A Deep Learning-Based Risk Prediction and Intervention Terminal for Industry-Education Integration Projects

Abstract

This application discloses a deep learning-based industry-education integration project risk prediction and intervention terminal, including a processor, memory, communication module, display module, data access module, identity recognition module, deep learning inference module, risk analysis module, intervention instruction generation module, and feedback collection module. The data access module accesses teaching execution data from the school side, participation data from the enterprise side, task execution data from the student side, mentor guidance data, and project outcome data. The processor performs unified format processing on the project data, configures project number, subject number, task node number, and time window number, generates a project time-series state matrix, and constructs a project risk correlation graph. It also updates the influence weights of edge relationships based on feedback delay, task rework, missing results, and progress deviation. The intervention instruction generation module generates structured electronic intervention instructions and pushes them to the responsible user terminal. The feedback collection module collects execution feedback and forms risk prediction samples.

Description

Technical Field

[0001] This invention belongs to the field of industry-education integration technology, and specifically relates to a risk prediction and intervention terminal for industry-education integration projects based on deep learning. Background Technology

[0002] Industry-education integration projects typically involve multiple stakeholders and stages, including schools, enterprises, students, on-campus mentors, enterprise mentors, and project outcomes. During project operation, various types of data are generated, such as course execution, enterprise feedback, student tasks, mentor guidance, interim results, and resource input. Current industry-education integration project management methods primarily rely on manual data entry, periodic checks, or simple indicator statistics. While these methods can record project progress, material submissions, and personnel participation, they struggle to promptly identify the risk transmission relationships between different stakeholders. For example, delayed feedback from enterprise mentors may further impact student task modifications, interim result submissions, and final acceptance; however, existing systems often only allow for manual intervention after risks have already manifested.

[0003] Therefore, a Chinese patent discloses a project risk prediction method and system based on the project R&D process, application number CN202011342989.6. This patented technology can break down the project R&D plan into multiple project tasks, abstract benchmark indicator attributes based on the project tasks, and then combine project R&D process data and a trained risk prediction algorithm model to predict the project risk type and key risk points, thereby improving the efficiency of R&D project risk management. Although this patent can predict the project risk type and key risk points, it does not form a structured electronic intervention instruction for responsible users, and it is difficult to associate and store the risk probability, processing content, and execution feedback before and after intervention as risk prediction samples. Therefore, those skilled in the art provide a deep learning-based industry-education integration project risk prediction and intervention terminal to solve the problems mentioned in the background technology. Summary of the Invention

[0004] This invention provides a deep learning-based terminal for risk prediction and intervention in industry-education integration projects, which can at least partially improve the above-mentioned problems; To achieve the above objectives, the present invention adopts the following solution: A deep learning-based industry-education integration project risk prediction and intervention terminal includes a processor, memory, communication module, display module, data access module, identity recognition module, deep learning inference module, risk analysis module, intervention instruction generation module, and feedback collection module. The data access module is used to access project data during the operation of the industry-education integration project. The project data includes at least school-side teaching execution data, enterprise-side participation data, student-side task execution data, mentor guidance data, and project outcome data. The identity recognition module is used to identify the identity type of the logged-in user and determine the corresponding data viewing permissions and intervention instruction receiving permissions based on the identity type; The processor is used to perform format unification processing on the project data and configure project number, main number, task node number and time window number for the processed project data. The processor is also configured to generate a project time-series state matrix according to the time window number, wherein each row of the project time-series state matrix corresponds to a time window and each column corresponds to a project state feature. The processor is also used to construct a project risk association diagram based on the participating entities, task nodes, resource nodes and result nodes in the project, and update the influence weight of the corresponding edge relationships in the project risk association diagram based on the feedback delay time, number of task reworks, number of result missing times and project progress offset value within the current time window. The deep learning inference module is used to input the project time-series state matrix and the project risk association graph into the trained deep learning risk prediction model, and to obtain the project risk type and its risk probability in a multi-label output manner. The risk analysis module is used to determine the risk level based on the risk probability, and to determine the key risk causes in descending order of node contribution or project status characteristic contribution. The intervention instruction generation module is used to generate structured electronic intervention instructions based on the project risk type, risk level, key risk causes and responsible user information, and send the structured electronic intervention instructions to the corresponding responsible user terminal through the communication module; The feedback acquisition module is used to collect the execution feedback of the structured electronic intervention command, and to associate and store the execution feedback with the corresponding project time-series state matrix, project risk correlation diagram, project risk type and key risk cause to form a risk prediction sample.

[0005] Preferably, the processor converts the school-side teaching execution data, enterprise-side participation data, student-side task execution data, mentor guidance data, and project outcome data into progress-type features, collaboration-type features, quality-type features, resource-type features, and outcome-type features, respectively. The progress-related features include at least the planned project completion time, the actual completion time, and the project progress offset value. The collaborative features include at least the number of school-enterprise communications, the time of enterprise feedback, and the number of mentor guidance sessions. The quality-related characteristics include at least the quality of student task completion, the number of times tasks were reworked, and the assessment results; The resource-related features include at least enterprise resource input records, training equipment usage records, and project funding execution records; The characteristics of the results include at least the stage results submission records, the completeness of the acceptance materials, and the results transformation records.

