ultimately output the capital flow characteristics. The current capital flow graph and the graph convolutional network can capture the implicit correlations of capital flows.

[0109] S262, based on a preset time window, extracts teaching week scale features, semester scale features, and school year scale features from historical financial data, and based on the teaching week feature vector of each financial identifier, fuses the teaching week scale features, semester scale features, and school year scale features through a temporal attention mechanism to obtain multi-scale features corresponding to each financial identifier.

[0110] For example, the preset time window of the teaching week scale can be 20 weeks. The financial data of the first 20 weeks of the current cycle are subjected to sliding window statistics to extract statistical characteristics of capital fluctuations (teaching week scale characteristics), such as average capital flow, capital flow change rate, and capital peak within a week. The teaching week scale characteristics can reflect short-term capital flow trends.

[0111] The preset time window of the semester-scale feature can be two semesters. The fund usage of the past two semesters is extracted, and the total fund expenditure of the semester, the fund category distribution trend, and the peak fund expenditure period within the semester are calculated as the semester-scale feature. The semester-scale feature can reflect the fund usage pattern at the semester level.

[0112] The preset time window of the academic year scale feature can be three academic years. The financial data of colleges and universities in the past three years are extracted, and the annual funding growth rate, annual budget execution status and funding utilization stability (variance) are calculated as academic year scale features. The academic year scale feature can identify long-term funding trends.

[0113] The teaching week feature vectors for different financial identifiers may be at different time points. Therefore, we can calculate their attention weights for historical timescale features separately and generate their own fusion features. The role of the temporal attention mechanism is to enable the teaching week feature vectors of each financial identifier to match information at different timescales.

[0114] For each financial identifier: In order to ensure that different time scale features can be effectively combined, the temporal attention mechanism can be used to adaptively adjust the contribution (weight) of each time scale feature. The formula for calculating the weight corresponding to different time scale features through the temporal attention mechanism is: ,in, Indicates the The teaching week feature vector of the financial identifier, Represents different time scale characteristics, express The characteristic dimension of Represents the weights corresponding to different time scale features. The weights corresponding to different time scale features and the weighted sum of different time scale features are obtained. Multi-scale features corresponding to financial identifiers.

[0115] This step can simultaneously consider the short-term, medium-term, and long-term trends of capital changes, and adaptively adjust the influence of different time scales through the temporal attention mechanism to accurately extract the multi-scale features corresponding to each financial identifier.

[0116] S263 , based on the capital flow characteristics, the multi-scale characteristics corresponding to each financial identifier, and the amount feature vector of each financial identifier, a regression model is used to predict the amount corresponding to each five-level classification in the category prediction results to obtain an amount prediction result. The regression model is a machine learning model.

[0117] For example, the training set and validation set in step S250 are divided into multiple previous periods and corresponding next periods. Steps S261 and S262 convert all previous period financial data into capital flow features, multi-scale features corresponding to each financial identifier, and an amount feature vector for each financial identifier. The corresponding five-level classification annotations for each financial identifier in the next period serve as input data, and the amount annotations for each financial identifier in the next period serve as output data. Appropriate regression models can be selected, such as random forest regression, XGBoost regression, or LSTM time series regression.

[0118] Training Process: You can set initial parameters for Random Forest, XGBoost, or LSTM, such as tree depth, learning rate, and batch size. The capital flow features, multi-scale features, and amount feature vectors are combined and input into the regression model. The five-class classification prediction results are converted to one-hot encoding, enabling the model to learn the relationship between the five-class classification and the amount. The regression model output, namely the predicted amount, is then calculated. The error between the predicted value and the actual amount can be calculated as a loss function, such as mean squared error (MSE) or root mean squared error (RMSE). For XGBoost and LSTM, gradient descent can be used to optimize weights and gradually reduce prediction error. For Random Forest regression, the number of decision trees can be increased or decreased, and the maximum tree depth can be optimized to improve model stability. Training can be repeated for multiple epochs (e.g., 100) or tree growth processes until the loss function converges.

[0119] After the model training is completed, the capital flow characteristics, the multi-scale features corresponding to each financial identifier, the amount feature vector of each financial identifier, and each five-level classification in the category prediction results are input into the regression model, and the model outputs the predicted amount corresponding to each five-level classification.

[0120] This step combines the characteristics of capital flow to improve prediction accuracy. Multi-scale characteristics can enhance time dependence, so that the regression model can accurately predict the financial amount of the next cycle and provide scientific decision-making support for university financial management.

[0121] S270 , arranging the category prediction results and the amount prediction results in the time sequence of the next period of the current financial identification sequence to obtain a predicted financial identification sequence.

[0122] For example, the predicted five-level classifications and corresponding amounts may be arranged in chronological order of the next period of the current financial identification sequence to form a complete predicted financial identification sequence.

[0123] In one possible implementation, the university financial data management method further includes:

[0124] It should be noted here that the current financial identification sequence may not contain all the five-level classifications predicted for the next cycle, which will cause some of the predicted five-level classifications to not appear in the current cycle, resulting in the inability to directly calculate the budget execution rate and the inability to determine the current cumulative expenditure, thereby affecting the calculation of the remaining budget amount. There are two solutions: If a predicted five-level classification does not exist in the current cycle, you can look up historical data or similar classification data from the past N cycles to fill in the budget execution rate and cumulative expenditure; if a predicted five-level classification does not exist in the current cycle and there is no historical data at all, the full-year budget allocation rule is used, that is, the full-year budget is used as the remaining budget amount, and the budget execution rate is set to the initial value (such as 0%).

[0125] S201, calculating the budget execution rate and remaining budget amount corresponding to each five-level classification in the category prediction result.

