CN111312367A

CN111312367A - Campus personnel abnormal psychological prediction method based on self-adaptive cloud management platform

Info

Publication number: CN111312367A
Application number: CN202010389779.6A
Authority: CN
Inventors: 黄希; 聂贻俊; 刘翼
Original assignee: Chengdu Paiwo Zhitong Technology Co ltd
Current assignee: Chengdu Paiwo Zhitong Technology Co ltd
Priority date: 2020-05-11
Filing date: 2020-05-11
Publication date: 2020-06-19

Abstract

The invention discloses a campus personnel abnormal psychological prediction method based on a self-adaptive cloud management platform, which comprises the steps of firstly establishing a campus personnel abnormal psychological prediction model, matching personnel entering a school zone through the self-adaptive cloud management platform or newly allocating a unique identification code, and acquiring facial feature data through a camera; then, acquiring a plurality of activity conditions such as network use conditions, external behavior expressions, in-school footprints, consumption conditions, courseware conditions, collective activity participation conditions and the like of personnel in a school zone through a plurality of perception front ends; then, carrying out data analysis on the collected multiple activity conditions and extracting a group of key data sets according to each identity identification code in a classified manner; finally, sending the multiple groups of key data sets to an abnormal psychological prediction model of the school staff for prediction analysis, storing all prediction analysis results in a psychological prediction database, and labeling and alarming abnormal prediction analysis results in the prediction analysis; thereby improving campus security.

Description

Campus personnel abnormal psychological prediction method based on self-adaptive cloud management platform

Technical Field

The invention belongs to the technical field of big data processing, and particularly relates to a campus personnel abnormal psychological prediction method based on a self-adaptive cloud management platform.

Background

The campus construction all has been the problem that the education field was paid attention to all, and it is all chasing by each school to build a set of wisdom campus solution that is fit for self development for the campus.

According to the existing smart campus project, a campus mobile management system is constructed, and technologies such as big data and a cloud platform are utilized to build application based on life data analysis, so that a smart and safe smart campus is built. Through constructing the wisdom campus, the fine difficult problem of having solved campus safety of campus mobile management system has made things convenient for simultaneously school to carry out student's attendance record, has again enough to help mr and school to carry out daily arrangement of lessons and teaching activities. In the technology, the system combines the technologies of face recognition, high-precision indoor positioning, mobile internet, cloud computing and the like, solves the problems of ' signing on behalf of others ', ' attendance checking in ', mobile phone signing in ' and ' multi-person signing in ' simultaneously, also provides the functions of automatic timing roll calling, manual roll calling, online attendance checking, sales requesting, signature supplementing, sign in and sign out and the like, enables related personnel to look up attendance checking results at any time and any place, know attendance statistical analysis and transverse comparison data of various colleges, professions, grades, classes or individuals, and can obtain ' early warning ' reports under various mechanisms.

On the other hand, campus safety is always a problem which is valued by schools and parents, and violence and fighting behaviors in the campuses; the infringement behavior of the outside bad personnel to the students at school; a light-life behavior; self abuse behavior; the mental health problem is hidden behind the back of the pillow. How to use big data analysis technology to actively prevent and quickly and effectively deal with risks possibly occurring or already emerging due to abnormal psychology of people from the perspective of psychological crisis intervention becomes a subject which needs to be intensively researched at present.

Disclosure of Invention

The invention aims to provide a campus personnel abnormal psychological prediction method based on a self-adaptive cloud management platform, which is used for observing any one or more of network use conditions, external behavior expressions, campus footprints, consumption conditions, courseware conditions and collective activity participation conditions of the same personnel and comprehensively predicting the psychological state of personnel entering a school zone by utilizing a campus personnel abnormal psychological prediction model, so that the campus safety is improved.

The invention is mainly realized by the following technical scheme:

a campus personnel abnormal psychological prediction method based on a self-adaptive cloud management platform comprises the steps of firstly, establishing a campus personnel abnormal psychological prediction model, matching or newly allocating a unique identification code to personnel entering a school zone range through the self-adaptive cloud management platform, collecting face feature data through a camera, and storing the identification code and the face feature data which correspond to each other in a face database as face feature data; then, acquiring a plurality of activity conditions of personnel in the school zone through a plurality of sensing front ends comprising a camera, a campus card reading device and a network management server; then, a distributed server cluster of a network layer performs data analysis on the collected multiple activity conditions and extracts a group of key data sets according to each identity identification code in a classified manner; finally, sending the multiple groups of key data sets to an abnormal psychological prediction model of the school staff for prediction analysis, storing all prediction analysis results in a psychological prediction database, and labeling and alarming abnormal prediction analysis results in the prediction analysis; the activity condition of the personnel in the school district comprises any one or more of network use condition, external performance, school foot print, consumption condition, courseware condition and collective activity participation condition.

The invention has the beneficial effects that:

(1) the invention observes one or more of the network use condition, the external behavior expression, the campus footprint, the consumption condition, the courseware condition and the collective activity participation condition of the same person, and comprehensively predicts the psychological state of the person entering a school zone by using an abnormal psychological prediction model of the person in the school, thereby improving the campus safety;

(2) the method analyzes the acquired massive webpage operation data by operating the webpage of a certain user in a machine learning mode, and predicts the interests of the user within a period of time, so as to judge whether the tendencies of emotions such as negative, violent and extreme exist in the interests.

Drawings

FIG. 1 is a schematic diagram of the technical idea of acquiring the network usage situation, the external behavior expression, the campus footprint, the consumption situation, the courseware situation and the collective activity participation situation of the personnel according to the present invention.

Fig. 2 is a schematic diagram of technical ideas for analyzing each key data set by an intra-school personnel abnormal psychological prediction model.

FIG. 3 is a schematic diagram of a technical idea of a school personnel abnormal behavior scoring by an abnormal psychological prediction model.

Fig. 4 is a flowchart illustrating a method for predicting interests reflected by a user during web browsing.

Fig. 5 is a schematic flow chart of a face recognition and id code matching method.

Fig. 6 is a flow chart of a behavior feature extraction and anomaly classification scoring method according to the present invention.

Fig. 7 is a schematic flow chart of a behavior feature extraction and anomaly classification scoring method in target detection based on the YOLO model in the present invention.

Fig. 8 is a flow chart of an identification method specially for student classroom behavior according to the present invention.

FIG. 9 is a flowchart illustrating a method for analyzing personnel consumption in a school district.

FIG. 10 is a flowchart illustrating a method for analyzing personnel consumption in a school district.

