Method, system, medium and equipment for predicting overdue reporting of multi-factor special funds

Abstract

The present invention proposes a multi-factor special fund reporting overdue prediction method, system, medium and equipment, which belongs to the technical field of fund reporting progress supervision. The method obtains historical project information and a complete reporting path containing node information, extracts time-series evaluation factors including node status, reporting deadline, and fund amount, inputs the trained LSTM model element fusion features, and uses the element fusion features to train and optimize the overdue prediction model to minimize the loss. Subsequently, the current project information is obtained, the element fusion features are extracted and input into the model, the current project reporting path and nodes are predicted, the reporting time is calculated according to the node position, and compared with the warning threshold to realize overdue judgment and warning. The present invention overcomes the problems of difficulty in real-time monitoring, management and supervision of special fund reporting, and low prediction accuracy, and effectively improves the efficiency and accuracy of special fund reporting management.

Description

Technical Field

[0001] The present application relates to the technical field of fund reporting progress supervision, and in particular to a multi-factor special fund reporting overdue prediction method, system, medium and equipment. Background Art

[0002] In the field of special funds, the existing technologies, such as the Chinese invention application with application number 202211069650.2, entitled "A method and device for managing special funds", involve restrictions on the right to use them; such as the Chinese invention application with application number 201910528446.4, entitled "A comprehensive management platform system for special fund declaration", involve early warning of suspected duplicate declaration projects. However, the existing technologies have not proposed corresponding solutions to the problems of overdue declaration of special funds in the field of natural resources and project filing, and cannot realize real-time monitoring of the progress of declaration, which makes it difficult for regulatory authorities to intervene in the first place, thus affecting the management and supervision of the progress of declaration. Summary of the invention

[0003] In view of the deficiencies in the prior art, the purpose of the present invention is to provide a multi-factor special fund reporting overdue prediction method, system, medium and equipment, by obtaining historical project information and complete reporting path, extracting time-series evaluation factors (node ​​status, reporting period, fund amount), training overdue prediction model after LSTM model feature fusion, and performing overdue prediction on current project information. By comprehensively judging the time-series reporting status of funds with the help of multiple factors, overdue warning at a certain time point can be achieved, which facilitates the early intervention of superior regulatory authorities and promotes the completion of reporting work on schedule.

[0004] In order to achieve the above object, the present invention is implemented through the following technical solutions:

[0005] In a first aspect, the present invention provides a method for predicting overdue special fund reporting by multiple factors, comprising the following steps:

[0006] Obtain historical project information and historical frequent paths; the historical frequent paths include multiple historical complete reporting paths with node information;

[0007] Extracting time-series evaluation factors of historical project information, wherein the time-series evaluation factors include node status factor, reporting deadline factor, and funding amount factor;

[0008] Input the temporal evaluation factor into the trained LSTM model and perform feature fusion to obtain element fusion features;

[0009] The overdue prediction model is trained by using the feature fusion features, and the loss function is calculated based on the historical frequent path matching, and optimized to the minimum to obtain the trained overdue prediction model;

[0010] Obtain the current project information and extract the current element fusion features, input the trained overdue prediction model, and obtain the reporting path and path nodes to which the current project belongs; calculate the reporting time based on the node position, compare it with the warning threshold, determine whether it is overdue and issue a warning.

[0011] Furthermore, the historical project information includes the project status, deadline and funding amount of various types of special fund projects in history from issuance to completion, under any reporting status.

[0012] Furthermore, the acquisition process of the historical frequent paths is specifically as follows:

[0013] Obtaining historical log files, including the province where the historical project is located, the historical reporting progress of the responsible district or county, historical ranking, project type, and project scale based on amount;

[0014] Extract log features of historical log files, including reporting time features, regional divisions, type classification features, and project scale features;

[0015] Based on the log features, define adjacent, causal, parallel, and irrelevant order relationships, generate a footprint matrix, and construct a directed graph according to the footprint matrix and the order relationship;

[0016] Perform pruning operations based on directed graphs to count and filter out frequent paths;

[0017] The pruned frequent paths are connected and extended to obtain the historical frequent paths.

[0018] Furthermore, the extraction of time-series evaluation factors of historical project information specifically includes: extracting node status characteristics, reporting deadline characteristics and fund amount characteristics of historical project information, performing time series processing, and obtaining time-series evaluation factors.