[0006] Preferably, the project time-series state matrix is ​​generated using weeks, months, or project phases as time windows; The project status characteristics within each time window should include at least the project schedule deviation rate, average interval of enterprise feedback, on-time completion rate of student tasks, number of task reworks, frequency of mentor guidance, completeness rate of phase results, completion rate of school-enterprise meetings, and enterprise evaluation change value.

[0007] Preferably, the project risk association diagram includes main nodes, task nodes, resource nodes, outcome nodes, and evaluation nodes; The main nodes include schools, enterprises, students, on-campus mentors, and enterprise mentors; The task nodes include course tasks, practical training tasks, enterprise tasks, and phase acceptance tasks; The resource nodes include training equipment, enterprise positions, teaching resources, and project funding; The deliverables include phase deliverables, acceptance deliverables, and transformation deliverables. The evaluation nodes include internal school evaluation, enterprise evaluation, and student practical training evaluation.

[0008] Preferably, the nodes in the project risk association diagram are connected by edge relationships, which include task dependency relationships, guidance relationships, resource consumption relationships, deliverable relationships, evaluation and feedback relationships, and anomaly transmission relationships. The processor updates the influence weight of the edge relationship based on the deviation between the abnormal data in the current time window and the historical benchmark data. When the enterprise reports a delay time, number of task reworks, number of missing results, or project progress deviation value that exceeds the corresponding historical benchmark value, the processor increases the influence weight of the corresponding edge relationship. When the enterprise reports that the delay time, number of task reworks, number of missing results, or project progress deviation value has recovered to within the corresponding historical benchmark value, the processor reduces the influence weight of the corresponding edge relationship.

[0009] Preferably, the deep learning risk prediction model includes a temporal feature processing channel, a graph feature processing channel, a feature fusion layer, and a risk output layer; The time-series feature processing channel is used to extract time-change features from the project's time-series state matrix; The graph feature processing channel is used to extract node features and edge weight features from the project risk association graph; The feature fusion layer is used to fuse the time-varying features, node features, and edge weight features; The risk output layer is used to output the probability of risks such as project delay risk, insufficient enterprise participation risk, decline in student training quality risk, insufficient mentor collaboration risk, lack of phase results risk, failure of results transformation risk, and abnormal execution of funding or resources in a multi-label output mode.

[0010] Preferably, the risk analysis module compares the risk probability of each project risk type with a preset level threshold to determine low risk, medium risk, high risk, or severe risk. The risk analysis module is also used to rank the contribution of nodes and project status characteristics involved in risk prediction, and to identify the nodes ranked at the top of a preset number and their corresponding abnormal characteristics as key risk causes.

[0011] Preferably, the intervention instruction generation module stores an intervention strategy table, which records the correspondence between project risk type, risk level, key risk causes, responsible user type, and intervention instruction template; The structured electronic intervention instructions shall include at least the project number, task node number, risk source, responsible user, completion deadline, processing requirements, and feedback entry point; When the risk level is low, the structured electronic intervention instruction is a reminder instruction; When the risk level is medium risk, the structured electronic intervention instruction is a review instruction; When the risk level is high, the structured electronic intervention instruction is a school-enterprise coordination instruction; When the risk level is severe, the structured electronic intervention instruction is a project adjustment instruction or a rectification instruction.

[0012] Preferably, the execution feedback collected by the feedback collection module includes whether the intervention instruction was completed, the completion time, the processing content, the responsible user confirmation information, the risk probability before intervention, the risk probability after intervention, and whether the risk was triggered again; The processor generates intervention effect data based on the risk probability before and after the intervention, and adds the intervention effect data to the risk prediction sample; The risk prediction samples are used to update the training sample library of the deep learning risk prediction model or to adjust the risk level threshold. Before inputting project data into the deep learning risk prediction model, the processor de-identifies and encodes sensitive fields in student identity information, enterprise business information, and project outcome materials, and uses the de-identified subject number to participate in the construction of the project time-series state matrix and project risk association diagram.

[0013] The beneficial effects of this invention are as follows: This invention collects teaching execution data from schools, participation data from enterprises, task execution data from students, guidance data from mentors, and project outcome data through a data access module. The processor then configures the above data with project numbers, subject numbers, task node numbers, and time window numbers, enabling project data from different sources and in different formats to enter a unified data processing structure. This reduces the problem of scattered multi-source data and difficulty in participating in continuous calculations in existing technologies.

[0014] This invention further generates a project time-series state matrix according to a time window, so that state features such as project progress deviation rate, average interval of enterprise feedback, student task on-time completion rate, number of task reworks, frequency of tutor guidance, and completeness rate of stage results can be continuously expressed according to the project operation process, thereby facilitating the deep learning model to identify the trend of project risk changes over time.

