A big data analysis-based intelligent campus precision management system

Abstract

The present application relates to the technical field of wisdom campus information management, more specifically, to a wisdom campus precision management system based on big data analysis, which is used to solve the problem that the existing wisdom campus system relies on static or periodic update in user portrait construction, is difficult to reflect the behavior change of teachers and students in real time, and often models the entities such as students, teachers, classes, classrooms and equipment in isolation, lacks unified expression of membership, use and interaction relationship between cross-type entities, and leads to limited multi-granularity correlation analysis capability; the present application constructs a cross-entity heterogeneous graph covering students, teachers, classes, classrooms and equipment through a dynamic portrait construction module, dynamically models the membership, use and interaction relationship strength, fuses its own statistics and neighbor weighted features, generates three-level portraits of individuals, groups and scenes, and supports high-time-efficiency, multi-granularity risk early warning, precision management and personalized service.

Description

Technical Field

[0001] This invention relates to the field of smart campus information management technology, and more specifically, to a smart campus precision management system based on big data analysis. Background Technology

[0002] With the rapid development of new-generation information technologies such as the Internet of Things, cloud computing, artificial intelligence, and 5G, educational informatization has entered a new stage of smart campus construction from digital campuses. Universities and primary and secondary schools have generally deployed various information systems such as academic affairs, campus card, security monitoring, energy consumption monitoring, access control and attendance, and network logs, generating massive amounts of heterogeneous campus operation data from multiple sources. Existing campus management platforms generally suffer from data silos, with subsystems lacking effective integration and data not being fully mined and utilized. This results in management decisions still relying on experience-based judgments, making it difficult to achieve refined, intelligent, and forward-looking governance.

[0003] Referring to patent application CN110517171A, a precise and intelligent education platform based on a smart campus is disclosed. The platform includes: a) rapid development of data microservice programs based on a microservice platform architecture; b) an intelligent management system that collects and integrates data from the microservice programs and manages the data information in a standardized manner; c) establishing a student competency assessment model; d) conducting big data analysis of students; e) generating precise assessment reports of student abilities; and f) automatically proposing diagnostic and improvement suggestions based on the assessment reports and pushing the corresponding microservice programs to them. Compared with existing technologies, the smart campus platform of this invention adopts a big data-based microservice business system architecture, forming a smart campus integration solution and implementation capability with the smart campus school-based data center as the core, application portals, mobile portals, decision portals, and diagnostic and improvement portals as standards, and various microservice business applications as the foundation. Through big data analysis and comparison, it achieves precise assessment and diagnosis of students' comprehensive qualities. However, existing smart campus systems rely heavily on static or periodic updates for user profile construction, making it difficult to reflect changes in teacher and student behavior in real time. Furthermore, they often model entities such as students, teachers, classes, classrooms, and equipment in isolation, lacking a unified expression of the affiliation, usage, and interaction relationships between different types of entities. This limits multi-granularity correlation analysis capabilities, and profile maintenance often requires historical snapshots or full reconstruction, resulting in high computational costs and poor timeliness. Meanwhile, behavior assessments generally use fixed dimensions and static weight models, failing to integrate dynamic profiles, structured event flows, and preliminary anomaly labels. This makes it difficult to dynamically adjust strategies based on management stages, individual characteristics, or specific situations, and the scoring results lack interpretability, failing to provide dimensional contributions and weighting basis, thus restricting the transparency and effectiveness of risk warnings and precise interventions.

[0004] To address the aforementioned problems, this invention proposes a smart campus precision management system based on big data analysis. Summary of the Invention

[0005] The purpose of this invention is to provide a smart campus precision management system based on big data analysis. This addresses the shortcomings of existing smart campus systems, which often rely on static or periodic updates for user profile construction, making it difficult to reflect changes in teacher and student behavior in real time. Furthermore, these systems frequently model entities such as students, teachers, classes, classrooms, and equipment in isolation, lacking a unified expression of cross-type entity affiliation, usage, and interaction relationships. This limits multi-granularity correlation analysis capabilities, and profile maintenance often requires historical snapshots or full reconstruction, resulting in high computational costs and poor timeliness. Simultaneously, behavior assessments generally employ fixed dimensions and static weight models, failing to integrate dynamic profiles, structured event flows, and preliminary anomaly labels. This makes it difficult to dynamically adjust strategies based on management stages, individual characteristics, or specific situations, and the scoring results lack interpretability, failing to provide dimensional contribution and weight basis, thus hindering the transparency and effectiveness of risk warning and precise intervention.