[0019] Furthermore, the temporal evaluation factor is input into the trained LSTM model and feature fusion is performed to obtain the element fusion feature, which specifically includes:

[0020] The temporal evaluation factors are input into three LSTM modules to obtain the zoning element features, type element features and scale element features respectively;

[0021] Weights are assigned to the element features, and the element features are fused using a fusion channel method to obtain element fusion features.

[0022] Furthermore, the overdue prediction model is trained by utilizing the element fusion features, and the loss function is calculated based on the historical frequent path matching, and optimized to the minimum to obtain the trained overdue prediction model. The specific process is as follows:

[0023] Training the overdue prediction model, inputting the element fusion features into the overdue prediction model, and obtaining the reporting path and path nodes to which the historical items belong;

[0024] Calculate the loss function based on the matching degree between the reported paths and path nodes and the historical frequent paths;

[0025] When the loss is minimized, the trained overdue prediction model is obtained.

[0026] Furthermore, the current project information is obtained and the current element fusion features are extracted, and the trained overdue prediction model is input to obtain the reporting path and path nodes to which the current project belongs; the reporting time is calculated according to the node position, and compared with the warning threshold to determine whether it is overdue and issue a warning. The specific process is as follows:

[0027] Obtain the current project information, extract the current project time series evaluation factors, and obtain the current factor fusion features based on the LSTM model;

[0028] Input the current element fusion features into the trained overdue prediction model to obtain the reporting path and path nodes to which the current project belongs;

[0029] According to the position of the path node in the reporting path, the time interval between the current node and the final reporting completion node is calculated to obtain the reporting time. The reporting time is compared with the set warning threshold to determine whether it is overdue and issue a warning.

[0030] In a second aspect, the present invention provides a multi-factor special fund reporting overdue prediction system, comprising:

[0031] The information acquisition module is configured to acquire historical project information and historical frequent paths; the historical frequent paths include a plurality of historical complete reporting paths with node information;

[0032] A factor extraction and conversion module is configured to extract time-series evaluation factors of historical project information, wherein the time-series evaluation factors include node status factors, reporting deadline factors, and funding amount factors;

[0033] A feature extraction module is configured to input the temporal evaluation factor into the trained LSTM model and perform feature fusion to obtain an element fusion feature;

[0034] The model training module is configured to train the overdue prediction model using the element fusion features, calculate the loss function based on the historical frequent path matching, and optimize it to the minimum to obtain the trained overdue prediction model;

[0035] The prediction module is configured to obtain the current project information and extract the current element fusion features, input the trained overdue prediction model, obtain the reporting path and path nodes to which the current project belongs; calculate the reporting time according to the node position, compare it with the warning threshold, determine whether it is overdue and issue a warning.

[0036] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a multi-factor special fund reporting overdue prediction method described in the first aspect.

[0037] In a fourth aspect, the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of a multi-factor special fund reporting overdue prediction method described in the first aspect are implemented.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] (1) In order to effectively avoid fund management problems caused by delayed reporting, the present invention proposes a multi-factor special fund reporting overdue prediction method. By acquiring a variety of project-related evaluation factors, the dynamic changes in the time dimension are captured and the key links in the project reporting process are deeply analyzed, thereby effectively extracting the time series characteristics of the project; further using the LSTM model for feature fusion, the model's ability to understand the interaction between complex factors is effectively enhanced, and the project reporting progress and potential risks can be understood more accurately. Based on the historical frequent path, the current project is given an overdue warning. Since the historical frequent path reflects the regular process and common patterns of special fund project reporting, by matching the current project with the historical frequent path, the project reporting progress and expected path can be more effectively evaluated, thereby improving the accuracy of the prediction.

[0040] (2) By finding historical frequent paths and extracting processes, the present invention can better mine and capture data flows, cater to the characteristics of time series, and more accurately predict the overdue time of reporting. With the help of process mining technology, information can be extracted from the historical reporting log, and the time interval between the current reporting node and the final reporting completion stage can be counted to provide a reference for the setting of the warning threshold, comprehensively construct a prediction model, and improve the accuracy and precision of the prediction.

