Campus environment-oriented student psychological state monitoring and intervention decision method and system

Abstract

This invention discloses a method and system for monitoring and intervening in the psychological state of students in a campus environment, relating to the field of psychological monitoring technology. The method and system include: S1, preprocessing real-time campus data and historical auxiliary environmental data; S2, constructing a full-link temporal mapping and delay structure representation, outputting delay sequences and window constraints; S3, performing offline backhaul reverse order rearrangement and delay scale constraint screening based on the cumulative risk discrimination analysis results of event temporal misalignment; S4, performing quantile deviation replacement and interaction intensity gating constraints based on the group pollution index compression discrimination results; and S5, establishing an end-to-end intervention closed-loop archiving. This solves the problem of existing technologies that directly aggregate and fuse multi-source campus data according to a fixed time window without characterizing granularity differences and reporting delays, easily leading to event temporal misalignment and group false alarms, thereby reducing the reliability of early warnings.

Description

Technical Field

[0001] This invention relates to the field of psychological monitoring technology, specifically to a method and system for monitoring and intervening in the psychological state of students in a campus environment. Background Technology

[0002] In recent years, driven by policy and the digital transformation of education governance, campus mental health work has shifted from primarily "post-event handling" to a focus on "process management, collaborative intervention, and closed-loop evaluation." Schools have generally established mental health education and service systems, encompassing psychological testing, daily screening, feedback from homeroom teachers and moral education staff, psychological counseling and crisis intervention, and home-school communication. Simultaneously, supporting information platforms are continuously improving, forming multi-dimensional information records covering student growth, learning, and campus life. With the popularization of campus business systems and the widespread use of smart terminals, the collection and management of mental health-related information has gradually shifted from offline paper-based, scattered ledgers to online systems and structured management. Psychological counseling and treatment processes are also becoming increasingly standardized and procedural, emphasizing record keeping, access control, and follow-up visits. At the current technological level, the education and mental health service fields often employ methods such as data integration, statistical analysis, time series analysis, natural language processing, and risk assessment to characterize trends and provide early warnings of changes in students' mental states. The results are then used to assist teachers, mental health professionals, and administrators in conducting tiered responses, resource allocation, and intervention effectiveness evaluation. Against this backdrop, technologies related to mental health monitoring and intervention decision-making in the campus environment continue to develop, with the goal of improving the timeliness, coverage, and traceability of mental health services, and promoting the evolution of campus mental health work from experience-driven to data-supported and process-closed-loop.

[0003] For example, the invention patent with publication number CN120179422A discloses a student psychological data analysis and early warning system based on artificial intelligence. By comprehensively judging the complexity of multi-source student psychological datasets and slicing the data, and combining the performance of the graphics processor to match the compatibility complexity, the system achieves dynamic allocation of psychological data slices and graphics processors. On this basis, it completes the analysis and early warning output of student psychological data, and adjusts the processing efficiency of the graphics processor according to the processing delay, thereby improving the overall operating efficiency of the mental health monitoring platform in the process of multi-source psychological data analysis and early warning.

[0004] Existing campus psychological state analysis and early warning technologies mostly focus on the computational processing efficiency and early warning output process after data aggregation. They do not pay enough attention to the temporal consistency issues caused by factors such as differences in collection granularity, reporting delays, event supplementation delays, and constraints of work and rest windows in multi-source campus data. They often directly aggregate and merge data from different sources, with different update frequencies, and different accumulation mechanisms according to a fixed time window, making it difficult to stably align the sequence of events and causal chains. It is also difficult to distinguish between group periodic fluctuations and individual abnormal changes, which in turn leads to the deviation of the early warning trigger point, the concentration of false alarms, and the incomplete interpretation chain, affecting the usability and credibility of risk conclusions in campus handling scenarios.

[0005] Therefore, in response to the above problems, there is an urgent need for decision-making methods and systems for monitoring and intervening in the psychological state of students in the campus environment. Summary of the Invention

[0006] Technical problems to be solved

[0007] To address the shortcomings of existing technologies, this invention provides a method and system for monitoring and intervening in the psychological state of students in a campus environment. It solves the problem that existing technologies directly aggregate and fuse multi-source campus data according to a fixed time window without characterizing granular differences and reporting delays, which can easily lead to misalignment of event timing and false alarms, thereby reducing the reliability of early warnings.

[0008] Technical solution

[0009] To achieve the above objectives, this invention provides the following technical solution: a student psychological state monitoring and intervention decision-making method for campus environments, comprising: S1, collecting real-time campus data, acquiring historical environmental auxiliary data, and preprocessing the real-time campus data and historical environmental auxiliary data; S2, constructing a full-link time-series mapping and delay structure representation from event occurrence to data entry completion, and outputting delay sequences and window constraints to support time-series reconstruction; S3, performing event time-series misalignment cumulative risk discrimination analysis on the real-time campus data and historical environmental auxiliary data, and performing offline backhaul reverse order rearrangement and delay scale constraint screening based on the event time-series misalignment cumulative risk discrimination analysis results; S4, performing group pollution index compression discrimination on the real-time campus data and historical environmental auxiliary data, and performing quantile deviation replacement and interaction intensity gating constraints based on the group pollution index compression discrimination results; S5, mapping the fused evidence into executable intervention tasks and establishing an end-to-end intervention closed-loop archive.

[0010] Furthermore, the specific process of collecting real-time campus data and obtaining historical auxiliary environmental data is as follows: Collecting real-time campus data includes: student terminal touch event timestamp sequences, original content data of student terminal text messages, student terminal text message submission time sequences, student terminal data reporting request sending time sequences, server-side data reception time sequences, server-side database entry completion time sequences, bound entity identifier data, student daily attendance check-in time sequences, student daily attendance status data, student weekly grade release time sequences, student weekly grade numerical data, consultation appointment creation time sequences, consultation appointment start time sequences, psychological treatment record creation time sequences, psychological treatment record submission time sequences, psychological expression submission events, consultation appointment creation events, and psychological treatment record submission events; Obtaining historical auxiliary environmental data includes: school schedule data, class identifier data, historical touch event timestamp sequences, historical text message submission time sequences, historical reporting link original record data, historical attendance original record data, historical grade original record data, and historical consultation appointment and treatment original record data.

[0011] Furthermore, the specific preprocessing process for real-time campus data and historical environmental auxiliary data is as follows: A sliding window counting algorithm is used to generate a unit-time touch count sequence from the timestamp sequence of student terminal touch events, and outlier removal and smoothing are performed on the unit-time touch count sequence; a sliding window counting algorithm is used to generate a unit-time text submission count sequence from the student terminal text message submission time sequence, and outlier removal and smoothing are performed on the unit-time text submission count sequence; invalid text is removed from the original content data of student terminal text messages using text segmentation, character repetition rate filtering, and invalid symbol proportion discrimination algorithms, generating standardized text sequences and text length sequences; reverse order segment correction is performed on the student terminal data reporting request sending time sequence, server-side data receiving time sequence, and server-side database entry completion time sequence using timestamp monotonicity verification and offline return reverse order rearrangement algorithms; zero-mean unit variance standardization is performed on real-time campus data and historical environmental auxiliary data using mean-standard deviation standardization algorithms; and interval normalization is performed on real-time campus data and historical environmental auxiliary data using minimum-maximum value normalization algorithms.

[0012] Furthermore, the specific process of constructing a full-link time-series mapping and delay structure representation from event occurrence to database entry completion, and outputting delay sequences and window constraints to support time-series reconstruction, is as follows: Based on the student terminal touch event timestamp sequence, student terminal text message submission time sequence, student daily attendance check-in time sequence, student weekly grade release time sequence, consultation appointment creation time sequence, psychological treatment record creation time sequence, and psychological treatment record submission time sequence, combined with the student terminal data reporting request sending time sequence, server-side data receiving time sequence, and server-side database entry completion time sequence, a link alignment relationship is established between the event occurrence time, reporting request sending time, server receiving time, and database entry completion time; the reporting sending delay sequence, server receiving delay sequence, and database entry delay sequence are obtained by calculating the difference in the link alignment relationship, and window markers are generated for the event's time window based on the school's timetable data.