Fig. 11 is a flowchart illustrating a user behavior monitoring method.

FIG. 12 is a diagram of a virtualization architecture.

FIG. 13 is an information system level protection security technology design framework.

Detailed Description

First, it should be noted that three psychological states of a person are:

normal state (normalcy) -mental health;

unbalanced state (bias) -psychological sub-health;

unhealthy state (metamorphosis) — psychological disease.

The psychological disease states are mainly expressed as follows:

1. pain (stress): in the case of depression, patients are often in a very tragic condition. It is not happy, sad, all have no meaning, and want to suicide. Other psychological diseases can not cause the pain of the user but cause the pain of other people. For example, the patients do not suffer from the social personality, but lose the same situation, the instinct of moral judgment, and have no hesitation on the victims of others and stealing.

2. Malfunction (dysfunction) physical or social malfunction. Still in the case of depression, the patient may not have the desire to get up, be unable to learn, and not want to touch friends. Social isolation, etc.

3. Behavioral abnormalities (Deviance) are one of the most controversial, since abnormal behaviors are defined differently in different civilizations, but highly abnormal behaviors are considered to be psychosis if they are not accepted at all by the civilization in which they are located.

Then, when focusing on the mental state, we usually observe from the external appearance, that is, the abnormal behavior we need to pay attention to is as follows:

(1) drowning in networks, electronic novels, etc.;

(2) self-complaining of self-worries, self-speaking, persistent decline in social interests, inability to communicate, presence of social disabilities, etc.;

(3) frequent course escaping and unqualified scores of multiple lessons;

(4) impulsive behavior;

(5) encounter with major traumatic events, etc.

Secondly, the method disclosed by the invention is only a prediction method based on data analysis, not a diagnosis method, and the result is only used as a reference for sensing the potential danger in advance.

Moreover, the data extraction means disclosed by the invention is realized technically, and when the data extraction means is operated in an actual project, corresponding data can be legally obtained according to relevant laws and regulations of the country and the industry, so that the problem of invading the privacy of other people is not involved.

Based on the above description, the following description will be given, with reference to specific embodiments, to describe the campus personnel abnormal psychological prediction method based on the adaptive cloud management platform provided by the present invention, and with reference to technologies such as big data analysis and deep learning, the psychological health state of the personnel is predicted by a scientific means, so as to improve the campus security.

Example 1:

According to the technical scheme, a sub-project in a smart campus project carries out data analysis on 'network use condition', 'external behavior performance', 'school internal footprint', 'consumption condition', 'courseware condition' and 'collective activity participation condition' of personnel in a school zone respectively by utilizing partial functions of an adaptive cloud management platform, and therefore the possibility that an abnormal psychology exists in a certain person is comprehensively predicted. Once a person with a prominent psychological abnormality is found, close attention needs to be paid to the movement of the person.

For example:

person P: often browsing web pages with violent words on the web; at ordinary times, people like three and five groups and push 22228and 22228; when no people exist around, the guidepost and the dustbin on the roadside of the school park are damaged; … … are provided. The person P who performs the above may have a tendency to violence.

Person Q: surfing the Internet often in late night, and browsing the web pages with sadness, pain and other worlds on the Internet; at ordinary times, one person likes to walk alone and select some obscure routes; the patients often seek medical treatment or buy medical supplies such as alcohol, gauze and the like due to insomnia and small trauma; often lack of lessons; rarely participate in collective activities; … … are provided. People P with the above-mentioned manifestations may have a tendency to depression.

In the embodiment, the intra-school personnel abnormal psychological prediction model is trained based on a deep learning technology, and the prediction accuracy is improved by continuously converging the prediction result through machine learning.

Example 2:

the embodiment is optimized on the basis of embodiment 1, mainly aiming at persons who go in and go out of a campus and except students, analyzing data of the persons in a monitoring area, namely the network use condition, the external performance and the school internal footprint, and analyzing data collected by the three aspects of the network use condition, the external performance and the school internal footprint through an abnormal psychological prediction model of the school persons, so that the possibility of abnormal psychology of the school persons is predicted.

First, people who go in and out of a campus except students mainly have: staff, logistics transportation personnel, external visitors, etc. The teacher belongs to the campus personnel of cominging in and going out of multifrequency, and is the same with the student, and the school system is stored with comparatively perfect individual basic information, and simultaneously, self-adaptation cloud management platform adopts long-term identification code distribution strategy, can distribute a long-term, fixed, unique identification code for it, for example: and (6) job number. Logistics transportation personnel, belonging to school personnel who frequently go in and out; external visitors, belonging to low-frequency campus access personnel, adopt a temporary identification code distribution strategy, and distribute a temporary and unique identification code for the external visitors. The temporary identification code is usually randomly generated by numbers and letters, but the same random code cannot be allocated to different people for a long period of time, and is generated again if the random code appears.

Therefore, when a person X with non-student identity enters the campus, face feature data are recorded in a camera at a school gate, the camera at the school gate collects face feature data and sends the face feature data to a background server, and basic information of the person is searched to verify the identity.

If the person X is the identity of a teaching worker, the background server can push the searched identity identification code to an access control display screen of a school entrance, and meanwhile, the face image which is prestored in the background server and is associated with the identity identification code is pushed.

If the person X is the identities of non-students, non-teaching workers, such as 'logistics transport personnel' and 'external visitors', when the person X enters a school zone for the first time, the background server allocates a special identity identification code for the person X and creates a person file, and the person X acquires the face feature data of the person X and sends the face feature data to the background server for classification and storage.

When the personnel X surf the internet through the network monitored by the network management server in the monitoring area, the self-adaptive cloud management platform crawls the surfing data of the personnel X through the network management server to obtain the surfing time data, the surfing time period data, the webpage operation data and the like, so that the surfing time data statistics, the surfing time period data statistics and the webpage browsing interest and hobby prediction are carried out.

The data statistics of the internet surfing time is carried out according to different periods of year, month and day, and an internet surfing time statistical table is generated;

the data statistics of the internet surfing time period is carried out by taking a day as a cycle and a 24-hour system, and an internet surfing time period statistical table is generated;

the webpage browsing interest prediction is to predict the interest reflected by the user when browsing the webpage according to a set period and count the prediction result to generate a network information interest statistical table.