[0015] This invention also constructs a project risk association diagram based on participating entities, task nodes, resource nodes, and result nodes, and dynamically updates the influence weights of edge relationships based on feedback delay time, number of task reworks, number of missing results, and project progress deviation. As a result, the terminal can not only determine whether there is a risk in the project, but also reflect the risk transmission relationship between schools, enterprises, students, mentors, tasks, and results, overcoming the problem that existing static rule scoring methods are difficult to locate the source of risk.

[0016] This invention uses a deep learning risk prediction model to simultaneously process the project time-series state matrix and the project risk correlation graph. It outputs multiple risk probabilities, such as project delay risk, insufficient enterprise participation risk, decline in student training quality risk, missing stage results risk, and failure of results transformation risk, in a multi-label output manner, thereby improving the completeness and pertinence of risk prediction.

[0017] This invention also uses a risk analysis module to rank the contribution of node and project status characteristics, identify key risk causes, and an intervention instruction generation module to generate structured electronic intervention instructions that link risk type, risk level, risk source, responsible user, and feedback entry point, enabling risk prediction results to be directly converted into traceable processing tasks. The feedback collection module further collects risk probabilities and processing results before and after intervention to form risk prediction samples for subsequent model training or risk threshold adjustment, thereby achieving a closed-loop process of risk prediction, cause identification, intervention execution, and feedback updates. Attached Figure Description

[0018] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a connection block diagram of the terminal of the present invention. Detailed Implementation Example 1

[0019] The present invention will be further described below with reference to embodiments. It should be understood that the following embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. Without departing from the concept of the present invention, equivalent substitutions of various modules, additions or deletions of data fields, and conventional adjustments to the model structure made by those skilled in the art should all fall within the scope of protection of the present invention.

[0020] Please refer to Figure 1 This embodiment provides a deep learning-based industry-education integration project risk prediction and intervention terminal. This terminal is used to uniformly process school-side data, enterprise-side data, student-side data, mentor guidance data, and project outcome data during the operation of industry-education integration projects, forming a project time-series state matrix and project risk correlation diagram that can be input into a deep learning model. Based on the model output results, it generates structured electronic intervention instructions, enabling the project risk prediction results to form a traceable processing closed loop.

[0021] The terminal in this embodiment includes a processor, a memory, a communication module, a display module, a data access module, an identity recognition module, a deep learning inference module, a risk analysis module, an intervention command generation module, and a feedback acquisition module. The processor is connected to the memory, communication module, display module, data access module, identity recognition module, deep learning inference module, risk analysis module, intervention command generation module, and feedback acquisition module, respectively. The terminal can be a fixed terminal used by project managers, or a mobile terminal with a processor, memory, display screen, and communication functions.

[0022] The data access module is used to access project data during the operation of industry-education integration projects. Project data can come from the school's academic affairs system, enterprise project management system, student training system, mentor guidance record system, achievement management system, or manual input interface. After receiving the above data, the data access module sends the data to the processor for standardized format processing. The communication module is used for data communication with the project leader's terminal, the school mentor's terminal, the enterprise mentor's terminal, and the student's terminal. The display module is used to display the project risk type, risk probability, risk level, key risk causes, structured electronic intervention instructions, and the execution status of the intervention instructions.

[0023] The identity recognition module is used to identify the identity type of logged-in users. Identity types include project leaders, on-campus mentors, corporate mentors, students, and administrators. The identity recognition module can identify users through account passwords, SMS verification codes, facial recognition, or a unified identity authentication interface. The processor determines data viewing permissions and intervention instruction receiving permissions based on the identified identity type. For example, students can only view their own tasks, their own training records, and their own related intervention instructions; corporate mentors can view corporate tasks in projects they participate in, corporate feedback records, and related student task completion status; on-campus mentors can view student tasks, teaching execution data, and outcome materials in projects they are responsible for; project leaders can view the overall project risk results, risk causes, and intervention implementation status; and administrators maintain model parameters, level thresholds, and intervention strategy tables.

[0024] After the data is received, the processor first performs format standardization processing on the project data. Specifically, the processor converts the school-side teaching execution data, enterprise-side participation data, student-side task execution data, mentor guidance data, and project outcome data into progress-type features, collaboration-type features, quality-type features, resource-type features, and outcome-type features, respectively.

[0025] The data includes: school-side teaching execution data (course progress, practical training hours completion, on-campus mentor guidance records, student assessment records, and project phase acceptance records); enterprise-side participation data (enterprise mentor feedback records, enterprise task posting records, enterprise evaluation records, enterprise resource input records, and enterprise acceptance opinions); student-side task execution data (student task submission time, task completion quality, number of task reworks, practical training attendance records, phase result submission records, and skills assessment results); mentor guidance data (number of on-campus mentor guidance sessions, number of enterprise mentor feedback sessions, guidance time, mentored individuals, and guidance content); and project outcome data (phase result submission records, completeness of acceptance materials, software deliverables, sample deliverables, intellectual property deliverables, paper deliverables, and technology transfer records).

