Smart campus management system

Abstract

The invention discloses an intelligent campus management system, and particularly relates to the field of campus management, and the system comprises a behavior data acquisition module, a behavior scene recognition module, a behavior intention reasoning module, a campus digital twinborn simulation module, a self-adaptive behavior intervention module, and a self-evolution learning module. The behavior data acquisition module is used for collecting original behavior data, sequentially executing anomaly detection, data cleaning, data standardization and feature extraction, and outputting a complete behavior data set. Management of multi-source behavior data dynamic acquisition, behavior scene recognition, behavior intention reasoning, campus digital twinborn simulation, self-adaptive behavior intervention and self-evolution learning optimization is executed through the system, so that the problems that in the prior art, a single data source is relied on, a static rule is adopted, dynamic response ability and self-adaptive optimization ability are lacked, and the reliability of campus digital twinborn simulation is poor are solved. The campus management problem that student behavior diversity and campus environment dynamic change cannot be effectively handled is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of campus management, and more specifically, to a smart campus management system. Background Art

[0002] In the prior art, the publication number is CN119005809B: A smart campus personnel management system and method, which discloses a smart campus personnel management system that obtains students' location information and reservation behavior information through a campus identity card combined with positioning technology, and then calculates students' activity evaluation index and behavior evaluation index, and finally forms a reservation level management system. This solution attempts to achieve quantitative management of students' learning behaviors through a mathematical weight model to help schools with resource allocation and student behavior intervention; however, the design logic of this system is based on a static rule and fixed parameter model, that is, all students' behaviors are preset to fluctuate within certain rules, relying on limited indicators such as static stay duration, displacement change, and behavior times to reflect students' status; this static scenario assumption ignores the high-frequency volatility, diversity, and sudden events of students' behaviors in the actual campus environment (such as temporary course adjustments, club activities, health abnormalities, etc.), resulting in the system being unable to flexibly adjust management strategies in changing scenarios.

[0003] In the prior art, the publication number is CN116452379B: A smart campus management system based on big data, which discloses a video monitoring device combined with a wireless communication module and a central processing module. By performing identity recognition, face recognition, personnel trajectory tracking, and big data analysis on the video streams of each monitoring point, it realizes basic management such as student attendance, security analysis, and pedestrian flow prediction; its purpose is to replace the traditional sensor or identity card solution with monitoring data to reduce equipment costs and improve the accuracy and adaptability of data analysis; The fundamental design idea of the CN116452379B solution is still based on fixed rules and predefined analysis processes, mainly with a linear processing chain of "video image input - image enhancement - face recognition - data analysis - output management results", belonging to the category of typical static scenario assumptions, without considering the real-time impact of factors such as students' flow patterns, activity types, and sudden events on the management logic in the dynamic campus environment. It is difficult for the system to dynamically adapt to complex and changing management requirements during actual operation; In the prior art, the publication number is CN112365237B: A smart campus management system based on the Internet of Things, which discloses that the system consists of three levels: a data acquisition layer, a linkage management layer, and a statistics layer. By combining punch card data, dormitory access data, and electricity consumption data, it analyzes students' return to the dormitory situation, monitors the electricity safety in the dormitory, and gives early warnings for non-return to the dormitory or illegal electricity consumption behaviors; its innovation lies in the introduction of a multi-device linkage and electricity control module, attempting to improve the management's supervision ability of students' daily behaviors through Internet of Things technology; However, the core of the CN112365237B solution remains a typical static rule system. All judgment logics use fixed conditions such as "whether the check-in is successful", "whether the power consumption exceeds the threshold", and "whether to return to the dormitory at the specified time" as trigger points, completely based on single-point threshold comparison, ignoring the complex dynamic variability of students' behaviors and failing to adapt to the highly uncertain and behaviorally diverse scenarios in the actual operation of the campus. Therefore, this solution has serious deficiencies in terms of dynamic scenario adaptation ability.

[0004] In the prior art, the publication number CN108921258A: An intelligent management system for campus personnel and its management method discloses an intelligent management system for campus personnel and its method based on a wireless local area network, mobile terminal intelligent school badges, and wireless transmitters. Through the 2.4GHz radio frequency communication between the intelligent school badges worn by students and the wireless transmitters distributed at multiple points on campus, real-time location monitoring and recording of students' activity trajectories are achieved. The background analyzes the activity trajectories through a policy analysis server and can generate reports for teachers and parents to view. The goal of this system is to improve the efficiency of campus personnel management, reduce the management burden on teachers, and ensure the safety of students on campus. However, this technology still adopts the system logic of static rules. Its core management process uses "whether the positioning is successful", "whether to check in", and "whether to enter the preset area" as the main trigger conditions, completely relying on the pre-set signal strength threshold and matrix block position matching, which belongs to a typical fixed parameter combined with a linear decision system and cannot effectively adapt to the dynamic and complex behavior patterns and temporary changes (such as curriculum adjustments, club activities, temporary leaving school, etc.) in the campus environment. Based on the analysis of the four prior arts, the existing intelligent campus management systems generally rely on static rules and single data sources, lacking the dynamic fusion of multi-source behavior data and real-time intention reasoning, resulting in the inability to achieve flexible response and forward-looking management when facing the diversity and suddenness of students' behaviors and the dynamic changes in the campus environment. Therefore, this solution proposes an intelligent campus management system that integrates multi-source behavior data, has the ability of dynamic intention reasoning and digital twin simulation, to solve the problems of lagging response, insufficient interpretability, and lack of predictive management in the existing systems. Summary of the Invention

[0005] In order to overcome the above defects of the prior art, an embodiment of the present invention provides an intelligent campus management system, which manages through the system's dynamic acquisition of multi-source behavior data, behavior scenario recognition, behavior intention reasoning, campus digital twin simulation, adaptive behavior intervention, and self-evolution learning optimization, to solve the campus management problems in the prior art that rely on a single data source, adopt static rules, lack dynamic response ability and adaptive optimization ability, resulting in the inability to effectively cope with the diversity of students' behaviors and the dynamic changes in the campus environment.

[0006] To achieve the above-mentioned object, the present invention provides the following technical solutions: a smart campus management system, comprising a behavior data acquisition module, a behavior scenario recognition module, a behavior intention reasoning module, a campus digital twin simulation module, an adaptive behavior intervention module, and a self-evolution learning module; The behavior data acquisition module is used to collect raw behavior data, perform anomaly detection, data cleaning, data standardization and feature extraction in sequence, and output a complete behavior data set; The behavior scene recognition module performs the behavior scene recognition, historical behavior matching and behavior trend prediction processes in sequence based on the complete behavior data set, and outputs the current behavior scene status and behavior trend prediction sequence; The behavior intention reasoning module identifies behavior intention, evaluates abnormal behavior, calculates the confidence of behavior intention, and outputs standard behavior intention status data based on the behavior intention relationship matrix established based on the current behavior scenario status, behavior trend prediction sequence, and historical behavior data; The campus digital twin simulation module is based on standard behavior intention state data, combined with current behavior state data, personnel location data and facility usage data, to construct the virtual campus space state, perform space modeling and intervention strategy simulation, and output the target intervention strategy and virtual campus space state; The adaptive behavior intervention module is used to generate and issue intervention measures, evaluate the feedback effect, and determine whether to return to the campus digital twin simulation module to reorganize the strategy or output it to the self-evolution learning module; The self-evolution learning module is used to optimize the behavior intention relationship matrix and the behavior scenario category set, and update the behavior scenario recognition module and the behavior intention reasoning module to achieve continuous self-evolution.