[0041] (3) The present invention uses three independent LSTM modules to extract time features respectively. The extracted zoning elements, type elements and scale elements can better represent the time characteristics of the project process. At the same time, weighted and multi-feature channel fusion is used to establish a comprehensive evaluation index, which can comprehensively consider the importance of different features while maintaining the differences of the features, and realize the comprehensive judgment of the time series reporting status of funds.

[0042] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their description are used to explain the present invention but do not constitute a limitation of the present invention.

[0044] Figure 1 A main flow chart of a method for predicting overdue special fund reporting with multiple factors provided by an embodiment of the present invention;

[0045] Figure 2 A detailed flow chart of a method for predicting overdue special fund reporting with multiple factors provided by an embodiment of the present invention;

[0046] Figure 3 A schematic diagram of fusion element features provided by an embodiment of the present invention;

[0047] Figure 4 A training diagram of a pre-trained long short-term memory network model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0049] Special funds in the field of natural resources in government big data refer to special funds set up to support the protection, restoration, utilization and management of natural resources. It covers air pollution prevention and control, water pollution prevention and control, soil pollution prevention and control, key ecological protection and restoration, marine ecological protection and restoration, etc. The use of these funds must comply with relevant regulations and focus on efficiency and quality.

[0050] Time series is a set of observation data sorted by time, which implies the time-varying trend of the parameters. Like the real government supervision business data, government big data also conforms to the time series characteristics that supervision process data naturally possesses, and time series prediction methods are suitable for prediction and analysis.

[0051] Therefore, the present invention introduces time series into the field of special fund progress supervision, constructs an algorithm to explore the rules between parameters and time series, and estimates unknown information based on this.

[0052] Embodiment 1

[0053] Embodiment 1 of the present invention provides a method for predicting overdue special fund reporting based on multiple factors, such as Figure 1 As shown, Figure 2 A detailed process is provided, including the following steps:

[0054] S1: Obtain historical project information and historical frequent paths.

[0055] Among them, the historical project information includes the project status, deadline and funding amount of various types of special fund projects in history from issuance to completion, under any reporting status.

[0056] The historical project information is cleaned and preprocessed, sorted and verified, and redundant information is removed to ensure the completeness and accuracy of the data, providing a basis for subsequent data analysis and prediction.

[0057] The historical frequent paths include multiple historical complete reporting paths with node information; the acquisition process is specifically as follows:

[0058] S101, select a log file; obtain a historical log file, wherein the historical log file includes the province where the historical project is located, the historical reporting progress of the district or county in charge, the historical ranking, the project type, and the project scale based on the amount;

[0059] S102, feature extraction: extracting log features of historical log files, wherein the log features include reporting time features, regional divisions, type classification features, project scale features, etc.;

[0060] S103, selecting a mining algorithm and constructing a directed graph: defining adjacent, causal, parallel, and irrelevant order relationships based on the log features, generating a footprint matrix, and constructing a directed graph according to the footprint matrix and the order relationship;

[0061] S104, pruning operation: counting the paths of historical logs, screening out frequent paths, removing unnecessary paths, and pruning the generated frequent paths;

[0062] S105, connection operation: connecting the pruned frequent paths to further expand the paths;

[0063] S106, result output: output the final historical frequent path set.

[0064] In S101, the historical reporting progress of the province where the project is located and the district and county in charge of the project in the historical log file reflects the characteristics of the rules of the geographical area to which the project belongs; the historical ranking reflects the competitiveness and performance of the project among similar projects; the project type corresponds to different reporting requirements and approval processes; the project scale based on the amount reflects the scale and importance of the project.

[0065] These historical log files comprehensively consider factors such as the temporal and spatial characteristics of the fund reporting status, the difficulty of reporting, and the financial risks, which help to evaluate the fund reporting status more comprehensively and accurately.

[0066] In S103, through process mining, the Alpha algorithm is used to analyze the historical reporting progress data, extract the nodes in the reporting process and the relationship between the nodes, and form workflow information, which can reflect the node sequence that often appears in the reporting process. The node sequence not only reflects the order of reporting, but also implies the dynamic changes in the time dimension. This time series feature is crucial for predicting the reporting progress and overdue risk of the project. At the same time, the time series of the node sequence can match the time series of the time series evaluation factor, thereby providing a more accurate and comprehensive prediction basis for the prediction of the later overdue time.