[0026] After standardizing the format, the processor configures the processed project data with a project number, subject number, task node number, and time window number. The project number is used to distinguish different industry-education integration projects; the subject number is used to distinguish different subjects such as schools, enterprises, students, on-campus mentors, and enterprise mentors; the task node number is used to distinguish course tasks, practical training tasks, enterprise tasks, and phase acceptance tasks; and the time window number is used to indicate the week, month, or project phase to which the data belongs.

[0027] For example, if a project number for an industry-education integration project is P001, a corporate mentor's entity number is E-T01, the task node number for the second phase of the corporate task is T203, and the time window number for the third week is W03, then the corporate mentor's feedback data on the second phase of the corporate task in the third week can be recorded as P001-E-T01-T203-W03. Through this numbering method, data from different sources can be mapped to the same project, the same entity, the same task node, and the same time window, facilitating subsequent matrix processing and risk correlation analysis.

[0028] The processor aggregates project data according to time window numbers to generate a project time-series state matrix. Each row in the project time-series state matrix corresponds to a time window, and each column corresponds to a project state feature. Time windows can be divided by week, month, or project stage. For short-term training projects, time windows can be divided by week; for longer-term university-industry cooperation projects, time windows can be divided by month or project stage.

[0029] In one implementation, the project timeline state matrix is ​​represented as X=[xij], where i represents the i-th time window and j represents the j-th project state characteristic. Project state characteristics include project schedule deviation rate, average interval of enterprise feedback, on-time completion rate of student tasks, number of task reworks, frequency of mentor guidance, completeness rate of stage deliverables, completion rate of university-enterprise meetings, and changes in enterprise evaluation values.

[0030] Project schedule deviation rate can be calculated based on the actual completion progress and the planned completion progress. For example, the project schedule deviation rate is the difference between the actual completion progress and the planned completion progress divided by the planned completion progress. The average interval of enterprise feedback can be obtained by averaging the intervals between two consecutive enterprise feedback times within the same time window. The on-time completion rate of student tasks can be determined by the ratio of the number of tasks completed on time to the number of tasks that should be completed. The completeness rate of stage deliverables can be determined by the ratio of the number of submitted deliverables to the number of deliverables that should be submitted. The completion rate of university-enterprise meetings can be determined by the ratio of the number of meetings actually held to the number of meetings planned. The change value of enterprise evaluation can be determined by the difference between the enterprise evaluation score in the current time window and the enterprise evaluation score in the previous time window.

[0031] After generating the project time-series state matrix, the processor further constructs a project risk correlation diagram. This diagram represents the relationships between different stakeholders, tasks, resources, outcomes, and evaluations within the industry-education integration project. The project risk correlation diagram includes stakeholder nodes, task nodes, resource nodes, outcome nodes, and evaluation nodes.

[0032] The main nodes include schools, enterprises, students, on-campus mentors, and enterprise mentors. Task nodes include course tasks, practical training tasks, enterprise tasks, and phase acceptance tasks. Resource nodes include practical training equipment, enterprise positions, teaching resources, and project funding. Outcome nodes include phase deliverables, acceptance deliverables, and commercialized deliverables. Evaluation nodes include on-campus evaluation, enterprise evaluation, and student practical training evaluation.

[0033] The nodes in the project risk relationship diagram are connected by edges. These edges include task dependencies, guidance relationships, resource allocation relationships, deliverable relationships, evaluation and feedback relationships, and anomaly propagation relationships. For example, a task execution edge is established between student nodes and training task nodes; a guidance edge is established between on-campus mentor nodes and student nodes; a feedback edge is established between enterprise mentor nodes and enterprise task nodes; a resource allocation edge is established between training equipment nodes and training task nodes; a deliverable edge is established between stage result nodes and stage acceptance task nodes; and an evaluation and feedback edge is established between enterprise evaluation nodes and enterprise task nodes.

[0034] The processor updates the influence weights of corresponding edges in the project risk correlation graph based on the deviation between the abnormal data within the current time window and the historical baseline data. The historical baseline data can be the historical average of similar industry-education integration projects, or the average of the current project over the previous few time windows. The abnormal data within the current time window includes enterprise feedback delay time, number of task reworks, number of missing deliverables, and project schedule deviation.

[0035] When the enterprise feedback delay exceeds the corresponding historical baseline, the processor increases the influence weight of the feedback edge between the enterprise mentor node and the enterprise task node. When the number of task reworks exceeds the corresponding historical baseline, the processor increases the influence weight of the task execution edge between the student node and the task node. When the number of missing deliverables exceeds the corresponding historical baseline, the processor increases the influence weight of the deliverable delivery edge between the deliverable node and the phase acceptance task node. When the project schedule deviation exceeds the corresponding historical baseline, the processor increases the influence weight of the task dependency edge between task nodes.

[0036] Accordingly, when the enterprise reports that the delay time, number of task reworks, number of missing deliverables, or project schedule deviation value has returned to within the corresponding historical baseline value, the processor reduces the impact weight of the corresponding edge relationship. In this way, the project risk association diagram is not a fixed static diagram, but can be updated as the project's operating status changes.