[0006] The objective of this invention is achieved through the following technical solution: A smart campus precision management system based on big data analytics, integrated into the campus comprehensive management platform, includes: The campus event perception module is used to collect campus operation data from the daily operation of the campus, standardize and event-based process the collected campus operation data, and finally output a structured event stream. The dynamic profile building module constructs a cross-entity heterogeneous map covering students, teachers, classes, classrooms and equipment based on structured event flow, and generates multi-level dynamic profiles on this basis. When a statistically significant shift in the profile attribute distribution is detected, or a new behavioral pattern not covered by the current feature space is identified, incremental updates of the profile are triggered. The behavioral context analysis module, based on multi-level dynamic profiling, structured event flow, and preliminary anomaly labels, combined with multi-dimensional behavioral indicators, generates context scores for campus management scenarios. Based on campus context, object characteristics, and historical intervention feedback, it dynamically optimizes the scoring strategy and outputs interpretable context score results. The Behavioral Trend and Risk Warning Module, based on multi-level dynamic profiling and structured event flow, predicts individual behavioral trends, potential anomalies, and group situation evolution, and outputs risk warning information.

[0007] In a preferred embodiment of the present invention, the process of constructing a cross-entity heterogeneous map covering students, teachers, classes, classrooms, and devices based on a structured event flow in the dynamic profile construction module includes: If a structured event is obtained, and the structured event flow ends, a cross-entity heterogeneous graph covering students, teachers, classes, classrooms and equipment is generated. Based on the behavior type code, the predefined event parsing rules are queried to determine the main entity type and associated entity types. The identity of the main entity is determined based on the object's unique identifier. Based on the predefined event parsing rules, the unique identifier of the associated entity is derived by combining the object's unique identifier, the event timestamp and the original value. The identity of the associated entity is determined based on the unique identifier of the associated entity. The relationship type between the main entity node and the associated entity node is determined based on the behavior type code. The statistical indicators of the main entity node are updated based on the behavior type code, the original value, and the main entity type. At the same time, the statistical indicators of the associated entity node are updated based on the behavior type code, the original value, and the associated entity type. If the initial anomaly label is normal or abnormal, the association strength value of the connection between the main entity node and the associated entity node of the current relationship type is increased by a preset positive value or a preset negative value, respectively. The next structured event is obtained and the process is repeated until the structured event flow is completed, generating a cross-entity heterogeneous graph covering students, teachers, classes, classrooms and equipment.

[0008] In a preferred embodiment of the present invention, the process of generating a multi-level dynamic image based on a cross-entity heterogeneous atlas in the dynamic image construction module includes: Obtaining the statistical metrics of the target student or teacher node and merging the weighted statistical metrics of the target node and its neighbors is achieved through the following steps: S1: Traverse all adjacent connections of the target node and obtain the association strength of each connection and the statistical indicators of the corresponding neighboring nodes; S2: Multiply the statistical index of each neighbor node by its association strength to obtain a weighted index, and then accumulate the weighted index into the container. S3: After traversal, add the container's accumulated result to the target node's own statistical indicators item by item, truncate each item, and output the truncated result as an individual profile. Obtain the profiles of all members of a class or college and their individual profiles, sum up each indicator and divide by the number of members, and use the set of arithmetic mean of the results as the group profile. Obtain classroom nodes, device nodes, student or teacher nodes that have recently interacted with the area, and their individual profiles within the target area. Traverse all connections within the area, multiply the statistical indicators of the starting node of each connection by the association strength and sum them up to obtain the interaction activity contribution. Add the baseline of the main behavior to the interaction activity contribution item by item to generate a scene profile.

[0009] In a preferred embodiment of the present invention, the process of triggering incremental updates of the portrait in the dynamic portrait construction module when a statistically significant shift in the portrait attribute distribution is detected, or a new behavioral pattern not covered by the current feature space is identified, includes: Get the student or teacher node associated with the triggering event, read the number of check-ins, total consumption, number of abnormal events and space usage frequency of the student or teacher node, update the corresponding indicators according to the event content, and regenerate the individual profile of the student or teacher node. Get the class node or college node to which the updated individual profile belongs, aggregate the latest individual profiles of all student nodes or teacher nodes under the class node or college node, and regenerate the corresponding group profile. Retrieve the specified area node from the event log, retrieve the classroom node, device node, student node or teacher node that recently interacted with the area node, as well as the starting node and association strength value of all connections between the area node, regenerate the scene profile based on the updated node profile and association strength value, and output the individual profile, group profile and scene profile of this incremental update.

[0010] In a preferred embodiment of the present invention, the process of combining multi-level dynamic profiling, structured event flow, and preliminary anomaly tags with multi-dimensional behavioral indicators in the behavioral context analysis module includes: Iterate through each object to be analyzed to obtain its dynamic profile, group and scene, as well as the associated structured event stream and the preliminary anomaly label of each event. Filter matching events within a fixed time window, retain only records with preliminary anomaly labels, those that occur in a preset specific time interval or belong to a predefined key event category, and accumulate their counts to obtain the quantitative count values ​​of each dimension. Obtain the historical baseline distribution of the reference group to which the object belongs in each behavioral dimension, compare its quantified count value with the corresponding distribution, calculate the degree of standardized deviation relative to the normal state of the reference group, and output the standardized deviation value of each of the six dimensions as the quantification result after context adaptation.

