Scientific research project cost budgeting and cost management system

Abstract

The invention relates to the field of scientific research project management, and discloses a scientific research project cost budgeting and cost management system, which comprises a data processing module used for collecting and preprocessing cost data of a scientific research project and converting unstructured data into structured data; the cost classification module is used for carrying out dimensionality reduction and classification on the cost data based on principal component analysis and a support vector machine algorithm; the anomaly detection module is used for carrying out anomaly detection on the cost expenditure by adopting an ARIMA model and an isolated forest algorithm and marking abnormal cost; the budget management module is used for optimizing budget distribution by using linear programming and performing budget adjustment by using dynamic programming according to the actual expenditure condition; and the report generation module is used for generating a cost report of the scientific research project and providing visual display. The system improves the efficiency and accuracy of cost data processing, classification, anomaly detection, budget management and report generation through automatic and intelligent technical means, and ensures reasonable use and management of funds of scientific research projects.

Description

Technical Field

[0001] The present invention relates to the technical field of scientific research project management, in particular to a scientific research project cost budget and expense management system. Background Art

[0002] In the process of implementing scientific research projects, cost management and budget control are key links in project management. The funds for scientific research projects usually come from multiple sources and involve different types of expenses, such as equipment fees, material fees, travel expenses, etc. These expenses are distributed at different stages of the project, and the expenses are complex and diverse, which makes it difficult for traditional manual management methods to cope with them. Although the existing scientific research project budget and expense management system provides basic fund management functions, there are still many problems, such as low data processing efficiency, inaccurate expense classification, difficulty in timely detection of abnormal expenses, and inflexible budget execution.

[0003] Specifically, when dealing with a large amount of expense data from different sources, the existing system usually needs to rely on manual input and sorting, which is prone to data redundancy or errors, affecting expense classification and subsequent budget management. In addition, due to the dynamic nature of the allocation and use of funds for scientific research projects, the traditional budget management method lacks flexibility and cannot adjust the budget in time according to the actual progress of the project, resulting in inefficient use of funds, and even overspending or waste of funds.

[0004] In terms of anomaly detection, traditional audit processes are often conducted after the fact, and are unable to identify abnormal expense behaviors during project execution in real time, such as over-budgeting and duplicate reimbursement, which may cause abuse and risk of project funds. Existing anomaly detection methods are mostly based on simple rule matching, which is difficult to deal with complex expenditure patterns and potential violations, and fail to provide sufficient financial security for scientific research projects. Summary of the invention

[0005] In view of the deficiencies of the prior art, the present invention provides a scientific research project cost budget and expense management system, which improves the processing efficiency, classification accuracy, timeliness of anomaly detection, and flexibility and accuracy of budget management of expense data in scientific research projects through automation and intelligent means.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a scientific research project cost budget and expense management system, comprising:

[0007] Data processing module, used to collect and pre-process the cost data of scientific research projects and convert unstructured data into structured data;

[0008] an expense classification module, connected to the data processing module, for classifying the expense data based on data features;

[0009] An anomaly detection module, connected to the expense classification module, for detecting abnormal expenditures in scientific research project expenses;

[0010] A budget management module, connected to the anomaly detection module, for optimizing and dynamically adjusting the budget according to budget rules and actual expenditures;

[0011] The report generation module is connected to the budget management module and is used to generate a fund usage report for a scientific research project.

[0012] Preferably, the data processing module includes:

[0013] OCR recognition unit, used to convert paper bills and expense receipts into structured text data through OCR technology;

[0014] An NLP processing unit, used to extract key information from expense descriptions through natural language processing;

[0015] The data cleaning unit is used to clean and standardize the extracted data.

[0016] Preferably, the expense classification module includes:

[0017] PCA dimensionality reduction unit, used to perform principal component analysis on cost data and extract the main features of the data;

[0018] The SVM classification unit is used to classify the reduced-dimensional data based on the support vector machine algorithm and classify the expenses into different categories.

[0019] Preferably, the SVM classification unit uses a Gaussian kernel function for nonlinear classification, and the calculation formula of the Gaussian kernel function is:

[0020]

[0021] Among them, x i and x j is the cost feature vector after dimensionality reduction, and σ is the smoothing parameter of the kernel function.

[0022] Preferably, the anomaly detection module includes:

[0023] The time series analysis unit is used to model the historical expenditures through the ARIMA model to predict the trend of future expenditures;

[0024] Unsupervised learning unit for detecting outliers in cost data using the Isolation Forest algorithm;

[0025] The statistical detection unit is used to detect the abnormality of a single expense based on the z-score standardization method.

[0026] Preferably, the ARIMA model is used to predict future expense trends, and its time series prediction equation is:

[0027] Y t =c+φ1Y t-1 +…+φ p Y t-p +θ1∈ t-1 +…+θ q ∈ t-q +∈ t

[0028] Among them, Y t is the expenditure in period t, φ i and θ j is the model parameter, ∈ t is the noise term.