[0126] For example, find out from the current financial identification sequence which five-level classifications are the same as the five-level classifications in the category prediction results, and directly use the total budget amount of the five-level classification to subtract the cumulative expenditure amount of the same five-level classification in the current financial identification sequence to obtain the remaining budget amount, and the ratio of the cumulative expenditure amount of the same five-level classification in the current financial identification sequence to the total budget amount is the budget execution rate.

[0127] For five-level classifications that do not exist in the current financial identification sequence, if historical data is available, the total budget amount corresponding to the five-level classification in the past N periods and the expenditure amount of the historical periods can be used to calculate the historical average budget execution rate. The remaining budget amount can be calculated based on the total budget amount and the historical average budget execution rate.

[0128] For the five-level classification that does not exist in the current financial identification sequence, if there is no historical data at all, the remaining budget amount of the five-level classification is the annual budget, and the budget execution rate is 0%.

[0129] This step solves the problem that the current financial identification sequence lacks certain five-level classifications in the forecast results and cannot calculate the corresponding budget execution rate and remaining budget amount, making the budget adjustment more scientific and reasonable.

[0130] S202: If the budget execution rate exceeds the budget threshold, the predicted amount of the five-level classification that exceeds the budget threshold is attenuated according to the excess ratio of the budget execution rate to obtain the attenuated amount of the five-level classification that exceeds the budget threshold.

[0131] For example, the budget execution rate overrun ratio is the difference between the budget execution rate and the budget threshold. If the overrun ratio is large, the forecast amount can be reduced more significantly; if the overrun ratio is small, the forecast amount is only slightly adjusted. The forecast amount attenuation formula is: ,in, Indicates the decay amount, represents the initial forecast amount, Indicates the adjustment coefficient, which determines the attenuation strength (such as 0.5). represents the budget execution rate, Indicates the budget threshold (such as 80%). Exceed , the forecast amount is reduced proportionally.

[0132] S203 : Determine the budget upper limit of the five-level classification that exceeds the budget threshold based on the remaining budget amount corresponding to the five-level classification that exceeds the budget threshold.

[0133] For example, to prevent the forecast amount from still being overspent, even after attenuation, the value may still be large, so an upper limit can be set for the five-level classification that exceeds the budget threshold, that is, , Indicates the budget ceiling, 1.2 is an adjustable coefficient (which can be changed according to actual conditions), indicating that the maximum amount that can exceed the remaining budget is 20%. Indicates the remaining budget. If Too small, will also become smaller to prevent over-prediction.

[0134] S204 , based on the budget upper limit of the five-level classification that exceeds the budget threshold, amend the attenuation amount of the five-level classification that exceeds the budget threshold to obtain the amended amount of the five-level classification that exceeds the budget threshold.

[0135] For example, the calculated decay amount may still be greater than the budget cap. In this case, the decay amount can be revised, that is, the budget cap is determined as the revised amount. Even if the budget execution rate of a certain five-level classification does not exceed the budget threshold and does not need to be decayed, there may still be a situation where the initial forecast amount exceeds the budget cap. Similarly, the budget cap is determined as the revised amount for this five-level classification. ,in, Indicates the revised amount.

[0136] Assume that the current five-level classification is as follows:

[0137]

[0138] Table 3

[0139] The budget execution rate of Category A is 85 / 100=85%, and the remaining budget amount is 100-85=150,000 yuan; the budget execution rate of Category B is 90 / 150=60%, and the remaining budget amount is 150-90=600,000 yuan; assuming that the average budget execution rate of Category C in the past three cycles is 80%, the remaining budget amount is 200-200×80%=400,000 yuan, and the budget execution rate of Category D is 0%, and the remaining budget amount is 2.5 million yuan. Assuming the budget threshold is 80%, then only Category A exceeds the limit, and the forecast amount of Category A is attenuated, that is, y1=20×(1-0.5×(85%-80%))=19 (million yuan), and the budget upper limit of Category A is y2=1.2×15=18 (million yuan). Since y1>y2, the revised amount of Category A is 180,000 yuan; the budget upper limit of Category B is y2=1.2×60=72 (million yuan), and the forecast amount of Category B does not exceed the budget upper limit, so the revised amount of Category B is still the forecast amount of 300,000 yuan; the budget upper limit of Category C is y2=1.2×40=48 (million yuan), and the forecast amount of 500,000 yuan is greater than the budget upper limit of 480,000 yuan, so the revised amount of Category C is 480,000 yuan; and Category D does not need to be adjusted, and the budget amount is the final amount corresponding to Category D.

[0140] Through these steps, a process of dynamic constraints on the forecast amount is established, which is conducive to adjusting the forecast amount when the budget execution rate or forecast amount exceeds the limit, so as to avoid overvaluation or unreasonable forecasts.

[0141] S300: Acquire an actual financial data sequence, and determine an actual financial identification sequence based on the actual financial data sequence, wherein the actual financial data sequence is the financial data actually collected in the next period of the current financial identification sequence.

[0142] Illustratively, after the actual financial data of the next period is collected, an actual financial data sequence is constructed using the same method as step S100 , and an actual financial identification sequence is obtained based on the actual financial data sequence.

[0143] S400 , comparing the actual financial identification sequence with the predicted financial identification sequence, and constructing an actual capital flow diagram based on the actual financial identification sequence, determining whether there is abnormal financial data in the actual financial data sequence, and obtaining a determination result.

[0144] It can be understood that the fund flow diagram is used to visualize the flow of funds between different financial identifiers in order to detect abnormal fund flows.

[0145] For example, the error rate between the actual and predicted values can be calculated for each financial indicator. If the error rate is small (e.g., within ±5%), the prediction is relatively accurate. If the error rate is too large (e.g., exceeding ±20%), the actual financial indicator sequence may contain an anomaly, and the cause of the anomaly can be further analyzed. Alternatively, the mean squared error (MSE) between all actual and predicted values can be calculated for all financial indicators. The root mean square error (RMSE) can then be calculated from the mean squared error. If the RMSE is too large, the overall prediction error is high, indicating a possible anomaly.