[0007] In a preferred embodiment, the behavior data acquisition module collects raw behavior data by configuring a preset sensing device, the sensing device is composed of a position signal acquisition device, an action behavior detection device and an identity authentication reading device, to form a raw behavior data set; The original behavior data set is transmitted to the data anomaly detection process to determine whether there are missing values ​​and whether there are abnormal records of the sensor device status. If either of the two is true, the data acquisition process of the sensor device in the behavior data acquisition module is returned to re-execute data collection; Input the complete original behavior data into the data cleaning process, and remove the erroneous data according to the combination rules of the behavior data type and the device status code in the data cleaning process to form a cleaned data set; The cleaned data set is passed into the data standardization process, the field format is unified and the timestamp is aligned through the data type mapping table, and the standardized behavior data set is output; Input the standardized behavior data set into the feature extraction process, and extract the standard behavior feature set according to the behavior feature template library; the behavior feature template library includes behavior sequence patterns, behavior continuity windows, and behavior activity categories; Judge whether there are discontinuous time windows and whether there are duplicate behavior records in the standard behavior feature set. If either of them holds, return to the data acquisition process of the sensing device in the behavior data acquisition module to re-execute data collection. If neither holds, output the complete behavior data set as the input data for the behavior scene recognition module.

[0008] In a preferred embodiment, the behavior scene recognition module is used to input the complete behavior data set into the behavior scene recognition process, and use the classification algorithm based on the behavior state probability matrix to combine the current behavior activity category and the behavior activity duration to recognize the current behavior scene; Judge whether the current behavior activity category is consistent with the preset behavior scene category set and whether the behavior activity duration is within the continuity window range. If either of them is not satisfied, return to the feature extraction process of the behavior data acquisition module to recalibrate the behavior data; if satisfied, enter the historical behavior matching process; Input the current behavior scene into the historical behavior matching process, and output the behavior matching degree score through the combined comparison of the Euclidean distance scoring algorithm and the historical behavior frequency scoring algorithm between the current behavior scene and the historical behavior path sequence; Judge whether the behavior matching degree score is lower than the set score lower limit or whether the current behavior activity category deviates from the historical behavior category sequence. If either of them holds, return to the behavior scene recognition process of the behavior scene recognition module for reclassification. If neither holds, enter the dynamic behavior trend prediction process; Input the current behavior scene combined with the historical behavior trend into the dynamic behavior trend prediction process, and predict the future behavior state path through the behavior state transition probability model to form a behavior trend prediction sequence; Determine whether the path consistency between the current behavior scenario state and the behavior trend prediction sequence is less than the preset lower limit of the consistency score or whether the behavior state change rate is higher than the dynamic threshold of the behavior state change rate. If either of them holds, return to the behavior scenario recognition process of the behavior scenario recognition module for re-analysis. If neither holds, output the current behavior scenario state and the behavior trend prediction sequence as the input data for the behavior intention inference module; the dynamic threshold of the behavior state change rate is calculated by real-time analyzing the number of student behavior state changes within the current time window and the average frequency and standard deviation of behavior state changes within the same historical time period. Specifically: Extract the behavior state change frequency for the same time period from the historical behavior data, calculate the mean of this frequency as the reference value, and combine the standard deviation to set the adjustment coefficient to form the dynamic threshold in the current context; Subsequently, count the number of consecutive state changes in the current behavior state sequence. If the current number of changes exceeds this dynamic threshold, it is considered that the behavior fluctuates abnormally, and it is triggered to return to the behavior scenario recognition process of the behavior scenario recognition module for re-analysis; It should be noted that in the behavior scenario recognition process, the classification algorithm based on the behavior state probability matrix takes the behavior activity categories (including classroom attendance, club activities, restaurant activities, dormitory stay, teaching building detention, abnormal departure, etc.) and the behavior activity duration (referring to the continuous time length of the same behavior activity) extracted from the complete behavior data set as input variables, calculates the state occurrence probability and state transition probability of each behavior state in the historical behavior pattern library, and establishes the joint probability scoring matrix of the current behavior data and the historical behavior state sequence. Subsequently, combined with whether the current behavior activity category is an allowed category (such as whether it is during the class schedule time or after-school time) and whether the behavior activity duration exceeds the upper limit of the historical behavior duration, the classification probability of the current behavior is comprehensively generated; If it is detected that the behavior activity category is "abnormal entry and exit area activity" (such as leaving the dormitory area at unauthorized times or crossing the fence boundary signal area, which is determined to be the climbing-over-wall behavior scenario) and the behavior activity duration exceeds the abnormal detection time threshold, it is directly marked as the "climbing-over-wall behavior scenario", etc.; The set of behavior scenario categories includes but is not limited to classroom learning behavior scenarios, teaching building activity behavior scenarios, restaurant dining behavior scenarios, dormitory stay behavior scenarios, campus path activity behavior scenarios, club activity behavior scenarios, and abnormal behavior scenarios. Among them, the abnormal behavior scenarios include climbing-over-wall behavior, staying in unauthorized areas for a long time, unexpected high-frequency position changes, etc.; In addition, in the historical behavior matching process, the current behavior scenario is first converted into a behavior path vector, which is jointly composed of a continuous behavior activity category coding sequence and the standardized value of each behavior duration. Subsequently, the Euclidean distance is calculated with all the stored historical behavior path vectors in the historical behavior database to obtain the difference degree score between the current behavior and the historical behavior in the space-time dimension. The proportion of the occurrence frequency of the current behavior activity category in the historical data is statistically analyzed, and the regularity of the current behavior is quantified through a frequency scoring function. Finally, the Euclidean distance score and the behavior frequency score are weighted and fused according to the set weights, and the behavior matching degree score is output to determine whether the current behavior conforms to the historical normal behavior pattern. In addition, in the behavior state transition probability model of the dynamic behavior trend prediction process, the current behavior scenario and the historical behavior trend data are jointly converted into a state sequence vector. Using the pre-established behavior state transition probability matrix, which calculates the transition probability of each state to the next state by statistically analyzing the transition frequency of each behavior state to the subsequent state in the historical behavior sequence. With the current state as the input, by looking up the transition probability distribution corresponding to the current state in the transition probability matrix, the optimal behavior state path sequence within a future fixed time window is iteratively derived according to the highest probability or the multi-state joint probability path, forming a behavior trend prediction sequence to provide the basic data for subsequent anomaly judgment and intervention decision-making.