[0013] Furthermore, the specific process of performing cumulative risk discrimination analysis on real-time campus data and historical environmental auxiliary data for event time sequence misalignment is as follows: Time windows are obtained by filtering the time stamp sequence of student terminal touch events, the time sequence of student terminal text message submissions, the time sequence of student daily attendance check-in, the time sequence of student weekly grade releases, and the time sequence of psychological treatment record submissions according to window markers. Internal event set; for psychological expression submission events, consultation appointment creation events, and psychological treatment record submission events under the same bound subject identifier data, an event pair set is generated according to the event type combination rules; the time sequence of student terminal data reporting requests and the time of event occurrence are calculated by difference to obtain the first event. Article and the first The event reporting delay is calculated by determining the quantile interval of the historical reporting delay sequence based on the source type of the original record data of the historical reporting link. Source type, delay scale, and event Source type delay scale; calculate the first Event reporting delay and the first The difference in reporting transmission delay for each event is used to obtain the reporting transmission delay difference; the event is calculated. Source type, delay scale parameter, event The sum of the source type delay scale parameter and the zero-division protection constant yields the scale normalization denominator; the ratio of the reported transmission delay difference to the scale normalization denominator is calculated and the exponential function value is taken to obtain the exponential amplification; the exponential amplification is added by one and the natural logarithm is taken to obtain the exponential logarithmic compression; the exponential logarithmic compression of all event pairs within the event pair set is summed to obtain the event pair cumulative logarithm term; the absolute value of the median of the reported transmission delay set of the event set within the time window is calculated and added by one to obtain the window stability denominator; the ratio of the event pair cumulative logarithm term to the window stability denominator is reversed and the exponential function value is taken to obtain the misalignment suppression exponential term; the misalignment suppression exponential term is subtracted by one to obtain the event timing misalignment risk judgment value.

[0014] Furthermore, the specific process of offline backhauling reverse order rearrangement and delay scale constraint screening based on the cumulative risk discrimination analysis results of event time sequence misalignment is as follows: Real-time comparison of the event time sequence misalignment risk judgment value and the event time sequence misalignment risk judgment threshold: When the event time sequence misalignment risk judgment value is less than the event time sequence misalignment risk judgment threshold, output a time sequence fusion tag, and perform cross-source soft alignment fusion on the standardized text sequence, text length sequence, unit time touch count sequence, student daily attendance status data, student weekly grade data, consultation appointment start time sequence, and psychological treatment record submission time sequence under the same window constraint, and enter the group contamination index compression discrimination; when the event time sequence misalignment risk judgment value is greater than or equal to the event time sequence misalignment risk judgment value, the process is as follows: When the misplacement risk threshold is reached, an untrusted time sequence marker is output, and offline backhaul reverse order rearrangement and delay scale constraint screening are performed: the link delay is decomposed and verified based on the server receiving delay sequence and the data entry delay sequence, and the received but not yet entered and the data entry delayed completion segments are determined. The received but not yet entered segments meet the condition that the server-side data receiving time sequence has been generated and the server-side data entry completion time sequence is missing consecutive records. The data entry delayed completion segments meet the condition that the data entry delay has k consecutive records exceeding the upper quantile limit of the historical data entry delay distribution, and the event timestamps are re-arranged in segments with the segments as boundaries; for events whose data entry delay is outside the n upper quantile of the historical data entry delay distribution, an untrusted link marker is output and the data is filtered out from the time window. The internal event set is removed, and the psychological treatment records are submitted to the time sequence for pre-risk inference input and centrally isolated. Only the post-exposure interpretation evidence set is allowed to enter. The rearranged event set, the reporting and sending delay sequence, and the event time sequence misalignment risk judgment value trajectory are created and archived into the time sequence reconstruction database.

[0015] Furthermore, the specific process for compressing and discriminating the group pollution index using real-time campus data and historical environmental auxiliary data is as follows: A product-coupled mapping algorithm is used to generate a class interaction intensity sequence from the unit-time touch count sequence and the unit-time text submission count sequence; a historical unit-time touch count sequence and a historical unit-time text submission count sequence are generated based on the historical touch event timestamp sequence and the historical text message submission time sequence, and the historical class interaction intensity baseline sequence is obtained through the product-coupled mapping algorithm; the difference between the median of the class interaction intensity sequence and the median of the historical class interaction intensity baseline sequence is calculated and its absolute value is taken to obtain the absolute deviation term; the sum of the median of the historical class interaction intensity baseline sequence and constant one is calculated, and then divided by the zero protection constant is added to obtain the scale-stable denominator term; the ratio of the absolute deviation term to the scale-stable denominator term is calculated and its opposite is taken to obtain the negative deviation term; the exponential function value of the negative deviation term is taken and added to constant one to obtain the exponential plus term; the ratio of constant one and the exponential plus term is calculated to obtain the group pollution suppression discrimination value.

[0016] Furthermore, the specific process of replacing quantile deviation and applying interaction strength gating constraints based on the group pollution index compression discrimination results is as follows: Real-time comparison of the group pollution suppression discrimination value and the group pollution suppression discrimination threshold: When the group pollution suppression discrimination value is less than the group pollution suppression discrimination threshold, under the constraint of time-series fusionable label, the standardized text sequence, text length sequence, unit time touch count sequence, student daily attendance status data, student weekly grade data, consultation appointment start time sequence and psychological treatment record submission time sequence are fused to generate an individual risk sequence, which is then incorporated into the intervention decision-making process; When the group pollution suppression discrimination value is greater than or equal to the group pollution suppression discrimination threshold, output the group event window marker and enable degrouping fusion constraints: replace the interaction intensity scalar of each student with the quantile deviation in the class interaction intensity scalar distribution, and set the interaction intensity input of class synchronous migration to a gating item that does not participate in the individual intervention trigger judgment within the time window corresponding to the group event window marker. When triggering the judgment, reset the corresponding interaction intensity input weight to zero and prohibit entry into the comparison path. At the same time, bind the group event window marker with the school timetable data to create and archive it to the group event database.

[0017] Furthermore, the specific process of mapping fused evidence into executable intervention task orchestration and establishing end-to-end intervention closed-loop archiving is as follows: Based on time-series fusionable markers, time-series unreliable markers, and event time-series misalignment risk judgment values, combined with group contamination suppression discriminant values, group event window markers, and de-grouping fusion results, student psychological state monitoring results and intervention decision orchestration are formed: standardized text sequences and text length sequences, unit time touch count sequences, student daily attendance status data, student weekly grade numerical data, and consultation appointment start time sequences corresponding to the same bound subject identifier data are jointly organized to generate an intervention decision list, and trigger evidence items, trigger time windows, and receipt deadlines are bound to the intervention decision tasks; at the same time, time-series fusionable markers, time-series unreliable markers, group event window markers, and corresponding evidence items are created as closed-loop evidence packages and archived in the intervention closed-loop database for subsequent review and consistency verification.

[0018] The second aspect of this invention provides a student psychological state monitoring and intervention decision-making system for campus environments, comprising: a data acquisition and preprocessing module for acquiring real-time campus data and historical environmental auxiliary data, and preprocessing the real-time campus data and historical environmental auxiliary data; a granularity difference and reporting delay characterization module for constructing a full-link time-series mapping and delay structure representation from event occurrence to data entry completion, and outputting a delay sequence and window constraints to support time-series reconstruction; an event chain time-series reconstruction and misplacement risk discrimination module for performing event time-series misplacement cumulative risk discrimination analysis on real-time campus data and historical environmental auxiliary data, and performing offline back-transmission reverse order rearrangement and delay scale constraint screening based on the event time-series misplacement cumulative risk discrimination analysis results; a de-grouping fusion and group pollution suppression module for performing group pollution index compression discrimination on real-time campus data and historical environmental auxiliary data, and performing quantile deviation replacement and interaction intensity gating constraints based on the group pollution index compression discrimination results; and an intervention decision orchestration and closed-loop evidence archiving module for mapping fused evidence into executable intervention tasks and establishing end-to-end intervention closed-loop archiving.

