College accounting process-oriented information management system and method

Abstract

The invention relates to the technical field of intelligent financial management, and discloses an information management system and method for a college accounting process, and the method comprises the steps: collecting the historical financial data of each school to form a spatio-temporal data matrix; inputting the data matrix into a space-time attention network, and capturing a space-time dependency relationship of fund flow; analyzing the spatial-temporal characteristics by using a seasonal prediction model, and predicting seasonal fund demands; constructing a campus demand model and a risk distribution model, and forming a multi-campus comprehensive evaluation matrix; solving the multi-objective function through a space-time optimization algorithm, and outputting an optimal fund configuration scheme; fund configuration is dynamically adjusted through real-time monitoring; according to the method, the time rule and the space distribution characteristics of fund flow can be captured at the same time, periodic changes of college fund requirements are effectively processed, intelligent configuration and dynamic adjustment of funds in multiple schools are achieved, and the fund configuration efficiency and the risk control capability are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent financial management, and more specifically, it relates to an information management system and method for university accounting processes. Background Art

[0002] With the development of higher education, multi-campus operation has become a common choice for many universities. While this operation mode improves the utilization efficiency of educational resources, it also brings new challenges to university financial management. Currently, university accounting information management mainly adopts traditional budget allocation models and simple historical data analysis methods, lacking comprehensive consideration of the spatio-temporal characteristics of multi-campus capital flows, resulting in problems such as low capital allocation efficiency, coexistence of resource idleness and shortage.

[0003] Traditional university financial management systems usually use linear regression or simple time series analysis for capital forecasting. Although these methods are simple to implement, they cannot effectively capture the complex spatio-temporal dependence relationships of university capital flows. For example, one campus needs a large amount of funds for teaching equipment renewal at the beginning of the semester, while another campus faces concentrated expenditures on research projects in the middle of the semester. Such spatio-temporal differences severely limit the forecasting accuracy of traditional methods, thus affecting the scientific nature of financial decisions.

[0004] Existing technologies also have obvious deficiencies in dealing with the seasonal characteristics of university capital demands. University capital usage shows obvious semester periodic changes, such as peak equipment procurement during the start of the semester and concentrated project settlements at the end of the semester. However, traditional financial management systems lack the ability to effectively model such periodic patterns and cannot achieve forward-looking capital planning, resulting in capital allocation often lagging behind actual demands.

[0005] In addition, existing university financial systems also lack a perfect risk assessment mechanism and multi-objective optimization capabilities. During the capital allocation process, multiple objectives such as maximizing efficiency, minimizing risk, and controlling costs need to be considered simultaneously, and are restricted by various constraint conditions such as liquidity, budget, and policies. However, traditional methods often adopt static budget allocation methods, unable to dynamically adjust according to the real-time demand changes and risk situations of each campus, and lacking effective risk quantification models and diversification strategies, which affect the stability and scientific nature of university financial management. Summary of the Invention

[0006] The present invention provides an information management system and method for university accounting processes, solving the technical problems in related technologies that lack comprehensive consideration of the spatio-temporal characteristics of multi-campus capital flows, resulting in low capital allocation efficiency, resource idleness and shortage.

[0007] The present invention provides an information management method for university accounting processes, including the following steps: Collect the historical financial data of each campus of the university to form a spatio-temporal data matrix; Input the spatio-temporal data matrix into a spatio-temporal attention network to capture the spatio-temporal dependence relationship of capital flows and output spatio-temporal feature representations; Use a seasonal prediction model to analyze the spatio-temporal feature representations and predict seasonal capital demands; Based on the seasonal capital demands, analyze the capital demand characteristics of each campus, construct a campus demand model and a risk distribution model, form a multi-campus comprehensive evaluation matrix, and solve the multi-objective function through a spatio-temporal optimization algorithm to output an optimal capital allocation plan; Monitor the capital usage of each campus through a real-time monitoring module, and dynamically adjust the capital allocation based on the multi-campus comprehensive evaluation matrix and the spatio-temporal optimization algorithm to maintain the optimality of the allocation plan.

[0008] In a preferred embodiment, the seasonal prediction model includes a periodic feature extraction module and a trend prediction module. The periodic feature extraction module uses the Fourier transform method to identify the periodic patterns of capital usage, and the trend prediction module uses a long short-term memory network structure to learn the long-term trend changes of capital demands.

[0009] In a preferred embodiment, the spatio-temporal optimization algorithm uses a multi-objective optimization method, including an objective function construction module and a constraint handling module. The objective function construction module takes maximizing capital allocation efficiency, minimizing risk, and minimizing cost as optimization objectives, and the constraint handling module uses the Lagrange multiplier method to handle liquidity constraints, budget constraints, and policy constraints.