[0009] In a preferred embodiment, the behavior intention reasoning module is used to synchronously input the current behavior scenario state and the behavior trend prediction sequence into the behavior intention mapping process, calculate the similarity between the behavior state vector and the behavior intention vector based on the behavior intention relationship matrix, and generate a preliminary behavior intention. It should be noted that for the behavior intention relationship matrix, the system extracts the mapping relationship between the behavior state features and the behavior intention labels through historical behavior data samples, establishes a statistical weight matrix between the behavior state vector and the behavior intention vector, and forms the behavior intention relationship matrix for the behavior intention reasoning module to call during real-time reasoning. Judge whether the similarity between the behavior state vector and the behavior intention vector is lower than the preset similarity score lower limit or whether the number of behavior state changes within the behavior state continuity window is higher than the continuous state change upper limit. If either of them holds, return to the behavior trend prediction process of the behavior scenario recognition module to re-execute the prediction. If neither holds, enter the behavior anomaly recognition process. In the behavior anomaly recognition process, combined with the preset behavior intention categories and the behavior anomaly parameter set of the system, the anomaly risk score is calculated through the behavior anomaly multi-parameter joint scoring algorithm. The behavior intention categories include four categories: learning intention, escape management intention, social participation intention, and health anomaly intention. The behavior anomaly parameter set includes three categories: behavior frequency, behavior duration, and behavior location change rate. Determine whether the abnormal risk score is higher than the upper limit of the risk score and whether the duration of the behavior exceeds the upper limit of the behavior category continuity time window. If both conditions are met, synchronously output the behavior anomaly warning data; otherwise, enter the behavior intention confidence score process; Input the behavior intention and abnormal warning data into the behavior intention confidence score process, and output the behavior intention confidence based on the weighted average of the behavior data integrity score and the behavior intention historical consistency score; Combine the behavior intention, abnormal warning data, and behavior intention confidence into standard behavior intention status data and output it to the campus digital twin simulation module; It should be noted that in the behavior intention mapping process, a behavior state vector is extracted according to the current behavior scenario state and the behavior trend prediction sequence. The behavior state vector is composed of the behavior activity category code, the standardized value of the behavior duration, and the behavior activity frequency; at the same time, the corresponding behavior intention vector is extracted from the historical behavior intention model library. In practical applications, the behavior intention vector contains the standard behavior feature templates corresponding to typical intention categories (such as learning intention, avoiding management intention, social intention, etc.); then, the cosine value of the angle between the behavior state vector and each behavior intention vector is calculated through the cosine similarity algorithm to obtain the similarity score. The closer the similarity score value is to 1, the higher the matching degree between the current behavior state and the intention template. Finally, the behavior intention with the highest score is selected as the preliminary behavior intention output; In the behavior anomaly recognition process, according to the current behavior intention category (learning intention, avoiding management intention, social participation intention, health anomaly intention), determine the corresponding behavior anomaly parameter weight template in the behavior anomaly parameter set. The behavior anomaly parameter weight template is the normal range and its weight coefficient of three parameters: the predefined behavior frequency, behavior duration, and behavior position change rate for each intention category; then, calculate the deviation between the actual behavior frequency, actual behavior duration, and actual behavior position change rate in the current behavior data and the normal parameter range in the corresponding intention category template respectively to obtain three standardized deviation scores; finally, sum the three deviation scores after multiplying them by their respective weight coefficients to output a single abnormal risk score, which is used to quantify the abnormality degree of the current behavior. The higher the value, the greater the degree of deviation of the behavior from the normal behavior pattern; In the above-mentioned behavior intention confidence scoring process, a completeness analysis is performed on the current behavior data, the data coverage rate of the perception device and the continuity of the behavior data are calculated, and a behavior data completeness score is formed. At the same time, the historical behavior intention sequence of the same individual in a similar situation is retrieved from the historical behavior database, and the historical consistency score of the behavior intention is calculated using the matching frequency between the current preliminary behavior intention and the historical behavior intention sequence. Finally, the behavior data completeness score and the historical consistency score of the behavior intention are weighted and summed according to the preset weight parameters of the system, and a comprehensive behavior intention confidence is output, which is used to evaluate the reliability of the current behavior intention inference result.

[0010] In a preferred embodiment, the campus digital twin simulation module is used to input the standard behavior intention state data into the virtual campus state construction process, and through the state synchronization algorithm, combined with the current behavior state data, personnel location data and facility usage data, the virtual campus space state is generated in real time. Judge whether the virtual campus space state completely covers all currently configured perception device areas and whether the state synchronization delay is lower than the preset synchronization delay upper limit. If either of them is not satisfied, return to the feature extraction process of the behavior data acquisition module to re-execute data acquisition. If both are satisfied, enter the virtual space modeling process. In the virtual space modeling process, the virtual campus space state is input into the space distribution modeling process, and a space heat map is generated through the crowd flow distribution density and the device occupancy probability. The space heat map is input into the intervention strategy simulation process, and strategy combination simulations are carried out according to the preset strategy library in the system to form an intervention simulation result set; the strategy library includes three types: path guidance strategy, device shutdown strategy, and personnel reminder strategy. Judge whether the difference in the behavior anomaly warning probability between any two groups of results in the intervention simulation result set is less than the behavior anomaly difference tolerance upper limit, or whether the crowd distribution density after intervention exceeds the space distribution density upper limit. If either of them holds, return to the intervention strategy simulation process of the campus digital twin simulation module to recombine the strategies. If neither holds, enter the strategy screening process. In the strategy screening process, the target intervention strategy is determined through the weighted average of the behavior intervention expected completion score and the expected crowd distribution optimization score; the target intervention strategy and the virtual campus space state are output as the input of the adaptive behavior intervention module. It should be noted that in the strategy screening process, the optimized score of the expected population distribution is a comprehensive score calculated by analyzing the pedestrian flow density distribution results after simulating each group of candidate intervention strategies, based on the spatial uniformity of the population density in each area and the degree of pressure release in the hot spot areas; further, according to the simulation results, the pedestrian flow density in each main area is statistically analyzed, and the regional density variance is calculated as the spatial uniformity index. At the same time, the pedestrian flow reduction rate in the high-density hot spot areas is evaluated as the pressure release index, and then the two indexes are weighted and summed according to the preset weights to obtain the optimized score of the expected population distribution corresponding to the strategy; subsequently, the optimized scores of the population distribution of each group of intervention strategies and the corresponding expected completion scores of the behavior interventions are weighted and averaged together, and finally the intervention strategy with the highest weighted score is selected as the target intervention strategy.

[0011] In a preferred implementation manner, the adaptive behavior intervention module is used to input the target intervention strategy into the behavior intervention execution process, and generate corresponding intervention measure data through the mapping table of the intervention strategy. The intervention measure data includes behavior reminder data, device control data, and path guidance data; The intervention measure data is synchronously sent to the execution objects, which include faculty terminals, student terminals, and infrastructure control terminals; In the intervention result feedback process, it is judged whether the behavior state recovery rate in the intervention execution feedback is lower than the lower limit of the intervention target recovery rate or whether the number of abnormal feedbacks of the perception device execution state exceeds the upper limit of the device abnormal feedback number. If either of them holds, the intervention strategy simulation process of the campus digital twin simulation module is returned to recombine the strategy. If neither holds, the intervention effect evaluation process is entered; the behavior state recovery rate refers to the ratio of the number of preset normal behavior states after the intervention to the number of abnormal behavior states before the intervention; In the intervention effect evaluation process, a weighted comprehensive score of two parameters, namely the difference degree of the behavior state before and after the intervention and the abnormal change rate of the device state, is calculated to form the intervention effect score; the difference degree of the behavior state before and after the intervention is calculated by dividing the absolute value of the difference between the total number of behavior states before the intervention and the total number of behavior states after the intervention by the total number of behavior states before the intervention; the abnormal change rate of the device state is calculated by dividing the difference between the number of abnormal device states after the intervention and the number of abnormal device states before the intervention by the number of abnormal device states before the intervention; It is judged whether the intervention effect score is lower than the preset lower limit of the intervention effect score or whether the behavior state recovery rate is lower than the lower limit of the behavior state recovery rate. If either of them holds, the virtual space modeling process of the campus digital twin simulation module is returned to recalculate the spatial state. If neither holds, the intervention process record data and the behavior change data are output to the self-evolving learning module.