[0029] Preferably, the isolation forest algorithm in the unsupervised learning unit performs anomaly scoring according to the path length of the data point, and its anomaly scoring formula is:

[0030]

[0031] Among them, E(h(x i )) is the data point x i is the average path length, and c(N) is the normalization constant.

[0032] Preferably, the budget management module includes:

[0033] A linear programming unit is used to optimize budget allocation based on a linear programming method to meet budget constraints;

[0034] The dynamic adjustment unit is used to adjust the budget according to actual expenditure using dynamic programming methods to ensure the rational use of project funds.

[0035] Preferably, the linear programming unit uses a linear programming equation to perform budget optimization, and its optimization goal is:

[0036]

[0037] Among them, x ij is the amount allocated to the i-th category of expenses in the j-th time period, u ij is the utility coefficient.

[0038] The present invention also provides a scientific research project cost budget and expense management method, comprising the following steps:

[0039] Collect cost data of scientific research projects and perform structured preprocessing of the data through OCR recognition and natural language processing;

[0040] Dimensionality reduction of cost data based on principal component analysis;

[0041] Use the support vector machine algorithm to classify the reduced-dimensional expense data and identify different categories of expenses;

[0042] Use the ARIMA model and isolation forest algorithm to detect anomalies in expense expenditures and mark abnormal expenses;

[0043] Optimize budget allocations based on linear programming and adjust budgets based on actual spending using dynamic programming methods;

[0044] Generate expense reports for scientific research projects and present them visually.

[0045] The present invention provides a scientific research project cost budget and expense management system. It has the following beneficial effects:

[0046] 1. The present invention integrates OCR technology and natural language processing technology to automatically process paper bills and unstructured data, quickly converting them into structured expense data, greatly reducing the workload of manual input and data sorting, and avoiding errors in manual operations. At the same time, the standardized steps in the data preprocessing process ensure the accuracy of subsequent analysis and classification.

[0047] 2. The present invention uses principal component analysis (PCA) to reduce the dimensionality of high-dimensional cost data, effectively simplifying the complexity of the data, and combines the support vector machine (SVM) algorithm to achieve accurate automatic cost classification. This method not only improves the classification efficiency, but also can handle the complex distribution of different cost types in scientific research projects, ensuring the accuracy and reliability of the classification results.

[0048] 3. The present invention can detect abnormal expenditure behavior in scientific research projects in real time through the ARIMA time series analysis model and the isolation forest algorithm, and timely mark abnormal expenses such as over-budget and repeated reimbursement. Compared with traditional manual auditing, the automated anomaly detection method improves the accuracy and timeliness of fund monitoring and effectively prevents potential fund risks.

[0049] 4. The present invention uses linear programming to achieve the optimal allocation of the budget, and combines dynamic programming to flexibly adjust the budget during the project execution process to ensure reasonable allocation of funds and efficient use. Through the introduction of dynamic programming methods, the system can adjust the budget in real time according to actual expenditures, avoid waste of funds and overspending, and ensure the smooth progress of scientific research projects. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a schematic diagram of the system architecture of the present invention;

[0051] Figure 2 It is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] Please see attached Figure 1 The embodiment of the present invention provides a scientific research project cost budget and expense management system, which can automatically collect and process expense data in scientific research projects, and realize expense classification, anomaly detection, budget optimization and dynamic adjustment through intelligent algorithms, generate fund usage reports, and effectively improve the fund management efficiency and accuracy of scientific research projects. The system mainly includes a data processing module, an expense classification module, an anomaly detection module, a budget management module and a report generation module. The following is a detailed description of each module of the scientific research project cost budget and expense management system of the present invention.

[0054] Data processing module

[0055] The data processing module is the basic module of the cost budget and expense management system for scientific research projects. It is responsible for collecting and preprocessing expense data from multiple channels to ensure the accuracy and operability of subsequent classification, anomaly detection and budget management. The expense data of scientific research projects often come from a wide range of sources and in various formats, including paper bills, electronic invoices, reimbursement documents, and project budget documents. In order to enable these data to be effectively processed by subsequent algorithm modules, this module converts unstructured data into structured data through OCR (optical character recognition) technology, natural language processing (NLP) technology, and data cleaning and standardization technology, and performs appropriate cleaning and preprocessing.

[0056] In this embodiment, the functions of the data processing module are divided into the following steps:

[0057] 1. Data Collection

[0058] The data processing module receives multi-source cost data from scientific research projects. The data can come from scanned paper bills, electronic invoices, project reimbursement documents, etc. Since the formats of different data sources are not uniform, all raw data are first collected and stored in the database to be processed.

[0059] 2.OCR recognition

[0060] In this embodiment, OCR recognition is an important component of the data processing module. For the expense data from paper bills and scanned copies, the system uses optical character recognition (OCR) technology to convert unstructured image data into processable text data. OCR technology detects the edges and shapes of characters in the image and parses them into text. Specifically, the system uses an OCR engine such as Tesseract to process these scanned images and convert them into structured character information.

[0061] OCR Processing Flow:

[0062] 1) Binarize the scanned image to remove noise in the image and enhance the recognition of characters.