For the statistics of the internet surfing time and the internet surfing time period, the prior art is mature, and the details are not repeated. One main innovation point of the embodiment is that a machine learning mode is adopted to analyze collected massive webpage operation data through webpage operation of a certain user, and the interests and hobbies of the user in a period of time are predicted, so that whether the tendency of emotions such as negative emotions, violence and extreme emotions exists or not is judged from the interests and hobbies.

Further, the data analysis method for the external behavior expression of the person X is as follows: firstly, constructing a behavior analysis model capable of identifying the behavior of a person in a video and a face identification model capable of identifying the face characteristics in the video by using image space characteristics extracted by a convolutional neural network; then video data are collected through a plurality of cameras arranged in a teaching area and a public activity area, and the video server group in the distributed server cluster carries out face recognition and identity recognition code matching, behavior feature extraction, abnormal classification and classification scoring according to an abnormal behavior management strategy on the collected video data; secondly, grouping and integrating the results of face recognition and identity recognition code matching and the results of behavior feature extraction and abnormal classification scoring according to the identity recognition codes; and finally, the grouped and integrated data is used as a key data set of the behavior expression and is sent to an intra-school personnel abnormal psychological prediction model for prediction analysis.

Further, the data analysis method for the campus footprint of the person X mainly comprises the steps of extracting face recognition results extracted from video data collected by a plurality of cameras arranged in a teaching area and a public activity area, extracting position information of the cameras and identification codes corresponding to the face recognition results, and grouping all the position information according to the identification codes to form a key data set of the footprint.

Other parts of this embodiment are the same as embodiment 1, and thus are not described again.

Example 3:

the embodiment is optimized on the basis of the embodiment 1 or 2, and the embodiment mainly aims at the personnel R with the identity of the student, and performs data analysis on the network use condition, the external behavior expression, the school internal footprint, the consumption condition, the courseware condition and the collective activity participation condition of the personnel R with the identity of the student by using the self-adaptive cloud management platform, so as to perform comprehensive scoring.

It should be noted that the identity of the student in the creation of the present invention refers to the student registered in the system used by the present platform, and the identity of the student not registered in the system used by the present platform is also the non-student identity in the creation of the present invention.

Firstly, analyzing data of network use conditions of personnel in a school zone.

Analyzing the data of the network use condition of personnel in the school zone, and grouping and integrating the analysis results of the network use condition according to the identification codes only aiming at the data analysis of the network use condition which can be monitored by a network management server in the monitoring area; the grouped and integrated data is used as a key data set used by the network and is sent to an intra-school personnel abnormal psychological prediction model for prediction analysis;

the main tasks of data analysis of the network use condition of personnel in a school district comprise: data statistics of internet surfing time, data statistics of internet surfing time intervals and prediction of interest and hobbies of webpage browsing;

For example: when the distributed message processing system analyzes that the time of surfing the internet of a certain person R continuously for one week exceeds 6 hours every day, the person R is very likely to be drowned in a network, an electronic novel and the like, so that the person R is abnormally marked when being analyzed according to a network monitoring strategy.

When predicting the interests and hobbies reflected by the user during the web browsing, as shown in fig. 4, the following specific steps are adopted:

step S1: acquiring a user log, dividing a plurality of user logs associated with the same identification code into a group, and extracting an IP address in each user log; the acquired user log is sent to a distributed message processing system in a Json format;

step S2: acquiring a webpage access record from the acquired user log, and adding the webpage access record into a MySQL database;

step S3: downloading HTML pages according to the webpage access records, filtering out repeated webpages and recording the number of times of repetition, crawling webpage contents from corresponding URLs according to a crawling rule by the filtered webpages, storing the crawled contents in a classified manner, and transferring the crawled HTML page codes to a MongoDB database;

step S4: selecting a corresponding specific extraction algorithm for an explicit webpage source to extract key contents and selecting a corresponding generalized extraction algorithm for an implicit webpage source or a webpage source without a corresponding specific extraction algorithm to extract key contents according to different webpage sources; the key content comprises webpage information and operation information of a user on the webpage, wherein the operation information comprises browsing and/or searching and/or collecting and/or paying attention to and/or sharing and/or downloading;

step S5: extracting webpage keywords from the webpage information of each key content, and combining multiple groups of webpage keywords corresponding to multiple user logs associated with the same identity identification code to form an original situation data group;

step S6: distributing weights to each group of webpage keywords in the original situation data set according to the repetition times of the corresponding webpage, the repetition times of the webpage keywords and the operation information of the webpage by a user, and extracting to obtain a situation characteristic data set;

step S7: and performing emotion recognition and emotion classification on the context feature data set in the step S6 based on a preset context type training set and an iterative K-means algorithm.

Grouping and integrating emotion classification results according to the identification codes; and the grouped and integrated data is used as a key data set used by the network and is sent to an intra-school personnel abnormal psychological prediction model for prediction analysis.

And secondly, analyzing the data of the external behavior of the personnel in the school zone.

The data of the external behavior expression of the personnel in the school zone are analyzed based on an external behavior analysis system, and the specific analysis method comprises the following steps: firstly, constructing a behavior analysis model capable of identifying the behavior of a person in a video and a face identification model capable of identifying the face characteristics in the video by using image space characteristics extracted by a convolutional neural network; then video data are collected through a plurality of cameras arranged in a teaching area and a public activity area, and the video server group in the distributed server cluster carries out face recognition and identity recognition code matching, behavior feature extraction, abnormal classification and classification scoring according to an abnormal behavior management strategy on the collected video data; secondly, grouping and integrating the results of face recognition and identity recognition code matching and the results of behavior feature extraction and abnormal classification scoring according to the identity recognition codes; and finally, the grouped and integrated data is used as a key data set of the behavior expression and is sent to an intra-school personnel abnormal psychological prediction model for prediction analysis.

2.1, carrying out face recognition and identity recognition code matching.