[0146] When it is determined that there is an anomaly in the actual financial identifier sequence, an actual capital flow graph is constructed: each financial identifier in the actual financial identifier sequence can be determined as a node of the graph, the edge of the graph can represent the direction of capital flow, and the weight of the edge can represent the amount of capital flow.

[0147] The actual fund flow chart can be used to identify abnormal flow paths, such as when funds do not flow to the required nodes according to the established process. It can also be used to identify unusual transaction amounts, such as when the amount of a particular fund flow far exceeds or falls far below the historical average. It can also be used to identify sudden changes, such as the sudden appearance of a previously unseen fund flow path. If at least one of these abnormalities is present in the actual fund flow chart, it is determined that abnormal financial data exists in the actual financial data sequence.

[0148] The predicted financial indicator sequence serves as a baseline, helping finance personnel determine whether the actual financial indicator sequence contains anomalies. The prediction does not represent the final financial situation, but rather provides an expected value for comparison and anomaly detection. The predicted financial indicator sequence is only used to identify potential anomalies in financial data. The actual funds flow diagram constructed in conjunction with the actual financial indicator sequence is required to ultimately determine whether the actual financial data sequence contains anomalies and take appropriate management measures.

[0149] This step can timely discover and handle abnormal situations, and improve the transparency of financial management and risk prevention and control capabilities of colleges and universities.

[0150] In one possible implementation, S400 compares the actual financial identifier sequence with the predicted financial identifier sequence, constructs an actual funds flow diagram based on the actual financial identifier sequence, and determines whether there is abnormal financial data in the actual financial data sequence. The determination result includes:

[0151] S410 , comparing each financial identifier in the actual financial identifier sequence with each financial identifier in the predicted financial identifier sequence one by one, calculating the difference between each financial identifier in the actual financial identifier sequence and each financial identifier in the predicted financial identifier sequence, and obtaining a difference collection.

[0152] For example, it's important to note that when comparing actual and forecasted financial indicators, two situations may arise: First, a predicted five-level category may not actually exist. This means the category didn't actually occur during execution, perhaps due to budget adjustments, project cancellations, or unallocated funds. Directly calculating the difference in amounts is impossible because the actual amounts don't exist. Second, a five-level category may not actually exist. This means revenue or expenditure actually occurred despite the forecast being unforeseen. This may be due to unexpected expenditures, unplanned cash flows, or the forecasting model's failure to accurately capture the category. Since these situations cannot be identified using the difference in amounts, new methods are needed to identify anomalies. For five-level categories for which a difference in amounts cannot be calculated, historical data pattern matching (checking whether the category has occurred before) can be used to determine whether the financial data corresponding to the five-level category is anomaly.

[0153] Historical Data Pattern Matching: Calculate the average frequency of occurrence of the five-level classifications that exhibit at least one of the above conditions over the past N cycles. If the classification appears frequently in historical data but not in the current cycle, it may be an anomaly. If the classification has never appeared in historical data but suddenly appears in the current cycle, it may be an anomaly. For example, if Category A appears in all three cycles and appears in the predicted financial identifier sequence but not in the actual financial identifier sequence, the corresponding financial data may be an anomaly. If Category B does not appear in the past three cycles but does not appear in the predicted financial identifier sequence but does in the actual financial identifier sequence, the corresponding financial data may be an anomaly.

[0154] For actual financial indicators and forecast financial indicators that do not meet the above conditions, you can directly calculate the difference between the amount of the actual financial indicator and the amount of the forecast financial indicator corresponding to the actual financial indicator.

[0155] This step takes into account the possible errors between the prediction and the actual situation, and solves the problem of being unable to calculate the difference in amount due to the error, which is beneficial to the integrity of anomaly detection.

[0156] S420 , each difference in the difference set is compared with a difference threshold. If each difference in the difference set is within the difference threshold, it is determined that no abnormal financial data exists in the actual financial data sequence. The difference threshold is dynamically changed based on the university's academic calendar.

[0157] Understandably, the margin threshold can vary dynamically based on the university's academic calendar. For example, at the beginning of the semester (the first two weeks), budget execution fluctuates more, so a higher margin of error is permitted; mid-semester (weeks 3-12), expenditures are stable, so the threshold is lower; and towards the end of the semester (weeks 13-18), significant expenditures (such as equipment purchases) may increase.

[0158] For example, each difference value in the difference value set calculated in step S410 is compared with the corresponding difference threshold range. If all the difference values are within the difference threshold range, it can be determined that there is no abnormal financial data in the actual financial data sequence. If at least one difference value exceeds the difference threshold range, the next step is to analyze the capital flow pattern.

[0159] It should be noted that if the situation in step S410 exists and it is determined to be abnormal data through historical data pattern matching, the next step of analyzing the capital flow pattern will also be entered.

[0160] At step S430, if at least one difference in the difference set is outside the difference threshold, an actual funds flow graph is constructed based on the actual financial data sequence. The actual funds flow graph is analyzed for fund flow patterns. If any abnormal features are present in the actual funds flow graph, abnormal financial data is determined to exist in the actual financial data sequence. The actual funds flow graph is a three-layered graph, where the nodes represent the accounts corresponding to each financial identifier in the actual financial identifier sequence, and the edges represent the direction of fund flows. Abnormal features include a closed-loop fund chain, a single node with a fund outflow ratio exceeding 40%, and direct transactions with accounts on the high-risk list.

[0161] It can be understood that the first layer (fund source layer) of the three-layer fund flow diagram is the fund source account (such as the fiscal appropriation account, tuition income account, and scientific research funding account); the second layer (fund allocation layer) is the fund use account (such as the accounts of various colleges and departments, research centers, and administrative departments); the third layer (final use layer) is the final flow of funds (such as faculty salary payment, equipment procurement, and payment to external cooperative institutions).