[0067] S2: Extract the time-series evaluation factors of historical project information, which include node status factors, reporting deadline factors, and funding amount factors. Specifically:

[0068] (1) Based on the historical project-related information, the historical workflow is obtained. According to the project status changes, task completion status, progress updates and other information, the project duration and task completion rate are analyzed to obtain the project's node status factor.

[0069] (2) Based on the relevant information of the historically reported projects, extract the deadline or submission date field and convert it into a date format. According to the time difference between the deadline and the current date, obtain the project's reporting deadline factor.

[0070] (3) Based on the historical project-related information, extract the funding amount field and perform standardization processing to calculate the funding scale and funding utilization rate to obtain the funding amount factor of the project.

[0071] In order to facilitate the input of the model, the extracted evaluation factors are processed by time series, and the features are converted into time series data to obtain time series evaluation factors.

[0072] The time series analysis process specifically includes: determining the time period, determining the time period of the historical workflow, the reporting deadline, and the amount of funds, such as day, week, month, etc. Selecting a week as the unit for time series analysis.

[0073] Collect characteristic data, including relevant data on historical workflow, reporting deadline, and funding amount within each time period.

[0074] Create time series data. For each time period, arrange the data of historical workflow, reporting deadline, and fund amount in chronological order to form a time series data set. Each time point represents a sample, which contains three feature values ​​(node ​​status, reporting deadline, fund amount) and is associated with the corresponding timestamp.

[0075] S3: Input the time - sequenced evaluation factors into the trained LSTM model for feature fusion to obtain feature - fused features. The specific steps are as follows:

[0076] Input the time - sequenced evaluation factors into three LSTM modules to obtain zoning element features, type element features, and scale element features respectively;

[0077] Assign weights to the element features and use the fusion channel method to fuse the element features to obtain feature - fused features.

[0078] Specifically, input the processed time - series data into three LSTM modules. The LSTM modules process the data, learn the features of zoning elements, type elements, and scale elements respectively, and output relevant feature representations; fuse the features and use the fusion channel method to integrate the features extracted by the three LSTM modules.

[0079] During the specific work, it includes the following steps:

[0080] S301: Use the LSTM modules to extract zoning elements, type elements, and scale elements in the evaluation factors respectively. Each module is responsible for processing the time - series of one element. As Figure 3 shown, the same time - series is input into each LSTM module, and each LSTM module is trained separately. The input information enters the unit inside the LSTM structure, and the output information of the previous - step unit is input into the current unit;

[0081] S302: Divide the time - series data into multiple subsequences;

[0082] As Figure 4 shown, taking the time - series of zoning elements as an example, assume the time - series sample is , the sample label is , and a certain batch of subsequence samples is (m < n). Then the input information of the LSTM time unit at time t includes: the subsequence input information of this unit, the output information of the LSTM unit at t - 1, and the fusion information of the LSTM unit at t - 1. The output information includes: the output information of the LSTM unit at time t and the fusion information of the LSTM unit at time t.

[0083] The specific workflow after the sample is transmitted in the network is as follows:

[0084] (1) Select information and discard some information. The forget gate is controlled by the sigmoid function, based on the output information of the previous moment. and current input information , decide whether to let the fusion information learned at the last moment Pass or partially pass. Discard some information As shown in formula (1).

[0085] (1)

[0086] In the formula, is the sigmoid activation function, w is the corresponding weight, and b is the offset constant.

[0087] (2) Update information. First, the input gate uses the sigmoid function to determine the data that needs to be updated, then uses the tanh function to generate new candidate values ​​for fusion information, and finally merges the information to create new fusion information. This is shown in formulas (2), (3), and (4).

[0088] (2)

[0089] (3)

[0090] (4)

[0091] (3) Output fusion information. First, the output gate uses the sigmoid function to obtain an initial output, then uses the tanh function to scale it, and finally multiplies it with the output of the sigmoid function to output the final fusion information, as shown in formulas (5) and (6).

[0092] (5)

[0093] (6)

[0094] Expand the LSTM time unit in the time dimension, then the input sample Each moment corresponds to a time unit, which can form an m-layer LSTM feature extraction model, and finally obtain the feature output sequence .