[0011] In a preferred embodiment of the present invention, the process of generating a context score for a campus management scenario in the behavioral context analysis module includes: Obtain the context adaptation values ​​of various behavioral indicators in the behavior record, the basic importance coefficients in the configuration library, the individual correction factors in the object file, and the context adjustment factors in the current campus context. For each behavioral indicator, a preset mapping function is called to obtain the mapping output, and the three factors are multiplied to form a dynamic weight. The mapping output is multiplied by the dynamic weight to obtain the weighted contribution of this indicator to the score. The weighted contributions of all indicators are accumulated to form the final situational score, which is then output as the quantitative evaluation result of the object in the current campus management scenario.

[0012] In a preferred embodiment of the present invention, the process by which the behavioral context analysis module dynamically optimizes the scoring strategy based on the campus context, object characteristics, and historical intervention feedback, and outputs interpretable context scoring results includes: Obtain historical intervention feedback. Based on historical intervention feedback under the same campus context, for each combination of behavioral indicators and object characteristics, according to whether the intervention result is effective or ineffective, use the corresponding effective or ineffective importance values ​​respectively. Calculate individual correction factors based on the object's grade, gender, class, and cumulative number of historical abnormal events. Set situational adjustment factors in combination with the current campus context and management stage. Each behavioral indicator is obtained and matched with the current campus context according to preset rules to generate context adaptation values. These values ​​are then converted into standardized output values ​​through a mapping function. The dynamic weight is obtained by multiplying the corresponding importance value, individual correction factor, and context adjustment factor. This dynamic weight is then multiplied by the standardized output value to obtain the weighted contribution value of the indicator. The weighted contribution values ​​of all indicators are summed to obtain the final context score. The context score is then output, along with a detailed list, explanation of the weight calculation basis, and a summary of auxiliary information.

[0013] In a preferred embodiment of the present invention, the process of processing multi-level dynamic profiles and structured event flows in the behavior trend and risk warning module includes: For the target object, obtain its corresponding multi-layered dynamic profile and the latest structured events associated with it: If the target is an individual, take its most recent event record; if the target is a group, obtain the most recent events of each member in the group and aggregate behavioral features, combine profile and event information to construct a joint context.

[0014] In a preferred embodiment of the present invention, the process of predicting individual behavioral trends, potential anomalies, and group situation evolution in the behavioral trend and risk warning module includes: The individual-level profile sequence of the target individual within a continuous historical time window is obtained. The difference between the profiles at adjacent time points in the sequence is calculated to obtain the behavioral change. Time decay weights are assigned to each behavioral change, and the weighted sum is obtained to obtain a comprehensive change trend vector. The comprehensive change trend vector is superimposed on the latest profile of the individual as a prediction of future behavioral state. Obtain the historical behavioral baseline and fluctuation range of an individual within the same time window, subtract the predicted future behavior from the historical behavioral baseline to obtain the deviation amount, obtain the sensitivity threshold, and mark the existence of potential anomalies if the absolute value of the deviation amount is greater than the product of the threshold and the fluctuation range. Obtain a list of members of the target group, take the calculated future behavior state prediction and anomaly label for each member, aggregate the prediction results of all members to obtain the future behavior state of the group, aggregate all anomaly labels to obtain the abnormal risk level of the group, and the two together constitute the prediction result of the group situation evolution.

[0015] Compared with the prior art, the advantages of this invention are: (1) In this invention, a cross-entity heterogeneous graph covering students, teachers, classes, classrooms and equipment is constructed through a dynamic profile construction module. The strength of membership, usage and interaction relationships is dynamically modeled. The self-statistics and neighbor weighted features are integrated to generate three-level profiles of individuals, groups and scenes. Attribute offsets or new behavior patterns are detected in real time, triggering lightweight incremental updates and avoiding full reconstruction. The graph is graph-driven, rules are configurable, and no historical snapshots are required. It has interpretability, scalability and computational efficiency, and supports high-timeliness, multi-granularity risk warning, precise management and personalized services. (2) In this invention, the behavioral context analysis module integrates dynamic profiles, structured event streams and preliminary abnormal labels, and constructs context-adaptive behavioral indicators around six dimensions: attendance, classroom participation, daily routine, resource usage, abnormal frequency and social activity. Combined with basic weights, individual corrections and context adjustment factors, highly targeted scores are dynamically generated, which can be optimized in real time according to management stage, object characteristics and intervention effects, to achieve accurate assessment based on time, person and event, and output an interpretable list of dimension contributions and weights, improve decision-making transparency, and support smart campus scenarios such as risk warning, precise intervention and resource optimization. Attached Figure Description

[0016] Figure 1 This is a system block diagram of Embodiment 1 of the present invention; Figure 2 This is a system block diagram of Embodiment 2 of the present invention; Figure 3 This is a flowchart illustrating the steps of merging the weighted statistical indicators of the target node and its neighbors in this invention. Detailed Implementation