[0162] For example, each account corresponding to each financial data point is treated as a node. Directed edges between nodes are then connected based on the fund attributes of each financial data point. The amount flowing between nodes is used as the edge weight to construct an actual fund flow graph. In an actual fund flow graph, multiple financial identifiers may point to the same source account (such as a tuition income account), multiple financial identifiers may share the same destination account (such as a teacher salary account), and some financial identifiers may involve multiple fund flows, resulting in duplicate accounts. If these cases are not addressed, they may lead to duplicate nodes, making the fund flow graph structure unclear. The fragmented fund flow information affects the accuracy of fund anomaly detection. Therefore, when multiple financial identifiers point to the same account, only one node is retained in the graph, and the fund inflow and outflow information for that account are combined based on the total fund amount.

[0163] It can be understood that closed-loop funds are funds that flow repeatedly between multiple accounts, forming a closed-loop transaction (such as A→B→C→A). The funds are not actually used but transferred back and forth between multiple accounts, which may involve fund abuse or money laundering.

[0164] For example, depth-first search (DFS) or breadth-first search (BFS) can be used to identify closed-loop paths in the actual capital flow graph and calculate the closed-loop amount. If the closed-loop amount exceeds a certain proportion (such as 20%), it is considered abnormal.

[0165] It can be understood that the proportion of capital outflow from a single node exceeding 40% means that if the single capital outflow direction of an account exceeds 40%, there may be abnormal concentrated capital outflow, which may involve risky payment or financial fraud.

[0166] For example, the total outflow of funds from each node can be calculated, and the proportion of funds outflow from that node to a single account can be calculated. If the proportion is greater than 40%, the transaction may be risky. For example, if the funds flowing from account A to account X are 800,000 yuan, the funds flowing from account A to account Y are 400,000 yuan, and the funds flowing from account A to account Z are 400,000 yuan, the funds outflow from account A to account X account for 50%, exceeding the 40% threshold and being considered abnormal.

[0167] It is understandable that if there is a direct transaction with an account on the high-risk list or a fund flow involves a known high-risk account (such as a non-compliant company or a regulated account), there may be abnormal transactions.

[0168] For example, a high-risk account list is established (such as from financial regulatory authorities or internal audit). If an account directly transacts with an account on the high-risk account list, it is marked as a high-risk transaction.

[0169] If the actual capital flow chart does not contain any of the above abnormal features, the financial data is determined to be normal; if the capital flow chart contains any of the above abnormal features, the financial data is determined to be abnormal.

[0170] This step helps to rationalize the flow of funds, prevent abnormal fund transactions, and improve financial transparency. When multiple financial identifiers correspond to the same account, they are merged into one node, which helps to make the actual fund flow diagram clearer, reduce redundant data, and improve the accuracy of anomaly detection.

[0171] S500: If the judgment result indicates that abnormal financial data exists in the actual financial data sequence, perform abnormal management on the abnormal financial data.

[0172] For example, before conducting exception management, it's possible to confirm that the abnormal data actually exists, rather than being misjudged due to factors like data entry errors or forecasting model errors. Finance personnel can review the vouchers, accounts, and transaction records for the abnormal data to confirm its authenticity and identify any emergencies (such as epidemics or budget changes) that may have impacted financial data. They can also check for data entry errors (such as over- or under-entry of amounts or incorrect dates) and technical issues within the financial system (such as data import or calculation errors). If a system or data entry error is confirmed, the data can be corrected without entering the exception management process.

[0173] If abnormal data does exist, it can be classified. Based on the abnormal data, the abnormal type (income type or expenditure type) can be determined, and the abnormal situation in the cash flow diagram corresponding to the abnormal data can be determined. Based on the abnormality type and situation, the degree of abnormality can be determined (minor, moderate, or severe), and appropriate handling methods can be taken based on the degree of abnormality.

[0174] If it is a minor anomaly, it can be recorded in the financial report and subsequent data can be closely monitored to prevent the anomaly from continuing to expand; if it is a moderate anomaly, the source of the abnormal data can be traced (such as checking relevant invoices, contracts, and bank statements), and the relevant financial management personnel and project leaders can be contacted to confirm the cause of the anomaly. The amount of the abnormal data can be monitored. If the abnormal amount continues to rise, it will be upgraded to a serious anomaly. If it is a serious anomaly, the financial audit department can be notified to check the approval process and capital flow of the financial data, and investigate whether it involves any illegal operations (such as budget overspending, misappropriation of funds, false reimbursement, etc.).

[0175] This step ensures the identification, analysis, and processing of abnormal financial data, making university financial management more transparent and efficient and reducing financial risks.

[0176] In one possible implementation, at S500, if the determination result indicates that abnormal financial data exists in the actual financial data sequence, performing abnormality management on the abnormal financial data includes:

[0177] S510: If the judgment result indicates that abnormal financial data exists in the actual financial data sequence, an abnormal capital flow graph is constructed with the abnormal financial data as the center.

[0178] For example, the structure of the abnormal funds flow diagram is: the first layer (source of funds layer), the initial inflow point of funds, such as government grant accounts, tuition accounts, and scientific research funding accounts; the second layer (intermediate account layer), funds are allocated to school-level management accounts or department accounts (such as financial centers, college and department funds accounts); the third layer (business flow layer), funds are allocated to specific purpose accounts (such as laboratory equipment procurement, teacher salaries, and administrative expenses); the fourth layer (external transaction layer), funds flow from the school to external partners, suppliers, or personal accounts; the fifth layer (final funds flow layer), the accounts where the funds finally land (such as personal salary accounts, corporate bank accounts, and cooperative institutions).

[0179] Focusing on abnormal financial data, we can trace back the source of funds, track their destination, establish a fund flow path, and confirm that all accounts involved in the transaction are tracked. For example, the first layer: fiscal appropriation accounts → the second layer: school-level scientific research fund accounts → the third layer: research institute procurement accounts → the fourth layer: backup supplier accounts → the fifth layer: personal accounts.