[0010] In a preferred embodiment, the real-time monitoring module includes a data collection sub-module and an anomaly detection sub-module. The data collection sub-module connects to the API interface of the university's financial system to obtain the capital flow data of each campus in real time, and the anomaly detection sub-module uses the statistical process control method to set the upper and lower limits of the control of capital usage deviations.

[0011] In a preferred embodiment, the historical financial data includes capital inflow data, capital outflow data, and budget execution data.

[0012] In a preferred embodiment, the periodic feature extraction module can automatically detect semester cycles, monthly cycles, and weekly cycles, and the trend prediction module contains 256 hidden units.

[0013] In a preferred embodiment, the objective function construction module uses a weighted summation method to construct a comprehensive objective function, and the constraint conditions processed by the constraint handling module include liquidity constraints, budget constraints, and policy constraints.

[0014] In a preferred embodiment, the data update frequency of the data collection sub-module is once per hour, and the anomaly detection sub-module triggers an alarm when the actual fund usage exceeds 15% plus or minus the predicted value.

[0015] In a preferred embodiment, the time attention layer contains 8 attention heads, each with a dimension of 64, and the spatial attention layer takes each campus as a graph node and the fund flow relationship between campuses as graph edges.

[0016] In a preferred embodiment, an information management system for university accounting processes is used to execute an information management method for university accounting processes, including: A data collection module for collecting historical financial data of each campus of a university to form a spatio-temporal data matrix; A spatio-temporal feature extraction module for capturing the spatio-temporal dependence relationship of fund flows through a spatio-temporal attention network and outputting a spatio-temporal feature representation; A seasonal prediction module for analyzing the spatio-temporal feature representation and predicting seasonal fund demands; A comprehensive evaluation module for analyzing the fund demand characteristics of each campus based on the seasonal fund demands, constructing a campus demand model and a risk distribution model, and forming a multi-campus comprehensive evaluation matrix; A fund allocation module for solving a multi-objective function through a spatio-temporal optimization algorithm, outputting an optimal fund allocation plan, and monitoring the fund usage of each campus through a real-time monitoring module, and dynamically adjusting the fund allocation based on the multi-campus comprehensive evaluation matrix and the spatio-temporal optimization algorithm to maintain the optimality of the allocation plan.

[0017] The beneficial effects of the present invention are as follows: Through the application of the spatio-temporal attention mechanism, this method can simultaneously capture the time pattern and spatial distribution characteristics of fund flows, significantly improving the accuracy of fund flow prediction and solving the technical problem that traditional methods lack consideration of spatio-temporal dimensions.

[0018] The introduction of the seasonal prediction model enables this method to effectively handle the periodic characteristics of university fund demands, improving the forward-looking and timeliness of fund allocation and solving the technical problem that existing systems are difficult to handle seasonal fluctuations.

[0019] The application of the dynamic balance optimization algorithm realizes the intelligent allocation of multi-campus funds, making optimal allocations according to the real-time demands and risk status of each campus, and solving the limitations of traditional static budget allocation methods.

[0020] The establishment of the real-time monitoring and adjustment mechanism ensures that the fund allocation strategy can be dynamically adjusted according to the actual situation, improving the stability of university financial management and the risk control ability. Description of the Drawings

[0021] Figure 1 It is a flowchart of an information management method for the accounting process in universities according to the present invention; Figure 2 It is a bar chart of the comparative analysis of the prediction accuracy according to the present invention; Figure 3 It is a line chart of the changing trend of the capital allocation efficiency according to the present invention; Figure 4 It is a radar chart of the system performance indicators according to the present invention. Detailed implementation manners

[0022] Now, the subject matter described herein will be discussed with reference to exemplary embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein. Without departing from the scope of protection of the content of this specification, changes can be made to the functions and arrangements of the elements discussed. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described in some examples can also be combined in other examples.

[0023] In at least one embodiment of the present invention, an information management method for the accounting process in universities is disclosed. As Figure 1 shown, it includes the following steps: Step 1: Collect the historical financial data of each campus of the university to form a spatio-temporal data matrix; Specifically, it includes the following steps: Step 1.1: Collect the historical financial data of each campus of the university, including capital inflow data, capital outflow data, and budget execution data, to form a spatio-temporal data matrix: ; Among them, represents the collected financial data matrix; represents the set of real numbers; is the length of the time dimension, indicating the number of historical data time points collected; represents the number of campuses; is the input feature dimension, indicating the number of types of financial data collected for each campus at each time point.

[0024] This matrix contains the complete time series and spatial distribution information of the university's financial data, providing a data basis for subsequent analysis.