[0017] 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. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0018] Example 1: As Figure 1 and Figure 3 As shown, this invention proposes a smart campus precision management system based on big data analysis, integrated into the campus comprehensive management platform, to achieve intelligent sensing, dynamic analysis, and precise decision support for scenarios such as teaching, security, energy consumption, and personnel flow, including: The campus event perception module is used to collect campus operation data from the daily operation of the campus, including course attendance records, access control records (including entry and exit directions), consumption transaction records, dormitory return records, classroom interaction logs, and building electricity meter readings. The module performs standardization and event processing on the collected campus operation data, including multi-source heterogeneous timestamp alignment, cross-source subject identity alignment, behavioral semantic standardization coding, and preliminary anomaly marking based on preset business rules. Finally, it outputs a structured event stream, with each event containing a unique object identifier, event timestamp, behavior type code, original value, and preliminary anomaly label. By integrating multi-source data such as courses, access control, consumption, dormitories, classrooms, and energy consumption through the campus event perception module, and generating structured event streams through time alignment, identity unification, and semantic standardization, and preliminarily identifying anomalies based on rules, the module achieves comprehensive and real-time perception of the campus's operational status. Its output is of high quality, scalable, and compliant, providing solid data support for security early warning, resource optimization, and intelligent services.

[0019] The dynamic profile building module constructs a cross-entity heterogeneous map covering students, teachers, classes, classrooms and equipment based on structured event flow, and generates multi-level dynamic profiles on this basis. When a statistically significant shift in the profile attribute distribution is detected, or a new behavioral pattern not covered by the current feature space is identified, incremental updates of the profile are triggered. The process of constructing a cross-entity heterogeneous graph covering students, teachers, classes, classrooms, and devices based on structured event streams in the dynamic profile construction module includes: A structured event is obtained, which includes a unique object identifier, an event timestamp, a behavior type code, raw values, and a preliminary anomaly label. If the structured event stream ends, a cross-entity heterogeneous graph covering students, teachers, classes, classrooms, and equipment is generated. Based on the behavior type code, predefined event parsing rules are queried to determine the main entity type and associated entity types. Both belong to one of the five entity types: students, teachers, classes, classrooms, and equipment. The identity of the main entity is determined based on the unique object identifier. If there is no corresponding main entity node in the graph, the main entity node is created and its statistical indicators are initialized. Based on the predefined event parsing rules, the unique identifier of the associated entity is derived by combining the unique object identifier, the event timestamp, and the raw values. If the derivation result is invalid, the current structured event is skipped. The identity of the associated entity is determined based on the unique identifier of the associated entity. If there is no corresponding associated entity node in the graph, an associated entity node is created and its statistical indicators are initialized. The relationship type between the main entity node and the associated entity node is determined based on the behavior type code. The relationship type is one of the following: membership, usage, or interaction. If there is no connection of the current relationship type between the main entity node and the associated entity node, a connection of the current relationship type is created and the association strength value of the connection is set to an initial positive value. The statistical indicators of the main entity node are updated based on the behavior type code, the original value, and the main entity type. At the same time, the statistical indicators of the associated entity node are updated based on the behavior type code, the original value, and the associated entity type. If the initial anomaly label is normal or abnormal, the association strength value of the connection between the main entity node and the associated entity node of the current relationship type will be increased by a preset positive value or a preset negative value respectively. The adjusted association strength value shall not be lower than the preset lower limit and shall not be higher than the preset upper limit. The next structured event will be obtained and processed repeatedly until the structured event flow is processed and a cross-entity heterogeneous graph covering students, teachers, classes, classrooms and equipment will be generated. The process of generating multi-level dynamic profiles based on cross-entity heterogeneous maps in the dynamic profile construction module includes: Obtain the statistical metrics of the target student or teacher node, initialize a weighted accumulator container with the same structure (each item is zero), and merge the weighted statistical metrics of the target node and its neighbors through the following steps: S1: Traverse all adjacent connections of the target node and obtain the association strength of each connection and the statistical indicators of the corresponding neighboring nodes; S2: Multiply the statistical index of each neighbor node by its association strength to obtain a weighted index, and then accumulate the weighted index into the container. S3: After traversal, add the container's accumulated result to the target node's own statistical indicators item by item, and truncate each item: if the result is less than the preset minimum threshold, set it to the preset minimum threshold; if the result is greater than the preset maximum threshold, set it to the preset maximum threshold; otherwise, retain the original value and output the truncated result as an individual profile. Obtain the profiles of all members of a class or college and their individual profiles, sum up each indicator and divide by the number of members, and use the set of arithmetic mean of the results as the group profile. Obtain classroom nodes, device nodes, student nodes or teacher nodes that have recently interacted with the area and their individual profiles within the target area. If there are no relevant nodes, all items of the main behavior baseline are zero. Otherwise, the average value of each item of the individual profile is used to obtain the main behavior baseline. Traverse all the connections in the area, multiply the statistical indicators of the starting node of each connection by the association strength and accumulate them to obtain the interaction activity contribution. Add the main behavior baseline and the interaction activity contribution item by item to generate the scene profile. In the dynamic profile building module, when a statistically significant shift in the profile attribute distribution is detected, or a new behavioral pattern not covered by the current feature space is identified, the incremental profile update process includes: Get the student or teacher node associated with the triggering event, read the number of check-ins, total consumption, number of abnormal events and space usage frequency of the student or teacher node, update the corresponding indicators according to the event content, and regenerate the individual profile of the student or teacher node. Get the class node or college node to which the updated individual profile belongs, aggregate the latest individual profiles of all student nodes or teacher nodes under the class node or college node, and regenerate the corresponding group profile. This process directly uses the latest individual profile data and does not rely on historical snapshots or intermediate caches. Get the specified area node from the event log, get the classroom node, device node, student node or teacher node that recently interacted with the area node, as well as the starting node and association strength value of all connections of the area node, regenerate the scene profile based on the updated node profile and association strength value, and output the individual profile, group profile and scene profile of this incremental update. The advantages of the dynamic profile building module are as follows: Based on structured event streams, it efficiently constructs cross-entity heterogeneous graphs covering students, teachers, classes, classrooms, and equipment. It dynamically models the membership, usage, and interaction relationships between entities through association strength, and integrates its own statistical indicators and neighbor weighted information to generate dynamic profiles at three levels: individual, group, and scene. This module supports real-time perception of behavioral changes. When a significant shift in attribute distribution or the emergence of new behavioral patterns is detected, it can accurately trigger incremental updates to the profile, avoiding full reconstruction and improving efficiency and response speed. At the same time, its graph-driven, rule-configurable, and lightweight update mechanism does not rely on historical snapshots, and it combines interpretability, scalability, and computational economy, providing timely and multi-granular data support for precise management, risk warning, and personalized services.