[0063] 2) Identify the text area through edge detection algorithm and extract each character.

[0064] 3) Use the OCR algorithm to convert the characters in the image into corresponding text and perform field parsing according to the preset format.

[0065] The result after OCR processing contains key field information in the invoice or bill, such as amount, supplier name, invoice number, date, etc.

[0066] 3. Natural Language Processing (NLP)

[0067] In this embodiment, NLP is used to further parse the expense description information in the structured text. Some reimbursement documents and bills may contain text description fields, such as "purchase of equipment" or "conference travel" and other expense purpose descriptions. In order to accurately classify and understand the contents of these fields, the system uses natural language processing technology to parse these texts and extract key information.

[0068] NLP processing steps:

[0069] 1) First, perform text preprocessing, remove irrelevant words and punctuation marks, and perform word segmentation on the expense description.

[0070] 2) Use part-of-speech tagging (POS Tagging) technology to tag each word and identify key terms in expense descriptions, such as "purchase", "equipment", "travel", etc.

[0071] 3) Apply a rule-based parser to extract key expense category related information and standardize it for processing by subsequent modules.

[0072] 4. Data cleaning and standardization

[0073] In this embodiment, data cleaning is an important step to ensure data accuracy. The data generated by OCR and NLP processing may contain noise or inconsistent data formats. The system uses data cleaning technology to ensure that all data formats are unified and non-redundant.

[0074] Data cleaning steps:

[0075] 1) Normalize the identified amount data, for example, convert different currency symbols and number formats into a unified numerical form.

[0076] 2) Format the date field and convert different date formats (such as "20xx / xx / xx" and "xx-xx-20xx") into a standard format.

[0077] 3) Remove duplicate data and invalid fields. For example, if noise characters or duplicate data are generated during the OCR recognition process, the system will correct them based on context rules or algorithms.

[0078] 5. Data structuring and feature extraction

[0079] In this embodiment, the data processing module further converts the cleaned data into a structured form format. In this step, the system generates a specific feature vector for each processed expense data for use by the subsequent classification and detection modules. These features may include important fields such as amount, supplier, invoice date, expense description, and reimbursement person.

[0080] Feature representation method:

[0081] 1) The system creates a multidimensional feature vector for each expense record. Assume that the feature set X contains n data points and each data point x i Contains d-dimensional features (such as amount, supplier, etc.), then each feature vector can be expressed as:

[0082] x i =[x i1 ,x i2 ,...,x id ]

[0083] 2) Among these features, numerical features (such as amount) are represented by raw numerical values, while text features (such as expense description) are represented by word vectors or encodings (such as based on TF-IDF or word embedding technology).

[0084] In a preferred embodiment, the feature vector x i Representation:

[0085] x i =[amount i, date i, supplier i, description code i ,...]

[0086] Through this feature representation, the system can convert various attributes of expenses into a unified numerical vector, which is convenient for subsequent machine learning model processing.

[0087] The above processing steps ensure the standardization, accuracy and consistency of expense data, providing a solid foundation for subsequent intelligent analysis and budget management.

[0088] Expense Classification Module

[0089] The expense classification module extracts features from preprocessed expense data and combines it with machine learning algorithms, especially the support vector machine (SVM) algorithm, to achieve accurate classification of different expense categories. Since the expense types in scientific research projects are complex and diverse (such as equipment expenses, travel expenses, material expenses, etc.), the classification module needs to take into account the diversity of data features and the accuracy of the classification model. In order to improve efficiency and accuracy, the system uses principal component analysis (PCA) to reduce the data dimension and uses SVM for classification in the feature space after dimension reduction to ensure the accuracy of classification.

[0090] In this embodiment, the expense classification module is divided into three main steps: feature extraction, data dimension reduction and support vector machine classification:

[0091] 1. Feature extraction

[0092] In this embodiment, the first step of expense classification is to extract features from the preprocessed expense data for subsequent classification tasks. These features may include numerical features (such as expense amount, number of reimbursements) and text features (such as expense description, supplier name). The system needs to uniformly represent features of different formats as numerical vectors for subsequent processing.

[0093] Feature vector construction: As in the data processing module above, a multi-dimensional feature vector is created for each expense record. Text features such as expense descriptions are encoded using word vector methods (such as TF-IDF or word embedding), while numerical features (such as amount) retain their original values.

[0094] Let the vocabulary be V, then the cost description text D i The TF-IDF vector is represented as:

[0095] TF-IDF(D i )=[tfidf(w1),tfidf(w2),…,tfidf(w V )]

[0096] Among them, w j represents the j words in the vocabulary, tfidf(w j ) represents word w j In describing D iTF-IDF value in .

[0097] Features such as the amount and number of reimbursements remain in their original numerical input without conversion.

[0098] In this way, the system uniformly vectorizes the features of all cost data, constructs a multi-dimensional feature representation of the costs, and provides input for subsequent classification.