[0019] Beneficial effects

[0020] The present invention has the following beneficial effects:

[0021] (1) This invention constructs a full-link time-series mapping and delay structure characterization from event occurrence to database entry completion, and performs time-series consistency constraints and reconstruction correction before fusion, thereby achieving the effect of aligning and fusing multi-source campus data under unified time semantics, effectively solving the problem of event time-series misalignment caused by direct aggregation of multi-source data according to fixed windows in the prior art.

[0022] (2) This invention introduces an event timing misalignment cumulative risk discrimination mechanism and performs offline backhaul reverse order rearrangement and link trust screening when the risk is triggered, thereby achieving the effect of isolating abnormal links and supplementary delayed records and preserving evidence, effectively solving the problem that the information supplemented after the event is misused as a clue for early warning in the prior art.

[0023] (3) This invention enables the suppression of the impact of group synchronous migration on individual risk conclusions by enabling degrouping fusion constraints and group pollution suppression discrimination in group fluctuation scenarios, thereby effectively solving the problem of concentrated outbreaks of early warning caused by periodic group events in the prior art.

[0024] (4) This invention organizes fusionable evidence into an executable list of intervention tasks and structurally binds the triggering basis of the tasks, thereby realizing the effect of transforming the early warning results into an executable intervention process that can be implemented. This effectively solves the problem that risk scores in the prior art cannot directly support the scheduling and collaborative execution of disposal.

[0025] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0026] Figure 1 This is a flowchart of the student psychological state monitoring and intervention decision-making method for campus environment according to the present invention;

[0027] Figure 2 This is a structural diagram of the student psychological state monitoring and intervention decision-making system for campus environment according to the present invention;

[0028] Figure 3 This is a comparison chart of the event pair reporting delay difference and scale synthesis quantity according to the present invention;

[0029] Figure 4 This is a flowchart of the group pollution suppression discrimination and degrouping fusion decision-making process of the present invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] Please see Figures 1-4This invention provides a technical solution: a student psychological state monitoring and intervention decision-making method for campus environments, comprising the following steps: S1, collecting real-time campus data and obtaining historical environmental auxiliary data, and preprocessing the real-time campus data and historical environmental auxiliary data; S2, constructing a full-link time-series mapping and delay structure representation from event occurrence to data entry completion, and outputting delay sequences and window constraints to support time-series reconstruction; S3, performing event time-series misalignment cumulative risk discrimination analysis on the real-time campus data and historical environmental auxiliary data, and performing offline backhaul reverse order rearrangement and delay scale constraint screening based on the event time-series misalignment cumulative risk discrimination analysis results; S4, performing group pollution index compression discrimination on the real-time campus data and historical environmental auxiliary data, and performing quantile deviation replacement and interaction intensity gating constraints based on the group pollution index compression discrimination results; S5, mapping the fused evidence into executable intervention tasks and establishing an end-to-end intervention closed-loop archive.

[0032] Specifically, the process of collecting real-time campus data and obtaining historical auxiliary environmental data is as follows: Real-time campus data includes: timestamp sequences of student terminal touch events, directly collected from touchscreen-driven event logs to construct minute-level interactive event sequences; original content data of student terminal text messages and time sequences of student terminal text message submissions, directly collected from the text submission interface to construct minute-level expressive event sequences; student terminal text message submission events, composed of bound subject identifier data, original text message content data, and text message submission time sequences, used to encapsulate the original text message content data and text message submission time sequences under the same subject constraint into computable event records for subsequent event set and event pair set generation; time sequences of student terminal data reporting requests, directly collected from terminal reporting queue logs to construct reporting and sending delay sequences; server-side data receiving time sequences and server-side database entry completion time sequences, directly collected from server receiving logs and database writing logs to construct server receiving delay sequences and database entry delay sequences; bound subject identifier data, directly collected from identity authentication logs and card swiping device logs to form event subject ownership consistency constraints; event record data structure; and event records... The recorded data structure includes at least binding subject identifier data, event type data, event occurrence time data, student terminal data reporting request sending time data, server-side data receiving time data, server-side data entry completion time data, and event payload data. The event payload data includes at least the original text message content data or touch event log field data, used to unify the expression of events from different sources and support the construction of link alignment relationships and delay sequence calculation; student daily attendance check-in time sequence and student daily attendance status data, including the number of times students arrived at school, left school, requested leave, and were late, directly collected by the campus attendance system. The system collects data on the following: daily behavioral event sequences; weekly student grade release time sequences and weekly student grade numerical data, directly collected by the academic affairs system, for constructing weekly behavioral event sequences; consultation appointment creation time sequences, consultation appointment start time sequences, psychological treatment record creation time sequences, and psychological treatment record submission time sequences, directly collected by the consultation management system, for constructing sequences of accumulated treatment events and supplementary delayed evidence; and psychological expression submission events, consultation appointment creation events, and psychological treatment record submission events, directly collected from the student terminal psychological service submission portal and the relevant business interfaces of the consultation management system's acceptance logs and call logs.

[0033] Historical environmental auxiliary data is acquired, including: school timetable data, including class time boundaries, break time boundaries, lunch break time boundaries, and evening self-study time boundaries, directly obtained from the academic affairs configuration, used for event attribution time window generation and window marking constraints; class identification data, directly obtained from the student registration system, used for group baseline construction and degrouping comparison set generation; historical touch event timestamp sequences and historical text message submission time sequences, directly obtained from terminal and server historical logs, used for historical interaction baseline generation; historical reporting link raw record data, including historical data reporting request sending time sequences, historical server-side data receiving time sequences, and historical server-side database entry completion time sequences, directly obtained from historical logs, used for delay distribution characterization and threshold generation; historical attendance raw record data, historical grade raw record data, and historical consultation appointment and handling raw record data, directly obtained from business system historical records, used for trend comparison and threshold quantile generation.

[0034] This implementation plan completes the unified collection and structured aggregation of multi-source real-time campus data and historical environmental auxiliary data, forming a full-scale observation input covering minute-level interactions and expressions, daily behaviors, weekly academic activities, and the disposal and sedimentation links. It constructs link timestamp evidence and solidifies and binds subject identification data constraints, providing time window generation and window marking constraint inputs supported by school timetable data and class identification data. It simultaneously accumulates historical touch and historical text baselines, historical reporting link delay distribution and historical business trend comparison data, providing a traceable, interpretable, and consistent basic data foundation for subsequent granularity difference and reporting delay characterization, event chain time sequence reconstruction, event time sequence misalignment risk identification, group pollution suppression identification and intervention decision-making.

[0035] Specifically, the preprocessing process for real-time campus data and historical environmental auxiliary data is as follows: A unit-time touch count sequence is generated from the student terminal touch event timestamp sequence using a sliding window counting algorithm. The unit time is the counting period, determined by the minimum time interval boundary of the school's timetable data. The integer division value of the minimum time interval boundary is used as the counting period length, ranging from 30 to 90 seconds. The unit-time touch count sequence is obtained by counting the touch event timestamps within each counting period. Anomaly removal and smoothing are performed on the unit-time touch count sequence to reduce the impact of repeated triggering and momentary jitter on interaction density statistics. Similarly, a unit-time text submission count sequence is generated from the student terminal text message submission time sequence using a sliding window counting algorithm. Anomaly removal and smoothing are performed on the unit-time text submission count sequence. The unit-time and unit-time touch count sequences use the same counting period length. The unit-time text submission count sequence is obtained by counting the text message timestamps within each counting period. The submission time points are counted to reduce the impact of short-term screen refreshes on expression density statistics. Invalid text is removed from the original content data of student terminal text messages using text segmentation, character repetition rate filtering, and invalid symbol proportion discrimination algorithms, generating standardized text sequences and text length sequences to form interpretable expression evidence and quantify length features. Reverse segment correction is performed on the student terminal data reporting request sending time sequence, server-side data receiving time sequence, and server-side database entry completion time sequence using timestamp monotonicity verification and offline return reverse order rearrangement algorithms to ensure that link timestamps meet monotonicity constraints. Zero-mean, unit-variance standardization is performed on real-time campus data and historical environmental auxiliary data using mean-standard-deviation standardization algorithms to unify dimensions and provide stable input for subsequent cross-source discrimination. Interval normalization is performed on real-time campus data and historical environmental auxiliary data using minimum-maximum value normalization algorithms to provide stable numerical input for subsequent event time-series misalignment risk discrimination and group pollution suppression discrimination.