[0025] Step 1.2: Construct a spatio-temporal attention network to capture the spatio-temporal dependence relationship of capital flow through an attention calculation model: ; Among them, represents the attention calculation model; Represents the query matrix, which contains the financial data features to be analyzed currently; Represents the key-value matrix, which contains the key features of historical financial data; Represents the numerical matrix, which contains the actual financial data values; Represents the transpose multiplication of the query matrix and the key-value matrix, which is used to calculate the correlation strength between different time points or different campuses; Represents the square root of the dimension of the key-value vector, which is used as a scaling factor to prevent the dot product result from being too large and causing the softmax gradient to disappear; Represents the normalized exponential function, which converts the correlation strength into attention weights in the form of a probability distribution; the calculation result of the entire formula is the weighted numerical matrix, which reflects the financial data feature representation considering the spatio-temporal dependence relationship.

[0026] The spatio-temporal attention network includes a time attention layer and a space attention layer.

[0027] The time attention layer adopts a multi-head attention structure, which contains 8 attention heads, and the dimension of each attention head is 64, which is used to capture the dependence relationship between different time steps.

[0028] The space attention layer adopts a graph attention network structure, which takes each campus as a graph node and the fund flow relationship between campuses as a graph edge, and learns the spatial dependence relationship between campuses through graph convolution operations.

[0029] In a specific college financial management scenario, the application example of the spatio-temporal attention network is as follows: When processing the monthly fund flow data of a multi-campus university, the time attention layer can identify the peak periods of fund use in September at the beginning of the semester and January at the end of the semester, and the space attention layer can discover the fund allocation pattern between the main campus and the branch campuses, so as to provide an accurate feature representation for subsequent fund prediction.

[0030] In some embodiments, the spatio-temporal attention network can also adopt an adaptive weight adjustment strategy to dynamically adjust the attention weights according to the fund scale of different campuses. For example, for the main campus with a large fund scale, a higher attention weight can be assigned, while for the branch campus with a small fund scale, a relatively lower weight can be adopted, so as to improve the overall prediction accuracy.

[0031] In addition, in some embodiments, the space attention layer can further consider the geographical distance factor between campuses, and the weight of the fund flow relationship between campuses with a shorter distance is higher, while the weight between campuses with a longer distance is relatively lower, so that the actual fund allocation cost can be better reflected.

[0032] Step 1.3, input the spatio-temporal data matrix into the spatio-temporal attention network, and output the spatio-temporal feature representation: ; Among them, represents the spatio-temporal feature representation; is the length of the time dimension, representing the number of historical data time points collected; represents the number of campus areas; is the output feature dimension, representing the feature vector dimension of each campus area at each time point after being processed by the spatio-temporal attention network. These features contain dependency information in both the time and space dimensions, providing a more accurate feature representation for subsequent seasonal prediction.

[0033] Step 2: Input the spatio-temporal data matrix into the spatio-temporal attention network to capture the spatio-temporal dependencies of fund flows and output the spatio-temporal feature representation; Specifically, it includes the following steps: Step 2.1: Based on the historical fund usage patterns, construct a seasonal prediction model, which uses the time attention algorithm to analyze the semester periodic characteristics.

[0034] The seasonal prediction model is an innovative improvement of this application. This model includes a periodic feature extraction module and a trend prediction module.

[0035] The periodic feature extraction module uses the Fourier transform method to identify the periodic patterns of fund usage and can automatically detect semester cycles, monthly cycles, and weekly cycles.

[0036] The trend prediction module adopts the Long Short-Term Memory (LSTM) structure, which contains 256 hidden units and is used to learn the long-term trend changes of fund demand.

[0037] In a specific application scenario, when predicting the fund demand of a certain university in the next semester, the periodic feature extraction module can identify periodic patterns such as the equipment procurement peak in the two weeks before the start of each semester, the daily operation stable period in the middle of the semester, and the project settlement concentration period at the end of the semester in the university's history. The trend prediction module can then combine the school's development plan to predict the overall growth trend of fund demand.

[0038] In some embodiments, the seasonal prediction model can also integrate an external factor impact assessment module, which can analyze the impact of external factors such as policy changes, economic environment fluctuations, and emergencies on the fund demand of universities. For example, when a new scientific research funding policy is introduced by the country, this module can automatically adjust the predicted value of scientific research funds.

[0039] In addition, in some embodiments, the trend prediction module can adopt an ensemble learning method, combine the prediction results of multiple LSTM networks, and improve the stability and accuracy of the prediction through weighted average or voting.

[0040] Step 2.2. Analyze the spatio-temporal feature representation using the seasonal prediction model and predict the seasonal fund demand through the calculation formula: ; wherein, represents the prediction result of the seasonal fund demand, including the predicted fund demand amounts for each campus in the future time period; represents the time attention function, which is used for weighted calculation of different time features; represents the historical pattern vector, with the dimension of , represents the length of the historical data, represents the pattern feature dimension, including the periodic pattern information in the historical financial data of the university; represents the semester cycle vector, with the dimension of , represents the number of semester cycles, represents the cycle feature dimension, recording the fund fluctuation characteristics at key time points such as the beginning, middle, and end of the semester; represents the external influence factor vector, with the dimension of , represents the number of external factors, represents the factor feature dimension, including the influence degree of external factors such as policy changes, economic environment fluctuations, and emergencies on the fund demand.