As shown in fig. 5, the face recognition and identity recognition code matching specifically includes the following steps:

step A1: preprocessing video data collected by a plurality of cameras arranged in a teaching area and a public activity area, dividing each n frames of a preprocessed video stream into an image group, wherein the 1 st frame of each image group is a key frame, the 2 nd to the nth frames are non-key frames, detecting the positions of all human faces and the positions of facial key points of each human face in the video frame by adopting an MTCNN algorithm for each image group key frame, and aligning the positions of the facial key points of each human face;

step A2: extracting a face feature actual value of each face at the position of a face key point of each face by using a face recognition model;

step A3: converting the actual face feature value of each face obtained in the step A2 into a hash feature value;

step A4: searching the hash characteristic values of the face to be recognized, which are obtained in the step A3, in a face database in which a plurality of face characteristic data sets are prestored, screening out a plurality of candidate hash characteristic values, using the obtained candidate hash characteristic values as indexes, inquiring face characteristic actual values corresponding to the candidate hash characteristic values in the face database, and using the inquired face characteristic actual values as candidate face characteristic actual values;

step A5: calculating the similarity between the actual face feature value of the face to be recognized and the actual candidate face feature value obtained in the step A4, taking the face corresponding to the actual candidate face feature value with the similarity exceeding a set similarity threshold as a candidate face recognition result, and extracting an identity recognition code corresponding to the candidate face recognition result;

step A6: respectively tracking the face recognition result in the 1 st frame of each image group in the non-key frame of each image group through a visual tracking algorithm, and storing the face tracking result of the nth frame of each image group and the identity recognition code corresponding to the face tracking result;

step A7: comparing the face tracking result of the nth frame of the previous image group with the face recognition result of the 1 st frame of the next image group from the first image group according to the playing sequence of the image groups in the video stream, and taking the face as the face recognition result of the next image group if the spatial position matching is consistent and the identification codes are consistent; and if the spatial position matching is inconsistent or the identity recognition codes are inconsistent, taking the face with higher similarity to the face characteristic value of the face to be recognized as the face recognition result of the next image group.

And 2.2, performing behavior feature extraction and abnormal classification and scoring.

The first mode is as follows:

the behavior feature extraction and the abnormal classification scoring, as shown in fig. 6, specifically include the following steps:

step B11: preprocessing video data acquired by a plurality of cameras arranged in a teaching area and a public activity area, grouping the preprocessed video streams into a group of 16 frames, and dividing a continuous monitoring video into a plurality of groups of small segments;

step B12: sampling 16 frames of images in each group of small segments according to the principle of 1 frame at intervals to obtain 8 sampled images, and sending the 8 sampled images into a 2D convolution network for prediction processing to obtain 8 2D characteristic graphs;

step B13: storing each 8 2D feature maps as a group of feature maps, randomly sampling the 5 groups of feature maps according to the proportion of 1:1:2:4:8 when the full 5 groups of feature maps are collected, randomly sampling 1 feature map in each of two groups of feature maps obtained firstly in the 5 groups of feature maps, randomly sampling 2 feature maps in a group of feature maps obtained thirdly, randomly sampling 4 feature maps in a group of feature maps obtained fourthly, and sampling all 8 feature maps in a group of feature maps obtained latest;

step B14: sending 16 feature maps obtained by sampling from the five groups of feature maps in the step B13 into a 3D convolution network for abnormal classification and scoring; when the abnormal classification is scored, extracting time domain sequence features and space domain sequence features from 16 feature maps sent into a 3D convolutional network, then simultaneously performing regularization processing on the time domain features and the space domain features, inputting a shared weight layer to extract time domain feature scores and space domain feature scores, then fusing the time domain feature scores and the space domain feature scores to obtain prediction space-time feature classification score vectors for predicting motion categories in a monitored video, and finally sequencing the generated prediction space-time feature classification score vectors from large to small, wherein the category index corresponding to the prediction space-time feature classification score vector with the largest value represents the motion category in the monitored video;

step B15: and discarding the five groups of sampled feature maps, and continuing to extract subsequent videos to perform the processing of the steps B11-B15.

The second mode is as follows:

the behavior feature extraction and the abnormal classification scoring are performed, target detection is performed based on a YOLO model, and as shown in FIG. 7, the method specifically comprises the following steps:

step B21: skipping frames of video data collected by a plurality of cameras arranged in a teaching area and a public activity area to extract a picture to be analyzed;

step B22: inputting the picture to be analyzed into a convolutional neural network, and extracting an abstract feature map through 5 initiation modules and 6 convolutional layers;

step B23: dividing the abstract feature diagram into a plurality of grids, performing regression calculation on feature data in each small grid, and extracting blob data of an object and blob data of a character from the extracted blob data of the abstract feature diagram based on a YOLO model;

step B24: extracting an object characteristic diagram and a behavior characteristic diagram from the abstract characteristic diagram obtained in the step B22 according to the blob data of the object and the blob data of the person;

step B25: respectively inputting the extracted object features and behavior features into a convergence layer of the space pyramid model, unifying the sizes of object feature graphs and behavior feature graphs with different sizes, and then performing feature fusion;

step B26: and calculating the matching degree of the feature graph after feature fusion and the behavior sample label by adopting a Softmax algorithm, and sequencing according to the sequence of the matching degree values from large to small, wherein the sample label with the maximum matching degree value represents the action category in the monitoring video.

Further, also based on the YOLO model, another identification method specially for the classroom behavior of the student is provided, as shown in fig. 8, specifically including the following steps:

step B31: acquiring and classifying images, and constructing a data set by the images; the method specifically comprises the steps of preprocessing video data collected by a plurality of cameras arranged in a teaching area and a public active area, dividing a continuous monitoring video into N video segments by taking 16 frames as a group, respectively extracting a frame of image from each video segment as an input image, and constructing a data set, wherein N is an integer greater than 1;

step B32: dividing a data set into a training set, a verification set and a test set according to a proportion, acquiring the position information of students in an image by using a Yolo v3 detection algorithm, cutting the images of the students by using Opencv according to the position information of the students, and uniformly zooming the images to finish the preprocessing of the images;

the method for acquiring the position information of the student in the image by using the Yolo v3 detection algorithm specifically comprises the following steps:

step B321: the images in the training set are transmitted to a Yolo v3 detection framework, the Yolo v3 detection framework divides the images into S-S grids, each grid is responsible for object detection in the grid area, and the target object category of the grid is output;

step B322: defining training labels, defining vectors for each grid

And then:

wherein the content of the first and second substances,

indicating whether the target object is contained;

representing a midpoint of the target object;

respectively representing the height and width of the bounding box;

sequentially indicating whether target objects of 1 st to nth classes in a training set are contained;

step B323: extracting features in each grid through forward operation in a convolutional neural network, identifying each grid through a positioning algorithm and a classification technology, and outputting the upper left corner coordinate of a target object image in each grid and the width and height of a cutting frame;

step B324: combining the S x S grid labels to obtain a target output;

step B33: expanding a training set in a data capacity expansion mode;

the data expansion mode comprises any one or more of affine change, turnover change, translation change, scale change, contrast change, noise disturbance, gray value setting to be zero, partial pixel value setting to be zero, median blurring, mean blurring and color change; but the data expansion mode is not limited to the above mode;

for example: obtaining more data by using the existing data space coordinate transformation relation;

the spatial coordinates are transformed as follows:

wherein the content of the first and second substances,

is the coordinates of the pixels in the original image;

is the coordinates of the pixels in the transformed image;

t is an affine transformation matrix;

step B34: training by using a ResNet50 network structure trained on an ImageNet data set as a pre-training model, finely adjusting the network according to a verification result of a verification set, and using a classification model obtained by training for later-stage student image behavior recognition;

when model training is carried out, an activation function is set firstly:

and calculating a loss function according to the activation function:

finally, calculating the partial derivative of the loss function to each parameter in the grid by using a back propagation algorithm, and updating all parameter values by using a gradient descent method;

step B35: and testing the test set by using the trained classification model, observing the difference between the image classification condition and the actual condition to obtain the classification effect of the model, and storing the classification result.