[0036] This implementation plan completes the discretization and robustness processing of minute-level interactive event sequences and minute-level expression event sequences, forming computable feature inputs such as unit-time touch count sequences, unit-time text submission count sequences, standardized text sequences, and text length sequences. This reduces the interference of unstable factors such as repeated triggering, instantaneous jitter, and short-term screen refresh on the statistical results. It also completes the link timestamp consistency repair to ensure that the event chain meets the monotonicity constraint and reconstructability. Furthermore, it completes the dimensional unification and numerical domain stabilization of real-time campus data and historical environmental auxiliary data, providing an aligned, comparable, and reproducible basic input for event time sequence misalignment risk judgment and group pollution suppression judgment.

[0037] Specifically, the process of constructing a full-link time-series mapping and delay structure representation from event occurrence to database entry completion, and outputting delay sequences and window constraints to support time-series reconstruction, is as follows: Based on the student terminal touch event timestamp sequence, student terminal text message submission time sequence, student daily attendance check-in time sequence, student weekly grade release time sequence, consultation appointment creation time sequence, psychological treatment record creation time sequence, and psychological treatment record submission time sequence, combined with the student terminal data reporting request sending time sequence, server-side data receiving time sequence, and server-side database entry completion time sequence, a link alignment relationship is established between the event occurrence time, reporting request sending time, server receiving time, and database entry completion time. The link alignment relationship is based on the binding subject identifier data and event type data of the event record data structure forming a subject type alignment key, and the event occurrence time data and student terminal... The time data of the terminal data reporting request sending time constitutes the time consistency alignment key. Simultaneously, a unique request identifier is introduced as a precise alignment key to handle duplicate submissions of the same bound subject identifier data within the same counting period, ensuring the uniqueness of the pairing object for difference calculation. The link alignment relationship is calculated to obtain the reporting sending delay sequence, server receiving delay sequence, and database entry delay sequence. The difference calculation uses a one-to-one correspondence between the event occurrence time data of the same event record and the student terminal data reporting request sending time data, server-side data receiving time data, and server-side database entry completion time data to avoid delay distortion caused by cross-event mismatch. The server receiving delay sequence and database entry delay sequence are used to perform delay decomposition rearrangement and link reliability filtering. Window markers are generated for the event's time window based on the school's timetable data; these window markers are used to generate the time window. The event set and event pair set are constrained to be paired only within the same work window, and are used to ensure consistency between the trigger window and the archive window of the intervention task. The window mark is synchronously written into the event record data structure as the belonging index field for subsequent reordering and consistent retrieval of archives, thereby forming the delay distribution parameters and alignment constraint inputs for subsequent time series reconstruction.

[0038] This implementation plan establishes link alignment relationships and computable delay representations covering multi-source events, precipitates unified delay distribution parameters such as reported transmission delay sequences, server reception delay sequences, and database entry delay sequences, supports constraint inputs for delay decomposition and rearrangement and link reliability screening, generates window markers based on school timetable data and solidifies event attribution time windows, enabling event sets and event pair sets within time windows to have window pairing constraints and boundary consistency constraints, establishes a consistency index foundation for intervention task trigger windows and archiving windows, and outputs an alignment constraint input base for subsequent event chain time sequence reconstruction and misplacement risk assessment.

[0039] Specifically, the process of performing cumulative risk discrimination analysis on event time sequence misalignment for real-time campus data and historical environmental auxiliary data is as follows: Time windows are obtained by filtering the timestamp sequence of student terminal touch events, the time sequence of student terminal text message submissions, the time sequence of daily student attendance check-in, the time sequence of weekly student grade releases, and the time sequence of psychological intervention record submissions according to window markers. An internal event set; for psychological expression submission events, consultation appointment creation events, and psychological treatment record submission events under the same bound subject identifier data, an event pair set is generated according to the event type combination rule. The event type combination rule is that only adjacent event types are paired within the same window marker, i.e., only event pairs of psychological expression submission events and consultation appointment creation events, and event pairs of consultation appointment creation events and psychological treatment record submission events are generated. The time sequence of student terminal data reporting requests and the time of event occurrence are calculated by the difference. Article and the first The reporting delay of an event is determined by the event occurrence time and the student terminal data reporting request sending time, which represents the time when the event is submitted to the server by the terminal. The difference between the two represents the reporting delay and serves as the input for timing misalignment judgment and rearrangement constraints. The quantile interval of the historical reporting delay sequence is calculated based on the source type of the original historical reporting link data to obtain the event. Source type, delay scale, and event The source type delay scale is defined as at least one of the following: touch event, text message submission event, attendance check-in event, grade release event, consultation appointment creation event, and psychological treatment record submission event. The quartile interval adopts the interquartile range, which is the difference between the 75th and 25th quartiles of the historically reported transmission delay sequence. This is used to characterize typical fluctuation bandwidth with robust dispersion and reduce the dominant role of long-tail supplementary samples in the delay scale. When the proportion of long-tail supplementary samples exceeds a preset ratio, the quartile interval is switched to the difference between the 90th and 10th quartiles to enhance sensitivity to tail expansion. The switching result is written into the time series reconstruction database as the value caliber of the source type delay scale for event i and the source type delay scale for event j.

[0040] Calculate the first Event reporting delay and the first The difference in reporting transmission delay for each event is used to obtain the reporting transmission delay difference, which characterizes the relative displacement strength of adjacent event pairs within the same window marker on the reporting link to support subsequent misalignment sensitivity amplification; the event is calculated. Source type, delay scale parameter, event The sum of the source type delay scale parameter and the division-by-zero protection constant yields the scale-normalized denominator, explicitly incorporating the delay dispersion differences of different source types in the acquisition and transmission links into the scale constraint to avoid a single source type dominating the judgment. The ratio of the reported transmission delay difference to the scale-normalized denominator is calculated and the exponential function value is taken to obtain the exponential amplification, used for nonlinear separation of the positive and negative directions of the delay difference and improving the response sensitivity to tail-misplaced samples. The exponential amplification is incremented by one and the natural logarithm is taken to obtain the exponential logarithmic compression, suppressing the numerical explosion caused by extreme delay differences and keeping the cumulative term within a controllable dynamic range. The exponential logarithmic compression of all event pairs within the event pair set is summed to obtain the event pair cumulative compression. The logarithmic term is used to form the overall misalignment evidence strength for all event pairs within the window mark and to achieve additive aggregation of multiple constraint pairs. The absolute value of the median of the reported transmission delay set of the event set within the time window is calculated and incremented by one to obtain the window stability denominator, which is used to characterize the typical link delay level within the window mark with robust statistics and reduce the interference of outliers on the scaling benchmark. The inverse of the ratio of the cumulative logarithmic term of the event pair to the window stability denominator is taken as the exponential function value to obtain the misalignment suppression index term, which maps the window-level misalignment evidence to a monotonically bounded suppression quantity to form a stable correspondence with subsequent discrimination thresholds. The misalignment suppression index term is subtracted by one to obtain the event time-series misalignment risk judgment value, with the specific calculation formula as follows:

[0041] ;

[0042] In the formula, The value represents the risk assessment of event timing misalignment, characterizing the overall strength of the disruption of the event chain sequence constraints within the window marker by link delay disturbances; i, j represent event pair indices, where This indicates the index of the previous event in the event pair. This indicates the index of the next event in the event pair; the previous and next events are determined by the event type combination rules, which include at least the sequential constraints that the psychological expression submission event precedes the consultation appointment creation event and the consultation appointment creation event precedes the psychological treatment record submission event. This is used to transform the prior directional causal order of the business into a computable directional constraint of the event pair to reduce the reverse pairing caused by correlation alone. Indicates time window Internal event set, time window Determined by window labels, it is used to assign events to class time windows, break time windows, lunch break windows, or evening self-study time windows, and to constrain events to enter the event set only within the same window label to avoid incorrect pairing across schedule windows. It also suppresses pseudo-misplacement caused by cross-window splicing at the boundary of the schedule and ensures that the event pair set is generated only in the semantically consistent time domain. This represents a set of event pairs. Event pairs are generated only within a subset of events with consistent window labels to avoid pairing post-event events with pre-event events from different time periods. This restricts the pairing space to the candidate domain of the same window to reduce combinatorial explosion and improve the interpretability of event pair constraints. Indicates the first The event reporting delay characterizes the link lag between the time the event occurs and the time the terminal sends the data, and serves as a direct observation for misalignment determination. Indicates the first The delay in reporting each event constitutes the relative delay difference between event pairs to characterize the direction of sequential disturbances within the pair; Indicates an event Source type delay scale, characterizing the delay fluctuation scale of the source type to which event i belongs in the historical link to provide a robust scaling benchmark for the denominator of the scaling; Indicates an event Source type delay scale, which characterizes the delay fluctuation scale of the source type of event j in the historical link to form a symmetrical scale constraint with event i; This represents the median operator, used to characterize the lag level within the window label with robust location statistics and reduce the impact of long-tailed samples on the baseline term; This represents the division-to-zero protection constant, which is the smallest resolvable positive value corresponding to the timestamp resolution of the historical delay sequence. Its value ranges from 0.0005 to 0.001. It is used to maintain numerical stability and avoid the discontinuity of judgment caused by division-to-zero when the scale-normalized denominator or the median term is close to zero.