[0020] The behavioral context analysis module, based on multi-level dynamic profiling, structured event flow, and preliminary anomaly labels, combined with multi-dimensional behavioral indicators, generates context scores for campus management scenarios. Based on campus context, object characteristics, and historical intervention feedback, it dynamically optimizes the scoring strategy and outputs interpretable context score results. The behavioral context analysis module, based on multi-level dynamic profiling, structured event streams, and preliminary anomaly labels, combines multi-dimensional behavioral indicators in the following process: Iterate through each object to be analyzed to obtain a dynamic profile of the individual, the group to which it belongs, and the scene in which it is located, as well as the associated structured event stream and the preliminary abnormal label of each event. Based on six behavioral dimensions, namely attendance regularity, classroom participation, work and rest stability, rational use of resources, frequency of abnormal events, and social activity, filter and match events within a fixed time window, and retain only records with preliminary abnormal labels, those that occur in a preset specific time interval, or those that belong to a predefined key event category. Accumulate their numbers to obtain the quantitative count value of each dimension. Obtain the historical baseline distribution of the reference group to which the object belongs in each behavioral dimension, compare its quantified count value with the corresponding distribution, calculate the degree of standardized deviation relative to the normal state of the reference group, and output the standardized deviation value of each of the six dimensions as the quantification result after context adaptation; The process of generating context scores for campus management scenarios in the behavioral context analysis module includes: Obtain the context adaptation values ​​of various behavioral indicators in the behavior record, the basic importance coefficients in the configuration library, the individual correction factors in the object file, and the context adjustment factors in the current campus context. For each behavioral indicator, a preset mapping function is called to obtain the mapping output. The three factors are then multiplied to form a dynamic weight. The mapping output is multiplied by the dynamic weight to obtain the weighted contribution of this indicator to the score. The weighted contributions of all indicators are summed to form the final context score. ,in It is an object Contextual rating, It is the total number of behavioral indicators. It is the first Context-appropriate values ​​for behavioral indicators It is a pre-defined mapping function used to convert context-fit values ​​into contribution values ​​to the score. It is the first In the current campus context, these indicators Dynamic weights under; Secondly, weight Generated by the following formula: ,in It is the first The basic importance coefficient of each indicator is predetermined by the campus management objectives. It is an object The individual correction factor is calculated based on the cumulative number of historical anomalous events. It is a situational moderating factor, based on the current campus situation. Confirm and output this scenario score as a quantitative evaluation result of the object in the current campus management scenario; The process by which the behavioral context analysis module dynamically optimizes the scoring strategy based on the campus context, the characteristics of the target audience, and historical intervention feedback, and outputs interpretable contextual scoring results includes: Obtain historical intervention feedback. Based on historical intervention feedback under the same campus context, for each combination of behavioral indicators and object characteristics, according to whether the intervention result is effective or ineffective, use the corresponding effective or ineffective importance value respectively. Calculate individual correction factors according to the object's grade, gender, class and the cumulative number of historical abnormal events. Set situational adjustment factors in combination with the current campus context (such as semester stage, exam cycle, holidays) and management stage. Each behavioral indicator is obtained and matched with the current campus context according to preset rules to generate context adaptation values. These values ​​are then converted into standardized output values ​​through a mapping function. The dynamic weight is obtained by multiplying the corresponding importance value, individual correction factor, and context adjustment factor. This dynamic weight is then multiplied by the standardized output value to obtain the weighted contribution value of the indicator. The weighted contribution values ​​of all indicators are then summed to obtain the final context score. Output the situational score, along with a detailed list (including the fit degree value of each behavioral dimension, the mapped output value, the dynamic weights and the contribution value obtained by their product), an explanation of the basis for the weight calculation (including the type of intervention result corresponding to the importance value, the method of determining the individual correction factor based on the number of historical abnormal events, and the management stage on which the situational adjustment factor is based), and a summary of auxiliary information (including the subject's grade, gender, class, number of historical abnormal events, and the time, type, and result of the most recent interventions). The advantages of the behavioral context analysis module are as follows: it deeply integrates multi-level dynamic profiles, structured event flows, and preliminary anomaly tags. Around six dimensions—attendance, classroom participation, daily routines, resource usage, anomaly frequency, and social activity—it constructs standardized behavioral indicators adapted to specific contexts. Furthermore, by integrating basic importance coefficients, individual correction factors, and context adjustment factors, it dynamically generates highly targeted context scores. This module not only supports real-time adjustments to scoring strategies based on campus management stages (e.g., exam weeks, holidays), object characteristics (e.g., grade level, historical anomaly records), and historical intervention effects, achieving precise assessments "time-based, person-based, and event-based," but also significantly improves the transparency and credibility of scoring results by outputting an interpretable list containing contribution values ​​for each dimension, weight calculation logic, and intervention history. This effectively supports smart campus management scenarios such as risk warning, precise intervention, and resource optimization.