[0099] 2. Data Dimensionality Reduction (Principal Component Analysis, PCA)

[0100] In this embodiment, the expense classification module uses the principal component analysis (PCA) method to reduce the dimension of the high-dimensional expense feature vector to reduce the computational complexity and avoid overfitting of the model. PCA maps the original high-dimensional data to a low-dimensional space through linear transformation and retains the main information in the data.

[0101] PCA selects eigenvectors by maximizing the variance of the data. Given the covariance matrix of the cost data, PCA needs to perform eigenvalue decomposition on the covariance matrix and select the eigenvector corresponding to the maximum eigenvalue as the principal component after dimensionality reduction.

[0102] For the feature matrix Its covariance matrix is:

[0103]

[0104] Among them, X is the preprocessed feature matrix, N is the number of samples, and d is the feature dimension.

[0105] By calculating the eigenvalues ​​and eigenvectors of the covariance matrix, the system retains the eigenvectors corresponding to the first k largest eigenvalues ​​as new principal components, reducing the high-dimensional cost data to k dimensions. The selection criterion for the principal component is the cumulative variance contribution rate, and generally 95% of the variance contribution rate is retained.

[0106] W=[v1,v2,…,v k ]

[0107] Among them, v1, v2, …, v k is the eigenvector corresponding to the largest eigenvalue.

[0108] For the reduced-dimensional data Z, the following transformation formula is used:

[0109] Z=XW

[0110] in, is the dimension reduction matrix, is the original feature matrix, is the feature matrix after dimensionality reduction.

[0111] Through PCA dimensionality reduction, the system can effectively reduce the redundant information of features, reduce the data dimension, and ensure that the classification algorithm can run efficiently in a low-dimensional space.

[0112] 3. Support Vector Machine (SVM) Classification

[0113] In this embodiment, the feature data after PCA dimension reduction is input into a support vector machine (SVM) classifier for automatic classification of expense categories. The SVM algorithm classifies expense data into different budget categories by finding the optimal separating hyperplane. Since there are many types of expenses in scientific research projects, SVM is particularly suitable for handling multi-category classification problems.

[0114] Given a cost feature dataset in is the feature vector after dimensionality reduction, y i ∈{1,2,…,C} is the cost category label. The goal of SVM is to find a hyperplane Make different types of cost points separate as much as possible.

[0115] The optimization objective function of SVM is:

[0116]

[0117] Among them, w is the weight vector of the classification hyperplane, b is the bias term, ξ i is a slack variable, and C is a penalty parameter used to control the penalty intensity for misclassification.

[0118] For linearly inseparable expense data, the system uses Gaussian kernel function (RBF kernel) for nonlinear mapping, mapping the data into high-dimensional space to improve classification accuracy. The Gaussian kernel function is defined as:

[0119]

[0120] Among them, x i and x j is the cost feature vector after dimensionality reduction, and σ is the smoothing parameter of the kernel function.

[0121] SVM classification steps:

[0122] 1) Input the feature data Z after dimension reduction.

[0123] 2) Train the classifier through the SVM algorithm to find the optimal separation hyperplane.

[0124] 3) For the newly submitted expense data, use the trained SVM model to classify it and output the classification results.

[0125] 4. Classification Results

[0126] In this embodiment, the expense classification module classifies each expense into a preset expense category (such as equipment expense, material expense, travel expense, etc.) according to the classification result. The classification result will be passed to the anomaly detection module and the budget management module as the basis for subsequent processing.

[0127] The expense classification module achieves efficient classification of complex expense data in scientific research projects through a combination of feature extraction, PCA dimensionality reduction and support vector machine (SVM) classification. This module ensures the accuracy of the classification results and provides an accurate basis for expense classification for subsequent anomaly detection and budget management.

[0128] Anomaly Detection Module

[0129] The anomaly detection module is used to monitor the expenditure of scientific research projects in real time and automatically identify potential abnormal behaviors, such as cost overruns, duplicate reimbursements, and unreasonable high expenditures. This module combines multiple algorithms, including time series analysis models, self-learning isolation forest algorithms, and statistical z-score detection methods, to flexibly identify different types of anomalies and provide guarantees for the financial management of scientific research projects. The anomaly detection module operates based on the output results of the expense classification module and provides anomaly information for subsequent budget management and report generation modules.

[0130] In this embodiment, the anomaly detection module is divided into three core steps: time series analysis, isolation forest algorithm and z-score detection. It aims to comprehensively identify abnormal cost behaviors in scientific research projects from the perspectives of time, pattern and statistics by combining multiple methods.

[0131] 1. Time Series Analysis (ARIMA Model)

[0132] In this embodiment, the anomaly detection module first uses the ARIMA (autoregressive integrated moving average) model to perform a time series analysis on the expenditure of scientific research projects. The expenditure of scientific research projects has obvious time dependence, so the ARIMA model can be used to perform time series forecasting on the expenditure and identify expenditure anomalies in the future.

[0133] The ARIMA model combines the ideas of autoregression (AR) and moving average (MA) to process time-correlated expense data. The model formula is:

[0134] Y t =c+φ1Y t-1 +…+φ p Y t-p +θ1∈ t-1 +…+θ q ∈ t-q +∈ t

[0135] Among them, Y t is the expenditure in period t, φ1,…,φ p are the autoregressive coefficients, θ1,…,θ q is the moving average coefficient, ∈ t is white noise, and c is a constant.