[0095] S303: The features of the zoning elements, type elements, and scale elements extracted by the three LSTM modules are integrated. Before integration, each feature is multiplied by a weight to represent the importance of each feature to the prediction result, and a comprehensive evaluation index is established.

[0096] Among them, after the zoning elements, type elements and scale elements are extracted through LSTM features respectively, the feature output at the last moment can be used as a fusion of multi-factor time series information.

[0097] The monitoring model based on multi-factor time series analysis proposed in this embodiment includes three long and short-term memory feature extraction modules, which respectively extract the zoning elements, type elements and scale elements in the evaluation factors. Finally, the evaluation utility of the three modules is integrated through the fusion channel method, while maintaining the individual differences of the three modules in the evaluation. This method can comprehensively judge the time series reporting status of funds and realize the overdue warning at a certain time point.

[0098] S4: Use the element fusion features to train the overdue prediction model, calculate the loss function based on the historical frequent path matching, and optimize it to the minimum to obtain the trained overdue prediction model. Specifically:

[0099] S401: training an overdue prediction model, inputting the element fusion features into the overdue prediction model, and obtaining the reporting path and path nodes to which the historical items belong;

[0100] S402: Calculate the loss function according to the matching degree between the obtained reported path and path nodes and the historical frequent paths;

[0101] S403: When the loss is minimal, a trained overdue prediction model is obtained.

[0102] The matching refers to comparing the reported paths and path nodes of the historical items with the nodes and node positions contained in each path in the historical frequent path set to determine whether there is similarity or consistency between them. The overdue prediction model is based on the matching process to learn the relationship between the element fusion features and the historical frequent paths and their nodes. It should be understood that the model training process can be implemented by those skilled in the art in combination with actual conditions.

[0103] In this embodiment, since both the factor fusion features and the historical frequent paths are time-series, which means that they contain dynamic change information in the time dimension, and the historical frequent paths reflect the common patterns and trends in the project reporting process, through path matching, the model can capture these time-series features and learn these patterns, and make full use of them to improve the accuracy of the prediction and better predict future reporting behavior and overdue risks.

[0104] It should be understood that when training the model, all samples are divided into a training set and a test set according to a certain ratio. The training set is responsible for training the model, and the test set is responsible for evaluating the training effect of the model.

[0105] S5: Get the current project information and extract the current element fusion features, input the trained overdue prediction model, and obtain the reporting path and path nodes to which the current project belongs; calculate the reporting time according to the node position, compare it with the warning threshold, determine whether it is overdue and issue a warning. The specific process is:

[0106] S501: Obtain current project information, extract the current project time series evaluation factor, and obtain the current factor fusion feature based on the LSTM model;

[0107] S502: Input the current element fusion feature into the trained overdue prediction model to obtain the reporting path and path nodes to which the current project belongs;

[0108] S503: According to the position of the path node in the reporting path, the time interval between the current node and the final reporting completion node is calculated to obtain the reporting time, and the reporting time is compared with the set warning threshold to determine whether it is overdue and issue a warning.

[0109] Specifically, according to the statistical characteristics of the time interval from the current reporting stage to the final reporting completion stage in the historical data, such as the average time, a threshold of reporting delay exceeding a specified number of days is set as the condition for triggering an early warning. The collected and monitored real-time reporting progress data is input into the model to obtain the prediction results. It is determined whether the current reporting status meets the conditions for triggering an early warning. If the predicted number of days overdue for reporting exceeds the set threshold, an early warning is triggered.

[0110] The multi-factors proposed in this invention are mainly reflected in two aspects: factor fusion characteristics and historical frequent paths.

[0111] First, multi-factor consideration of project information-evaluation factor-element fusion features: The time-series evaluation factors (node ​​status factor, reporting deadline factor, and funding amount factor) extracted based on historical project information reflect the reporting status and progress of the project from different dimensions. These factors are input into the LSTM model for feature fusion to obtain more comprehensive and in-depth element fusion features. This multi-factor fusion method can capture the complexity and diversity of the project reporting process, thereby improving the accuracy of the prediction model. At the same time, due to the consideration of multiple factors, the model can show stronger robustness when facing projects in different situations.