[0037] In one implementation, the influence weight of an edge relationship can be updated based on the deviation ratio between the current outlier and the historical benchmark. If the current outlier is higher than the historical benchmark, the influence weight of the corresponding edge relationship is increased according to the deviation ratio; if the current outlier returns to within the historical benchmark, the influence weight of the corresponding edge relationship is decreased according to a preset attenuation ratio. The influence weight can be limited to between 0 and 1, or other numerical ranges can be set according to the actual project needs.

[0038] The deep learning inference module is used to load the trained deep learning risk prediction model. The deep learning risk prediction model includes a temporal feature processing channel, a graph feature processing channel, a feature fusion layer, and a risk output layer.

[0039] The temporal feature processing channel receives the project's temporal state matrix and extracts the temporal variation features between different time windows. This channel can employ recurrent neural networks, long short-term memory networks, or Transformer encoders. The graph feature processing channel receives the project risk correlation graph and extracts node features and edge weight features from it. This channel can employ graph convolutional networks or graph attention networks. The feature fusion layer concatenates, weights, or maps the temporal variation features, node features, and edge weight features. The risk output layer outputs the risk probabilities of different project risk types using a multi-label output method.

[0040] Project risk types include project delay risk, insufficient corporate participation risk, decline in student training quality risk, insufficient mentor collaboration risk, missing interim results risk, failure of results transformation risk, and abnormal execution of funding or resources risk. The risk output layer outputs a risk probability for each type of project risk, allowing multiple risk type predictions to be obtained simultaneously for the same project within the same time window.

[0041] The training samples for the deep learning risk prediction model are derived from historical industry-education integration project data and historical intervention feedback data. The inputs to the training samples include a historical project time-series state matrix and a historical project risk correlation graph. The labels for the training samples include project delays, insufficient enterprise participation, declining student training quality, insufficient mentor collaboration, missing interim results, failed results transformation, and abnormal funding or resource execution. Each label can be 0 or 1 to indicate whether the corresponding risk has occurred, or a value between 0 and 1 to represent the degree of risk.

[0042] During model training, the historical project time-series state matrix is ​​input into the time-series feature processing channel, and the historical project risk association graph is input into the graph feature processing channel. After passing through the feature fusion layer, the risk output layer outputs the predicted probabilities of each risk type. The processor or training device adjusts the model parameters based on the error between the predicted probabilities and the actual risk labels. When the loss value is lower than a preset value, or the prediction accuracy on the validation set meets the preset requirements, the trained deep learning risk prediction model is obtained. The trained model is stored in memory and called by the deep learning inference module.

[0043] The risk analysis module determines the risk level based on the risk probabilities output by the deep learning inference module. The module compares the risk probability of each project's risk type with preset level thresholds. For example, a risk probability below the first threshold is classified as low risk; a risk probability greater than or equal to the first threshold but lower than the second threshold is classified as medium risk; a risk probability greater than or equal to the second threshold but lower than the third threshold is classified as high risk; and a risk probability greater than or equal to the third threshold is classified as severe risk.

[0044] In one implementation, the first threshold is 0.4, the second threshold is 0.6, and the third threshold is 0.8. That is, a risk probability below 0.4 is considered low risk, a risk probability between 0.4 and 0.6 is considered medium risk, a risk probability between 0.6 and 0.8 is considered high risk, and a risk probability greater than or equal to 0.8 is considered severe risk. These thresholds can be adjusted by the administrator based on different schools, companies, or project types.

[0045] The risk analysis module is also used to identify key risk causes. Specifically, the risk analysis module ranks the nodes and project status features participating in this risk prediction by their contribution, and identifies the top-ranked nodes and their corresponding abnormal features as key risk causes. The contribution can be determined based on the attention weights in the deep learning risk prediction model, the change in risk probability before and after removing a certain node or project status feature, or the node weights output from the graph feature processing channel.

[0046] For example, the deep learning inference module outputs a project delay risk probability of 0.78, an insufficient enterprise participation risk probability of 0.81, and a missing interim deliverable risk probability of 0.73. The risk analysis module, based on contribution ranking, determines that the average interval of enterprise feedback, the number of task reworks, and the completeness rate of interim deliverables have a significant impact on these risk probabilities. Therefore, delayed enterprise feedback, abnormal number of task reworks, and missing interim deliverable materials are identified as key risk causes.

[0047] The intervention instruction generation module is used to generate structured electronic intervention instructions based on project risk type, risk level, key risk causes, and responsible user information. The module stores an intervention strategy table, which records the correspondence between project risk type, risk level, key risk causes, responsible user type, and intervention instruction templates.

[0048] After the risk analysis module outputs the project risk type, risk level, and key risk causes, the intervention instruction generation module matches the corresponding intervention instruction template from the intervention strategy table and fills in the project number, task node number, risk source, responsible user, completion deadline, processing requirements, and feedback entry point to generate a structured electronic intervention instruction. This structured electronic intervention instruction, as a data object generated by the terminal, is sent to the corresponding responsible user's terminal via the communication module.