[0136] Model application steps:

[0137] 1) Cost data fitting: Based on historical cost expenditure data, select appropriate ARIMA model parameters (p, d, q) and use the maximum likelihood estimation method to fit the model parameters.

[0138] 2) Forecast future expenditures: Use the fitted ARIMA model to forecast expenditures for several time periods in the future and generate a forecast value range.

[0139] 3) Anomaly detection: If the actual expenditure Y at a certain moment t If it exceeds the forecast range, the expenditure is marked as abnormal.

[0140] For a certain period of time, the actual expenditure Y t ,if:

[0141]

[0142] in, is the predicted value, σ is the standard deviation of the prediction error, then the expenditure Y t Considered an anomaly.

[0143] 2. Isolation Forest

[0144] In this embodiment, the anomaly detection module uses the isolation forest algorithm to perform unsupervised anomaly detection on the cost data. Isolation forest is an unsupervised learning method based on a tree structure. It can effectively detect anomalies in the data by randomly selecting features and split points to build multiple decision trees. This method is particularly suitable for identifying outliers, that is, abnormal cost behaviors that are significantly different from other normal data points.

[0145] The Isolation Forest Algorithm measures the degree of isolation of data points by constructing multiple random trees (Isolation Tree). The core idea of ​​the algorithm is that the more isolated the data points (i.e. outliers), the shorter the path in the tree structure, and therefore they are more likely to be detected as anomalies.

[0146] Algorithm steps:

[0147] 1) Random splitting: Randomly select features of the cost data and build multiple isolated trees through random splitting points.

[0148] 2) Path length calculation: For each cost data point x i , calculate the average path length in each isolated tree. The shorter the path, the more likely the data point is an outlier.

[0149] 3) Anomaly score calculation: Combine the average path lengths of multiple isolated trees to calculate the anomaly score of the data point. i ) is defined as:

[0150]

[0151] Among them, E(h(x i )) is the data point x i is the average path length, and c(n) is a normalization constant.

[0152] Algorithm application:

[0153] 1) Training Isolation Forest Model: The Isolation Forest model is trained using unsupervised learning.

[0154] 2) Calculate the anomaly score: For each cost data point, calculate its anomaly score s(x i ) is close to 1, indicating that the data point may be an abnormal cost.

[0155] 3. z-score statistical test

[0156] In this embodiment, in order to quickly detect abnormal behavior of a single expense, the system also uses the z-score statistical method for detection. The z-score method identifies abnormally high or low expenses by calculating the standardized deviation of a single expense relative to the average. This method is particularly suitable for quickly screening expenses in a specific time period.

[0157] The z-score is used to measure the degree of deviation of a data point from the overall mean, and its calculation formula is:

[0158]

[0159] Among them, x i is a single fee, μ is the mean of the fee, and σ is the standard deviation of the fee. The z-score indicates how much the fee differs from the overall mean.

[0160] Abnormal judgment: When the z-score exceeds the set threshold (such as |z i |>2 or |z i |>3), the system marks the expense as abnormal. This method can quickly screen out expenses that deviate from normal spending patterns and provide them to financial personnel for further review.

[0161] 4. Processing of abnormal detection results

[0162] In this embodiment, the output of the anomaly detection module is the anomaly mark and anomaly score of each expense. These results will be passed to the system's budget management module and report generation module for reference and decision-making by scientific research project managers. The system will not only generate anomaly alerts, but also provide detailed abnormal expenditure reports, showing the anomaly score of each expenditure and possible abnormal reasons (such as over-budget, duplicate reimbursement, etc.).

[0163] The anomaly detection module can comprehensively monitor the expenditure of scientific research projects and identify potential abnormal behaviors in a timely manner through the comprehensive application of ARIMA time series analysis, isolation forest algorithm and z-score statistical method. Through multi-angle and multi-method detection means, this module ensures the compliance and rational use of scientific research project funds, and provides strong support for the management of scientific research project funds.

[0164] Budget management module

[0165] The budget management module is used to reasonably allocate research project funds and make dynamic adjustments based on the project's budget rules and actual expenditures. This module uses linear programming and dynamic programming algorithms to ensure that during the project execution, the allocation of funds can maximize project needs and avoid overspending or waste of funds. The budget management module directly relies on the output results of the aforementioned expense classification and anomaly detection modules, and incorporates the classified expenses and detected anomalies into the decision-making process of budget adjustment, thereby realizing intelligent fund management.

[0166] In this embodiment, the budget management module is divided into two parts: linear programming is used for optimal allocation of the initial budget, and dynamic programming is used for real-time adjustment of the budget during project execution.

[0167] 1. Linear programming to optimize budget allocation

[0168] In this embodiment, the budget management module first uses a linear programming method to perform initial budget allocation. In the start-up phase of a scientific research project, the budget management module needs to reasonably allocate various expense categories (such as equipment costs, material costs, travel expenses, etc.) under a limited total budget to ensure that various expenditures can meet project requirements.