By analyzing the data of the external behavioral manifestations (behavior behaviors) of the persons in the school zone, the system mainly pays attention to whether the person R has behaviors of unintelligibly listening to or disturbing the classroom order in class, whether the person R has behavior of spamming or cheating in examination, whether the person R has behaviors of violence, cheating or being cheated by others indoors and outdoors, whether the person R has behaviors of self-speaking, self-complaining, self-disability and even delightful in a monitoring area, and the like. If there are behaviors such as long-term non-earnestly listening, classroom order disturbance, class absence, self-speaking and self-complaint and self-advice, or behaviors such as cheating, violence, deceiving others or being deceived, self-disabled and even revitalized by others, the corresponding item is labeled when analyzed according to the abnormal behavior management strategy, and finally the item is scored due to the abnormal behavior.

And thirdly, analyzing the data of the school footmarks of the personnel in the school zone.

The campus footprint analysis system is used for forming a track data group classified according to identification codes by calling position data, identification codes and other identification data in the video monitoring system and the campus one-card system, and therefore data in the track data group are analyzed. The video monitoring system collects video data through a plurality of cameras serving as sensing front ends; the campus card-through system collects all electronic data of campus cards through a campus card reading device serving as a sensing front end.

Specifically, the data analysis method for the campus footprint of the personnel in the campus area extracts the position information and the corresponding identification code of the campus card by acquiring the use condition of the campus card; on the other hand, the position information of the cameras and the identification codes corresponding to the face identification results are extracted through the face identification results extracted from the video data collected by the cameras arranged in the teaching area and the public activity area, and all the position information forms a footprint key data set according to the identification code group.

And analyzing the data of the school foot print of the personnel in the school district, mainly for capturing the abnormal action track of the personnel R. For example: when a person R is in the vicinity of a dangerous goods storage warehouse in a school zone during a night sleep period and stays for a long time, the person R will mark the dangerous goods storage warehouse when analyzing according to an abnormal behavior management strategy.

And fourthly, analyzing the data of the consumption condition of the personnel in the school zone.

Specifically, the data analysis method for the consumption condition of personnel in the school district only analyzes the consumption data of shopping and hospitalizing on a designated card reader for the campus card; the campus card is pre-stored with a unique identification code;

the consumption behavior analysis system calls the consumption situation data of the personnel in the data collation area in the campus card system for analysis, and as shown in fig. 9, the specific analysis method includes the following contents:

step C1: the card reader is accessed to a cash register system of the distributed server cluster, and consumption data of all campus cards are sent to a consumption analysis server group in the distributed server cluster;

step C2: analyzing and marking the consumption data from two aspects of the consumption condition of the control type articles and the consumption condition of the larger amount by the consumption analysis server group; wherein the content of the first and second substances,

when the consumption analysis server group analyzes and marks all consumption data from the consumption conditions of the control type articles, firstly, the consumption analysis server group extracts a consumption article list from the consumption data, compares the names of the articles in the consumption article list with the data in a prestored control type article list table, if the names of the articles in the consumption article list do not appear in the control type article list table, the consumption data are not marked, and if one or more article names in the consumption article list appear in the control type article list table, the consumption data are marked;

when the consumption analysis server group analyzes and marks the consumption data from the consumption condition of larger amount, firstly, the consumption analysis server group extracts the consumption amount in the consumption data and compares the consumption amount with a preset consumption limit strategy, if the consumption amount accords with the consumption limit strategy, the consumption data is not marked, and if the consumption amount does not accord with the consumption limit strategy, the consumption data is marked;

step C3: and forming a group of consumption behavior key data sets by the marked consumption data and the identification codes associated with the consumption data, and storing the consumption behavior key data sets in the consumption behavior key data sets.

And analyzing the data of the school foot prints of the personnel in the school district, mainly aiming at capturing the abnormal consumption behaviors of the personnel R. For example: the person R often makes medical treatment injured or often purchases simple medical supplies such as alcohol and gauze, or suddenly pays a large amount relative to the usual consumption level, and the person R may mark the expense when analyzing according to the consumption restriction policy.

And fifthly, analyzing data of the courseware situation and the collective activity participation situation of the personnel in the school zone.

Calling data in a data proofreading area in a student management platform by adopting a courseware and collective activity analysis system to analyze data of courseware conditions and collective activity participation conditions of personnel in the school zone; the student management platform prestores the identity identification codes of students and inputs course selection information, attendance information, score information, collective activity participation information and violation performance recording information corresponding to the identity identification codes of the students;

as shown in fig. 10, the data analysis method for the consumption situation of the personnel in the school zone specifically includes the following steps:

step D1: the activity server group in the distributed server cluster respectively calls course selection information, attendance information and collective activity participation information corresponding to the identification codes from the student management platform;

step D2: comparing the fetched course selection information with a pre-stored course selection management strategy, if the course selection information does not accord with the course selection management strategy, marking the course selection information, otherwise, not marking the course selection information;

comparing the taken attendance information with a prestored attendance management strategy, if the attendance information does not accord with the attendance management strategy, marking the attendance information, otherwise, not marking the attendance information;

comparing the called result information with a prestored result management strategy, if the result information does not accord with the result management strategy, marking the result information, otherwise, not marking the result information;

comparing the called illegal expression record information with a pre-stored daily expression management strategy, if the illegal expression record information does not accord with the daily expression management strategy, marking the illegal expression record information, otherwise, not marking the illegal expression record information;

comparing the called collective activity participation information with a collective activity management strategy, if the collective activity participation information does not accord with the collective activity management strategy, marking the collective activity participation information, otherwise, not marking;

step D3: and forming a group of key data sets with liveness by using the marked course selection information and/or attendance information and/or achievement information and/or collective activity participation information and/or violation performance recording information and the associated identification codes.