[0041] Step 2.3. Output the prediction results of the seasonal fund demand for each campus in the future time period ; wherein, is the prediction time length, indicating how long in the future needs to be predicted; represents the number of campuses, indicating the total number of different campuses owned by the university; represents the set of real numbers, indicating that the prediction result is a real number matrix; Each element in the matrix represents the predicted fund demand amount for a specific campus at a specific future time point, in yuan. This prediction result will be used as the input data for the subsequent multi-campus demand analysis and risk assessment.

[0042] Step 3. Analyze the spatio-temporal feature representation using the seasonal prediction model and predict the seasonal fund demand; Specifically, it includes the following steps: Step 3.1. Based on the prediction result of the seasonal fund demand , analyze the fund demand characteristics of each campus and construct a campus demand model: ; Among them, represents the campus demand model; represents the number of campuses; represents the set of real numbers; represents the demand feature dimension, which is the feature vector dimension of each campus in the campus demand model.

[0043] This model is obtained through function calculation: ; Among them, represents the campus demand model; represents the demand feature extraction function, which comprehensively analyzes various features using a weighted fusion strategy. This function can automatically identify and integrate demand feature information from different sources; represents the seasonal fund demand prediction result, with a dimension of , which contains the predicted fund demand amounts for each campus within the future time period; represents the historical demand pattern, with a dimension of , represents the historical data time length, which records the past fund usage rules and patterns of each campus; represents the campus priority vector, which reflects the importance and priority levels of each campus in the overall school development strategy; The demand feature dimension output by the model contains multiple key indicators, specifically including: Demand intensity coefficient , demand urgency , demand elasticity coefficient , demand stability etc.

[0044] This model not only quantifies the fund demand intensity of each campus but also accurately depicts the time distribution characteristics of the demand. For example, for teaching campuses, the model can identify that their fund demands are mainly concentrated at the beginning of the semester (equipment renewal and textbook procurement) and during the semester (daily operations), with relatively low demand elasticity; while for research campuses, the fund demands may be more discrete and volatile, closely related to the research project cycle.

[0045] The model uses exponential smoothing and seasonal adjustment algorithms to process time series data to ensure accurate modeling of the demand time distribution for various campuses.

[0046] Step 3.2, use a risk quantification algorithm to evaluate the financial risk status of each campus and construct a risk distribution model: ; Among them, represents the risk distribution model; Indicates the number of campus areas. Indicates the number of risk indicators.

[0047] This model includes a comprehensive assessment of liquidity risk, credit risk, and operational risk.

[0048] Specifically, liquidity risk is evaluated through the funds flow ratio and the cash coverage ratio, reflecting the campus's ability to meet short-term funding needs: ; ; Among them, Indicates the funds flow ratio; Indicates the current assets of the campus, including assets that can be quickly liquidated such as cash, short-term investments, and accounts receivable; Indicates the current liabilities of the campus, including debts that need to be repaid in the short term such as short-term borrowings and accounts payable; Indicates the cash coverage ratio; Indicates the cash held by the campus and short-term investments that can be quickly converted into cash of a known amount; Indicates the total monthly recurring expenses of the campus.

[0049] Credit risk is quantified through the historical default rate and the debt coverage ratio, measuring the campus's ability to repay debts and its creditworthiness: ; Among them, Indicates the historical default rate, reflecting the frequency of the campus's past failure to fulfill its financial obligations on time; Indicates the net income of the campus, that is, the remaining funds after subtracting the total expenses from the total revenue; Indicates the total amount of debts that the campus needs to repay, including principal and interest.

[0050] Operational risk is evaluated through the frequency of internal control failures and the process complexity index to reflect the potential financial risks in the campus's internal management process.

[0051] The risk quantification algorithm uses the Analytic Hierarchy Process (AHP) to determine the weights of each risk indicator and generates the probability distribution under different risk scenarios through Monte Carlo simulation, finally forming a risk distribution model: ; Among them, Indicates the risk distribution model; Indicates the funds flow ratio; Indicates the cash coverage ratio; represents the historical default rate; represents the historical default rate; represents the frequency of internal control failure; represents the process complexity index; represents the risk index weight vector. This model can accurately characterize the vulnerability of each campus in different risk dimensions and provide a risk control basis for subsequent capital allocation decisions.

[0052] Step 3.3, combine the campus demand model and the risk distribution model to form a multi-campus comprehensive evaluation matrix: ; where is the number of campuses, representing the total number of different campuses owned by the university; is the number of evaluation indicators, including multi-dimensional evaluation indicators such as the intensity of capital demand, the urgency of demand, the liquidity risk index, the credit risk score, and the operational risk level; represents the set of real numbers, indicating that the evaluation result is a real number matrix; Each element in the matrix represents the quantitative score of a specific campus on a specific evaluation indicator, providing a decision-making basis for subsequent optimal capital allocation.