[0021] Example 2: The technical solution of this embodiment of the invention differs from that of Example 1 in that: like Figure 2 As shown, the behavior trend and risk warning module, based on multi-level dynamic profiles and structured event streams, predicts individual behavior trends, potential anomalies, and group situation evolution, and outputs risk warning information. The process of processing multi-level dynamic profiles and structured event flows in the Behavioral Trends and Risk Warning module includes: For the target object, obtain its corresponding multi-layered dynamic profile, including applicable parts of the individual layer, group layer, or scene layer, and obtain the latest structured events associated with it: If the target is an individual, take its most recent event record; if the target is a group, obtain the most recent events of each member in the group and aggregate behavioral features, fuse profile and event information, and construct a joint context. The process of predicting individual behavioral trends, potential anomalies, and group situation evolution in the Behavioral Trends and Risk Warning module includes: The individual-level profile sequence of the target individual within a continuous historical time window is obtained. The difference between the profiles at adjacent time points in the sequence is calculated to obtain the behavioral change. Time decay weights are assigned to each behavioral change, and the weighted sum is obtained to obtain a comprehensive change trend vector. The comprehensive change trend vector is superimposed on the latest profile of the individual as a prediction of future behavioral state. Obtain the historical behavioral baseline and fluctuation range of an individual within the same time window, subtract the predicted future behavior from the historical behavioral baseline to obtain the deviation amount, obtain the sensitivity threshold, and mark the existence of potential anomalies if the absolute value of the deviation amount is greater than the product of the threshold and the fluctuation range. Obtain a list of members of the target group, take the calculated future behavior state prediction and anomaly label for each member, aggregate the prediction results of all members to obtain the future behavior state of the group, aggregate all anomaly labels to obtain the abnormal risk level of the group, and the two together constitute the prediction result of the group situation evolution. The process of outputting risk warning information in the Behavioral Trends and Risk Warning module includes: Obtain the risk score and importance weight of the target object, multiply them to obtain the weighted risk value, obtain the threshold boundaries and corresponding level codes in ascending order, match the level codes according to the interval of the weighted risk value, and obtain the description of abnormal behavior indicators (including the behavioral dimensions and direction of change of the predicted deviation), the name of the associated object and its unique identifier, the start and end timestamps of the risk time window, the profile level identifier and the event type identifier. The risk warning information is composed of the level code, abnormal behavior indicators, associated object name, associated object unique identifier, risk time window start timestamp, risk time window end timestamp, profile level identifier and event type identifier, and is output through the standard interface. By integrating multi-level dynamic profiles and the latest events through behavioral trend and risk warning modules, and modeling individual behavioral evolution trends using time decay weighting, the system accurately predicts future states and identifies significant anomalies. It simultaneously aggregates individual prediction results to achieve collaborative extrapolation of group status and risk levels. Combining configurable scoring and tiered thresholds, it outputs structured warnings that include risk level, anomaly dimensions, object information, and time windows. This system is forward-looking, interpretable, and operable, providing precise, layered, and real-time risk decision support for campus safety.