[0169] Linear programming principle: The budget management problem can be modeled as a linear programming problem, the goal is to maximize the use of funds under the total budget constraint. Specifically, suppose there are m different expense categories, n time periods, and the amount allocated to each category of expense in each time period is x. ij , the utility coefficient is u ij The goal is to maximize the total utility of the budget, with the constraint that the cost allocation cannot exceed the total budget.

[0170] Objective function:

[0171]

[0172] Among them, x ij is the amount allocated to the i-th category of expenses in the j-th time period, u ij It is the utility coefficient, which indicates the contribution of each type of expense to the overall project during that time period.

[0173] Constraints:

[0174] 1. Total budget constraint: For each time period, the total allocation amount cannot exceed the available budget B for that period j :

[0175]

[0176] 2. Non-negativity constraint: The amount allocated to each type of expense must be non-negative:

[0177]

[0178] Optimization process:

[0179] 1) Based on the overall budget B of the project and the requirements of different expense categories, the system initializes the expense allocation plan and defines the utility coefficient u for each category of expense ij .

[0180] 2) Use a linear programming algorithm (such as the simplex method or the interior point method) to solve the above linear programming problem and obtain the optimal budget allocation plan.

[0181] 3) The output result is the optimal allocation amount x for each type of expense in each time period ij .

[0182] 2. Dynamically adjust the budget

[0183] In this embodiment, the budget management module uses a dynamic programming algorithm to adjust the budget in real time during the project execution. As the project progresses, the actual expenditure may deviate from the initial budget allocation. Dynamic programming continuously adjusts the future budget allocation in a recursive manner to ensure the rationality of fund use.

[0184] Dynamic programming ensures that the budget decisions made in each time period are optimal by gradually optimizing the budget allocation in each stage, and that the allocation of funds in subsequent time periods can also be optimal within the budget constraints.

[0185] State definition: Let V j (x) is the optimal value function when the remaining budget is x in the jth time period. The system needs to select the allocation plan x at each stage. ij, so that the total utility of the current and subsequent time periods is maximized.

[0186] The recursive relationship of dynamic programming is:

[0187]

[0188] Among them, x ij is the amount allocated to the i-th category of expenses in the j-th period, u ij is the corresponding utility coefficient, V j+1 (xx ij ) represents the remaining budget xx ij Optimal value in subsequent stages.

[0189] Adjustment process:

[0190] 1) Real-time monitoring of actual expenditures: The system continuously tracks the actual expenditures of scientific research projects and compares whether there are deviations between current expenditures and budget allocations.

[0191] 2) Dynamically adjust the budget: When the system detects a spending deviation, dynamic planning readjusts the subsequent budget based on the actual spending situation to ensure that the capital utility is maximized within the remaining budget.

[0192] 3. Trigger mechanism for budget adjustment

[0193] In this embodiment, the budget adjustment trigger mechanism works in conjunction with the anomaly detection module. When the anomaly detection module identifies overspending or other abnormal expenditures, the budget management module will re-trigger the budget adjustment process based on the detection results.

[0194] Abnormal trigger: If the expenditure of a certain type of expenses exceeds expectations or exceeds the budget limit, the system will immediately trigger the dynamic programming algorithm to reallocate the budget for future stages to avoid overspending or affecting the use of funds in other parts of the project.

[0195] Automatic adjustment process:

[0196] 1) Obtain abnormal expenditure information: The system obtains the output information of the anomaly detection module and identifies the specific overspending category and overspending amount.

[0197] 2) Budget adjustment: According to the overspending situation, the budget for future stages is adjusted. The system first reduces the future allocation amount of the overspending expense category and reallocates it to other categories that are not overspending.

[0198] 4. Output of budget management results

[0199] In this embodiment, the final output of the budget management module is the optimal allocation plan for each type of expense at each stage and the dynamically adjusted budget. The system passes these allocation results to the report generation module to generate a summary report on the use of project funds. In addition, managers can also view the current budget execution status and adjustment plans for future stages in real time through the system interface.

[0200] The budget management module ensures the rational allocation and dynamic adjustment of scientific research project funds through the combination of linear programming and dynamic programming. Linear programming is used to optimize the initial budget allocation, while dynamic programming ensures timely adjustment of the budget during project execution to maximize the benefits of the use of funds for scientific research projects. By linking with the anomaly detection module, the budget management module can dynamically respond to sudden overspending or abnormal expenditures in the project to ensure the robustness of fund management.

[0201] Report generation module

[0202] The report generation module generates detailed fund usage reports based on the data output by other modules of the system. This module integrates the results of the expense classification, anomaly detection and budget management modules to form a complete report on the use of scientific research project funds, which is convenient for managers, financial personnel and auditors of scientific research projects to conduct a comprehensive review and analysis of project funds. The report generation module can not only generate regular fund usage reports, but also support dynamic visualization, such as charts, trend charts, etc., to provide intuitive fund usage. In addition, this module also provides report export function, which supports the generation of standardized file formats (such as Excel, PDF) for further operation and archiving.