[0053] Step 4, based on the seasonal capital demand, analyze the capital demand characteristics of each campus, construct a campus demand model and a risk distribution model, form a multi-campus comprehensive evaluation matrix, and solve the multi-objective function through a spatio-temporal optimization algorithm to output the optimal capital allocation plan; Specifically, it includes the following steps: Step 4.1, construct a spatio-temporal optimization algorithm, which takes the multi-campus comprehensive evaluation matrix as input and considers liquidity constraints and capital transfer costs; The spatio-temporal optimization algorithm is the core innovation of this application. This algorithm adopts a multi-objective optimization method, including an objective function construction module and a constraint condition processing module.

[0054] The objective function construction module maximizes the capital allocation efficiency, minimizes the risk, and minimizes the cost as optimization objectives, and constructs a comprehensive objective function using the weighted summation method.

[0055] The constraint condition processing module uses the Lagrange multiplier method to process multiple constraint conditions such as liquidity constraints, budget constraints, and policy constraints.

[0056] In a specific university financial management scenario, when a multi-campus university needs to allocate 10 million yuan of special funds among the main campus, the east campus, and the west campus, the spatio-temporal optimization algorithm can comprehensively consider the urgency, risk level, and capital transfer costs of each campus and automatically calculate the optimal allocation plan: 4 million yuan is allocated for equipment renewal in the main campus, 3.5 million yuan for infrastructure construction in the east campus, and 2.5 million yuan for the initiation of scientific research projects in the west campus. At the same time, it is ensured that each campus maintains a working capital reserve of not less than 500,000 yuan.

[0057] In some embodiments, the spatio-temporal optimization algorithm can also adopt a dynamic weight adjustment strategy to automatically adjust the weights of various optimization objectives according to the degree of capital tightness in different periods. For example, in a period when funds are relatively abundant, the weight of the risk minimization objective can be increased, while in a period when funds are tight, the weight of the efficiency maximization objective can be increased.

[0058] In addition, in some embodiments, the constraint condition processing module can adopt an adaptive constraint relaxation method. When strict constraint conditions lead to no feasible solution, some non-critical constraints can be appropriately relaxed to ensure that the algorithm can output a feasible capital allocation plan.

[0059] Step 4.2, apply the spatio-temporal optimization algorithm to achieve the optimal allocation of funds through a calculation formula: ; Among them, represents the optimal fund balance plan, which is a matrix with a dimension of , represents the number of campuses, represents the prediction time length, and each element in the matrix represents the amount of funds that should be allocated to a specific campus at a specific time point; represents the spatio-temporal optimization function, which is used to calculate the optimal fund allocation plan by comprehensively considering various factors; represents the campus demand model, with a dimension of , represents the demand feature dimension, which includes feature information such as the fund demand intensity, urgency, and elasticity coefficient of each campus; represents the risk distribution model, with a dimension of , represents the number of risk indicators, which includes evaluation indicators such as the liquidity risk, credit risk, and operational risk of each campus; represents the liquidity constraint vector; represents the fund transfer cost matrix, including handling fees, time costs, and operation costs, etc.

[0060] Step 4.3, output the optimal fund allocation plan , which specifies the amount of funds allocated to each campus in the future time period. Among them, represents the optimal fund allocation plan matrix, which contains the final fund allocation decision calculated by the system; represents the set of real numbers, indicating that the values in the allocation plan are real numbers; $N$ represents the number of campuses, which is the total number of different campuses owned by a university; $T$ represents the length of the prediction time period, indicating how far into the future the fund allocation plan is needed.

[0061] Step 5: Monitor the fund usage of each campus through the real-time monitoring module. Based on the multi-campus comprehensive evaluation matrix and the spatio-temporal optimization algorithm, dynamically adjust the fund allocation to maintain the optimality of the allocation plan; Specifically, it includes the following steps: Step 5.1: Construct a real-time monitoring module to continuously collect data on the fund usage of each campus and changes in the external environment.

[0062] The real-time monitoring module includes a data collection sub-module and an anomaly detection sub-module.

[0063] The data collection sub-module connects to the API interface of the university's financial system to obtain real-time data on the inflow and outflow of funds, budget execution progress, and changes in external policies for each campus. The data update frequency is once per hour.

[0064] The anomaly detection sub-module uses statistical process control methods to set the upper and lower control limits for fund usage deviations. When the actual fund usage exceeds ±15% of the predicted value, an early warning is triggered.

[0065] In specific applications, when the actual fund expenditure of a certain campus exceeds 20% of the predicted value within 3 consecutive hours, the real-time monitoring module will automatically send out an early warning signal and start the dynamic adjustment algorithm to recalculate the fund allocation plan to ensure that the balance of the overall fund pool is not affected.