[0112] Secondly, multi-factor considerations for obtaining historical frequent paths: When obtaining historical frequent paths, the present invention considers multiple factors such as the province where the project is located, the historical reporting progress of the district and county in charge, historical rankings, project types, and project scales based on amounts. These factors together constitute the "footprints" of the project reporting process, reflecting the reporting patterns and trends of the projects at different stages and under different conditions. By analyzing these footprints, historical frequent paths can be identified more accurately, thereby providing more reliable data support for training overdue prediction models. This multi-factor analysis method helps the model learn more realistic and effective reporting rules, further improving the accuracy of the prediction.

[0113] The above two types of data containing multiple factors are used together in the training of the overdue prediction model. Through accurate matching and analysis of project nodes and frequent path nodes, a deep understanding and comprehensive capture of the project reporting process is achieved. This joint use not only enriches the input information of the model, but also enables the prediction model to more comprehensively reflect the various complex factors and potential laws in the project reporting process by integrating features of different dimensions.

[0114] Considering multiple factors enables the forecasting model to fully consider the diversity and uncertainty of project reporting when it is constructed, thereby enhancing the model's interpretability. The model no longer relies solely on a single or limited number of factors for forecasting, but is able to comprehensively consider multiple interrelated and mutually influential factors to generate more accurate and reliable forecast results. This comprehensive improvement enables the model to more accurately predict the project's reporting progress and overdue risks, thereby helping to take appropriate measures in advance to avoid or mitigate the negative impact of overdue.

[0115] Embodiment 2

[0116] Embodiment 2 of the present invention provides a multi-factor special fund reporting overdue prediction system, including:

[0117] The information acquisition module is configured to acquire historical project information and historical frequent paths; the historical frequent paths include a plurality of historical complete reporting paths with node information;

[0118] A factor extraction and conversion module is configured to extract time-series evaluation factors of historical project information, wherein the time-series evaluation factors include node status factors, reporting deadline factors, and funding amount factors;

[0119] A feature extraction module is configured to input the temporal evaluation factor into the trained LSTM model and perform feature fusion to obtain an element fusion feature;

[0120] The model training module is configured to train the overdue prediction model using the element fusion features, calculate the loss function based on the historical frequent path matching, and optimize it to the minimum to obtain the trained overdue prediction model;

[0121] The prediction module is configured to obtain the current project information and extract the current element fusion features, input the trained overdue prediction model, obtain the reporting path and path nodes to which the current project belongs; calculate the reporting time according to the node position, compare it with the warning threshold, determine whether it is overdue and issue a warning.

[0122] Embodiment 3

[0123] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the multi-factor special fund reporting overdue prediction method as described in the above-mentioned embodiment 1 are implemented.

[0124] Embodiment 4

[0125] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the method for predicting overdue reporting of multi-factor special funds as described in the first embodiment above are implemented.

[0126] The steps or modules involved in the above embodiments 2 to 4 correspond to those in embodiment 1. For the specific implementation, please refer to the relevant description of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0127] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multi-factor special fund reporting overdue prediction method, characterized in that: include: Obtain historical project information and historical frequent paths; the historical frequent paths include multiple historical complete reporting paths with node information; Extract the time-series evaluation factors of historical project information, which include node status factor, reporting deadline factor, and funding amount factor. Specifically: Based on the historical reporting of project related information, historical workflows are obtained. According to the project status changes, task completion status, and progress update information, project duration and task completion rate are analyzed to obtain the project's node status factor. Based on the relevant information of historical reported projects, extract the deadline or submission date field and convert it into a date format; obtain the reporting deadline factor of the project based on the time difference between the deadline and the current date; Based on the historically reported project-related information, the fund amount field is extracted and standardized, and the fund scale and fund utilization rate are calculated to obtain the project's fund amount factor; Input the temporal evaluation factor into the trained LSTM model and perform feature fusion to obtain element fusion features; The overdue prediction model is trained by using the feature fusion features, and the loss function is calculated based on the historical frequent path matching, and optimized to the minimum to obtain the trained overdue prediction model; Obtain the current project information and extract the current element fusion features, input the trained overdue prediction model, and obtain the reporting path and path nodes to which the current project belongs; calculate the reporting time based on the node position, compare it with the warning threshold, determine whether it is overdue and issue a warning.