[0180] The hierarchical structure of the abnormal fund flow diagram helps to track the source, flow and scope of influence of funds.

[0181] S520, identifying abnormal fund aggregation nodes and key propagation paths in the abnormal fund flow graph, and obtaining graph analysis results.

[0182] It can be understood that an abnormal fund convergence node refers to an account where a large amount of funds flow into the abnormal fund flow diagram, which may involve the abnormal convergence of large amounts of funds in a single account (such as an external company account suddenly receiving a large amount of funds, which is inconsistent with the historical fund flow pattern), the confluence of multiple accounts (such as funds from different departments flowing to the same external supplier, which may pose a risk of concentrated fund transfer), and long-term fund stagnation (that is, funds entering an account do not move for a long time, which may indicate abnormal accumulation or potential fund retention).

[0183] For example, the proportion of capital inflow can be calculated. If the proportion of total capital inflow of a certain account in the capital flow diagram is greater than 50%, it will be marked as an abnormal convergence node; the abnormal capital residence time can be calculated. If the capital inflow is not spent for a long time after the inflow, and exceeds more than 2 times the historical average residence time, it will be marked as suspicious.

[0184] It can be understood that a key transmission path refers to how abnormal funds flow between different accounts, involving a transfer chain involving multiple accounts. Abnormal transmission patterns may include: chain transfers, where funds flow continuously between multiple accounts (A→B→C→D), ultimately ending up in individual or high-risk accounts; reflux, where funds are transferred from A to B and then back to A, forming a reflux path; and fund splitting, where large funds are split into multiple small transactions and flowed to different accounts to circumvent supervision.

[0185] For example, the number of layers of fund transfer can be tracked. If funds are continuously transferred in a path of more than 3 layers, it will be marked as suspicious; the proportion of single transaction amount can be calculated. If, after funds from an account are transferred out, the single transaction amount accounts for more than 40% of the total funds in the account, it may be an abnormal path.

[0186] This step can identify high-risk accounts where funds are concentrated, uncovering potential abuse, money laundering, or misappropriation. It can also detect abnormal fund flows across multiple levels, preventing illegal fund operations and reducing financial risk. Graph analysis enables more intuitive, efficient, and systematic management and auditing of abnormal fund flows, contributing to financial security and compliance.

[0187] S530: Execute exception management on the abnormal financial data according to the graph analysis result.

[0188] For example, the risk level in the graph analysis results can be identified and corresponding measures implemented accordingly. High-risk levels correspond to abnormal features such as large, unusual transactions at fund aggregation nodes. For example, a single account absorbing more than 50% of the funds can result in an immediate account freeze and a compliance investigation. Medium-risk levels correspond to abnormal features such as key transmission paths involving multiple transfers, such as funds transferred across more than four levels of accounts. Internal audits can be conducted to track transaction vouchers. Low-risk levels correspond to abnormal fluctuations in fund usage, such as a transaction exceeding the budget by 50%. This can result in an account warning and subsequent monitoring.

[0189] For cases where funds in key transmission paths eventually flow into high-risk accounts, the account will be frozen immediately, a report will be submitted to the financial audit department, and the capital chain will be traced and transaction vouchers will be checked.

[0190] The risk grading process in this step can improve management efficiency, freeze high-risk accounts, and prevent capital losses.

[0191] Optionally, at S530, performing exception management on the abnormal financial data according to the graph analysis result, including:

[0192] S531: If the concentration of abnormal funds at the aggregation node exceeds 50% or there are multiple loops between nodes in the key transmission path, high-risk management will be implemented for the source account of the abnormal financial data. Abnormal management includes high-risk management, medium-risk management, and low-risk management. High-risk management includes freezing the source account of the abnormal financial data.

[0193] For example, if the concentration of abnormal fund aggregation nodes is greater than 50%, that is, a large amount of funds are concentrated in a single account, forming a high-risk fund accumulation; there are multiple inter-node loops in the key transmission path, that is, funds are transferred back and forth between multiple accounts, which may involve money laundering or illegal fund transfers. If any of the above conditions are met, it is judged to be high risk. Execute high-risk management: immediately freeze the source account of the abnormal fund aggregation node in the fund flow diagram to prevent further transfers; while the account is frozen, suspend the account's fund operations and notify the relevant financial management department; submit the frozen account to the internal audit department or judicial supervision agency for review.

[0194] High-risk management can prevent further capital outflow.

[0195] S532: If a critical transmission path involves multiple nodes or the capital outflow from a single node accounts for more than 40%, implement medium-risk management for the critical transmission path. Medium-risk management includes real-time monitoring of capital flows along the critical transmission path.

[0196] For example, a key transmission path involves multiple nodes, meaning that funds flow from a single source account to multiple accounts, which may pose a risk of fund splitting to evade regulation; a single node's fund outflow ratio is greater than 40%, meaning that funds primarily flow to a certain account, which may involve abnormal fund allocation or profit transfer. If any of the above conditions are met, the risk is determined to be medium. Implement medium-risk management: In the financial management system, set up transaction monitoring for key transmission paths to capture abnormal fund flows; if fund flows exceed the set threshold (such as a single transaction amount exceeding 1 million yuan or multiple transfers from the same source within 48 hours), trigger an alarm and record transaction details; regularly review fund flows for key transmission paths to ensure compliance with regulations.

[0197] Medium-risk management can help track capital dynamics.

[0198] S533: If the abnormal fund flow chart involves an account on the high-risk list, implement low-risk management. Low-risk management includes sending an early warning message to the regulatory authorities.