[0066] Step 5.2: When the monitoring data shows a significant deviation between the actual situation and the predicted result, trigger the dynamic adjustment algorithm to recalculate the fund allocation plan; Specifically, the dynamic adjustment algorithm is an adaptive optimization algorithm that determines whether to start the adjustment process through the formula: ; where $D$ represents the dynamic adjustment algorithm; $\Delta$ represents the deviation matrix between the actual fund usage and the predicted value; $S$ represents the time sensitivity vector; $R$ is the resource availability matrix, indicating the feasibility of fund allocation between campuses; $f$ represents the decision function.

[0067] When the condition is met, the system determines it as a significant deviation and triggers an adjustment. Among them, $i$ represents campus in the time period The absolute value of the deviation within, that is, the absolute percentage difference between the actual fund usage and the predicted value; Represents the campus The deviation threshold, which is the minimum deviation percentage that triggers an adjustment. Different campuses can set different thresholds to reflect their specific risk tolerance; Represents the campus During the time period The duration of continuous deviation, which measures the length of time the deviation persists; Represents the minimum duration window, which is the shortest duration requirement for triggering an adjustment, preventing unnecessary adjustments due to temporary fluctuations; Represents the logical "AND" operation, indicating that the adjustment mechanism will only be triggered when both conditions are met: the deviation exceeds the threshold and the duration exceeds the time window.

[0068] This dual-trigger mechanism that combines the deviation magnitude and duration avoids frequent adjustments caused by short-term random fluctuations, effectively filters out noise signals, and improves the stability and reliability of the fund allocation system.

[0069] The dynamic adjustment algorithm adopts a hierarchical response strategy and takes different intensity adjustment measures according to the degree of deviation: When the deviation is between 10% and 15%, a mild adjustment strategy is adopted, and only a small amount of funds are redistributed between adjacent campuses; When the deviation is between 15% and 25%, a moderate adjustment strategy is adopted, allowing cross-regional fund allocation and the use of short-term fund buffer pools; When the deviation exceeds 25%, a strong adjustment strategy is adopted, starting a global fund rebalancing and possibly using strategic reserve funds.

[0070] In addition, the dynamic adjustment algorithm also has a learning ability. By recording historical adjustment effects and actual deviation patterns, it continuously optimizes the trigger conditions and adjustment parameters to improve the self-adaptability of the algorithm.

[0071] Step 5.3, based on the calculation results of the dynamic adjustment algorithm, the system outputs the adjusted fund allocation strategy: ; Among them, Represents the adjusted fund allocation strategy; Represents the number of campuses, Represents the remaining predicted time length.

[0072] The adjustment strategy is generated through the formula: ; Among them, Denotes the adjusted fund allocation plan, where each element in the matrix represents the adjusted fund allocation amount for a specific campus during the remaining time; Denotes the dynamic adjustment function, which adopts an incremental optimization method and only makes local adjustments to the campuses and time points with deviations exceeding the threshold, avoiding large fluctuations in the overall plan; Denotes the original optimal allocation plan; Denotes the deviation matrix between the actual execution and the predicted value; Denotes the risk tolerance threshold vector, and different adjustment trigger thresholds are set according to the risk tolerance of each campus.

[0073] This adjustment strategy not only includes the reallocation of funds among campuses but also three buffer mechanisms: Short-term liquidity buffer pool , used to cope with the fund fluctuations within 7 days; Among them, Denotes the fund size of the short-term liquidity buffer pool; Denotes the short-term buffer coefficient, which is automatically adjusted according to the historical fund volatility; Denotes the variance of the deviation matrix, reflecting the fluctuation degree between the actual use of funds and the predicted value.

[0074] Medium-term strategy adjustment pool , used to cope with the demand changes within 1 - 3 months; Among them, Denotes the fund size of the medium-term strategy adjustment pool; Denotes the medium-term buffer coefficient, which is automatically adjusted according to the historical demand volatility; Denotes the covariance of the campus demand matrix, reflecting the correlation and fluctuation amplitude of the fund demands among different campuses.

[0075] Long-term risk hedging pool , used to cope with systematic risks; Among them, Denotes the fund size of the long-term risk hedging pool; Denotes the long-term buffer coefficient, which is automatically adjusted according to the systematic risk assessment results; Denotes the risk distribution model, which quantifies the exposure degree of each campus in different risk dimensions.

[0076] The parameters of these three buffer pools , , Are the short, medium, and long-term buffer coefficients respectively, and the system will automatically adjust according to the volatility characteristics analyzed from historical data to ensure that the buffer pool size matches the actual risk level.