[0043] In this embodiment, Table 1 is a comparison table of reporting delay and scale synthesis quantity for window-labeled consistent event pairs. It records in detail the reporting delay of the i-th event, the reporting delay of the j-th event, the source type delay scale synthesis quantity, the exponential logarithmic compression quantity, and the cumulative logarithmic term of the event pair within the time window t for different event pair indices. This is used to quantify the relative displacement strength and the level of accumulative misplacement evidence of the event pair in the reporting link under the window-labeled consistency constraint. Specifically: for event pair index (1, 2), the reporting delay for the i-th event is 0.12, the reporting delay for the j-th event is 0.18, the source type delay scale composite is 0.181, the exponential logarithmic compression is 0.541, and the cumulative logarithmic term for the event pair is 0.541; for event pair index (2, 3), the reporting delay for the i-th event is 0.18, the reporting delay for the j-th event is 0.25, the source type delay scale composite is 0.221, the exponential logarithmic compression is 0.547, and the cumulative logarithmic term for the event pair is 1.088; for event pair index (3, 4), the reporting delay for the i-th event is 0.25, the reporting delay for the j-th event is 0.40, the source type delay scale composite is 0.321, the exponential logarithmic compression is 0.487, and the cumulative logarithmic term for the event pair is 1.57. 5; The reporting delay for the i-th event corresponding to event pair index (4, 5) is 0.40, the reporting delay for the j-th event is 0.33, the source type delay scale composite amount is 0.351, the exponential logarithmic compression amount is 0.798, and the cumulative logarithmic term for the event pair is 2.373; The reporting delay for the i-th event corresponding to event pair index (5, 6) is 0.33, the reporting delay for the j-th event is 0.45, the source type delay scale composite amount is 0.271, the exponential logarithmic compression amount is 0.499, and the cumulative logarithmic term for the event pair is 2.872; The reporting delay for the i-th event corresponding to event pair index (6, 7) is 0.45, the reporting delay for the j-th event is 0.52, the source type delay scale composite amount is 0.361, the exponential logarithmic compression amount is 0.601, and the cumulative logarithmic term for the event pair is 3.473.

[0044]

[0045] like Figure 3The figure shows a comparison of the difference in reporting transmission delay between event pairs and the scale synthesis quantity. Combined with Table 1, it can be seen that the relative difference between the reporting transmission delay of the i-th event and the j-th event under different event pair indices exhibits a directional change, and forms a comparable expression of the normalized difference under the constraint of the source type delay scale synthesis quantity. Specifically, the normalized difference corresponding to the reporting transmission delay difference of event pair index (3, 4) is at a relatively prominent level, reflecting that the relative displacement of the link within the same window label is more easily amplified by the exponential amplification and contributes to the accumulation of misalignment evidence. Although the reporting transmission delay of the i-th event and the j-th event show an inverse difference in event pair index (4, 5), it still forms a considerable normalized difference under the effect of the source type delay scale synthesis quantity, reflecting a sensitive response to the direction of the sequential constraint disturbance. The normalized differences of event pair indices (1, 2), (2, 3), and (6, 7) are in a relatively convergent range, indicating that their link displacement strength contributes more robustly to the accumulation of misalignment under the same window label constraint. Overall, Figure 3 It intuitively presents the comparable differences in the reporting delay difference of event pairs under the constraint of scale composition, which can serve as a visual support for the formation process of the cumulative logarithm of event pairs in the event timing misalignment risk judgment value.

[0046] In this implementation scheme, the structured construction and directional sequence constraints of the event set and event pair set within the time window are solidified to form a candidate domain and an interpretable event pair constraint system oriented towards the window marker. A standardized input with the same caliber for reporting transmission delay and source type delay scale is generated. The output is a misplacement evidence aggregation result with explicit suppression capability for the difference in delay dispersion of different source types. A monotonically bounded event time sequence misplacement risk judgment value is constructed to characterize the comprehensive strength of the event chain sequence constraints within the window marker being destroyed by link delay disturbances. A core discriminant quantity that can be used for time sequence misplacement judgment, reverse order rearrangement constraint and link credibility screening decision is provided.

[0047] Specifically, the process of offline backhauling and delay scale constraint screening based on the cumulative risk discrimination analysis results of event time sequence misalignment is as follows: real-time comparison of the event time sequence misalignment risk judgment value and the event time sequence misalignment risk judgment threshold:

[0048] When the event timing misalignment risk judgment value is less than the event timing misalignment risk judgment threshold, the output timing fusion mark is used to solidify the validity of the window constraint between the event set and the event pair set within the time window t to ensure that the subsequent fusion input domain meets the window mark consistency. The standardized text sequence, text length sequence, unit time touch count sequence, student daily attendance status data, student weekly grade data, consultation appointment start time sequence and psychological treatment record submission time sequence are fused across sources under the same window constraint. The phase drift caused by the sampling granularity difference of different source types is suppressed by the alignment residual minimization strategy and the comparability of fusion features is maintained. Then, it enters the group contamination index compression discrimination.

[0049] When the event timing misalignment risk judgment value is greater than or equal to the event timing misalignment risk judgment threshold, a timing untrustworthy flag is output, triggering a link trustworthiness review of the event set within the time window t to prevent misaligned samples from entering the fusion input domain and causing false alarm propagation. Offline backhaul reverse order rearrangement and delay scale constraint screening are performed: the link delay is decomposed and verified based on the server received delay sequence and the database entry delay sequence. The decomposed delay components are used to locate link congestion and write lag-dominant segments to support the boundary determination of subsequent segment rearrangement. Segments that have been received but not yet entered into the database and those that have been entered into the database but are later written are identified. Segments that have been received but not yet entered into the database satisfy the following conditions: The conditions for continuous records where the server-side data reception time series has been generated and the server-side data entry completion time series is missing are met. For data entry delay supplementation segments, the condition for consecutive records where the data entry delay exceeds the upper quantile limit of the historical data entry delay distribution (k ranges from 3 to 10) is met. The segment boundary is determined by the continuity interruption point. Event timestamps are reordered segmentally using the segment as the boundary, and within each segment, events are sorted by their occurrence time using the server-side data reception time series as the sorting key. For events where the data entry delay is outside the n upper quantile of the historical data entry delay distribution, an untrusted link is output and the event is processed within the time window. The internal event set is removed by using upper quantile thresholds to robustly truncate long-tail inbound delay anomalies to prevent extreme write events from dominating the judgment. The historical inbound delay sequence is calculated by subtracting the historical server-side inbound completion time sequence from the historical server-side data reception time sequence, forming a reproducible link performance baseline to support consistent removal decisions across different time periods. The corresponding threshold value is determined within the upper quantile range using a quantile calculation algorithm, with n ranging from 90% to 99%. The psychological treatment records are submitted to the time sequence for centralized isolation from pre-risk inference input, allowing them to enter only the post-interpretation evidence set. The rearranged event set, the reported transmission delay sequence, and the event timing misalignment risk judgment value trajectory are created and archived in the time series reconstruction database, forming a traceable reconstruction version and adjudication evidence to support subsequent consistency verification and responsibility determination.