[0199] For example, if the transaction accounts involved in the funds flow are on the regulator's list of high-risk accounts but do not form abnormal fund concentrations or critical transmission paths, the risk is determined to be low. Implementing low-risk management: Transactions involving high-risk accounts are flagged in the financial system for subsequent analysis; transaction reports are automatically generated and submitted to financial regulators, including transaction time, amount, source and destination of funds, and account information (such as account opening institution and bank information). If transactions continue to occur in high-risk accounts, the risk can be upgraded to medium or high risk management. Low-risk management can improve risk prevention capabilities.

[0200] The above steps are conducive to transparent capital flow, secure transactions, and efficient financial management, providing strong support for the financial management of universities, enterprises, governments and other institutions.

[0201] In a possible implementation, exception management also includes:

[0202] S501, determine the source department corresponding to the abnormal financial data based on the abnormal funds flow diagram, generate an audit list of the related departments of the abnormal financial data, and implement transaction limit control measures for the accounts on the audit list.

[0203] For example, the abnormal funds flow chart can be used to identify the account with the largest inflows and trace its source. If the account's inflows account for more than 50%, the department to which it belongs may be the source of the abnormal financial data. Based on the hierarchical structure of the funds flow chart, the source department of the account's funds is determined. If the account receives funds from multiple departments, all related departments are marked as potential source departments. All direct transaction partners of the source department are added to the audit list. If a department experiences abnormal funds flow for three consecutive periods, it will be subject to enhanced audit. The audit list may include the department name, the abnormal accounts involved, the main inflow and outflow accounts, the abnormal transaction amount, whether there is circular trading, and whether there is any connection to the high-risk list. For high-risk accounts on the audit list (abnormal fund pooling and circular trading), the daily transaction limit is reduced to 20%, and the single transaction amount cannot exceed 500,000 yuan. For medium-risk accounts on the audit list (single fund outflow greater than 40%), the daily transaction limit is reduced to 50%, and the number of transactions per week is limited to 10. Low-risk accounts on the audit list (accounts on the high-risk list) only require monitoring, without limit adjustments.

[0204] S502, determine whether there are three or more levels of circular transactions in the abnormal capital flow diagram. If there are three or more levels of circular transactions in the abnormal capital flow diagram, start blockchain evidence storage and transmit it to the regulatory authorities.

[0205] For example, the funds path in the abnormal funds flow diagram can be traced to check whether the transaction chain is closed. If funds circulate between more than three accounts and the inflow ratio exceeds 50%, it is determined to be a circular transaction. The abnormal funds flow path, transaction amount, and transaction account can be recorded, and a blockchain hash value can be generated to facilitate data integrity. Asymmetric encryption algorithms can be used to protect transaction information security, and only regulatory authorities are authorized to access the stored evidence data. The stored evidence data is uploaded to the chain of custody in real time and synchronized with financial regulatory agencies.

[0206] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0207] Corresponding to the university financial data management method described in the above embodiment, the embodiment of the present application also provides a university financial data management system, and each unit of the system can implement each step of the university financial data management method.

[0208] The system includes:

[0209] The current financial identification sequence determination unit is configured to obtain a current financial data sequence and determine a current financial identification sequence based on the current financial data sequence. The current financial data sequence is obtained by chronologically arranging the financial data of universities within a certain period. The current financial identification sequence is used to represent the financial identification corresponding to each financial data in the current financial data sequence.

[0210] The prediction unit is used to predict each financial identifier of the next period of the current financial identifier sequence according to the current financial identifier sequence to obtain a predicted financial identifier sequence.

[0211] The actual financial identification sequence determination unit is used to obtain an actual financial data sequence and determine an actual financial identification sequence based on the actual financial data sequence, wherein the actual financial data sequence is the financial data actually collected in the next cycle of the current financial identification sequence.

[0212] The abnormality judgment unit is used to compare the actual financial identification sequence with the predicted financial identification sequence, and to construct an actual capital flow diagram based on the actual financial identification sequence, to judge whether there is abnormal financial data in the actual financial data sequence, and to obtain a judgment result.

[0213] The exception management execution unit is configured to execute exception management on the abnormal financial data if the judgment result indicates that abnormal financial data exists in the actual financial data sequence.

[0214] It should be noted that the information interaction, execution process, etc. between the above-mentioned units are based on the same concept as the method embodiment of this application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0215] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0216] An embodiment of the present application also provides a university financial data management device. The university financial data management device of this embodiment includes: at least one processor, at least one memory, and a computer program stored in the at least one memory and runnable on the at least one processor. When the processor executes the computer program, the university financial data management device implements the steps of any of the above-mentioned university financial data management method embodiments, or implements the functions of each unit in the above-mentioned system embodiments.

[0217] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to implement the present application. The one or more modules / units may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in the university financial data management device.

[0218] The university financial data management device can be a computing device such as a desktop computer, laptop, PDA, or cloud server. The university financial data management device can include, but is not limited to, a processor and memory. It can also include more or fewer components, or a combination of certain components, or different components. For example, it can also include input and output devices, network access devices, buses, etc.

[0219] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0220] In some embodiments, the memory may be an internal storage unit of the university financial data management device, such as a hard drive or memory of the device. In other embodiments, the memory may also be an external storage device of the device, such as a plug-in hard drive, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash memory card, etc. equipped with the device. Furthermore, the memory may include both the internal storage unit of the university financial data management device and an external storage device. The memory is used to store an operating system, application programs, a boot loader, data, and other programs, such as the program code of the computer program. The memory may also be used to temporarily store data that has been output or is about to be output.

[0221] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.

[0222] An embodiment of the present application provides a computer program product. When the computer program product is run on a university financial data management device, the university financial data management device is enabled to implement the steps of any of the above-mentioned method embodiments.