[0077] The system evaluates the effectiveness of the adjusted configuration strategy daily, calculates the risk dispersion index, and when is below 0.75, it triggers the rebalancing mechanism of the fund pool: ; Among them, represents the risk dispersion index, which is used to measure the degree of risk dispersion among campuses. The value range is [0, 1], and the closer the value is to 1, the higher the degree of risk dispersion; represents the number of campuses, which represents the total number of different campuses owned by the university; represents the th campus risk score, which is a quantitative indicator that comprehensively considers liquidity risk, credit risk, and operational risk; represents the average value of the risk scores of all campuses; represents the product of the number of campuses and the square of the average risk, which is used as a standardization factor; -0.75 is the risk dispersion threshold set by the system. When is below this value, it indicates that the risk is too concentrated in some campuses, and it is necessary to trigger the rebalancing mechanism of the fund pool to reallocate the risk.

[0078] This multi-level dynamic adjustment ensures that the fund pool can still maintain a stable balance state in the face of changes in campus needs and external environmental impacts. At the same time, through the risk dispersion strategy at different risk levels, it effectively reduces the concentration risk and improves the resilience and adaptability of the overall fund allocation.

[0079] Application example of this implementation method: Application scenario: A comprehensive university has three campuses: the main campus, the east campus, and the west campus, with a total fund scale of about 200 million yuan. The school faces problems such as large differences in fund requirements among different campuses, obvious seasonal fund fluctuations, and room for improvement in fund allocation efficiency. The school decides to adopt the information management method for the accounting process of universities provided in this application to optimize fund management.

[0080] Example of the implementation process: Data collection and processing stage: The system collected the school's financial data for the past 3 years, including the monthly fund inflow and outflow records, budget execution, equipment procurement expenditures, personnel expenses, etc. of each campus, totaling 108 months of data.

[0081] Construct these data into a spatio-temporal data matrix, with the time dimension , the number of campuses , and the input feature dimension .

[0082] Spatiotemporal Feature Extraction Phase: The spatiotemporal attention network analyzes historical data. The temporal attention layer identifies obvious seasonal features in the school's fund usage: the fund expenditure reaches the annual peak in September at the beginning of each school year, accounting for 18% of the total annual expenditure; The fund expenditures at the end of the semester in January and July are relatively low, accounting for 6% and 8% of the total annual expenditure respectively.

[0083] The spatial attention layer discovers that there is a frequent fund allocation relationship between the main campus and the east campus, with the allocation frequency being 2 - 3 times per month, while the west campus is relatively independent with a lower allocation frequency.

[0084] Seasonal Prediction Phase: The seasonal prediction model predicts the fund demand for the next school year based on historical patterns.

[0085] The prediction results show that the fund demand for the main campus in September of the next school year will reach 36 million yuan, an 8% increase compared to the same period of the previous year; the fund demands for the east campus and the west campus in September are 24 million yuan and 18 million yuan respectively.

[0086] At the same time, the model predicts that due to the commissioning of the newly built experimental building, the fund demand for the east campus in the fourth quarter will increase by 15% compared to the historical same period.

[0087] Optimization and Allocation Phase: The spatiotemporal optimization algorithm comprehensively considers the predicted demands, risk assessments, and fund transfer costs of each campus, and formulates a fund allocation plan for the next school year.

[0088] The algorithm recommends allocating 32 million yuan to the main campus, 21 million yuan to the east campus, and 16 million yuan to the west campus in advance at the end of August. At the same time, to cope with unexpected demands, 5 million yuan of emergency funds is reserved in the main campus.

[0089] Real - time Monitoring Phase: During the actual implementation process, the real - time monitoring module discovers that the actual expenditure of the east campus in October is 22% higher than the predicted value, exceeding the warning threshold of 15%.

[0090] The system automatically triggers the dynamic adjustment algorithm, and allocates 2 million yuan from the emergency funds of the main campus to the east campus, ensuring the normal operation of the east campus.

[0091] Verification of Technical Effects: Effect of Improving the Prediction Accuracy of Fund Flow: By comparing the prediction results of the traditional prediction method and the method of this application, the prediction accuracy of the traditional linear regression method is 73%, while the prediction method using the spatiotemporal attention network has an accuracy rate of 89%, and the prediction accuracy is improved by 16 percentage points.

[0092] Specifically: During the 36 - month verification period, the traditional method has 10 months with a prediction error exceeding 20%, while the method of this application only has 2 months with a prediction error exceeding 20%.

[0093] Effect of improving the efficiency of fund allocation: Before implementing the method of this application, the school had 4.2 cases of fund shortage or idle per month on average, with an average of 1.5 million yuan involved in each case, and the fund utilization rate was 82%.

[0094] After implementing the method of this application, the number of cases of fund shortage or idle per month decreased to 1.8 on average, with an average of 800,000 yuan involved in each case, and the fund utilization rate increased to 91%.

[0095] The overall efficiency of fund allocation increased by 40%, and the annual fund cost was saved by approximately 1.8 million yuan.