In summary, as shown in fig. 1, fig. 2, and fig. 3, the network management server, the camera, and the campus card reader are used as the perception front end of the perception layer of the adaptive cloud management platform to collect the appearance of the person in multiple dimensions, the data of corresponding categories are deeply analyzed according to the categories of 'network use condition', 'external behavior expression', 'school footprint', 'consumption condition', 'courseware condition' and 'collective activity participation' through a distributed message processing system, an external behavior analysis system, a school footprint analysis system, a consumption behavior analysis system and a courseware and collective activity analysis system, thereby obtaining a key data set used by the network, a key data set of the behavioral expression, a key data set of the footprint, a key data set of the consumption behavior and a key data set of the activity, the five types of data sets are classified into external expression behavior data groups of the same person according to the identity identification codes; and the external expression behavior data group is input into an abnormal psychological prediction model of the personnel in the school, data in a key data set used by the network, a key data set of behavior expression, a key data set of footprint, a key data set of consumption behavior and a key data set of activity are classified and analyzed according to an abnormal psychological attention strategy, a network use abnormal score, a behavior abnormal score, a footprint abnormal score, a consumption behavior abnormal score and an activity abnormal score are correspondingly obtained, and weights are assigned to the network use abnormal score, the behavior abnormal score, the footprint abnormal score, the consumption behavior abnormal score and the activity abnormal score according to the abnormal psychological attention strategy, so that the total score of the external abnormal behavior of the same personnel is calculated.

For example: network usage anomaly score

Behavioral behavior abnormality score

Abnormal footmark score

Consumption behavior abnormality score

Abnormal score of activeness

(ii) a Each item is assigned a weight of

(ii) a Therefore, the total Score of abnormal behavior Score of the uniform person R is calculated as follows:

；

in the usual case of the use of a magnetic tape,

the sum of (a) and (b) is 1. And for convenience of statistics, all scores are in a percentage system, so that the value of the total abnormal behavior Score of the final uniform personnel R is [0, 100 ]](ii) a Then, the school personnel abnormal psychology prediction model will grade the probability that there may be abnormal psychology for the personnel R according to the value of the total Score, for example: a stage (100-80), B stage (79-60), C stage (59-40), D stage (39-20), E stage (19-0). When there are a plurality of abnormalities in the index that is generally used to predict the abnormal psychology of the person R, the value of the total Score increases, and the corresponding abnormal psychology prediction level increases (highest class a, lowest class E). For a general management authority, the adaptive cloud management platform only displays the potential abnormal psychological prediction level of a certain person, and only a higher authority manager is likely to see the abnormal items of the person.

Note that, for the person X in example 2, the total abnormal behavior Score of the person X was calculated by the same method as described above. However, due to the limitation of the data acquisition mode of the front end, it is generally difficult to acquire data corresponding to consumption behaviors and activeness of the person X, and it is fully considered that the potential danger potential of the abnormal psychology of the person X as a non-student identity is relatively larger than that of the person R as a student identity, so for a certain key data set lacking data, the system defaults that the abnormal score of the key data set is full. For example: the intra-school personnel abnormal psychological prediction model only collects the key data set of network use, the key data set of behavior expression and the key data set of footprint of personnel X, but not collects the key data set of consumption behavior and the key data set of activeness of personnel X, and when percentage scoring is adopted, the consumption behavior abnormal score is defaulted

Is 100 points, activity degree abnormal score

100, and the network uses the anomaly score

Behavioral behavior abnormality score

Abnormal footmark score

And classifying and analyzing the key data sets used by the network, the key data sets of the behavioral performance and the key data sets of the footprint according to the abnormal psychological attention strategy, and then scoring.

Further, a scoring method for each key data set specifically includes comparing score points according to management strategies corresponding to each key data set, and if a certain data index collected in the key data set appears in the corresponding management strategy, scoring the data. For example: the key data set of the liveness can collect the characteristics of representing whether a certain course passes or not, and meanwhile, the score management strategy also has a corresponding item of 'course fails'. Generally, the value corresponding to the failing of a certain course is 1, and the value corresponding to the passing is 0; therefore, when the value of the feature of the liveness key data set for representing whether a certain course passes or not is 1, the item of 'course fails' is scored as t through the comparison of the result management strategy. When the number s of courses is not passed, the item "course is not passed" gets the score of s x t. The content of each specific management strategy is not an innovation point of the invention, and can be adjusted at any time according to the actual situation, but the main ideas for constructing each management strategy are consistent and are not repeated.

In this embodiment, the distributed message processing system, the external behavior analysis system, the campus footprint analysis system, the consumption behavior analysis system, and the courseware and collective activity analysis system are trained based on a deep learning technique, and the prediction accuracy is improved by continuously converging the prediction result through machine learning.

The rest of this embodiment is the same as embodiment 1 or 2, and therefore, the description thereof is omitted.

Example 4:

in this embodiment, optimization is performed on the basis of any one of embodiments 1 to 3, and the adaptive cloud management platform configures a master control center, a resource monitoring unit, a resource scheduling unit, a plurality of key data collection units, and a data disaster recovery service unit in a distributed server cluster;

the master control center collects the data processed by the plurality of key data collecting units, classifies the data according to a series of key data sets of the same identity identification code, performs abnormal shape psychological prediction analysis on the classified data through an abnormal psychological prediction model of the personnel in the school, stores all prediction analysis results in a psychological prediction database, and simultaneously marks, alarms and pushes the analysis results according to an abnormal psychological attention strategy; the resource monitoring unit receives a resource monitoring strategy input by the master control center, monitors the resources of each server in the self-adaptive cloud management platform in real time, and acquires the resource occupancy rate and the resource residual capacity of each server;

the resource scheduling unit calculates and calls corresponding resources according to the resource occupancy rate and the resource residual capacity of each server;

the resources comprise computing resources, storage resources and network resources of each server;

the key data collection unit receives data transmitted by the sensing front end, cleans, analyzes and stores the data, and calculates key data sets of all the characterization items in a classified manner;

the data disaster recovery service unit receives data acquired by the disaster recovery equipment information acquisition terminal; the data disaster recovery service unit comprises a disaster recovery user system server, an ISM server and a disaster recovery storage server;

the disaster recovery equipment information acquisition terminal: the data disaster recovery center room is respectively used for collecting the machine temperature, voltage, the working state of an electric fan, the environment temperature and humidity, the leakage conditions of water, acid and alkali liquid in the environment, the power consumption information, the power supply information, the case invasion information, the important system log information and the conditions of people entering and leaving the machine room;

the disaster recovery backup user system server can receive externally input disaster recovery service implementation parameters and receive a service state returned by the ISM server in real time;

the ISM server receives disaster recovery backup service implementation parameters input by the disaster recovery backup user system server, monitors acquisition parameters of a plurality of disaster recovery backup equipment information acquisition terminals, interactively accesses the disaster recovery backup storage server, monitors the change state of disaster recovery backup storage in real time, and records the disaster recovery backup service execution condition;

and the disaster recovery storage server is used for storing disaster recovery data information in real time.