[0012] In a preferred embodiment, the self-evolution learning module is used to synchronously input the intervention process record data and the behavior change data into the self-evolution learning data preparation process to form a self-evolution training data set; Input the self-evolution training data set into the behavior intention self-evolution optimization process. Through the error backpropagation algorithm, according to the difference between the behavior intention prediction output and the actual label, combined with the set target for improving the behavior intention prediction accuracy, adjust the behavior intention relationship matrix; wherein the error backpropagation algorithm calculates the difference between the behavior intention prediction output value and the actual label value, and adjusts the behavior intention relationship matrix in the gradient direction according to this difference; Judge whether the improvement amplitude of the behavior intention prediction accuracy in the result of the behavior intention self-evolution optimization process is lower than the lower limit of the self-evolution target improvement, and whether the decrease amplitude of the false alarm rate of the behavior anomaly early warning is lower than the lower limit of the false alarm rate decrease target. If either of them holds, return to the self-evolution learning data preparation process of the self-evolution learning module for continued training. If both are satisfied, enter the behavior scenario recognition optimization process; Input the self-evolution training data in the self-evolution training data set into the behavior scenario recognition optimization process, and optimize the behavior state determination rules of each behavior scenario category in the behavior scenario category set according to the comprehensive score of the behavior scenario recognition accuracy and the behavior anomaly rate; wherein the comprehensive score of the behavior anomaly rate is calculated by the ratio of the number of abnormal behavior detections to the total number of behavior detections; Judge whether the improvement amplitude of the behavior scenario recognition accuracy is lower than the lower limit of the behavior scenario recognition improvement target or whether the behavior anomaly rate is higher than the upper limit of the behavior anomaly target. If either of them holds, return to the self-evolution learning data preparation process of the self-evolution learning module for continued training. If both are satisfied, enter the self-evolution parameter synchronization process; In the self-evolution parameter synchronization process, synchronize the optimized behavior intention relationship matrix and the behavior state determination rules of each behavior scenario category in the behavior scenario category set to the behavior scenario recognition module and the behavior intention reasoning module to achieve continuous knowledge update.

[0013] The technical effects and advantages of the present invention: 1. By integrating the dynamic acquisition of multi-source behavior data and self-evolution learning, the present invention breaks through the limitations of the existing static rules and fixed threshold models, and realizes high-adaptability dynamic management of the diversity and suddenness of students' behaviors; 2. The system performs multi-level dynamic behavior scenario recognition by combining the behavior state probability matrix with the behavior activity category and duration, improving the real-time scenario classification accuracy and anomaly detection sensitivity under complex behavior patterns; 3. By introducing the behavior intention reasoning module, the present solution uses the behavior intention relationship matrix to realize the similarity analysis of the behavior state vector and the intention vector, identify the behavior intention of students, and improve the interpretability of behavior intention prediction; 4. Construct the virtual campus space state based on digital twin simulation, and combine the space heat map and strategy combination simulation to realize the prediction, evaluation and optimization of campus crowd distribution and behavior intervention strategies; 5. The self-evolving learning module in the system can continuously optimize the behavior intention relationship matrix and the set of behavior scenario categories, realize the adaptive adjustment of the model, improve the behavior recognition accuracy and reduce the abnormal false alarm rate, and has the ability of continuous knowledge update. Description of the Drawings

[0014] Figure 1 It is a schematic diagram of the system module of the present invention. Detailed Embodiments

[0015] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0016] Referring to the attached Figure 1 drawings, a smart campus management system according to an embodiment of the present invention includes a behavior data acquisition module, a behavior scenario recognition module, a behavior intention reasoning module, a campus digital twin simulation module, an adaptive behavior intervention module, and a self-evolving learning module; The behavior data acquisition module is used to collect original behavior data, sequentially perform anomaly detection, data cleaning, data standardization and feature extraction, and output a complete set of behavior data; The behavior scenario recognition module is based on the complete set of behavior data, and sequentially performs behavior scenario recognition, historical behavior matching and behavior trend prediction processes, and outputs the current behavior scenario state and the behavior trend prediction sequence; The behavior intention reasoning module identifies behavior intentions, evaluates abnormal behaviors and calculates the confidence of behavior intentions based on the behavior intention relationship matrix established by the current behavior scenario state, the behavior trend prediction sequence and historical behavior data, and outputs standard behavior intention state data; The campus digital twin simulation module constructs a virtual campus space state based on the standard behavior intention state data, combines the current behavior state data, personnel location data and facility usage data, performs space modeling and intervention strategy simulation, and outputs the target intervention strategy and the virtual campus space state; The adaptive behavior intervention module is used to generate and issue intervention measures, evaluate the feedback effect, and determine whether to return to the campus digital twin simulation module to recombine strategies or output to the self-evolving learning module; The self-evolution learning module is used to optimize the behavior intention relationship matrix and the set of behavior scenario categories, and update them to the behavior scenario recognition module and the behavior intention reasoning module to achieve continuous self-evolution.

[0017] The behavior data acquisition module collects the original behavior data by configuring the preset sensing devices. The sensing devices consist of a location signal acquisition device, an action behavior detection device, and an identity authentication reading device, forming a set of original behavior data. The set of original behavior data is transmitted to the data anomaly detection process to determine whether there are missing values and whether there are abnormal records of the sensing device status. If either of them holds, the data acquisition process of the sensing device of the behavior data acquisition module is returned to re-execute the data collection. The complete original behavior data is input into the data cleaning process, and the error data is removed according to the combination rule of the behavior data type and the device status code in the data cleaning process, forming a set of cleaned data. The set of cleaned data is passed into the data standardization process, and the field format is unified and the timestamps are aligned through the data type mapping table, and a set of standardized behavior data is output. The set of standardized behavior data is input into the feature extraction process, and a set of standard behavior features is extracted according to the behavior feature template library; the behavior feature template library includes behavior sequence patterns, behavior continuity windows, and behavior activity categories. It is judged whether there are discontinuous time windows in the set of standard behavior features and whether there are duplicate behavior records. If either of them holds, the data acquisition process of the sensing device of the behavior data acquisition module is returned to re-execute the data collection. If neither holds, the complete set of behavior data is output as the input data of the behavior scenario recognition module.