[0203] In this embodiment, the report generation module mainly includes three parts: data aggregation, visualization generation and report export, so as to ensure the integrity, accuracy and operability of the fund management report.

[0204] 1. Data aggregation

[0205] In this embodiment, the report generation module first constructs a comprehensive use of scientific research project funds by summarizing the data from the expense classification module, the anomaly detection module, and the budget management module. The data from each module provides multi-dimensional information such as expense classification, anomaly detection results, and budget execution. The report generation module constructs a scientific and reasonable report structure by integrating these data.

[0206] Data aggregation process:

[0207] Expense classification data: The report generation module obtains all classified expense data from the expense classification module, including the category of each expense (such as equipment expenses, material expenses, travel expenses, etc.), amount, date, supplier information, etc. These data will serve as the basis of the report to present the expense distribution of the project.

[0208] Anomaly detection data: The module obtains the output results of the anomaly detection module and marks all expenses detected as abnormal, including the anomaly category (such as overspending, duplicate reimbursement, etc.), abnormal amount, abnormal reason and the score of the detection algorithm (such as the anomaly score of the isolation forest).

[0209] Budget execution: Obtain the budget allocation results and actual expenditures of each expense category from the budget management module, especially the comparison information between the dynamically adjusted budget and expenditures. The budget execution will show the budget usage progress and the current remaining budget in each time period.

[0210] Assume x i represents the amount of the i-th expense, b j represents the budget amount of the jth category of expenses, then the total expenditure and budget execution summary formula is:

[0211]

[0212] Where n is the total number of expense items and m is the number of expense categories.

[0213] 2. Visualization Generation

[0214] In this embodiment, the report generation module supports intuitive display of fund usage through charts and visualization tools. In order to help managers of scientific research projects better understand and analyze fund usage, the module generates various forms of visualization charts, such as pie charts, bar charts, and trend charts.

[0215] Chart generation tool: The module uses JavaScript-based chart libraries (such as D3.js or Chart.js) to generate interactive charts. Through these charts, users can clearly see the project's funding distribution, expenditure trends of various types of expenses, and budget execution.

[0216] Chart Type:

[0217] Cost distribution pie chart: shows the distribution of different types of costs in scientific research projects. i is the total amount of the i-th category of expenses, X is the total project expenditure, then the proportion of each category of expenses in the pie chart is:

[0218]

[0219] Expenditure trend chart: shows the changes in expenditure in each time period, helping users identify the peak and trough periods of expenditure. Let the time period be t and the expenditure be x(t). The trend chart draws the expenditure curve based on the time series.

[0220] Budget Execution Bar Chart: Displays the budget usage for each expense category, showing the budget allocation, actual expenditure, and remaining budget for each category.

[0221] 3. Report export

[0222] In this embodiment, the report generation module supports exporting the summary data and visualization results into a standardized file format, which is convenient for managers to archive and further operate. The supported file formats include common office document formats such as Excel and PDF, and users can choose a suitable format for data storage and sharing.

[0223] Excel report export: The module converts the summarized expense data, budget execution status, anomaly detection results and other information into Excel table format by calling relevant APIs, and automatically generates multiple worksheets to display different types of fund data. Excel reports not only contain numerical data, but also embed visual charts to help users easily view graphical results in subsequent analysis.

[0224] PDF report export: The module uses the PDF generation library to export the fund report to PDF format, ensuring that the report format is fixed and easy to print and share. The PDF report contains complete information on the project's cost distribution, exception analysis, and budget execution, and is suitable for auditing and project evaluation.

[0225] Report content structure:

[0226] Overview page: displays the overall data of the project, such as total budget, total expenditure, remaining budget, etc.

[0227] Expense classification details: Display the total amount of each expense, expenditure progress, etc. by expense category.

[0228] Abnormal analysis report: lists all expenditures marked as abnormal, with detection results and explanations of abnormal reasons.

[0229] Visual charts: embed various expense distribution pie charts, budget execution bar charts and other charts in the report to provide an intuitive display of fund usage.

[0230] The report generation module provides a detailed analysis tool for the use of funds for scientific research projects through data aggregation, visualization generation and report export functions. This module can not only generate regular fund use reports, but also help users intuitively understand the flow of funds through a variety of visualization tools.

[0231] In general, the present invention uses OCR and NLP technology to perform structured processing on multi-source expense data, combines machine learning algorithms to achieve automated expense classification, and accurately identifies abnormal expenditures through time series analysis and unsupervised learning methods. The system uses linear programming and dynamic programming to optimize and dynamically adjust the budget to ensure the rational allocation and use of funds. The report generation module integrates the data of each module, provides detailed fund usage reports and visual displays, and improves the efficiency and accuracy of scientific research project fund management.

[0232] Please see attached Figure 2 The present invention also provides a scientific research project cost budget and expense management method, comprising the following steps:

[0233] S1: First, collect various expense data in scientific research projects, use OCR technology to convert paper receipts into structured text data, and parse and standardize the expense description information through natural language processing to lay the foundation for subsequent steps.