[0050] In this implementation plan, an online adjudication and diversion mechanism based on the risk judgment value of event time sequence misalignment is formed. When the output time sequence is fusionable, it ensures that cross-source features enter the fusionable input domain under the window marking constraint and maintains the consistency of the fusion caliber. When the output time sequence is untrustworthy, it completes the link delay decomposition verification, reverse segment reordering and link untrustworthy event elimination, and realizes the isolation constraint between the pre-risk inference input and the post-exposure interpretation evidence set. The reordering results and judgment trajectory are solidified and written into the time sequence reconstruction database, which improves the time sequence credibility of the early warning input, the interpretability of abnormal handling and the traceability of the whole process.

[0051] Specifically, the process of compressing and discriminating the group pollution index using real-time campus data and historical environmental auxiliary data is as follows: A product-coupled mapping algorithm is used to generate a class interaction intensity sequence from the unit-time touch count sequence and the unit-time text submission count sequence. This algorithm has a synergistic amplification characteristic for the same-direction changes of the two types of sequences and a suppression characteristic for single-source isolated surges, thereby improving the discriminability of group synchronization signals. This is used to map the synchronous activity of touch behavior and text expression into a same-scale intensity representation to enhance the distinguishability of group same-direction fluctuations. The class interaction intensity sequence maintains comparable time granularity under window labeling constraints to support robust cross-window comparisons. Historical unit-time touch count sequences and historical unit-time text submission count sequences are generated based on historical touch event timestamp sequences and historical text message submission time sequences. The historical unit-time sequences are generated with the same counting period length to ensure consistency between historical and current statistical standards and reduce the risk of misjudgment caused by baseline drift. A historical class interaction intensity baseline sequence is obtained through the product-coupled mapping algorithm. A comparable historical steady-state reference is formed using the same-caliber mapping result to support robust discrimination of group shifts.

[0052] The absolute deviation term is obtained by calculating the difference between the median of the class interaction intensity sequence and the median of the historical class interaction intensity baseline sequence. This is a robust position statistic that eliminates the interference of extreme individual samples on the overall class intensity characterization and characterizes the class's offset relative to the historical baseline. The sum of the median of the historical class interaction intensity baseline sequence and a constant is calculated, and then a division-by-zero protection constant is added to obtain a scale-stable denominator term. This maintains the scale term's continuity and differentiability in the interval where the historical baseline intensity is close to zero and improves the numerical stability of the discrimination in low-activity scenarios. The negative deviation term is obtained by calculating the ratio of the absolute deviation term to the scale-stable denominator term and taking its opposite. This maintains the scale term's continuity and differentiability in the interval where the historical baseline intensity is close to zero and improves the numerical stability of the discrimination in low-activity scenarios. The negative deviation term is then exponentially added to the constant to obtain an exponential addition term. This exponential form provides a gradual response to small offsets and a progressive enhancement to sustained offsets, while avoiding sensitive oscillations caused by the ratio term directly participating in threshold adjudication. The ratio of the constant and the exponential addition term is calculated to obtain the group contamination suppression discriminant value. The specific calculation formula is as follows:

[0053] ;

[0054] In the formula, This represents the group pollution suppression discriminant value, which characterizes the discriminant output after exponential compression of the robust shift of the class interaction intensity sequence relative to the historical class interaction intensity baseline sequence; It represents the class interaction intensity sequence and provides an instantaneous intensity representation of the synchronous activity level of classes within a window to participate in group offset discrimination; This represents the baseline sequence of historical class interaction intensity, providing historical reference intensity of the same caliber to support baseline comparison under cross-cycle fluctuations; The median operator is used to reduce the impact of extremely high-frequency individual interactions on the overall class intensity statistics and enhance robustness to group-wide structural shifts. This represents the division-to-zero protection constant, which is the smallest resolvable positive value corresponding to the timestamp resolution of the historical delay sequence. Its value ranges from 0.0005 to 0.001. It is used to maintain the continuity of the function when the denominator is close to zero and to avoid false triggering of the group event window marker caused by numerical instability.

[0055] In this implementation plan, a quantitative representation of the synchronous activity intensity at the class level with the same caliber and a historical steady-state comparison baseline are formed. The output class interaction intensity sequence and historical class interaction intensity baseline sequence are enhanced with the ability to distinguish the coordinated fluctuations of touch behavior and text expression. A group contamination suppression discriminant value based on median robust shift and exponential compression is constructed to stably characterize the structural migration intensity of the class as a whole relative to the historical baseline. This reduces the perturbation sensitivity of extreme individual samples and low-activity intervals to the discriminant output, and provides interpretable and reproducible core discriminant input for group event window marking triggering and degrouping gating constraints.

[0056] Specifically, the process of replacing quantile deviations and applying interaction intensity gating constraints based on the group pollution index compression discrimination results is as follows: Figure 4 The diagram shows the process flow of the fusion decision-making process for swarm contamination suppression discrimination and de-swarming, which compares the swarm contamination suppression discrimination value and the swarm contamination suppression discrimination threshold in real time.

[0057] When the group pollution suppression discriminant value is less than the group pollution suppression discriminant threshold, it is determined that the class interaction intensity sequence does not show observable structural migration relative to the historical class interaction intensity baseline sequence and maintains the statistical stability of the individual judgment input domain. Under the constraint of time-series fusionable label, the standardized text sequence, text length sequence, unit time touch count sequence, student daily attendance status data, student weekly grade data, consultation appointment start time sequence and psychological treatment record submission time sequence are fused to generate an individual risk sequence. The individual risk sequence is obtained by constructing a feature vector and calculating a robust comprehensive score after normalizing the above fused features according to a unified caliber within the time window t. The robust comprehensive score is formed by mapping the quantile limit value out-of-bounds count and the median deviation intensity to form a sequence output in the interval of 0 to 1. The fusion process maintains the alignment of features with the same caliber to reduce weight drift caused by different sampling granularities and enhance the interpretability of risk trajectories, and then enters the intervention decision orchestration.

[0058] When the group pollution suppression discriminant value is greater than or equal to the group pollution suppression discriminant threshold, it is determined that there is a pollution risk dominated by class-level synchronous fluctuations in the current time window, and the de-grouping constraint of individual triggering conditions is triggered. The group event window flag is output, and the de-grouping fusion constraint is enabled. The common factors of the class level are separated from the individual triggering input to avoid the misinterpretation of group synchronous activity as individual anomalies. The interaction intensity scalar of each student is replaced with the quantile deviation in the distribution of the interaction intensity scalar of the class. The quantile deviation is obtained by taking the value of the empirical cumulative distribution function of the student interaction intensity scalar in the set of interaction intensity scalars in the class. The value of the empirical cumulative distribution function is relative to extremely high frequencies. Interaction samples exhibit robust compression to reduce the dominant role of tail samples in triggering. The difference between the quantile and the median of the scalar set of interaction intensity within the class is used as the quantile deviation. Within the time window corresponding to the group event window marker, the interaction intensity input of synchronous migration in the class is set as a gating item that does not participate in the individual intervention triggering determination. When triggering determination, the gating item resets the corresponding interaction intensity input weight to zero and prohibits it from entering the threshold comparison path. The gating item only affects the triggering determination path and retains the original input for the post-event interpretation evidence set to support review. At the same time, the group event window marker is bound to the school timetable data to create and archive it to the group event database.

[0059] In this implementation plan, an online triage and adjudication mechanism based on the group pollution suppression discrimination value is formed. When the group pollution suppression discrimination value meets the group pollution suppression discrimination threshold constraint, the individual fusion path under the temporally fusionable label is maintained and the individual risk sequence is output into the intervention decision orchestration. When the group pollution suppression discrimination value triggers the threshold to exceed the limit, the group event window label is output and the degrouping fusion constraint is enabled. The interaction intensity input is replaced with the quantile deviation and the gating of the class synchronous migration interaction intensity is set to not participate in the individual intervention trigger judgment. The pollution isolation of individual risk judgment by the group synchronous fluctuation is completed. The binding relationship between the group event window label and the school timetable data is solidified and archived into the group event database, which improves the individual orientation of the intervention trigger, the interpretability of the adjudication and the traceability of evidence.