[0018] The behavior scenario recognition module is used to input the complete set of behavior data into the behavior scenario recognition process, and use the classification algorithm based on the behavior state probability matrix to combine the current behavior activity category and the behavior activity duration to identify the current behavior scenario. It is judged whether the current behavior activity category is consistent with the preset set of behavior scenario categories and whether the behavior activity duration is within the range of the continuity window. If either of them is not satisfied, the feature extraction process of the behavior data acquisition module is returned to recalibrate the behavior data; if satisfied, it enters the historical behavior matching process. The current behavior scenario is input into the historical behavior matching process, and the behavior matching degree score is output through the combined comparison of the Euclidean distance scoring algorithm and the historical behavior frequency scoring algorithm between the current behavior scenario and the historical behavior path sequence. It is judged whether the behavior matching degree score is lower than the set score lower limit or whether the current behavior activity category deviates from the historical behavior category sequence. If either of them holds, the behavior scenario recognition process of the behavior scenario recognition module is returned to re-classify. If neither holds, it enters the dynamic behavior trend prediction process. The current behavior scenario and the historical behavior trend are combined and input into the dynamic behavior trend prediction process. The future behavior state path is predicted through the behavior state transition probability model, and a behavior trend prediction sequence is formed. Determine whether the path consistency between the current behavior scenario state and the behavior trend prediction sequence is less than the preset lower limit of the consistency score or whether the behavior state change rate is higher than the dynamic threshold of the behavior state change rate. If either of them holds, return to the behavior scenario recognition process of the behavior scenario recognition module for re-analysis. If neither holds, output the current behavior scenario state and the behavior trend prediction sequence as the input data for the behavior intention inference module. The dynamic threshold of the behavior state change rate is calculated by real-time analyzing the number of student behavior state changes within the current time window and the average frequency and standard deviation of behavior state changes within the same historical time period. Specifically, the behavior state change frequency in the same time period is extracted from the historical behavior data, and the mean of this frequency is calculated as the reference value. Then, an adjustment coefficient is set in combination with the standard deviation to form the dynamic threshold in the current situation. Subsequently, the number of consecutive state changes in the current behavior state sequence is counted. If the current number of changes exceeds this dynamic threshold, it is considered that the behavior fluctuates abnormally, and the process is triggered to return to the behavior scenario recognition process of the behavior scenario recognition module for re-analysis. It should be noted that in the behavior scenario recognition process, the classification algorithm based on the behavior state probability matrix takes the behavior activity categories (including classroom attendance, club activities, restaurant activities, dormitory stay, teaching building stay, abnormal departure, etc.) and the behavior activity duration (referring to the continuous time length of the same behavior activity) extracted from the complete behavior data set as input variables, calculates the state occurrence probability and state transition probability of each behavior state in the historical behavior pattern library, and establishes a joint probability scoring matrix of the current behavior data and the historical behavior state sequence. Subsequently, in combination with whether the current behavior activity category is an allowed category (such as whether it is during the class schedule time or after-school time) and whether the behavior activity duration exceeds the upper limit of the historical behavior duration, the classification probability of the current behavior is comprehensively generated. If it is detected that the behavior activity category is "abnormal access to area activity" (such as leaving the dormitory area at unauthorized times or crossing the fence boundary signal area, which is determined as the behavior scenario of climbing over the wall) and the behavior activity duration exceeds the abnormal detection time threshold, it is directly marked as "behavior scenario of climbing over the wall", etc. The set of behavior scenario categories includes but is not limited to classroom learning behavior scenarios, teaching building activity behavior scenarios, restaurant dining behavior scenarios, dormitory stay behavior scenarios, campus path activity behavior scenarios, club activity behavior scenarios, and abnormal behavior scenarios. Among them, the abnormal behavior scenarios include climbing over the wall, staying in unauthorized areas for a long time, and unexpected high-frequency position changes, etc. In addition, in the historical behavior matching process, the current behavior scenario is first converted into a behavior path vector, which is composed of a continuous behavior activity category coding sequence and the normalized value of each behavior duration. Subsequently, the Euclidean distance is calculated with all the stored historical behavior path vectors in the historical behavior database to obtain the difference degree score between the current behavior and the historical behavior in the space-time dimension. The proportion of the behavior frequency of the current behavior activity category appearing in the historical data is counted, and the regularity of the current behavior is quantified through a frequency scoring function. Finally, the Euclidean distance score and the behavior frequency score are weighted and fused according to the set weights, and the behavior matching degree score is output to determine whether the current behavior conforms to the historical normal behavior pattern. In addition, in the behavior state transition probability model of the dynamic behavior trend prediction process, the current behavior scenario and the historical behavior trend data are jointly converted into a state sequence vector. Using the pre-established behavior state transition probability matrix, which calculates the transition probability of each state to the next state by statistically counting the transition frequencies of each behavior state to the subsequent state in the historical behavior sequence. With the current state as the input, by looking up the transition probability distribution corresponding to the current state in the transition probability matrix, according to the highest probability or the multi-state joint probability path, the optimal behavior state path sequence within a future fixed time window is iteratively deduced to form a behavior trend prediction sequence to provide the basic data for subsequent anomaly judgment and intervention decision-making.

[0019] The behavior intention inference module is used to synchronously input the current behavior scenario state and the behavior trend prediction sequence into the behavior intention mapping process, calculate the similarity between the behavior state vector and the behavior intention vector according to the behavior intention relationship matrix, and generate a preliminary behavior intention. It should be noted that for the behavior intention relationship matrix, the system extracts the mapping relationship between the behavior state characteristics and the behavior intention labels through historical behavior data samples, establishes a statistical weight matrix between the behavior state vector and the behavior intention vector, and forms the behavior intention relationship matrix for the behavior intention inference module to call during real-time inference. Judge whether the similarity between the behavior state vector and the behavior intention vector is lower than the preset similarity score lower limit or whether the number of behavior state changes within the behavior state continuity window is higher than the continuous state change upper limit. If either of them holds, return to the behavior trend prediction process of the behavior scenario recognition module to re-execute the prediction. If neither holds, enter the behavior anomaly recognition process. In the behavior anomaly recognition process, combined with the preset behavior intention categories and the behavior anomaly parameter set of the system, the anomaly risk score is calculated through the behavior anomaly multi-parameter joint scoring algorithm. The behavior intention categories include four categories: learning intention, avoiding management intention, social participation intention, and health anomaly intention. The behavior anomaly parameter set includes three categories: behavior frequency, behavior duration, and behavior location change rate. Judge whether the abnormal risk score is higher than the upper limit of the risk score and whether the duration of the behavior exceeds the upper limit of the continuous time window of the behavior category. If both conditions are met, synchronously output the behavior anomaly warning data; otherwise, enter the behavior intention confidence score process; Input the behavior intention and abnormal warning data into the behavior intention confidence score process, and output the behavior intention confidence based on the weighted average of the behavior data integrity score and the behavior intention historical consistency score; Combine the behavior intention, abnormal warning data and behavior intention confidence into the standard behavior intention status data and output it to the campus digital twin simulation module; It should be noted that in the behavior intention mapping process, a behavior state vector is extracted according to the current behavior scenario state and the behavior trend prediction sequence. The behavior state vector is composed of the behavior activity category code, the standardized value of the behavior duration and the behavior activity frequency; at the same time, the corresponding behavior intention vector is extracted from the historical behavior intention model library. In practical applications, the behavior intention vector contains the standard behavior feature templates corresponding to typical intention categories (such as learning intention, avoiding management intention, social intention, etc.); then, the cosine value of the angle between the behavior state vector and each behavior intention vector is calculated through the cosine similarity algorithm to obtain the similarity score. The closer the similarity score value is to 1, the higher the matching degree between the current behavior state and the intention template. Finally, the behavior intention with the highest score is selected as the preliminary behavior intention output; In the behavior anomaly recognition process, according to the current behavior intention category (learning intention, avoiding management intention, social participation intention, health anomaly intention), determine the corresponding behavior anomaly parameter weight template in the behavior anomaly parameter set. The behavior anomaly parameter weight template is the normal range and its weight coefficient of three parameters: predefined behavior frequency, behavior duration, and behavior position change rate for each intention category; then, calculate the deviation between the actual behavior frequency, actual behavior duration, and actual behavior position change rate in the current behavior data and the normal parameter range in the corresponding intention category template respectively to obtain three standardized deviation scores; finally, sum the three deviation scores after multiplying them by their respective weight coefficients to output a single abnormal risk score, which is used to quantify the abnormality degree of the current behavior. The higher the value, the greater the degree of deviation of the behavior from the normal behavior pattern; In the above-mentioned behavior intention confidence scoring process, the integrity of the current behavior data is analyzed, the data coverage rate of the perception device and the continuity of the behavior data are calculated, and a behavior data integrity score is formed. At the same time, the historical behavior intention sequence of the same individual in a similar situation is retrieved from the historical behavior database, and the historical consistency score of the behavior intention is calculated using the matching frequency between the current preliminary behavior intention and the historical behavior intention sequence. Finally, the behavior data integrity score and the historical consistency score of the behavior intention are weighted and summed according to the preset weight parameters of the system, and a comprehensive behavior intention confidence is output, which is used to evaluate the reliability of the current behavior intention reasoning result.