[0234] S2: Then, principal component analysis is used to reduce the dimension of the preprocessed expense data, extract key features, and reduce the dimension of the data in order to improve the efficiency of subsequent classification and analysis.

[0235] S3: Based on the data after dimensionality reduction, the support vector machine algorithm is used to classify different types of expenses, and automatically identify different types of expenses such as equipment expenses, material expenses, travel expenses, etc.

[0236] S4: Subsequently, the system uses the ARIMA model to perform time series analysis and combines it with the isolation forest algorithm to detect anomalies in the expense data, marking possible abnormal spending behaviors, such as over-budgeting and duplicate reimbursements.

[0237] S5: According to the budget situation of the project, use the linear programming method to optimally allocate the budget, and during the project implementation process, adjust the future budget allocation plan through dynamic programming based on the actual expenditure situation to ensure the rational use of funds.

[0238] S6: Finally, the system generates expense reports for scientific research projects, summarizes expenditure information and provides visual displays, making it easier for scientific research project managers to monitor and analyze the use of funds in real time.

[0239] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. Scientific research project cost budget and expense management system, characterized by: include: Data processing module, used to collect and pre-process the cost data of scientific research projects and convert unstructured data into structured data; an expense classification module, connected to the data processing module, for classifying the expense data based on data features; An anomaly detection module, connected to the expense classification module, for detecting abnormal expenditures in scientific research project expenses; A budget management module, connected to the anomaly detection module, for optimizing and dynamically adjusting the budget according to budget rules and actual expenditures; The report generation module is connected to the budget management module and is used to generate a fund usage report for a scientific research project.

2. The scientific research project cost budget and expense management system according to claim 1, characterized in that: The data processing module comprises: OCR recognition unit, used to convert paper bills and expense receipts into structured text data through OCR technology; An NLP processing unit, used to extract key information from expense descriptions through natural language processing; The data cleaning unit is used to clean and standardize the extracted data.

3. The scientific research project cost budget and expense management system according to claim 1, characterized in that: The expense classification module includes: PCA dimensionality reduction unit, used to perform principal component analysis on cost data and extract the main features of the data; The SVM classification unit is used to classify the reduced-dimensional data based on the support vector machine algorithm and classify the expenses into different categories.

4. The scientific research project cost budget and expense management system according to claim 3 is characterized in that: The SVM classification unit uses a Gaussian kernel function for nonlinear classification. The calculation formula of the Gaussian kernel function is: Among them, x i and x j is the cost feature vector after dimensionality reduction, and σ is the smoothing parameter of the kernel function.

5. The scientific research project cost budget and expense management system according to claim 1, characterized in that: The anomaly detection module comprises: The time series analysis unit is used to model the historical expenditures through the ARIMA model to predict the trend of future expenditures; Unsupervised learning unit for detecting outliers in cost data using the Isolation Forest algorithm; The statistical detection unit is used to detect the abnormality of a single expense based on the z-score standardization method.

6. The scientific research project cost budget and expense management system according to claim 5, characterized in that: The ARIMA model is used to predict future expense trends, and its time series prediction equation is: Y t =c+φ1Y t-1 +…+φ p Y t-p +θ1∈ t-1 +…+θ q ∈ t-q +∈ t Among them, Y t is the expenditure in period t, φ i and θ j is the model parameter, ∈ t is the noise term.

7. The scientific research project cost budget and expense management system according to claim 5, characterized in that: The isolation forest algorithm in the unsupervised learning unit performs anomaly scoring based on the path length of the data point, and its anomaly scoring formula is: Among them, E(h(x i )) is the data point x i is the average path length, and c(N) is the normalization constant.

8. The scientific research project cost budget and expense management system according to claim 1, characterized in that: The budget management module includes: A linear programming unit is used to optimize budget allocation based on a linear programming method to meet budget constraints; The dynamic adjustment unit is used to adjust the budget according to actual expenditure using dynamic programming methods to ensure the rational use of project funds.

9. The scientific research project cost budget and expense management system according to claim 8, characterized in that: The linear programming unit uses a linear programming equation to perform budget optimization, and its optimization goal is: Among them, x ij is the amount allocated to the i-th category of expenses in the j-th time period, u ij is the utility coefficient.

10. A scientific research project cost budget and expense management method, applied to the scientific research project cost budget and expense management system as claimed in any one of claims 1 to 9, characterized in that: The following steps are involved: Collect cost data of scientific research projects and perform structured preprocessing of the data through OCR recognition and natural language processing; Dimensionality reduction of cost data based on principal component analysis; Use the support vector machine algorithm to classify the reduced-dimensional expense data and identify different categories of expenses; Use the ARIMA model and isolation forest algorithm to detect anomalies in expense expenditures and mark abnormal expenses; Optimize budget allocations based on linear programming and adjust budgets based on actual spending using dynamic programming methods; Generate expense reports for scientific research projects and present them visually.

Images