[0022] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and its improved concept, should be covered within the scope of protection of the present invention.

Claims

1. A smart campus precision management system based on big data analysis, integrated into a campus comprehensive management platform, characterized in that: include: The campus event perception module is used to collect campus operation data from the daily operation of the campus, standardize and event-based process the collected campus operation data, and finally output a structured event stream. The dynamic profile building module constructs a cross-entity heterogeneous map covering students, teachers, classes, classrooms and equipment based on structured event flow, and generates multi-level dynamic profiles on this basis. When a statistically significant shift in the profile attribute distribution is detected, or a new behavioral pattern not covered by the current feature space is identified, incremental updates of the profile are triggered. The behavioral context analysis module, based on multi-level dynamic profiling, structured event flow, and preliminary anomaly labels, combined with multi-dimensional behavioral indicators, generates context scores for campus management scenarios. Based on campus context, object characteristics, and historical intervention feedback, it dynamically optimizes the scoring strategy and outputs interpretable context score results. The Behavioral Trend and Risk Warning Module, based on multi-level dynamic profiling and structured event flow, predicts individual behavioral trends, potential anomalies, and group situation evolution, and outputs risk warning information.

2. The smart campus precision management system based on big data analysis according to claim 1, characterized in that, The process of constructing a cross-entity heterogeneous graph covering students, teachers, classes, classrooms, and devices based on structured event streams in the dynamic profile construction module includes: If a structured event is obtained, and the structured event flow ends, a cross-entity heterogeneous graph covering students, teachers, classes, classrooms and equipment is generated. Based on the behavior type code, the predefined event parsing rules are queried to determine the main entity type and associated entity types. The identity of the main entity is determined based on the object's unique identifier. Based on the predefined event parsing rules, the unique identifier of the associated entity is derived by combining the object's unique identifier, the event timestamp and the original value. The identity of the associated entity is determined based on the unique identifier of the associated entity. The relationship type between the main entity node and the associated entity node is determined based on the behavior type code. The statistical indicators of the main entity node are updated based on the behavior type code, the original value, and the main entity type. At the same time, the statistical indicators of the associated entity node are updated based on the behavior type code, the original value, and the associated entity type. If the initial anomaly label is normal or abnormal, the association strength value of the connection between the main entity node and the associated entity node of the current relationship type is increased by a preset positive value or a preset negative value, respectively. The next structured event is obtained and the process is repeated until the structured event flow is completed, generating a cross-entity heterogeneous graph covering students, teachers, classes, classrooms and equipment.

3. The smart campus precision management system based on big data analysis according to claim 2, characterized in that, The process of generating multi-level dynamic profiles based on cross-entity heterogeneous maps in the dynamic profile construction module includes: Obtaining the statistical metrics of the target student or teacher node and merging the weighted statistical metrics of the target node and its neighbors is achieved through the following steps: S1: Traverse all adjacent connections of the target node and obtain the association strength of each connection and the statistical indicators of the corresponding neighboring nodes; S2: Multiply the statistical index of each neighbor node by its association strength to obtain a weighted index, and then accumulate the weighted index into the container. S3: After traversal, add the container's accumulated result to the target node's own statistical indicators item by item, truncate each item, and output the truncated result as an individual profile. Obtain the profiles of all members of a class or college and their individual profiles, sum up each indicator and divide by the number of members, and use the set of arithmetic mean of the results as the group profile. Obtain classroom nodes, device nodes, student or teacher nodes that have recently interacted with the area, and their individual profiles within the target area. Traverse all connections within the area, multiply the statistical indicators of the starting node of each connection by the association strength and sum them up to obtain the interaction activity contribution. Add the baseline of the main behavior to the interaction activity contribution item by item to generate a scene profile.

4. The smart campus precision management system based on big data analysis according to claim 3, characterized in that, In the dynamic profile building module, when a statistically significant shift in the profile attribute distribution is detected, or a new behavioral pattern not covered by the current feature space is identified, the incremental profile update process includes: Get the student or teacher node associated with the triggering event, read the number of check-ins, total consumption, number of abnormal events and space usage frequency of the student or teacher node, update the corresponding indicators according to the event content, and regenerate the individual profile of the student or teacher node. Get the class node or college node to which the updated individual profile belongs, aggregate the latest individual profiles of all student nodes or teacher nodes under the class node or college node, and regenerate the corresponding group profile. Retrieve the specified area node from the event log, retrieve the classroom node, device node, student node or teacher node that recently interacted with the area node, as well as the starting node and association strength value of all connections between the area node, regenerate the scene profile based on the updated node profile and association strength value, and output the individual profile, group profile and scene profile of this incremental update.