[0049] When the risk level is low, the structured electronic intervention instruction is a reminder instruction, used to prompt the responsible user to pay attention to the corresponding task milestones. When the risk level is medium, the structured electronic intervention instruction is a review instruction, used to require the responsible user to verify the corresponding data or task status. When the risk level is high, the structured electronic intervention instruction is a university-enterprise coordination instruction, used to require on-campus mentors, enterprise mentors, or project leaders to confirm task milestones and processing deadlines. When the risk level is severe, the structured electronic intervention instruction is a project adjustment or rectification instruction, used to require the responsible user to adjust or supplement the project plan, task arrangements, or deliverables.

[0050] For example, when the project risk type is insufficient enterprise participation risk, the risk level is high risk, and the key risk cause is delayed enterprise feedback, the intervention instruction generation module generates a university-enterprise coordination instruction for enterprise mentors and project leaders; when the project risk type is missing stage deliverables risk, the risk level is medium risk, and the key risk cause is low completeness of acceptance materials, the intervention instruction generation module generates a review instruction for on-campus mentors; when the project risk type is project delay risk, the risk level is severe risk, and the key risk cause is continuous delays in task nodes, the intervention instruction generation module generates a project adjustment instruction for project leaders.

[0051] The feedback collection module is used to collect execution feedback from structured electronic intervention instructions. Execution feedback includes whether the intervention instruction was completed, the completion time, the processed content, the responsible user's confirmation information, the pre-intervention risk probability, the post-intervention risk probability, and whether the risk was triggered again. After receiving the structured electronic intervention instruction, the responsible user submits the processed content and confirmation information through the feedback entry point. The feedback collection module then sends the submitted content to the processor.

[0052] The processor generates intervention effect data based on the pre-intervention risk probability and the post-intervention risk probability. If the post-intervention risk probability is lower than the pre-intervention risk probability, the processor records the feedback as a valid intervention sample; if several prognostic risk probabilities do not decrease, or if the same type of risk is triggered again in a subsequent time window, the processor records the feedback as an invalid intervention sample or an intervention sample pending review.

[0053] The processor associates and stores intervention effect data with the corresponding project time-series state matrix, project risk correlation diagram, project risk type, and key risk causes, forming risk prediction samples. These risk prediction samples are used to update the training sample library of the deep learning risk prediction model or to adjust risk level thresholds. Through this method, the terminal can form a data loop encompassing risk prediction, intervention execution, and feedback results.

[0054] To protect sensitive information in student personal data, company business information, and project deliverables, the processor anonymizes sensitive fields before inputting project data into the deep learning risk prediction model. Sensitive fields include student names, student ID numbers, company contact person names, company business data, original project deliverable file names, and unpublished deliverable content. The processor replaces these sensitive fields with the subject number, company number, deliverable number, or hash code, and uses the anonymized subject number in the construction of the project time-series state matrix and project risk association diagram. The original sensitive fields are stored in an access-controlled data table and are not directly input into the deep learning risk prediction model.

[0055] The following describes this embodiment using a specific application process: A university and a company are jointly conducting an industry-education integration project in intelligent manufacturing, with a project duration of 12 weeks. The project includes one project leader, two university mentors, two company mentors, and 30 students. Project tasks include coursework, company training, submission of interim results, and final acceptance. The terminal is configured with project number P001, and corresponding numbers are assigned to the university, company, students, university mentors, company mentors, coursework tasks, training tasks, company tasks, and interim acceptance tasks.

[0056] After the project started, the data access module received project data weekly. In the first and second weeks, the average interval between company feedback was 2 days, student tasks were reworked once, and the completion rate of phase deliverables was 90%. The processor used this data as one of the initial baseline data for the project. In the third week, the project data received by the data access module showed that the average interval between company feedback increased to 5 days, the number of student tasks reworked increased to 4, and the completion rate of phase deliverables decreased to 60%, indicating that the actual progress was lower than the planned progress.

[0057] The processor assigns the project data for the third week to time window number W03 and generates project status features including project schedule deviation rate, average interval of enterprise feedback, on-time completion rate of student tasks, number of task reworks, frequency of mentor guidance, completeness rate of phase deliverables, completion rate of university-enterprise meetings, and changes in enterprise evaluation values. The processor then arranges the project status features from the first to the third week in chronological order to generate a project time-series status matrix.

[0058] Meanwhile, the processor updates the project risk correlation graph based on the abnormal data from the third week. Because the average interval of enterprise feedback exceeds the historical baseline, the processor increases the influence weight of the feedback edge between enterprise mentor nodes and enterprise task nodes; because the number of student task reworks exceeds the historical baseline, the processor increases the influence weight of the task execution edge between student nodes and training task nodes; because the completeness rate of stage deliverables decreases, the processor increases the influence weight of the deliverable delivery edge between deliverable nodes and stage acceptance task nodes; and because the project schedule deviates, the processor increases the influence weight of the task dependency edge between relevant task nodes.