[0223] If the integrated unit is implemented as a software functional unit and sold or used as a standalone product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the process steps in the above-mentioned method embodiments by using a computer program to instruct the relevant hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium can include at least: any entity or system capable of carrying the computer program code to a university financial data management device, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signals, telecommunication signals, and software distribution media. Examples include USB flash drives, removable hard drives, magnetic disks, or optical disks. In some jurisdictions, based on legislation and patent practice, computer-readable media cannot be electric carrier signals or telecommunication signals.

[0224] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0225] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0226] In the embodiment provided in the present application, it should be understood that the disclosed university financial data management system, equipment and method can be implemented in other ways. For example, the university financial data management system and equipment embodiment described above are merely schematic, and for example, the division of the modules or units is merely a logical function division, and other division modes can be arranged during actual implementation, such as multiple units or components can be combined or can be integrated into another system, or some features can be ignored, or not executed. Another point, the coupling or direct coupling or communication connection between what is shown or discussed can be through some interfaces, the indirect coupling or communication connection of the system or unit can be electrical, mechanical or other forms.

[0227] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0228] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the scope of protection of the present application.

Claims

1. A method for managing financial data of a university, characterized in that: include: Obtain a current financial data sequence, and determine a current financial identification sequence based on the current financial data sequence; wherein the current financial data sequence is obtained by arranging the financial data of the university in chronological order within a certain period, the current financial identification sequence is used to represent the financial identification corresponding to each financial data in the current financial data sequence, and each financial identification in the current financial identification sequence includes a teaching cycle, a five-level classification, a funding attribute, and an amount. The teaching cycle is obtained by converting the time of the financial data into an academic year number, a semester stage, and a teaching week. The five-level classification is to divide the financial data into five levels: major category, medium category, minor category, special item, and sub-item according to the standards of the Ministry of Education. The funding attributes include the source type of funds and the destination type of funds. According to the current financial identification sequence, each financial identification of the next period of the current financial identification sequence is predicted to obtain a predicted financial identification sequence; wherein the predicted financial identification sequence includes a category prediction result and an amount prediction result, extracting the semester stage feature vector, teaching week feature vector, five-level classification feature vector, capital attribute feature vector and amount feature vector of each financial identification; based on the semester stage feature vector and five-level classification feature vector of each financial identification, using a classification model to predict the five-level classification of each financial identification of the next period of the current financial identification sequence to obtain a category prediction result; based on the teaching week feature vector, capital attribute feature vector and amount feature vector of each financial identification, using a regression model to predict the amount corresponding to each five-level classification in the category prediction result to obtain an amount prediction result; Acquire an actual financial data sequence, and determine an actual financial identification sequence based on the actual financial data sequence; wherein the actual financial data sequence is the financial data actually collected in the next period of the current financial identification sequence; Comparing the actual financial identification sequence with the predicted financial identification sequence, and constructing an actual capital flow diagram based on the actual financial identification sequence, determining whether there is abnormal financial data in the actual financial data sequence, and obtaining a determination result; If the judgment result indicates that abnormal financial data exists in the actual financial data sequence, performing abnormality management on the abnormal financial data; The teaching week feature vector, fund attribute feature vector, and amount feature vector of each financial identifier are used to predict the amount corresponding to each five-level classification in the category prediction result using a regression model to obtain an amount prediction result, including: Based on the fund attributes and fund attribute feature vectors of each financial identifier, a current fund flow graph is constructed, and based on the current fund flow graph, a graph convolutional network is used to generate fund flow features; wherein the nodes of the current fund flow graph are each financial identifier, and the edges of the current fund flow graph represent the flow direction of funds; Based on a preset time window, teaching week scale features, semester scale features, and academic year scale features are extracted from historical financial data. Based on the teaching week feature vector of each financial identifier, the teaching week scale features, the semester scale features, and the academic year scale features are fused through a temporal attention mechanism to obtain multi-scale features corresponding to each financial identifier. Based on the capital flow characteristics, the multi-scale characteristics corresponding to each financial identifier and the amount feature vector of each financial identifier, the regression model is used to predict the amount corresponding to each five-level classification in the category prediction results to obtain an amount prediction result; wherein, the regression model is a machine learning model.

2. The university financial data management method according to claim 1, wherein: The determining of a current financial identification sequence based on the current financial data sequence includes: Extracting key information of each financial data from the current financial data sequence according to an identification template; wherein the identification template includes a teaching period, a five-level classification, a fund attribute, and an amount; and the key information of each financial data includes the teaching period of each financial data, the five-level classification of each financial data, the fund attribute of each financial data, and the amount of each financial data; Filling the key information of each financial data into the identification template to obtain the financial identification corresponding to each financial data; The financial identifier corresponding to each financial data is arranged according to the arrangement order of each financial data in the current financial data sequence to obtain the current financial identifier sequence.

3. The university financial data management method according to claim 2, wherein: The step of predicting each financial identifier of the next cycle of the current financial identifier sequence according to the current financial identifier sequence to obtain a predicted financial identifier sequence includes: Encoding the teaching period of each financial identifier in the current financial identifier sequence to obtain a semester phase feature vector and a teaching week feature vector for each financial identifier; Performing category coding on the five-level classification of each financial identifier in the current financial identifier sequence to obtain a five-level classification feature vector for each financial identifier; wherein the five-level classification feature vector includes a major category feature vector, a medium category feature vector, a minor category feature vector, a special feature vector, and a sub-item feature vector; Converting the fund attribute in each financial identifier in the current financial identifier sequence into a fund attribute feature vector through a type mapping matrix; Normalizing the amount in each financial identifier in the current financial identifier sequence to obtain an amount feature vector for each financial identifier; Based on the semester stage feature vector and the five-class classification feature vector of each financial identifier, using the classification model to predict the five-class classification of each financial identifier in the next period of the current financial identifier sequence, to obtain a class prediction result; Based on the teaching week feature vector, fund attribute feature vector, and amount feature vector of each financial identifier, a regression model is used to predict the amount corresponding to each five-level classification in the category prediction result to obtain an amount prediction result; The category prediction results and the amount prediction results are arranged in the time sequence of the next cycle of the current financial identification sequence to obtain the predicted financial identification sequence.