[0020] The campus digital twin simulation module is used to input the standard behavior intention state data into the virtual campus state construction process, and through the state synchronization algorithm, combine the current behavior state data, personnel location data and facility usage data to generate the virtual campus space state in real time; Judge whether the virtual campus space state completely covers all currently configured perception device areas and whether the state synchronization delay is lower than the preset synchronization delay upper limit. If either of them is not satisfied, return to the feature extraction process of the behavior data acquisition module to re-execute the data acquisition. If both are satisfied, enter the virtual space modeling process; In the virtual space modeling process, the virtual campus space state is input into the space distribution modeling process, and a space heat map is generated through the crowd flow distribution density and the device occupancy probability; The space heat map is input into the intervention strategy simulation process, and strategy combination simulations are carried out according to the preset strategy library in the system to form an intervention simulation result set; the strategy library includes three types: path guidance strategy, device shutdown strategy, and personnel reminder strategy; Judge whether the difference in the behavior anomaly warning probability between any two groups of results in the intervention simulation result set is less than the behavior anomaly difference tolerance upper limit, or whether the crowd distribution density after intervention exceeds the space distribution density upper limit. If either of them holds, return to the intervention strategy simulation process of the campus digital twin simulation module to re-combine the strategies. If neither holds, enter the strategy screening process; In the strategy screening process, the target intervention strategy is determined through the weighted average of the behavior intervention expected completion score and the expected crowd distribution optimization score; the target intervention strategy and the virtual campus space state are output as the input of the adaptive behavior intervention module; It should be noted that in the strategy screening process, the expected population distribution optimization score is a comprehensive score calculated by analyzing the pedestrian flow density distribution results after simulating each group of candidate intervention strategies, based on the spatial uniformity of the population density in each region and the degree of pressure release in the hot spots; further, according to the simulation results, the pedestrian flow density in each main region is statistically analyzed, and the regional density variance is calculated as the spatial uniformity index. At the same time, the pedestrian flow reduction rate in the high-density hot spots is evaluated as the pressure release index, and then the two indexes are weighted and summed according to the preset weight to obtain the expected population distribution optimization score corresponding to the strategy; subsequently, the population distribution optimization score of each group of intervention strategies and the corresponding expected completion degree score of the behavior intervention are weighted and averaged together, and finally the intervention strategy with the highest weighted score is selected as the target intervention strategy.

[0021] The adaptive behavior intervention module is used to input the target intervention strategy into the behavior intervention execution process, generate corresponding intervention measure data through the mapping table of the intervention strategy, and the intervention measure data includes behavior reminder data, device control data, and path guidance data; The intervention measure data is synchronously sent to the execution objects, which include faculty terminals, student terminals, and infrastructure control terminals; In the intervention result feedback process, it is judged whether the behavior state recovery rate in the intervention execution feedback is lower than the lower limit of the intervention target recovery rate or whether the number of abnormal feedbacks of the perception device execution state exceeds the upper limit of the device abnormal feedback number. If either of them holds, the intervention strategy simulation process of the campus digital twin simulation module is returned to recombine the strategy. If neither holds, the intervention effect evaluation process is entered; the behavior state recovery rate refers to the ratio of the number of preset normal behavior states after the intervention to the number of abnormal behavior states before the intervention; In the intervention effect evaluation process, a weighted comprehensive score of two parameters, the difference degree of the behavior state before and after the intervention and the abnormal change rate of the device state, is calculated to form the intervention effect score; the difference degree of the behavior state before and after the intervention is calculated by dividing the absolute value of the difference between the total number of behavior states before the intervention and the total number of behavior states after the intervention by the total number of behavior states before the intervention; the abnormal change rate of the device state is calculated by dividing the difference between the number of abnormal device states after the intervention and the number of abnormal device states before the intervention by the number of abnormal device states before the intervention; It is judged whether the intervention effect score is lower than the preset lower limit of the intervention effect score or whether the behavior state recovery rate is lower than the lower limit of the behavior state recovery rate. If either of them holds, the virtual space modeling process of the campus digital twin simulation module is returned to recalculate the spatial state. If neither holds, the intervention process record data and the behavior change data are output to the self-evolution learning module.

[0022] The self-evolution learning module is used to synchronously input the intervention process record data and the behavior change data into the self-evolution learning data preparation process to form a self-evolution training data set; Input the self-evolving training dataset into the self-evolving optimization process of behavior intention. Through the error backpropagation algorithm, based on the difference between the predicted output of behavior intention and the actual label, combined with the set target of improving the accuracy of behavior intention prediction, adjust the behavior intention relationship matrix. Among them, the error backpropagation algorithm calculates the difference between the predicted output value of behavior intention and the actual label value, and adjusts the behavior intention relationship matrix in the gradient direction according to this difference. Judge whether the improvement amplitude of the behavior intention prediction accuracy in the result of the self-evolving optimization process of behavior intention is lower than the lower limit of the self-evolving target improvement, and whether the decrease amplitude of the false alarm rate of the behavior anomaly early warning is lower than the lower limit of the false alarm rate decrease target. If either of them holds, return to the self-evolving learning data preparation process of the self-evolving learning module for continued training. If both are satisfied, enter the behavior scenario recognition optimization process. Input the self-evolving training data in the self-evolving training dataset into the behavior scenario recognition optimization process, and optimize the behavior state determination rules of each behavior scenario category in the behavior scenario category set according to the comprehensive score of the behavior scenario recognition accuracy and the behavior anomaly rate. Among them, the comprehensive score of the behavior anomaly rate is calculated by the ratio of the number of abnormal behavior detections to the total number of behavior detections. Judge whether the improvement amplitude of the behavior scenario recognition accuracy is lower than the lower limit of the behavior scenario recognition improvement target or whether the behavior anomaly rate is higher than the upper limit of the behavior anomaly target. If either of them holds, return to the self-evolving learning data preparation process of the self-evolving learning module for continued training. If both are satisfied, enter the self-evolving parameter synchronization process. In the self-evolving parameter synchronization process, synchronize the optimized behavior intention relationship matrix and the behavior state determination rules of each behavior scenario category in the behavior scenario category set to the behavior scenario recognition module and the behavior intention reasoning module to achieve continuous knowledge update.