[0060] Specifically, the process of mapping fused evidence into executable intervention tasks and establishing an end-to-end intervention closed-loop archive is as follows: Based on time-series fusionable markers, time-series unreliable markers, and event time-series misalignment risk judgment values, combined with group contamination suppression discriminant values, group event window markers, and degrouping fusion results, student psychological state monitoring results and intervention decision-making are arranged: Joint evidence organization is performed on standardized text sequences and text length sequences, unit-time touch count sequences, student daily attendance data, student weekly grade data, and consultation appointment start time sequences corresponding to the same bound subject identifier data to generate an intervention decision list. Among these, the student daily attendance data change summary is obtained by comparing the student daily attendance data with historical attendance records; the student weekly grade data change summary is obtained by comparing the student weekly grade data with historical grade records; and the consultation appointment and treatment time sequence comparison is provided by the baseline interval provided by historical consultation appointment and treatment records, used to generate quantile threshold values ​​for intervention triggering conditions. The attendance status change summary and the grade change summary are subjected to boundary judgment to form the triggering conditions for intervention tasks. The intervention decision list includes at least the observation and follow-up tasks, the on-campus interview tasks, and the counseling appointment guidance tasks. The student's psychological state evidence includes at least the statistical characteristics of expression frequency and text length corresponding to the standardized text sequence, the occurrence and supplementation time sequence characteristics of counseling appointment and treatment records, and the functional impairment signs evidence formed by the attendance status change summary and the grade change summary. The evidence is used to characterize the fluctuation and persistence of the student's psychological state and drive the intervention decision diversion. The intervention decision tasks are bound to trigger evidence items, trigger time windows, and receipt deadlines. The trigger time window is determined by the window mark, and the corresponding window mark is synchronously written into the evidence package as an archived window index to ensure the consistency and traceability of "trigger window - evidence window - receipt window". At the same time, the time sequence fusion mark, the time sequence unreliable mark, the group event window mark, and the corresponding evidence items are created as a closed-loop evidence package and archived in the intervention closed-loop database for subsequent review and consistency verification.

[0061] This implementation plan establishes a multi-marker joint adjudication output and evidence closed-loop mechanism for intervention decision orchestration. Time-series fusionable markers, time-unreliable markers, event time-series misalignment risk judgment values, group pollution suppression judgment values, group event window markers, and de-grouping fusion results jointly constrain the boundaries of the monitoring input domain and the trigger input domain. It completes the joint evidence organization of multi-source evidence under the same bound subject identifier data and generates an intervention decision list. It completes the construction and boundary judgment of quantile limits based on historical attendance records, historical performance records, and historical consultation appointment and handling records to solidify intervention trigger conditions. It outputs executable tasks such as observation and follow-up tasks, on-campus interview tasks, and consultation appointment guidance tasks, and binds trigger evidence items, trigger time windows, and receipt deadlines. Window markers are entered into the evidence package index to ensure the consistency and traceability of trigger windows, evidence windows, and receipt windows. The closed-loop evidence package is archived in the intervention closed-loop database to support review and consistency verification.

[0062] like Figure 2 As shown, the second aspect of this invention provides a student psychological state monitoring and intervention decision-making system for campus environments, comprising: a data acquisition and preprocessing module for acquiring real-time campus data and historical environmental auxiliary data, and preprocessing the real-time campus data and historical environmental auxiliary data; a granularity difference and reporting delay characterization module for constructing a full-link time-series mapping and delay structure representation from event occurrence to data entry completion, and outputting a delay sequence and window constraints to support time-series reconstruction; an event chain time-series reconstruction and misplacement risk discrimination module for performing event time-series misplacement cumulative risk discrimination analysis on real-time campus data and historical environmental auxiliary data, and performing offline back-transmission reverse order rearrangement and delay scale constraint screening based on the event time-series misplacement cumulative risk discrimination analysis results; a de-grouping fusion and group pollution suppression module for performing group pollution index compression discrimination on real-time campus data and historical environmental auxiliary data, and performing quantile deviation replacement and interaction intensity gating constraints based on the group pollution index compression discrimination results; and an intervention decision orchestration and closed-loop evidence archiving module for mapping fused evidence into executable intervention tasks and establishing end-to-end intervention closed-loop archiving.

[0063] This implementation plan establishes an integrated processing link for monitoring and intervening in the psychological state of students in the campus environment. It achieves preprocessing of multi-source campus data with consistent caliber and full-link temporal reliability constraint modeling. It outputs delayed sequences and window marking constraint inputs that can be used for event chain temporal reconstruction. It constructs a closed-loop handling mechanism for event temporal misalignment risk judgment and offline backhaul reverse order rearrangement. It forms an individual triggering anti-pollution capability with group pollution suppression discrimination and de-grouping fusion gating. It completes the generation of intervention decision list, trigger window consistency binding and closed-loop evidence package archiving, and improves the credibility of early warning conclusions, the interpretability of decisions and the traceability of the whole process.