4. The university financial data management method according to claim 3, wherein: The five-class classification of each financial identifier in the next period of the current financial identifier sequence is predicted using a classification model based on the semester stage feature vector and the five-class classification feature vector of each financial identifier to obtain a category prediction result, including: For each financial identifier, the following operations are performed: based on the hierarchical attention mechanism, the weight of each eigenvector in the five-level classification feature vector is calculated, and the weight of each eigenvector and each eigenvector are weighted summed to obtain a classification comprehensive feature vector; the classification weight corresponding to the semester stage eigenvector is selected from the classification weight matrix; wherein the classification weight matrix is pre-set and includes the classification weights corresponding to the eigenvectors of different semester stages; Based on the classification weight and classification comprehensive feature vector of each financial identifier, the classification model is used to predict the five-level classification of each financial identifier in the next cycle of the current financial identifier sequence to obtain a category prediction result; wherein, the classification model is a machine learning model.

5. The university financial data management method according to claim 3, wherein: The method further comprises: Calculate the budget execution rate and remaining budget amount corresponding to each five-level classification in the category forecast results; If the budget execution rate exceeds the budget threshold, the amount of the five-level classification that exceeds the budget threshold is reduced according to the excess ratio of the budget execution rate to obtain the reduced amount of the five-level classification that exceeds the budget threshold; Determining a budget cap for the five-level classification that exceeds the budget threshold based on the remaining budget amounts corresponding to the five-level classification that exceeds the budget threshold; Based on the budget upper limit of the five-level classification that exceeds the budget threshold, the attenuation amount of the five-level classification that exceeds the budget threshold is corrected to obtain the corrected amount of the five-level classification that exceeds the budget threshold.

6. The university financial data management method according to claim 1, characterized in that: The actual financial identifier sequence is compared with the predicted financial identifier sequence, and an actual capital flow diagram is constructed based on the actual financial identifier sequence to determine whether there is abnormal financial data in the actual financial data sequence, and obtain a determination result, including: Comparing each financial identifier in the actual financial identifier sequence with each financial identifier in the predicted financial identifier sequence one by one, calculating the difference between each financial identifier in the actual financial identifier sequence and each financial identifier in the predicted financial identifier sequence, and obtaining a set of difference values; Comparing each difference in the difference set with a difference threshold, and if each difference in the difference set is within the difference threshold, determining that no abnormal financial data exists in the actual financial data sequence; wherein the difference threshold changes dynamically according to the teaching calendar of the university; If at least one difference in the difference collection is not within the difference threshold, the actual funds flow graph is constructed based on the actual financial data sequence, and the funds flow pattern analysis is performed on the actual funds flow graph. If there is any abnormal feature in the actual funds flow graph, it is determined that abnormal financial data exists in the actual financial data sequence; wherein, the actual funds flow graph is a three-layer funds flow graph, the nodes of the actual funds flow graph are the accounts of each financial identifier in the actual financial identifier sequence, the edges of the actual funds flow graph represent the flow direction of funds, and abnormal features include a closed-loop funds chain, a single node funds outflow accounting for more than 40%, and direct transactions with high-risk list accounts.

7. The university financial data management method according to claim 1, characterized in that: If the judgment result indicates that abnormal financial data exists in the actual financial data sequence, performing abnormality management on the abnormal financial data includes: If the judgment result indicates that abnormal financial data exists in the actual financial data sequence, an abnormal capital flow diagram is constructed with the abnormal financial data as the center; wherein the abnormal capital flow diagram is a five-layer capital flow diagram; Identify abnormal fund aggregation nodes and key propagation paths in the abnormal fund flow graph to obtain graph analysis results; Exception management is performed on the abnormal financial data according to the graph analysis results.

8. The university financial data management method according to claim 7, wherein: The performing abnormality management on the abnormal financial data according to the graph analysis result includes: If the concentration of the abnormal fund aggregation node exceeds 50% or there are multiple inter-node loops in the key propagation path, high-risk management is performed on the source account of the abnormal financial data; wherein the abnormal management includes high-risk management, medium-risk management and low-risk management, and the high-risk management includes freezing the source account of the abnormal financial data; If the key transmission path involves multiple nodes or the capital outflow of a single node accounts for more than 40%, medium risk management will be implemented for the key transmission path; wherein, the medium risk management includes real-time monitoring of the capital flow on the key transmission path; If the abnormal funds flow diagram involves an account on the high-risk list, low-risk management is implemented; wherein, the low-risk management includes sending an early warning message to the regulatory authorities.

9. A university financial data management system, characterized in that: Used to implement the university financial data management method according to any one of claims 1 to 8, the university financial data management system comprises: An acquisition unit is configured to acquire a current financial data sequence and determine a current financial identification sequence based on the current financial data sequence; wherein the current financial data sequence is obtained by chronologically arranging the financial data of colleges and universities within a certain period, and the current financial identification sequence is used to represent the financial identification corresponding to each financial data in the current financial data sequence; A prediction unit, configured to predict each financial identifier of a next period of the current financial identifier sequence based on the current financial identifier sequence to obtain a predicted financial identifier sequence; An actual financial identification sequence determining unit, configured to obtain an actual financial data sequence and determine an actual financial identification sequence based on the actual financial data sequence; wherein the actual financial data sequence is the financial data actually collected in the next period of the current financial identification sequence; a judgment unit, configured to compare the actual financial identification sequence with the predicted financial identification sequence, construct an actual capital flow diagram based on the actual financial identification sequence, judge whether there is abnormal financial data in the actual financial data sequence, and obtain a judgment result; An exception management unit is configured to perform exception management on the abnormal financial data if the judgment result indicates that abnormal financial data exists in the actual financial data sequence.

Images