[0059] The deep learning inference module inputs the project time-series state matrix formed from the first to the third week and the updated project risk correlation graph into the trained deep learning risk prediction model. The model outputs the risk probability of each project risk type in a multi-label manner, where the probability of project delay risk is 0.78, the probability of insufficient enterprise participation is 0.81, the probability of missing stage results is 0.73, and the probability of declining student training quality is 0.66.

[0060] The risk analysis module compares the above risk probabilities with preset level thresholds, identifying insufficient enterprise participation as a serious risk, project delays and missing stage deliverables as high risks, and declining student training quality as a high risk. The module further ranks the contribution of node and project status characteristics, determining that the average interval of enterprise feedback, the number of task reworks, and the completeness rate of stage deliverables are the project status characteristics with the highest contributions, and identifying delayed enterprise feedback, abnormal number of task reworks, and missing stage deliverable materials as the key risk causes.

[0061] The intervention instruction generation module matches the corresponding intervention instruction template from the intervention strategy table based on the project risk type, risk level, key risk causes, and responsible user information, generating three structured electronic intervention instructions. The first structured electronic intervention instruction is sent to the enterprise mentor's terminal, requiring the mentor to supplement the third week's enterprise task feedback within a specified period. The second structured electronic intervention instruction is sent to the university mentor's terminal, requiring the mentor to review the reasons for student task rework and submit their handling opinions. The third structured electronic intervention instruction is sent to the project leader's terminal, requiring the project leader to organize university-enterprise coordination and confirm the next stage acceptance milestone.

[0062] In the fifth week, the feedback collection module collected data showing that corporate mentors had supplemented their task feedback, on-campus mentors had submitted explanations for task rework, and the project leader had adjusted the phase acceptance schedule. The processor recalculated the project status characteristics for the fifth week and again called the deep learning risk prediction model for prediction. The prediction results showed that the probability of insufficient corporate participation decreased to 0.42, the probability of project delay decreased to 0.51, and the probability of missing phase deliverables decreased to 0.39.

[0063] The processor generates intervention effect data based on the risk probability changes in weeks three and five, and stores this intervention effect data in association with the project time-series state matrix, project risk correlation diagram, project risk type, and key risk causes from week three, forming a new risk prediction sample. This risk prediction sample is added to the training sample library for subsequent model retraining or risk level threshold adjustment.

[0064] Through the above implementation methods, this invention can uniformly convert multi-source project data in industry-education integration projects into a computable time-series state matrix and project risk correlation diagram, and simultaneously output the risk probabilities of multiple risk types through a deep learning risk prediction model. Compared with methods relying solely on manual inspection or static indicator scoring, this invention can dynamically update the influence weights between nodes based on the project's operational status, generate structured electronic intervention instructions based on the risk prediction results, and then use execution feedback to form risk prediction samples, thereby achieving integrated processing of project risk prediction, risk cause identification, intervention execution, and feedback updates.

[0065] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A deep learning-based risk prediction and intervention terminal for industry-education integration projects, characterized in that, It includes a processor, memory, communication module, display module, data access module, identity recognition module, deep learning inference module, risk analysis module, intervention instruction generation module, and feedback collection module; The data access module is used to access project data during the operation of the industry-education integration project. The project data includes at least school-side teaching execution data, enterprise-side participation data, student-side task execution data, mentor guidance data, and project outcome data. The identity recognition module is used to identify the identity type of the logged-in user and determine the corresponding data viewing permissions and intervention instruction receiving permissions based on the identity type; The processor is used to perform format unification processing on the project data and configure project number, main number, task node number and time window number for the processed project data. The processor is also configured to generate a project time-series state matrix according to the time window number, wherein each row of the project time-series state matrix corresponds to a time window and each column corresponds to a project state feature. The processor is also used to construct a project risk association diagram based on the participating entities, task nodes, resource nodes and result nodes in the project, and update the influence weight of the corresponding edge relationships in the project risk association diagram based on the feedback delay time, number of task reworks, number of result missing times and project progress offset value within the current time window. The deep learning inference module is used to input the project time-series state matrix and the project risk association graph into the trained deep learning risk prediction model, and to obtain the project risk type and its risk probability in a multi-label output manner. The risk analysis module is used to determine the risk level based on the risk probability, and to determine the key risk causes in descending order of node contribution or project status characteristic contribution. The intervention instruction generation module is used to generate structured electronic intervention instructions based on the project risk type, risk level, key risk causes and responsible user information, and send the structured electronic intervention instructions to the corresponding responsible user terminal through the communication module; The feedback acquisition module is used to collect the execution feedback of the structured electronic intervention command, and to associate and store the execution feedback with the corresponding project time-series state matrix, project risk correlation diagram, project risk type and key risk cause to form a risk prediction sample.