[0064] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0065] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for monitoring and intervening in the psychological state of students in a campus environment, characterized by: Includes the following steps: S1 collects real-time campus data, obtains historical environmental auxiliary data, and preprocesses the real-time campus data and historical environmental auxiliary data. S2 constructs the end-to-end time-series mapping and delay structure representation from event occurrence to database entry completion, and outputs the delay sequence and window constraints that support time-series reconstruction; S3 performs cumulative risk discrimination analysis on real-time campus data and historical environmental auxiliary data, and performs offline back-up reverse order rearrangement and delay scale constraint screening based on the results of the cumulative risk discrimination analysis on cumulative risk of event time sequence misalignment. S4 performs group pollution index compression and discrimination on real-time campus data and historical environmental auxiliary data, and performs quantile deviation replacement and interaction intensity gating constraints based on the group pollution index compression and discrimination results. S5 maps fused evidence into actionable intervention tasks and establishes an end-to-end intervention closed-loop archive. The specific process of performing cumulative risk discrimination analysis on event time sequence misalignment of real-time campus data and historical environmental auxiliary data is as follows: The time window is obtained by filtering the student terminal touch event timestamp sequence, student terminal text message submission time sequence, student daily attendance check-in time sequence, student weekly grade release time sequence, and psychological treatment record submission time sequence according to window markers. Internal event set; for psychological expression submission events, consultation appointment creation events, and psychological treatment record submission events under the same bound subject identifier data, an event pair set is generated according to the event type combination rules; the time sequence of student terminal data reporting requests and the time of event occurrence are calculated by difference to obtain the first event. Article and the first The event reporting delay is calculated by determining the quantile interval of the historical reporting delay sequence based on the source type of the original record data of the historical reporting link. Source type, delay scale, and event Source type delay scale; The specific formula for calculating the risk assessment value of event timing misalignment is as follows: ； In the formula, This indicates the risk assessment value for event timing misalignment; i, j represent the event pair indices, where This indicates the index of the previous event in the event pair. This indicates the index of the next event in the event pair; Indicates time window Internal event set, time window Determined by window markers; Represents a set of event pairs; Indicates the first Event reporting delay; Indicates the first Event reporting delay; Indicates an event Source type delay scale; Indicates an event Source type delay scale; Represents the median operator; Represents the division-by-zero protection constant; The specific process of offline backhaul reverse order rearrangement and delay scale constraint screening based on the cumulative risk discrimination analysis results of event time sequence misalignment is as follows: Real-time comparison of event timing misalignment risk assessment values ​​and event timing misalignment risk assessment thresholds: When the event timing misalignment risk judgment value is less than the event timing misalignment risk judgment threshold, the output timing fusion mark is used to perform cross-source soft alignment fusion on the standardized text sequence, text length sequence, unit time touch count sequence, student daily attendance status data, student weekly grade data, consultation appointment start time sequence and psychological treatment record submission time sequence under the same window constraint, and enter the group contamination index compression judgment. When the event timing misalignment risk assessment value is greater than or equal to the event timing misalignment risk assessment threshold, an untrusted timing flag is output, and offline backhaul reverse order rearrangement and delay scale constraint screening are performed: the link delay is decomposed and verified based on the server receiving delay sequence and the data entry delay sequence, and the received but not yet entered and the data entry delayed completion segments are determined. The received but not yet entered segments meet the condition that the server-side data receiving time sequence has been generated and the server-side data entry completion time sequence is missing consecutive records. The data entry delayed completion segments meet the condition that the data entry delay has k consecutive records exceeding the upper quantile limit of the historical data entry delay distribution, and the event timestamps are re-arranged in segments with the segments as boundaries; for events whose data entry delay is outside the n upper quantile of the historical data entry delay distribution, an untrusted link flag is output and the data is processed from the time window. The internal event set is removed, and the psychological treatment records are submitted to the time sequence for pre-risk inference input and centrally isolated. Only the post-interpretation evidence set is allowed to enter. The rearranged event set, the reporting and sending delay sequence and the event time sequence misalignment risk judgment value trajectory are created and archived into the time sequence reconstruction database. The specific process for compressing and judging the population pollution index using real-time campus data and historical environmental auxiliary data is as follows: A product-coupled mapping algorithm is used to generate a class interaction intensity sequence from the unit time touch count sequence and the unit time text submission count sequence; a historical unit time touch count sequence and a historical unit time text submission count sequence are generated based on the historical touch event timestamp sequence and the historical text message submission time sequence, and a historical class interaction intensity baseline sequence is obtained through the product-coupled mapping algorithm. The specific formula for calculating the group pollution inhibition discriminant value is as follows: ; In the formula, Indicates the group pollution suppression discriminant value; Represents the sequence of class interaction intensity; This represents the baseline sequence of historical class interaction intensity; Represents the median operator; Represents the division-by-zero protection constant; The specific process of performing quantile deviation replacement and interaction intensity gating constraints based on the compressed discrimination results of the population pollution index is as follows: Real-time comparison of quorum contamination suppression discriminant value and quorum contamination suppression threshold: When the group pollution suppression discrimination value is less than the group pollution suppression discrimination threshold, under the constraint of time-series fusionable label, the standardized text sequence, text length sequence, unit time touch count sequence, student daily attendance status data, student weekly grade data, consultation appointment start time sequence and psychological treatment record submission time sequence are fused to generate an individual risk sequence, which is then incorporated into the intervention decision-making process. When the group pollution suppression discrimination value is greater than or equal to the group pollution suppression discrimination threshold, output the group event window marker and enable degrouping fusion constraints: replace the interaction intensity scalar of each student with the quantile deviation in the class interaction intensity scalar distribution, and set the interaction intensity input of class synchronous migration to a gating item that does not participate in the individual intervention trigger judgment within the time window corresponding to the group event window marker. When triggering the judgment, reset the corresponding interaction intensity input weight to zero and prohibit entry into the comparison path. At the same time, bind the group event window marker with the school timetable data to create and archive it to the group event database. The specific process of mapping fused evidence into executable intervention task orchestration and establishing an end-to-end intervention closed-loop archive is as follows: Based on time-series fusionable markers, time-series unreliable markers, and event time-series misalignment risk judgment values, combined with group contamination suppression discriminant values, group event window markers, and degrouping fusion results, a student psychological state monitoring result and intervention decision arrangement are formed: Standardized text sequences and text length sequences, unit-time touch count sequences, student daily attendance data, student weekly grade data, and consultation appointment start time sequences corresponding to the same bound subject identifier data are jointly organized to generate an intervention decision list. Trigger evidence items, trigger time windows, and receipt deadlines are bound to the intervention decision tasks. Simultaneously, time-series fusionable markers, time-series unreliable markers, group event window markers, and corresponding evidence items are created as a closed-loop evidence package and archived in the intervention closed-loop database for subsequent review and consistency verification.

2. The student psychological state monitoring and intervention decision-making method for campus environment as described in claim 1, characterized in that: The specific process of collecting real-time campus data and obtaining historical auxiliary environmental data is as follows: Collect real-time campus data, including: timestamp sequences of student terminal touch events, original content data of student terminal text messages, time sequences of student terminal text message submissions, time sequences of student terminal data reporting requests sent, time sequences of server-side data reception, time sequences of server-side data entry completion, binding entity identifier data, time sequences of student daily attendance check-in, time sequences of student daily attendance status, time sequences of student weekly grade releases, time sequences of student weekly grade values, time sequences of consultation appointment creation, time sequences of consultation appointment start, time sequences of psychological treatment record creation, time sequences of psychological treatment record submission, psychological expression submission events, consultation appointment creation events, and psychological treatment record submission events. Obtain historical auxiliary environmental data, including: school schedule data, class identification data, historical touch event timestamp sequence, historical text message submission time sequence, historical reporting link original record data, historical attendance original record data, historical grade original record data, and historical consultation appointment and handling original record data.

3. The student psychological state monitoring and intervention decision-making method for campus environment according to claim 1, characterized in that: The specific process for preprocessing real-time campus data and historical environmental auxiliary data is as follows: Using a sliding window counting algorithm, a unit-time touch count sequence is generated from the timestamp sequence of touch events on student terminals, and outlier removal and smoothing are performed on the unit-time touch count sequence. Similarly, a unit-time text submission count sequence is generated from the time sequence of text message submissions on student terminals, and outlier removal and smoothing are performed on the unit-time text submission count sequence. Invalid text is removed from the original content data of student terminal text messages using text segmentation, character repetition rate filtering, and invalid symbol proportion discrimination algorithms, generating standardized text sequences and text length sequences. Reverse segment correction is performed on the time sequences of student terminal data reporting requests, server-side data reception, and server-side database entry completion using timestamp monotonicity verification and offline return reverse order rearrangement algorithms. Finally, zero-mean, unit-variance standardization is performed on real-time campus data and historical environmental auxiliary data using mean-standard-deviation standardization algorithms. The minimum-maximum value normalization algorithm is used to perform interval normalization processing on real-time campus data and historical environmental auxiliary data.

4. The student psychological state monitoring and intervention decision-making method for campus environment according to claim 1, characterized in that: The specific process of mapping and representing the entire timeline from the occurrence of the construction event to the completion of data entry, and outputting the delay sequence and window constraints that support timeline reconstruction, is as follows: Based on the timestamp sequence of student terminal touch events, the timestamp sequence of student terminal text message submissions, the timestamp sequence of student daily attendance check-in, the timestamp sequence of student weekly grade releases, the timestamp sequence of consultation appointment creation, the timestamp sequence of psychological treatment record creation, and the timestamp sequence of psychological treatment record submission, combined with the timestamp sequence of student terminal data reporting request sending, the timestamp sequence of server-side data reception, and the timestamp sequence of server-side data entry completion, a link alignment relationship is established between the event occurrence time, the timestamp sequence of reporting request sending, the timestamp sequence of server reception, and the timestamp sequence of data entry completion. The reporting delay sequence, the server reception delay sequence, and the data entry delay sequence are obtained by calculating the difference between the link alignment relationships, and window markers are generated for the event's time window based on the school's timetable data.

5. A student psychological state monitoring and intervention decision-making system for a campus environment, employing the student psychological state monitoring and intervention decision-making method for a campus environment as described in any one of claims 1-4, characterized in that, include: The data acquisition and preprocessing module is used to acquire real-time campus data, obtain historical environmental auxiliary data, and preprocess the real-time campus data and historical environmental auxiliary data. The granularity difference and reporting delay characterization module is used to construct the end-to-end time-series mapping and delay structure representation from event occurrence to database entry completion, and outputs the delay sequence and window constraints that support time-series reconstruction. The event chain time sequence reconstruction and misplacement risk discrimination module is used to perform event time sequence misplacement cumulative risk discrimination analysis on real-time campus data and historical environmental auxiliary data, and to perform offline back-up reverse order rearrangement and delay scale constraint screening based on the event time sequence misplacement cumulative risk discrimination analysis results; The de-grouping fusion and group pollution suppression module is used to compress and discriminate the group pollution index of real-time campus data and historical environmental auxiliary data, and to perform quantile deviation replacement and interaction intensity gating constraints based on the group pollution index compression and discrimination results. The intervention decision orchestration and closed-loop evidence archiving module is used to map fused evidence into executable intervention task orchestration and establish end-to-end intervention closed-loop archiving.

Images