[0096] As Figures 2 to 4 shown, they are respectively the comparative analysis of prediction accuracy; the change trend of fund allocation efficiency; and the system performance indicators.

[0097] The embodiments of the present invention have been described above. However, these embodiments are not limited to the above specific implementation manners. The above specific implementation manners are merely illustrative rather than restrictive. Under the inspiration of these embodiments, those of ordinary skill in the art can also make more equivalent embodiments in various forms, all of which fall within the protection scope of these embodiments.

Claims

1. An information management method for the accounting process in universities, characterized in that, Including the following steps: Collect the historical financial data of each campus of the university to form a spatio-temporal data matrix; Input the spatio-temporal data matrix into the spatio-temporal attention network to capture the spatio-temporal dependence relationship of fund flows and output the spatio-temporal feature representation; Use the seasonal prediction model to analyze the spatio-temporal feature representation and predict the seasonal fund demand; Based on the seasonal fund demand, analyze the fund demand characteristics of each campus, construct the campus demand model and the risk distribution model, form a multi-campus comprehensive evaluation matrix, and solve the multi-objective function through the spatio-temporal optimization algorithm to output the optimal fund allocation plan; Monitor the fund usage of each campus through the real-time monitoring module, and dynamically adjust the fund allocation based on the multi-campus comprehensive evaluation matrix and the spatio-temporal optimization algorithm to maintain the optimality of the allocation plan.

2. The information management method for the accounting process in colleges and universities according to claim 1, wherein The seasonal prediction model includes a periodic feature extraction module and a trend prediction module. The periodic feature extraction module uses the Fourier transform method to identify the periodic patterns of fund usage, and the trend prediction module uses the long short-term memory network structure to learn the long-term trend changes of fund demand.

3. The information management method for the accounting process in colleges and universities according to claim 1, characterized in that, The spatio-temporal optimization algorithm uses the multi-objective optimization method, including an objective function construction module and a constraint condition processing module. The objective function construction module takes maximizing the fund allocation efficiency, minimizing the risk, and minimizing the cost as the optimization objectives, and the constraint condition processing module uses the Lagrange multiplier method to handle the liquidity constraint, budget constraint, and policy constraint.

4. The information management method for the accounting process in colleges and universities according to claim 1, characterized in that The real-time monitoring module includes a data collection sub-module and an anomaly detection sub-module. The data collection sub-module connects to the API interface of the university's financial system to obtain the fund flow data of each campus in real time, and the anomaly detection sub-module uses the statistical process control method to set the upper and lower limits of the control of the fund usage deviation.

5. The information management method for the accounting process in colleges and universities according to claim 1, characterized in that, The historical financial data includes fund inflow data, fund outflow data, and budget execution data.

6. The information management method for the accounting process in colleges and universities according to claim 2, characterized in that The periodic feature extraction module can automatically detect the semester cycle, monthly cycle, and weekly cycle, and the trend prediction module contains 256 hidden units.

7. The information management method for the accounting process in colleges and universities according to claim 3, characterized in that, The objective function construction module uses the weighted summation method to construct the comprehensive objective function, and the constraint conditions processed by the constraint condition processing module include the liquidity constraint, budget constraint, and policy constraint.

8. The information management method for the accounting process in colleges and universities according to claim 4, characterized in that, The data update frequency of the data collection sub-module is once per hour, and the anomaly detection sub-module triggers an alarm when the actual fund usage exceeds plus or minus 15% of the predicted value.

9. The information management method for the accounting process in colleges and universities according to claim 1, characterized in that, The time attention layer contains 8 attention heads, and the dimension of each attention head is 64. The space attention layer takes each campus as a graph node and the fund flow relationship between campuses as a graph edge.

10. An information management system for the accounting process in colleges and universities, which is used to execute an information management method for the accounting process in colleges and universities according to any one of claims 1-9, characterized in that, Including: A data collection module for collecting the historical financial data of each campus of the university to form a spatio-temporal data matrix; A spatio-temporal feature extraction module for capturing the spatio-temporal dependence relationship of fund flows through the spatio-temporal attention network and outputting the spatio-temporal feature representation; A seasonal prediction module for analyzing the spatio-temporal feature representation and predicting the seasonal fund demand; A comprehensive evaluation module for analyzing the fund demand characteristics of each campus based on the seasonal fund demand, constructing the campus demand model and the risk distribution model, and forming a multi-campus comprehensive evaluation matrix; The fund allocation module is used to solve multi-objective functions through a spatio-temporal optimization algorithm, output the optimal fund allocation plan, and monitor the fund usage of each campus through a real-time monitoring module. Based on the multi-campus comprehensive evaluation matrix and the spatio-temporal optimization algorithm, it dynamically adjusts the fund allocation to maintain the optimality of the allocation plan.

Images