2. The deep learning-based industry-education integration project risk prediction and intervention terminal according to claim 1, characterized in that, The processor converts school-side teaching execution data, enterprise-side participation data, student-side task execution data, mentor guidance data, and project outcome data into progress-type features, collaboration-type features, quality-type features, resource-type features, and outcome-type features, respectively. The progress-related features include at least the planned project completion time, the actual completion time, and the project progress offset value. The collaborative features include at least the number of school-enterprise communications, the time of enterprise feedback, and the number of mentor guidance sessions. The quality-related characteristics include at least the quality of student task completion, the number of times tasks were reworked, and the assessment results; The resource-related features include at least enterprise resource input records, training equipment usage records, and project funding execution records; The characteristics of the results include at least the stage results submission records, the completeness of the acceptance materials, and the results transformation records.

3. The deep learning-based industry-education integration project risk prediction and intervention terminal according to claim 1, characterized in that, The project time-series state matrix is ​​generated using weekly, monthly, or project phases as time windows. The project status characteristics within each time window should include at least the project schedule deviation rate, average interval of enterprise feedback, on-time completion rate of student tasks, number of task reworks, frequency of mentor guidance, completeness rate of phase results, completion rate of school-enterprise meetings, and enterprise evaluation change value.

4. The deep learning-based industry-education integration project risk prediction and intervention terminal according to claim 1, characterized in that, The project risk correlation diagram includes main nodes, task nodes, resource nodes, outcome nodes, and evaluation nodes; The main nodes include schools, enterprises, students, on-campus mentors, and enterprise mentors; The task nodes include course tasks, practical training tasks, enterprise tasks, and phase acceptance tasks; The resource nodes include training equipment, enterprise positions, teaching resources, and project funding; The deliverables include phase deliverables, acceptance deliverables, and transformation deliverables. The evaluation nodes include internal school evaluation, enterprise evaluation, and student practical training evaluation.

5. A deep learning-based industry-education integration project risk prediction and intervention terminal according to claim 4, characterized in that, The nodes in the project risk association diagram are connected by edge relationships, which include task dependency relationships, guidance relationships, resource consumption relationships, deliverable relationships, evaluation and feedback relationships, and anomaly transmission relationships. The processor updates the influence weight of the edge relationship based on the deviation between the abnormal data in the current time window and the historical benchmark data. When the enterprise reports a delay time, number of task reworks, number of missing results, or project progress deviation value that exceeds the corresponding historical benchmark value, the processor increases the influence weight of the corresponding edge relationship. When the enterprise reports that the delay time, number of task reworks, number of missing results, or project progress deviation value has recovered to within the corresponding historical benchmark value, the processor reduces the influence weight of the corresponding edge relationship.

6. A deep learning-based industry-education integration project risk prediction and intervention terminal according to claim 1, characterized in that, The deep learning risk prediction model includes a temporal feature processing channel, a graph feature processing channel, a feature fusion layer, and a risk output layer. The time-series feature processing channel is used to extract time-change features from the project's time-series state matrix; The graph feature processing channel is used to extract node features and edge weight features from the project risk association graph; The feature fusion layer is used to fuse the time-varying features, node features, and edge weight features; The risk output layer is used to output the probability of risks such as project delay risk, insufficient enterprise participation risk, decline in student training quality risk, insufficient mentor collaboration risk, lack of phase results risk, failure of results transformation risk, and abnormal execution of funding or resources in a multi-label output mode.

7. A deep learning-based industry-education integration project risk prediction and intervention terminal according to claim 1, characterized in that, The risk analysis module compares the risk probability of each project risk type with a preset level threshold to determine low risk, medium risk, high risk or severe risk. The risk analysis module is also used to rank the contribution of nodes and project status characteristics involved in risk prediction, and to identify the nodes ranked at the top of a preset number and their corresponding abnormal characteristics as key risk causes.

8. A deep learning-based industry-education integration project risk prediction and intervention terminal according to claim 1, characterized in that, The intervention instruction generation module stores an intervention strategy table, which records the correspondence between project risk type, risk level, key risk causes, responsible user type and intervention instruction template. The structured electronic intervention instructions shall include at least the project number, task node number, risk source, responsible user, completion deadline, processing requirements, and feedback entry point; When the risk level is low, the structured electronic intervention instruction is a reminder instruction; When the risk level is medium risk, the structured electronic intervention instruction is a review instruction; When the risk level is high, the structured electronic intervention instruction is a school-enterprise coordination instruction; When the risk level is severe, the structured electronic intervention instruction is a project adjustment instruction or a rectification instruction.

9. A deep learning-based industry-education integration project risk prediction and intervention terminal according to claim 1, characterized in that, The execution feedback collected by the feedback collection module includes whether the intervention instruction was completed, the completion time, the processing content, the responsible user confirmation information, the risk probability before intervention, the risk probability after intervention, and whether the risk was triggered again. The processor generates intervention effect data based on the risk probability before and after the intervention, and adds the intervention effect data to the risk prediction sample; The risk prediction samples are used to update the training sample library of the deep learning risk prediction model or to adjust the risk level threshold. Before inputting project data into the deep learning risk prediction model, the processor de-identifies and encodes sensitive fields in student identity information, enterprise business information, and project outcome materials, and uses the de-identified subject number to participate in the construction of the project time-series state matrix and project risk association diagram.

Images