5. The smart campus precision management system based on big data analysis according to claim 1, characterized in that, The behavioral context analysis module, based on multi-level dynamic profiling, structured event streams, and preliminary anomaly tags, combines multi-dimensional behavioral indicators in the following process: Iterate through each object to be analyzed to obtain its dynamic profile, group and scene, as well as the associated structured event stream and the preliminary anomaly label of each event. Filter matching events within a fixed time window, retain only records with preliminary anomaly labels, those that occur in a preset specific time interval or belong to a predefined key event category, and accumulate their counts to obtain the quantitative count values ​​of each dimension. Obtain the historical baseline distribution of the reference group to which the object belongs in each behavioral dimension, compare its quantified count value with the corresponding distribution, calculate the degree of standardized deviation relative to the normal state of the reference group, and output the standardized deviation value of each of the six dimensions as the quantification result after context adaptation.

6. The smart campus precision management system based on big data analysis according to claim 5, characterized in that, The process of generating context scores for campus management scenarios in the behavioral context analysis module includes: Obtain the context adaptation values ​​of various behavioral indicators in the behavior record, the basic importance coefficients in the configuration library, the individual correction factors in the object file, and the context adjustment factors in the current campus context. For each behavioral indicator, a preset mapping function is called to obtain the mapping output, and the three factors are multiplied to form a dynamic weight. The mapping output is multiplied by the dynamic weight to obtain the weighted contribution of this indicator to the score. The weighted contributions of all indicators are accumulated to form the final situational score, which is then output as the quantitative evaluation result of the object in the current campus management scenario.

7. The smart campus precision management system based on big data analysis according to claim 6, characterized in that, The process by which the behavioral context analysis module dynamically optimizes the scoring strategy based on the campus context, object characteristics, and historical intervention feedback, and outputs interpretable context scoring results includes: Obtain historical intervention feedback. Based on historical intervention feedback under the same campus context, for each combination of behavioral indicators and object characteristics, according to whether the intervention result is effective or ineffective, use the corresponding effective or ineffective importance values ​​respectively. Calculate individual correction factors based on the object's grade, gender, class, and cumulative number of historical abnormal events. Set situational adjustment factors in combination with the current campus context and management stage. Each behavioral indicator is obtained and matched with the current campus context according to preset rules to generate context adaptation values. These values ​​are then converted into standardized output values ​​through a mapping function. The dynamic weight is obtained by multiplying the corresponding importance value, individual correction factor, and context adjustment factor. This dynamic weight is then multiplied by the standardized output value to obtain the weighted contribution value of the indicator. The weighted contribution values ​​of all indicators are summed to obtain the final context score. The context score is then output, along with a detailed list, explanation of the weight calculation basis, and a summary of auxiliary information.

8. The smart campus precision management system based on big data analysis according to claim 1, characterized in that, The process of processing multi-level dynamic profiles and structured event flows in the behavior trend and risk warning module includes: For the target object, obtain its corresponding multi-layered dynamic profile and the latest structured events associated with it: If the target is an individual, take its most recent event record; if the target is a group, obtain the most recent events of each member in the group and aggregate behavioral features, combine profile and event information to construct a joint context.

9. A smart campus precision management system based on big data analysis according to claim 8, characterized in that, The process of predicting individual behavioral trends, potential anomalies, and group situation evolution in the behavioral trend and risk warning module includes: The individual-level profile sequence of the target individual within a continuous historical time window is obtained. The difference between the profiles at adjacent time points in the sequence is calculated to obtain the behavioral change. Time decay weights are assigned to each behavioral change, and the weighted sum is obtained to obtain a comprehensive change trend vector. The comprehensive change trend vector is superimposed on the latest profile of the individual as a prediction of future behavioral state. Obtain the historical behavioral baseline and fluctuation range of an individual within the same time window, subtract the predicted future behavior from the historical behavioral baseline to obtain the deviation amount, obtain the sensitivity threshold, and mark the existence of potential anomalies if the absolute value of the deviation amount is greater than the product of the threshold and the fluctuation range. Obtain a list of members of the target group, take the calculated future behavior state prediction and anomaly label for each member, aggregate the prediction results of all members to obtain the future behavior state of the group, aggregate all anomaly labels to obtain the abnormal risk level of the group, and the two together constitute the prediction result of the group situation evolution.

10. A smart campus precision management system based on big data analysis according to claim 9, characterized in that, The process of outputting risk warning information in the behavioral trend and risk warning module includes: Obtain the risk score and importance weight of the target object, multiply them to obtain the weighted risk value, obtain the threshold boundaries and corresponding level codes in ascending order, match the level codes according to the interval of the weighted risk value, and obtain the description of abnormal behavior indicators, the name of the associated object and its unique identifier, the start and end timestamps of the risk time window, the profile level identifier and the event type identifier. The risk warning information is structured by combining the level code, abnormal behavior indicators, associated object name, associated object unique identifier, risk time window start timestamp, risk time window end timestamp, profile level identifier, and event type identifier, and output through a standard interface.

Images