2. A method for predicting overdue special fund reporting based on multiple factors as claimed in claim 1, characterized in that: The historical project information includes the project status, deadline and funding amount of various types of special fund projects in history from issuance to completion, under any reporting status.

3. A method for predicting overdue special fund reporting based on multiple factors as claimed in claim 1, characterized in that: The acquisition process of the historical frequent paths is specifically as follows: Obtaining historical log files, including the province where the historical project is located, the historical reporting progress of the responsible district or county, historical ranking, project type, and project scale based on amount; Extract log features of historical log files, including reporting time features, regional divisions, type classification features, and project scale features; Based on the log features, define adjacent, causal, parallel, and irrelevant order relationships, generate a footprint matrix, and construct a directed graph according to the footprint matrix and the order relationship; Perform pruning operations based on directed graphs to count and filter out frequent paths; The pruned frequent paths are connected and extended to obtain the historical frequent paths.

4. A method for predicting overdue special fund reporting with multiple factors as claimed in claim 1, characterized in that: The temporal evaluation factor is input into the trained LSTM model and feature fusion is performed to obtain the element fusion feature, specifically including: The temporal evaluation factors are input into three LSTM modules to obtain the zoning element features, type element features and scale element features respectively; Weights are assigned to the element features, and the element features are fused using a fusion channel method to obtain element fusion features.

5. A method for predicting overdue special fund reporting based on multiple factors as claimed in claim 1, characterized in that: The overdue prediction model is trained by using the element fusion feature, and the loss function is calculated based on the historical frequent path matching, and the trained overdue prediction model is obtained by optimizing it to the minimum. The specific process is as follows: Training the overdue prediction model, inputting the element fusion features into the overdue prediction model, and obtaining the reporting path and path nodes to which the historical items belong; Calculate the loss function based on the matching degree between the reported paths and path nodes and the historical frequent paths; When the loss is minimized, the trained overdue prediction model is obtained.

6. A method for predicting overdue special fund reporting with multiple factors as claimed in claim 1, characterized in that: The current project information is obtained and the current element fusion features are extracted, and the trained overdue prediction model is input to obtain the reporting path and path nodes to which the current project belongs; the reporting time is calculated according to the node position, and compared with the warning threshold to determine whether it is overdue and issue a warning. The specific process is as follows: Obtain the current project information, extract the current project time series evaluation factors, and obtain the current factor fusion features based on the LSTM model; Input the current element fusion features into the trained overdue prediction model to obtain the reporting path and path nodes to which the current project belongs; According to the position of the path node in the reporting path, the time interval between the current node and the final reporting completion node is calculated to obtain the reporting time. The reporting time is compared with the set warning threshold to determine whether it is overdue and issue a warning.

7. A multi-factor special fund reporting overdue prediction system, characterized in that: include: The information acquisition module is configured to acquire historical project information and historical frequent paths; the historical frequent paths include a plurality of historical complete reporting paths with node information; The factor extraction conversion module is configured to extract the time-series evaluation factors of historical project information, wherein the time-series evaluation factors include node status factors, reporting deadline factors, and funding amount factors. Specifically: Based on the historical reporting of project related information, historical workflows are obtained. According to the project status changes, task completion status, and progress update information, project duration and task completion rate are analyzed to obtain the project's node status factor. Based on the relevant information of historical reported projects, extract the deadline or submission date field and convert it into a date format; obtain the reporting deadline factor of the project based on the time difference between the deadline and the current date; Based on the historically reported project-related information, the fund amount field is extracted and standardized, and the fund scale and fund utilization rate are calculated to obtain the project's fund amount factor; A feature extraction module is configured to input the temporal evaluation factor into the trained LSTM model and perform feature fusion to obtain an element fusion feature; The model training module is configured to train the overdue prediction model using the element fusion features, calculate the loss function based on the historical frequent path matching, and optimize it to the minimum to obtain the trained overdue prediction model; The prediction module is configured to obtain the current project information and extract the current element fusion features, input the trained overdue prediction model, obtain the reporting path and path nodes to which the current project belongs; calculate the reporting time according to the node position, compare it with the warning threshold, determine whether it is overdue and issue a warning.

8. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in a multi-factor special fund reporting overdue prediction method as described in any one of claims 1 to 6.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the multi-factor special fund reporting overdue prediction method as described in any one of claims 1-6 are implemented.

Images