The data disaster recovery service unit also provides a running state monitoring function of the physical host and the virtual server, and the virtualization management software monitors the running state and the resource consumption of the virtual machine and the physical host; the remote monitoring of the physical server and the virtual server comprises physical CPU, memory, storage, Network utilization rate and use condition monitoring and on-off state monitoring. When the management server is different from the virtual server and the physical server in the LAN, the monitoring of the running state of the server can be supported. The monitoring is to ensure accuracy and timeliness.

In another specific embodiment, the total control center of the adaptive cloud management platform adopts a virtualization architecture, and the structure of the virtualization architecture is shown in fig. 12. At this time, the server virtualization software is developed based on the mature KVM in the industry, provides rich virtualization capability and perfect management function, and realizes a perfect virtualization solution including a virtual machine and a management platform. The problem of resource integration is solved by using virtualization software, various resources such as calculation, storage, networks and the like are standardized in the integration process, and the resources are divided into smaller resource units which can be better scheduled, so that the capability of fully utilizing hardware resources in the scheduling process is achieved.

Furthermore, first, the system of the grandmaster center supports server virtualization. The virtual machine can realize all functions of the physical machine, such as having its own resources (CPU, memory, storage, network card), and can specify an individual IP address, MAC address, and the like. Each virtual machine may support virtual multi-path cpu (vmsmp) technology to meet the requirements of high-load application environments. The server virtualization supports the utilization of the NUMA characteristic of the CPU, so that the memory access performance of the virtual machine is optimal, and the service performance is improved; and the method supports automatic NUMA balance, optimizes system scheduling tasks and memory allocation algorithm, and obtains better performance. Secondly, the system supports virtualization of the mainstream Intel network card, and a user can add 8 network cards at most in a virtual machine. And the bandwidth limit of the network card in and out can be adjusted according to the needs of the customer, and the maximum value of the bandwidth is 1000 Mbps. The client can flexibly configure the virtual machine network card according to the requirement so as to meet different service scenes. The DPDK technique provided by Intel is used. Thirdly, the system supports connecting various storage devices such as iSCSI, NAS and FC interfaces as virtualized storage. And supports LVM, NFS, RAW. Compatible with the magnetic array and model of the mainstream brand on the market. The method can be used for deploying multiple sets of storage with flexibility, easy expansion and ultrahigh performance aiming at different resource pools.

At this time, a plurality of functions such as off-line adjustment of the CPU resources of the virtual machine, off-line adjustment of the memory resources of the virtual machine, off-line adjustment of the network card resources of the virtual machine, off-line adjustment of the hard disk resources of the virtual machine, on-line adjustment of the CPU resources of the virtual machine, on-line adjustment of the memory resources of the virtual machine, on-line adjustment of the network card resources of the virtual machine, and on-line adjustment of the hard disk resources of.

Other parts of this embodiment are the same as any of embodiments 1 to 3, and thus are not described again.

Example 5:

the embodiment is optimized on the basis of any one of embodiments 1 to 4, and some information security solutions are provided in the embodiment.

The security threat of the informatization of colleges and universities mainly comes from the following aspects:

(1) threats from cyber attacks can cause our servers or workstations to crash.

(2) Threats from information theft, causing secret leakage, illegal access of internal servers, destroying the integrity of transmitted information or being directly counterfeited.

(3) And computer virus threats from the public network cause the server or the workstation to be infected by the computer virus, so that the system is crashed or crashed, and even the network is crashed.

(4) Management and operating personnel lack safety knowledge. Because the information and network technology is developed rapidly and the application and the safety technology of the information are relatively lagged, when a user introduces and adopts the safety equipment and the system, the user lacks comprehensive and deep training and learning, the importance and the technical recognition of the information safety are insufficient, the safety equipment system is easy to be arranged, and the safety equipment system cannot play a correct role. If a limitation is originally required for some communication and operation, the network is set to a fully open state for convenience, and the like, so that a network vulnerability occurs.

(5) And lightning stroke. Because a network system relates to a plurality of network devices, terminals, lines and the like which are transmitted through communication cables, the network system is very easy to be struck by lightning to cause chain reaction, so that the whole network is paralyzed, the devices are damaged, and serious consequences are caused.

In daily office work of teaching staff, information transmission, resource sharing and the like need to be carried out in real time by each district and the on-business teaching staff, and mutual business is increasingly dependent on a network. However, due to the openness of the internet and the limitation of the original design of the communication protocol, all information is transmitted in plaintext, so that the security problem of the internet is increasingly serious, illegal access, network attack, information stealing and the like frequently occur, potential safety hazards are brought to normal operation of schools, and even inestimable loss is caused.

For physical security of a network system, main protection modes include firewall and physical isolation, risk analysis and vulnerability scanning, emergency response, virus prevention and control, access control, security audit, intrusion detection, source routing filtering, degradation use, data backup and the like.

In order to ensure the security of the platform and the data of each system, as shown in fig. 13, information security is ensured by adopting a hierarchical system interconnection mode.

Furthermore, a master control center of the self-adaptive cloud management platform operates by adopting a private network, and an attack protection system is arranged in the master control center to monitor user behaviors;

as shown in fig. 11, the method for monitoring the user behavior is as follows:

step P1: after receiving a user login request, an attack protection system establishes an ID for each user and collects the application type operated by the user and the user behavior of the operation mode;

step P2: the data is subjected to solution analysis, feature extraction and data integration processing through a data rule established by an attack protection system, a recording module generates the processed data into a user behavior information record, a regular expression matching module compares and matches the extracted user information features with a user behavior rule base, a matching result is evaluated and analyzed, and whether the user has abnormal violation behaviors or not is judged: if illegal operation exists, the attack protection system gives an early warning and protection to the user according to the strategy response and informs the service management personnel, so that the service management personnel can closely pay attention to the dynamic and recent related behaviors; if no illegal operation exists, the attack protection system generates a behavior audit log of the user and stores the behavior audit log in an audit database.