[0023] It should be noted as a whole that the intelligent campus management system provided by the present invention is an innovative solution proposed based on in-depth analysis of the problems of static rules, single data source, and inability to dynamically respond to behavior diversity and suddenness existing in the prior art. The system architecture consists of a behavior data acquisition module, a behavior scenario recognition module, a behavior intention reasoning module, a campus digital twin simulation module, an adaptive behavior intervention module, and a self-evolving learning module. The behavior data acquisition module uses configured position signal acquisition devices, action behavior detection devices, and identity authentication reading devices to collect the original behavior data of students on campus in real time. After the collected data undergoes anomaly detection, cleaning, and standardization, standard behavior feature sets such as behavior sequence patterns, behavior continuity windows, and behavior activity categories are extracted to form a complete behavior data set, providing a high-quality data basis for subsequent analysis. Subsequently, the behavior scenario recognition module inputs the complete behavior data set, and uses a classification algorithm based on the behavior state probability matrix, combined with the current behavior activity category and the duration of the behavior activity, to recognize the current behavior scenario; further, through the combined comparison of the Euclidean distance scoring algorithm and the behavior frequency scoring algorithm of the historical behavior path, the behavior matching degree score is output, and the future behavior state path is predicted through the behavior state transition probability model to form a behavior trend prediction sequence; a multi-level judgment logic is set in the behavior scenario recognition module, and the dynamic threshold mechanism is used to monitor the behavior state change and detect anomalies in real time to ensure the dynamic adaptability of the behavior scenario analysis; The behavior intention reasoning module synchronously inputs the current behavior scenario state and the behavior trend prediction sequence, calculates the similarity between the behavior state vector and the behavior intention vector through the behavior intention relationship matrix established by the system based on historical behavior data, and outputs the preliminary behavior intention; at the same time, the behavior intention reasoning module combines the behavior intention category and the behavior anomaly parameter set, calculates the anomaly risk score, and identifies abnormal behaviors; the behavior intention confidence is calculated by weighted calculation of the behavior data integrity score and the historical consistency score of the behavior intention, and the standard behavior intention state data is output to provide input for the subsequent digital twin simulation; The campus digital twin simulation module generates the virtual campus space state in real time based on the standard behavior intention state data, combined with the current behavior state data, personnel location data and facility usage data; through spatial distribution modeling, a campus space heat map is formed, and strategies such as path guidance, equipment shutdown and personnel reminder in the strategy library are input for intervention strategy simulation; multiple groups of simulation results are evaluated through the difference in behavior anomaly warning probability and the standard of crowd distribution density, and finally the target intervention strategy and the virtual campus space state are output for guiding actual intervention; The adaptive behavior intervention module receives the target intervention strategy, generates intervention measure data such as behavior reminder, equipment control and path guidance by using the intervention strategy mapping table, and synchronously sends it to the faculty terminal, student terminal and infrastructure control terminal; the adaptive behavior intervention module continuously monitors the intervention results in the system, evaluates the behavior state recovery rate and the abnormal change of the equipment state, if the effect does not meet the standard, it returns to the campus digital twin simulation module to re-optimize the strategy, if it meets the standard, it outputs the intervention process record data and the behavior change data for the self-evolution learning module to use; The self-evolution learning module forms a self-evolution training data set based on the intervention process record data and the behavior change data, and uses the error backpropagation algorithm to optimize the behavior intention relationship matrix; at the same time, according to the behavior scenario recognition accuracy and the anomaly rate, the behavior state determination rules of each behavior scenario category in the behavior scenario category set are optimized; in addition, in the actual system application, the self-evolution learning module can improve the prediction accuracy and reduce the abnormal false alarm rate through multiple rounds of iterative learning, and synchronously update to the dynamic behavior scenario recognition module and the behavior intention reasoning module after meeting the set threshold, and finally realize the continuous self-evolution and knowledge update of the system.

[0024] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A smart campus management system, including a behavior data acquisition module, a behavior scene recognition module, a behavior intention reasoning module, a campus digital twin simulation module, an adaptive behavior intervention module, and a self-evolution learning module, characterized in that: The behavior data acquisition module is used to collect raw behavior data, perform anomaly detection, data cleaning, data standardization and feature extraction in sequence, and output a complete behavior data set; The behavior scene recognition module performs the behavior scene recognition, historical behavior matching and behavior trend prediction processes in sequence based on the complete behavior data set, and outputs the current behavior scene status and behavior trend prediction sequence; The behavior intention reasoning module identifies behavior intention, evaluates abnormal behavior, calculates the confidence of behavior intention, and outputs standard behavior intention status data based on the behavior intention relationship matrix established based on the current behavior scenario status, behavior trend prediction sequence, and historical behavior data; The campus digital twin simulation module is based on standard behavior intention state data, combined with current behavior state data, personnel location data and facility usage data, to construct the virtual campus space state, perform space modeling and intervention strategy simulation, and output the target intervention strategy and virtual campus space state; The adaptive behavior intervention module is used to generate and issue intervention measures, evaluate the feedback effect, and determine whether to return to the campus digital twin simulation module to reorganize the strategy or output it to the self-evolution learning module; The self-evolution learning module is used to optimize the behavior intention relationship matrix and the behavior scenario category set, and update the behavior scenario recognition module and the behavior intention reasoning module to achieve continuous self-evolution.

2. A smart campus management system according to claim 1, characterized in that: The behavior data acquisition module collects raw behavior data by configuring preset sensing devices. The sensing devices are composed of position signal acquisition devices, action behavior detection devices, and identity authentication reading devices to form a collection of raw behavior data. The original behavior data set is transmitted to the data anomaly detection process to determine whether there are missing values ​​and whether there are abnormal records of the sensor device status. If either of the two is true, the data acquisition process of the sensor device in the behavior data acquisition module is returned to re-execute data collection; Input the complete original behavior data into the data cleaning process, and remove the erroneous data according to the combination rules of the behavior data type and the device status code in the data cleaning process to form a cleaned data set; The cleaned data set is passed into the data standardization process, the field format is unified and the timestamp is aligned through the data type mapping table, and the standardized behavior data set is output; Input the standardized behavior data set into the feature extraction process, and extract the standard behavior feature set based on the behavior feature template library; The behavior feature template library includes behavior sequence patterns, behavior continuity windows, and behavior activity categories; Determine whether there are discontinuous time windows in the standard behavior feature set and whether there are repeated behavior records. If either of them is true, return to the data acquisition process of the perception device of the behavior data acquisition module to re-execute data collection. If neither is true, output the complete behavior data set as the input data of the behavior scene recognition module.

3. A smart campus management system according to claim 2, characterized in that: The behavior scenario recognition module is used to input the complete behavior data set into the behavior scenario recognition process, and use the classification algorithm based on the behavior state probability matrix to combine the current behavior activity category and the behavior activity duration to recognize the current behavior scenario; Judge whether the current behavior activity category is consistent with the preset behavior scenario category set and whether the behavior activity duration is within the continuous window range. If either of them is not satisfied, return to the feature extraction process of the behavior data acquisition module to recalibrate the behavior data; if satisfied, enter the historical behavior matching process; Input the current behavior scenario into the historical behavior matching process, and output the behavior matching degree score through the combined comparison of the Euclidean distance scoring algorithm and the historical behavior frequency scoring algorithm between the current behavior scenario and the historical behavior path sequence; Judge whether the behavior matching degree score is lower than the set score lower limit or whether the current behavior activity category deviates from the historical behavior category sequence. If either of them is true, return to the behavior scenario recognition process of the behavior scenario recognition module for reclassification. If both are not true, enter the dynamic behavior trend prediction process; Input the current behavior scenario and the historical behavior trend into the dynamic behavior trend prediction process, and predict the future behavior state path through the behavior state transition probability model to form a behavior trend prediction sequence; Judge whether the path consistency between the current behavior scenario state and the behavior trend prediction sequence is less than the preset consistency score lower limit or whether the behavior state change rate is higher than the behavior state change rate dynamic threshold. If either of them is true, return to the behavior scenario recognition process of the behavior scenario recognition module for re-analysis. If both are not true, output the current behavior scenario state and the behavior trend prediction sequence as the input data of the behavior intention inference module.

4. A smart campus management system according to claim 3, wherein: The behavior intention inference module is used to synchronously input the current behavior scenario state and the behavior trend prediction sequence into the behavior intention mapping process, calculate the similarity between the behavior state vector and the behavior intention vector according to the behavior intention relationship matrix, and generate a preliminary behavior intention; Judge whether the similarity between the behavior state vector and the behavior intention vector is lower than the preset similarity score lower limit or whether the number of behavior state changes within the behavior state continuous window is higher than the continuous state change upper limit. If either of them is true, return to the behavior trend prediction process of the behavior scenario recognition module to re-perform the prediction. If both are not true, enter the behavior anomaly recognition process; In the behavior anomaly recognition process, combine the preset behavior intention categories and the behavior anomaly parameter set of the system, and calculate the anomaly risk score through the behavior anomaly multi-parameter joint scoring algorithm; The behavior intention categories include four categories: learning intention, avoiding management intention, social participation intention, and health anomaly intention; the behavior anomaly parameter set includes three categories: behavior frequency, behavior duration, and behavior location change rate; Judge whether the anomaly risk score is higher than the risk score upper limit and whether the behavior duration exceeds the upper limit of the behavior category continuous time window. If both are true, synchronously output the behavior anomaly warning data; Otherwise, enter the behavior intention confidence score process; Input the behavior intention and anomaly warning data into the behavior intention confidence score process, and output the behavior intention confidence based on the weighted average of the behavior data integrity score and the behavior intention historical consistency score; Combine the behavior intention, anomaly warning data, and behavior intention confidence into the standard behavior intention status data, and output it to the campus digital twin simulation module.

5. The intelligent campus management system according to claim 4, characterized in that: The campus digital twin simulation module is used to input the standard behavior intention status data into the virtual campus status construction process, and combine the current behavior status data, personnel location data, and facility usage data through the status synchronization algorithm to generate the virtual campus space status in real time; Judge whether the virtual campus space status completely covers all the currently configured sensing device areas and whether the status synchronization delay is lower than the preset synchronization delay upper limit. If either of them is not satisfied, return to the feature extraction process of the behavior data acquisition module to re-execute the data acquisition. If both are satisfied, enter the virtual space modeling process; In the virtual space modeling process, input the virtual campus space status into the space distribution modeling process, and generate the space heat map through the crowd flow distribution density and the device occupancy probability; Input the space heat map into the intervention strategy simulation process, and perform strategy combination simulation based on the preset strategy library in the system to form an intervention simulation result set; the strategy library includes three types: path guidance strategy, device shutdown strategy, and personnel reminder strategy; Judge whether the difference in the behavior anomaly warning probability between any two groups of results in the intervention simulation result set is less than the behavior anomaly difference tolerance upper limit, or whether the crowd distribution density after intervention exceeds the space distribution density upper limit. If either of them holds, return to the intervention strategy simulation process of the campus digital twin simulation module to re-combine the strategies. If both do not hold, enter the strategy screening process; In the strategy screening process, determine the target intervention strategy through the weighted average of the behavior intervention expected completion score and the expected crowd distribution optimization score; Output the target intervention strategy and the virtual campus space status as the input of the adaptive behavior intervention module.

6. The intelligent campus management system according to claim 5, characterized in that: The adaptive behavior intervention module is used to input the target intervention strategy into the behavior intervention execution process, and generate the corresponding intervention measure data through the mapping table of the intervention strategy. The intervention measure data includes behavior reminder data, device control data, and path guidance data; Synchronously send the intervention measure data to the execution objects, which include faculty terminals, student terminals, and infrastructure control terminals; In the intervention result feedback process, judge whether the behavior status recovery rate in the intervention execution feedback is lower than the intervention target recovery rate lower limit or whether the number of abnormal feedbacks on the execution status of the sensing device exceeds the device abnormal feedback number upper limit. If either of them holds, return to the intervention strategy simulation process of the campus digital twin simulation module to re-combine the strategies. If both do not hold, enter the intervention effect evaluation process; In the intervention effect evaluation process, calculate the weighted comprehensive score of two parameters, the behavior status difference degree before and after the intervention and the device status abnormal change rate, to form the intervention effect score; Determine whether the intervention effect score is lower than the preset lower limit of the intervention effect score or whether the behavior state recovery rate is lower than the lower limit of the behavior state recovery rate. If either of them holds, return to the virtual space modeling process of the campus digital twin simulation module to recalculate the space state. If neither holds, output the intervention process record data and the behavior change data to the self-evolving learning module.

7. A smart campus management system according to claim 6, characterized in that: The self-evolving learning module is used to synchronously input the intervention process record data and the behavior change data into the self-evolving learning data preparation process to form a self-evolving training data set; Input the self-evolving training data set into the behavior intention self-evolving optimization process. Through the error backpropagation algorithm, according to the difference between the behavior intention prediction output and the actual label, and in combination with the set target for improving the behavior intention prediction accuracy, adjust the behavior intention relationship matrix; Judge whether the improvement amplitude of the behavior intention prediction accuracy in the result of the behavior intention self-evolving optimization process is lower than the lower limit of the self-evolving target improvement and whether the decrease amplitude of the false alarm rate of the behavior anomaly warning is lower than the lower limit of the false alarm rate decrease target. If either of them holds, return to the self-evolving learning data preparation process of the self-evolving learning module to continue training. If both are satisfied, enter the behavior scenario recognition optimization process; Input the self-evolving training data in the self-evolving training data set into the behavior scenario recognition optimization process, and optimize the behavior state determination rules of each behavior scenario category in the behavior scenario category set according to the comprehensive score of the behavior scenario recognition accuracy and the behavior anomaly rate; Judge whether the improvement amplitude of the behavior scenario recognition accuracy is lower than the lower limit of the behavior scenario recognition improvement target or whether the behavior anomaly rate is higher than the upper limit of the behavior anomaly target. If either of them holds, return to the self-evolving learning data preparation process of the self-evolving learning module to continue training. If both are satisfied, enter the self-evolving parameter synchronization process; In the self-evolving parameter synchronization process, synchronize the optimized behavior intention relationship matrix and the behavior state determination rules of each behavior scenario category in the behavior scenario category set to the behavior scenario recognition module and the behavior intention reasoning module.

Images