Other parts of this embodiment are the same as any of embodiments 1 to 4, and thus are not described again.

The above description is only a preferred embodiment of the present invention, and is not intended to limit the present invention in any way, and all simple modifications and equivalent variations of the above embodiments according to the technical spirit of the present invention are included in the scope of the present invention.

Claims

1. A campus personnel abnormal psychological prediction method based on a self-adaptive cloud management platform is characterized in that firstly, an intra-school personnel abnormal psychological prediction model is established, a unique identification code is matched or newly distributed to personnel entering a school zone range through the self-adaptive cloud management platform, face feature data are collected through a camera, and the identification code and the face feature data which correspond to each other one by one are stored in a face database as a face feature data group; then, acquiring a plurality of activity conditions of personnel in the school zone through a plurality of sensing front ends comprising a camera, a campus card reading device and a network management server; then, a distributed server cluster of a network layer performs data analysis on the collected multiple activity conditions and extracts a group of key data sets according to each identity identification code in a classified manner; finally, sending the multiple groups of key data sets to an abnormal psychological prediction model of the school staff for prediction analysis, storing all prediction analysis results in a psychological prediction database, and labeling and alarming abnormal prediction analysis results in the prediction analysis;

the activity condition of the personnel in the school district comprises any one or more of network use condition, external performance, school foot print, consumption condition, courseware condition and collective activity participation condition.

2. The campus personnel abnormal psychological prediction method based on the adaptive cloud management platform according to claim 1, characterized in that, the data analysis of the network use condition of personnel in the campus area is only performed on the network data which can be monitored by the network management server in the monitoring area; the network management server sends the collected video data to a distributed message processing system, the distributed message processing system analyzes the network use condition according to a network monitoring strategy, and the analysis results of the network use condition are grouped and integrated according to the identification codes; the grouped and integrated data is used as a key data set used by the network and is sent to an intra-school personnel abnormal psychological prediction model for prediction analysis;

the main data analysis tasks of the network use condition of personnel in the school zone according to the network monitoring strategy comprise: data statistics of internet surfing time, data statistics of internet surfing time intervals and prediction of interest and hobbies of webpage browsing;

the webpage browsing interest prediction means that the interest reflected by the user during webpage browsing is predicted according to a set period, and the prediction result is counted to generate a network information interest statistical table;

the data statistics of the internet surfing time, the data statistics of the internet surfing time period and the prediction of the web browsing interests and hobbies are carried out;

when the interest and hobbies reflected by a user in webpage browsing are predicted, the following specific steps are adopted:

step S1: acquiring a user log, dividing a plurality of user logs associated with the same identification code into a group, and extracting an IP address in each user log;

3. The campus personnel abnormal psychological prediction method based on the adaptive cloud management platform according to claim 1, characterized in that the data analysis method for the external behavior expression of personnel in the school zone is as follows: firstly, constructing a behavior analysis model capable of identifying the behavior of a person in a video and a face identification model capable of identifying the face characteristics in the video by using image space characteristics extracted by a convolutional neural network; then video data are collected through a plurality of cameras arranged in a teaching area and a public activity area, and the video server group in the distributed server cluster carries out face recognition and identity recognition code matching, behavior feature extraction, abnormal classification and classification scoring according to an abnormal behavior management strategy on the collected video data; secondly, grouping and integrating the results of face recognition and identity recognition code matching and the results of behavior feature extraction and abnormal classification scoring according to the identity recognition codes; and finally, the grouped and integrated data is used as a key data set of the behavior expression and is sent to an intra-school personnel abnormal psychological prediction model for prediction analysis.

4. The adaptive cloud management platform-based campus personnel abnormal psychological prediction method according to claim 3, wherein the face recognition and identification code matching specifically comprises the following steps:

5. The campus personnel abnormal psychological prediction method based on the adaptive cloud management platform according to claim 3, wherein the behavior feature extraction and abnormal classification scoring specifically comprises the following steps:

6. The campus personnel abnormal psychological prediction method based on the adaptive cloud management platform as claimed in claim 3, wherein the behavior feature extraction and abnormal classification are scored, and target detection is performed based on a YOLO model, specifically comprising the following steps:

7. The campus personnel abnormal psychological prediction method based on the adaptive cloud management platform as claimed in claim 1, wherein the data analysis method of the campus footprint of the personnel in the school zone extracts the position information and the corresponding identification code of the campus card by acquiring the use condition of the campus card; on the other hand, the position information of the cameras and the identification codes corresponding to the face identification results are extracted through the face identification results extracted from the video data collected by the cameras arranged in the teaching area and the public activity area, and all the position information forms a footprint key data set according to the identification code group.

8. The campus personnel abnormal psychological prediction method based on the adaptive cloud management platform as claimed in claim 1, wherein the data analysis method for the consumption condition of personnel in the school zone only analyzes the consumption data of shopping and hospitalizing on a designated card reader of a campus card; the campus card is pre-stored with a unique identification code;

the data analysis method for the consumption condition of the personnel in the school zone specifically comprises the following steps:

step C3: forming a group of consumption behavior key data sets by the marked consumption data and the identification codes related to the consumption data, and storing the consumption behavior key data sets in the consumption behavior key data sets;

the data analysis method for the class-work situation and the collective activity participation situation of the personnel in the school zone needs to call the data in the student management platform; the student management platform prestores the identity identification codes of students and inputs course selection information, attendance information, score information, collective activity participation information and violation performance recording information corresponding to the identity identification codes of the students;

comparing the called illegal expression record information with a pre-stored daily expression management strategy, if the illegal expression record information does not accord with the attendance management strategy, marking the illegal expression record information, otherwise, not marking the illegal expression record information;

9. The campus personnel anomaly psychology prediction method based on the adaptive cloud management platform according to any one of claims 1 to 8, wherein the adaptive cloud management platform configures a master control center, a resource monitoring unit, a resource scheduling unit, a plurality of key data collection units, and a data disaster recovery service unit in a distributed server cluster;

10. The campus personnel abnormity psychology prediction method based on the adaptive cloud management platform according to claim 9, characterized in that a master control center of the adaptive cloud management platform operates by adopting a private network, and an attack protection system is arranged in the master control center to monitor user behaviors;

the method for monitoring the user behavior comprises the following steps: