CN116611022B

CN116611022B - Intelligent campus education big data fusion method and platform

Info

Publication number: CN116611022B
Application number: CN202310444301.2A
Authority: CN
Inventors: 马朝辉; 何凯英; 郭婷
Original assignee: Shenzhen Lexing Smart Industry Co ltd
Current assignee: Shenzhen Lexing Smart Industry Co ltd
Priority date: 2023-04-21
Filing date: 2023-04-21
Publication date: 2024-04-26
Anticipated expiration: 2043-04-21
Also published as: CN116611022A

Abstract

The invention relates to the technical field of education data prediction processing, in particular to a method and a platform for intelligent campus education big data fusion. The method comprises the following steps: acquiring student educational activity data; data standardization is carried out according to the student education activity data, so that standard real-time data is obtained; fusion construction is carried out according to standard real-time data, so that a student behavior model is constructed; performing behavior anomaly prediction according to the student behavior model so as to obtain anomaly behavior data; and carrying out intelligent event tracing evaluation according to the abnormal behavior data so as to obtain detailed information of the abnormal behavior for tracking the abnormal events of the intelligent campus education. According to the invention, the student education activity data is obtained, standardized and fused, the student behavior model is quickly constructed, the multi-source data is integrated to realize comprehensive analysis, and the efficiency of student education management is improved.

Description

Intelligent campus education big data fusion method and platform

Technical Field

The invention relates to the technical field of education data prediction processing, in particular to a method and a platform for intelligent campus education big data fusion.

Background

The intelligent campus is an intelligent campus work, study and life integrated environment based on the Internet of things, the integrated environment uses various application service systems as carriers to fully integrate teaching, scientific research, management and campus life, and the blueprint is depicted as follows: the method has the advantages of ubiquitous network learning, integration of innovative network scientific research, transparency, high efficiency school administration, colorful campus culture, convenience for the life of the surrounding campuses, and in short, the method needs to be used as a safe, stable, environment-friendly and energy-saving campus. Artificial intelligence (ARTIFICIAL INTELLIGENCE, AI for short) refers to techniques and applications that allow robots to have mental intelligence like robots. Artificial intelligence includes a number of fields that utilize large amounts of data and algorithms to help computers simulate human mental activities and achieve autonomous decisions and actions. The future artificial intelligence has very broad prospects. With the continuous development of technology and the continuous expansion of application scenes, the artificial intelligence is widely applied in various fields. How to combine artificial intelligence with smart campus educational data processing becomes a problem.

Disclosure of Invention

The invention provides a method and a platform for fusing big data of intelligent campus education to solve at least one technical problem.

The application provides a method for fusing big data of intelligent campus education, which comprises the following steps:

Step S1: acquiring student educational activity data;

Step S2: data standardization is carried out according to the student education activity data, so that standard real-time data is obtained;

step S3: fusion construction is carried out according to standard real-time data, so that a student behavior model is constructed;

step S4: performing behavior anomaly prediction according to the student behavior model so as to obtain anomaly behavior data;

step S5: and carrying out intelligent event tracing evaluation according to the abnormal behavior data so as to obtain detailed information of the abnormal behavior for tracking the abnormal events of the intelligent campus education.

According to the method, student education activity data are obtained, standardized and fused, a student behavior model is quickly built, comprehensive analysis is realized by integrating multi-source data, efficiency of student education management is improved, abnormal behavior can be timely found out and evaluation tracing is conducted by utilizing the student behavior model to conduct behavior abnormal prediction, accordingly bad behavior is prevented and restrained, the student behavior is carefully analyzed through analysis and excavation of intelligent campus education big data, accurate education decision advice is provided, education strategies are improved, abnormal behavior can be timely found out and detailed information tracking is conducted through intelligent event tracing evaluation, and data safety is further enhanced.

In one embodiment of the present disclosure, the standard real-time data includes first standard real-time data and second standard real-time data, and step S2 specifically includes:

step S21: data cleansing is carried out on student education activity data, so that cleansing activity data are generated;

step S22: performing distributed calculation according to the cleaning activity data, thereby obtaining distributed activity data;

Step S23: judging whether the distribution activity data is larger than or equal to preset distribution threshold data or not;

Step S24: when the distribution activity data is determined to be greater than or equal to the preset distribution threshold data, performing first standardized calculation according to the cleaning activity data, so as to generate first standard real-time data;

Step S25: and when the distribution activity data is determined to be smaller than the preset distribution threshold data, performing second standardized calculation according to the cleaning activity data, so as to generate second standard real-time data, wherein the first standardized calculation is different from the data standardized calculation adopted by the second standardized calculation.

According to the method, the standardized method of the data can be determined more accurately through cleaning and distributed computing of the student educational activity data, so that the accuracy of data standardization is improved, different standardized computing methods can be selected according to actual conditions through judging the distributed threshold data, unnecessary data processing processes are avoided, the data processing flow is optimized, the construction of a student behavior model and abnormal behavior prediction can be performed more rapidly through generating standard real-time data, and the efficiency of data analysis is improved.

In one embodiment of the present specification, step S3 is specifically:

Step S31: performing data preprocessing according to the standard real-time data, thereby obtaining preprocessed data;

Step S32: extracting time sequence activity characteristics and time sequence position characteristics according to the preprocessing data, thereby acquiring the time sequence activity characteristics and the time sequence position characteristics;

step S33: identifying through a student activity risk identification model according to the time sequence activity characteristics, so as to obtain student activity risk information;

Step S34: performing depth association coupling according to the time sequence position characteristics and the time sequence activity characteristics so as to obtain real-time activity information;

Step S35: marking the real-time activity information by using the student activity risk information, so as to obtain risk level real-time activity data;

Step S36: and carrying out fusion construction according to the risk level real-time activity data, thereby constructing a student behavior model.

According to the embodiment, through data preprocessing, time sequence activity feature extraction and application of the student activity risk identification model on standard real-time data, the activity risk of a student can be identified more accurately, so that abnormal behaviors of the student can be found early, real-time activity information can be obtained through deep association coupling of time sequence position features and time sequence activity features, the real-time activity information is marked by utilizing the student activity risk information, risk level real-time activity data are obtained, real-time student behavior information is provided, and through fusion construction by utilizing the risk level real-time activity data, a more accurate student behavior model can be constructed, and more accurate basis is provided for subsequent education management and decision.

In one embodiment of the present specification, step S32 is specifically:

Step S321: performing time sequence conversion according to the preprocessing data to obtain time sequence activity record data;

step S322: extracting time sequence activity characteristics according to the time sequence activity record data, thereby obtaining time sequence activity characteristics;

Step S323: and performing shot image position correction according to the position information in the time sequence activity record data, thereby obtaining time sequence position characteristics.

According to the embodiment, through time sequence conversion of the preprocessed data and time sequence feature extraction and position information correction of the time sequence activity record data, more accurate and complete time sequence activity features and position features can be obtained, so that the accuracy and efficiency of subsequent data analysis and decision generation are improved, student activity data are better analyzed, potential rules and trends are found, the learning effect and teaching quality of students are improved, the position information of the students can be more accurately positioned and recorded through shot image position correction, and the accuracy of the position information is improved.

In one embodiment of the present specification, the step of capturing an image position correction is specifically:

step S301: the corresponding cameras are controlled to perform image acquisition operation through student position information in the time sequence activity record data, so that student image information is obtained;

step S302: carrying out student image recognition according to the student image information so as to obtain student image recognition information;

Step S303: when the student image identification information is determined to contain true student image identification information, extracting time sequence position features according to time sequence activity record data, so as to obtain time sequence position features;

Step S304: when the student image identification information is determined to contain false student image identification information, controlling a camera in the campus to acquire images and conduct student image identification, iterating until the student image identification information is determined to contain true student image identification information, conducting relative position calibration on time sequence activity record data by utilizing current student position information and conducting time sequence position feature extraction, and therefore obtaining time sequence position features.

According to the embodiment, the real student position information is determined according to the student image identification information, and the relative position calibration is carried out on the time sequence activity record data, so that the position information of the students can be reflected more accurately, the accuracy of the position information is improved, the detailed information such as the position and the movement track of the student activity can be obtained through time sequence position feature extraction, further detailed student behavior information is provided for subsequent education management and decision, the position calibration and feature extraction are carried out only when the student image identification information contains the real information in consideration of student privacy protection, excessive position tracking and monitoring are avoided, and the student privacy protection is improved.

In one embodiment of the present specification, the step of constructing the student activity risk recognition model in step S33 specifically includes:

Step S331: acquiring standard time sequence activity data, wherein the standard time sequence activity data comprises standard activity time data, standard activity place data and standard activity type data, and the standard time sequence activity data comprises legal time sequence activity data and abnormal time sequence activity data;

Step S332: extracting time sequence activity characteristics according to the standard time sequence activity data, so as to generate standard time sequence activity characteristics;

Step S333: optimizing denoising according to the standard time sequence activity characteristic, so as to obtain a denoising time sequence activity characteristic;

step S334: performing dimension reduction normalization according to the denoising time sequence activity feature so as to obtain a normalized time sequence activity feature;

step S335: dividing according to the normalized time sequence activity characteristics so as to generate training time sequence activity characteristics and testing time sequence activity characteristics;

Step S336: selecting the condition characteristic according to the training time sequence activity characteristic, thereby obtaining the condition time sequence activity characteristic;

Step S337: generating an optimal characteristic node according to the conditional time sequence activity characteristics, so as to construct a time sequence activity characteristic decision tree model;

step S338: pruning is carried out after optimization according to the time sequence activity characteristic decision tree model, so that an optimized time sequence activity characteristic decision tree model is obtained;

Step S339: and performing error iteration on the optimized time sequence activity feature decision tree model according to the test time sequence activity features, thereby obtaining a student activity risk identification model.

The embodiment can more accurately identify the activity risk of students by constructing a student activity risk identification model, thereby finding the abnormal behaviors of the students early, providing more refined student behavior characteristics, processing the original data through the steps of standard time sequence activity data, time sequence activity characteristic extraction, denoising and normalization, reducing random fluctuation in the data and interference caused by abnormal shops on the model, providing more refined student behavior characteristics, providing more detailed basic data for subsequent student risk identification and education management, adaptively constructing a decision tree model through the steps of condition characteristic selection, optimal characteristic node generation and the like, improving the accuracy and reliability of the model, wherein the method can identify the most relevant characteristics of the student activity risk by adopting condition characteristic selection, reducing characteristic redundancy, the model complexity is reduced, the model accuracy is improved, the campus education student data identification and tracking have more practical significance, compared with other deep learning algorithms, the method has good interpretation, the relation between the feature and the target variable is intuitively represented by the existing data generation tree structure, other deep learning algorithms, such as a neural network algorithm, usually a black box model, are difficult to interpret the working principle, in the practical training process, the training speed of the optimization generation decision tree model method adopted by the method is usually faster, the optimization generation decision tree model method is suitable for data information with more determined time-space features, the data is recursively segmented and the optimal feature is selected according to the greedy algorithm, the calculation resources and time for hardware and software are reduced, the other deep learning algorithms are adopted, a large amount of calculation resources and time are often needed to optimize a complex network model, for specific intelligent campus construction, the data volume of a data set is often only thousands to tens of thousands, the optimization generation decision tree model method adopted by the method is good in performance on the data volume, other deep learning algorithms, such as a neural network algorithm, often need a large amount of training data to achieve good performance, the method can better process the condition of missing data, other deep learning algorithms can need additional preprocessing steps when processing missing data, and on the aspect of processing student education condition tracking in the intelligent campus, the optimization generation decision tree algorithm method adopted by the method can react more efficiently and has higher processing speed than other deep learning algorithms on the aspect of the problem of high complexity and nonlinear characteristics.

In one embodiment of the present disclosure, step S34 is specifically:

Performing time sequence series connection according to the time sequence position characteristics and the time sequence activity characteristics, so as to obtain characteristic time series connection data;

and performing feature selection according to the feature time series data, so as to obtain real-time activity information.

According to the embodiment, the time sequence position features and the time sequence activity features are subjected to time sequence series connection and feature selection, so that real-time activity information can be obtained more quickly, and the efficiency of data analysis is improved. Through serial connection of time sequence position characteristics and time sequence activity characteristics, more detailed student behavior information can be obtained, including information of multiple dimensions such as time, position, frequency, duration of activity and the like, more detailed student behavior information is provided for subsequent education management and decision.

In one embodiment of the present specification, step S36 is specifically:

Step S361: carrying out data division according to the risk level real-time activity data so as to obtain training risk level real-time activity data and testing risk level real-time activity data;

Step S362: selecting minimum cost conditions according to the training risk level real-time activity data, so as to obtain the condition risk level real-time activity data;

Step S363: constructing a decision tree according to the real-time activity data of the conditional risk level, thereby constructing a real-time activity decision tree model of the risk level;

Step S364: and carrying out iterative model evaluation on the risk level real-time activity decision tree model according to the test risk level real-time activity data, thereby determining a student behavior model.

According to the embodiment, the minimum cost condition selection and the decision tree construction are carried out aiming at the training set data, so that the risk level of the student behavior can be determined more accurately, the accuracy of risk level judgment is improved, a real-time active decision tree model of the risk level is constructed, a more detailed student behavior model comprising the activity behavior of the student, the risk level and related characteristic information thereof can be provided, a calculation basis is provided for identifying the depth data relationship in the student educational data, the decision tree model is subjected to iterative evaluation by utilizing test data, the student behavior model can be optimized and perfected continuously, and the efficiency and the accuracy of student behavior management are improved.

In one embodiment of the present disclosure, the iterative model evaluation uses a historical corrected student activity evaluation calculation formula for model evaluation, where the historical corrected student activity evaluation calculation formula is specifically:

G is a student behavior model evaluation index, a is an initial term, alpha is a weight coefficient of student behavior model accuracy, delta is student behavior model accuracy, beta is a weight coefficient correction term, h is student behavior model history accuracy, t is student behavior model total accuracy, gamma is a weight coefficient of student behavior risk coefficient, w is student behavior risk coefficient, t _p is a real example number, f _p is a false negative example number, and u is a correction term of student behavior model evaluation index.

The embodiment provides a historical correction student activity evaluation calculation formula which fully considers the initial term a, the weight coefficient alpha of the student behavior model accuracy, the student behavior model accuracy delta, the weight coefficient correction term beta, the student behavior model historical accuracy, the student behavior model total accuracy, the weight coefficient of the student behavior risk coefficient, the real number of cases, the false number of cases and the interaction relation among each other to form a functional relationAccording to the method, the device and the system, the evaluation index G of the student behavior model can be obtained through the formula, the accuracy of the student behavior model can be evaluated, the student behavior model can be optimized through adjusting weight coefficients of different parameters, the basis for adjusting the student behavior model is provided, an initial item a represents an evaluation value of the student behavior model at the beginning, for a newly built model, the weight coefficient alpha of the student behavior model accuracy is 0, the influence degree of the accuracy on the evaluation index is measured, the student behavior model accuracy delta reflects the accuracy of the student behavior model, the weight coefficient correction item beta is used for considering the influence degree of the historical accuracy and the total accuracy on the evaluation index, the student behavior model historical accuracy h represents the accuracy in a period of time, the long-term stability of the model is measured, the total accuracy t of the student behavior model represents the accuracy of the whole model, the overall performance of the model is reflected, the weight coefficient gamma of the student behavior risk coefficient is used for measuring the influence degree of the evaluation index, the student behavior risk coefficient w reflects the risk degree of the behavior is reflected, and the student behavior model is corrected through the correction item of the student behavior model to realize the iterative correction of the student behavior model.

The application provides a large data fusion platform for intelligent campus education, which comprises the following steps:

At least one processor; and

A memory communicatively coupled to the at least one processor; wherein,

The memory stores a computer program executable by the at least one processor to enable the at least one processor to perform a smart campus education big data fusion method as described above.

The method has the advantages that the method can monitor and manage student behaviors more comprehensively through the steps of acquiring student educational activity data, standardizing the data, constructing a student behavior model and the like, improves management efficiency and accuracy, provides an abnormal behavior early warning and tracking function, can timely discover and early warn the abnormal behaviors of students through behavior abnormal prediction and intelligent event traceability evaluation, provides detailed abnormal behavior condition information, facilitates intervention and tracking of education managers, enhances the capability of individual education services of the students, can deeply understand the behavior characteristics and requirements of the students through construction of the student behavior model, provides more individual education services, promotes comprehensive development of the students, improves scientificity of education decisions, and can obtain more objective, comprehensive and accurate education data through collection and analysis of a large amount of data to help the education managers to make more scientific and effective education decisions.

Drawings

Other features, objects and advantages of the application will become more apparent upon reading of the detailed description of a non-limiting implementation, made with reference to the accompanying drawings in which:

FIG. 1 is a flow chart showing the steps of a smart campus education big data fusion method according to an embodiment;

FIG. 2 is a flow chart illustrating the steps of a standard real-time data acquisition method according to one embodiment;

FIG. 3 is a flow chart illustrating steps of a student behavior model construction method of an embodiment;

FIG. 4 is a flow chart illustrating steps of a method for extracting timing activity features and timing location features of an embodiment;

FIG. 5 is a flowchart showing the steps of a captured image position correction operation according to an embodiment;

FIG. 6 is a flow chart showing the steps of a student activity risk identification model building method of an embodiment;

FIG. 7 is a flow chart illustrating steps of a student behavior model generation method according to one embodiment.

Detailed Description

The following is a clear and complete description of the technical method of the present patent in conjunction with the accompanying drawings, and it is evident that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, are intended to fall within the scope of the present invention.

Furthermore, the drawings are merely schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus a repetitive description thereof will be omitted. Some of the block diagrams shown in the figures are functional entities and do not necessarily correspond to physically or logically separate entities. The functional entities may be implemented in software or in one or more hardware modules or integrated circuits or in different networks and/or processor methods and/or microcontroller methods.

It will be understood that, although the terms "first," "second," etc. may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another element. For example, a first element could be termed a second element, and, similarly, a second element could be termed a first element, without departing from the scope of example embodiments. The term "and/or" as used herein includes any and all combinations of one or more of the associated listed items.

Referring to fig. 1 to 7, the application provides a smart campus education big data fusion method, which comprises the following steps:

Step S1: acquiring student educational activity data;

Specifically, educational data of the classroom performance, the job completion, the examination performance of the students, and personal information of the students such as age, sex, and personal data of the home background are collected, for example, by the educational management system.

Specifically, for example, data from different sources and different types is standardized in format, such as data from different educational systems or platforms are converted into the same format or structure, various types of data are classified and managed uniformly according to standard codes, such as behavior data of students are classified and coded according to a certain coding rule, quality of the data is standardized, such as data acquisition time is standardized by formulating a data entry specification and a checking mechanism, and time stamps of different data sources are converted into a standard time format uniformly.

Specifically, for example, the standard real-time data is trained and modeled by using a machine learning method to predict the behavior of the student, for example, the data such as the student performance, the homework submitting situation and the like are trained by using an algorithm such as a Support Vector Machine (SVM) and the like, so as to obtain a learning model of the student.

Specifically, for example, by mining association rules among student behavior data to determine whether student behaviors are abnormal, if a plurality of students are found to have some abnormal behavior at the same time, it can be inferred that the behavior is an abnormal behavior, and further tracking and processing are performed.

Specifically, for example, by using a data mining method, relevance analysis is performed on abnormal behavior data to find out the relationship and root cause between abnormal events, such as relevance analysis is performed on student education data of student's achievements, attendance situations and work completion situations, and to find out the cause of possible abnormal behavior.

Specifically, for example, duplicate removal processing is performed on data that may have duplicate, such as that multiple examination results of the same student are kept for only one piece of latest data, processing is performed on data that has missing values, such as that using interpolation, average filling, and the like, the missing values are complemented, processing is performed on data that has abnormal values, such as that according to business rules or statistical methods, whether the data is abnormal or not is judged, and removal or replacement is performed when necessary, and formatting processing is performed on data that is not standard in data format, such as that data types such as date and time, numbers, and the like are formatted into standard form.

Specifically, complex relationship networks are computed and analyzed, for example, using a distributed graph processing framework. For example, social networks between students are analyzed and modeled.

Specifically, for example, in this process, a distribution threshold value data is preset, and when the examination score is greater than or equal to the value, a first standardized calculation is required to be performed on the data; when the examination score is smaller than the value, second standardized calculation is needed to be carried out on the data, the examination score data of 200 students are obtained under the assumption that the distribution threshold value of a certain examination is 80 minutes, and cleaning processing is carried out on the examination score data, wherein the cleaning processing comprises duplicate removal, missing value processing, abnormal value processing and data formatting, and cleaning activity data is obtained after cleaning. Performing distributed calculation on the cleaning activity data to obtain distributed activity data, for example, sequencing and counting the data through a MapReduce framework to obtain the distribution condition of student achievements; according to the judgment result, if the distribution activity data is greater than or equal to 80 minutes, a first standardized calculation is required to be performed on the data. For example, the examination results are normalized by z-score to obtain first standard real-time data. If the distributed activity data is less than 80 minutes, a second normalization calculation is required for the data. For example, the examination score is normalized in min-max to obtain second standard real-time data.

And analyzing and processing the generated standard real-time data, for example, displaying the variation trend, distribution condition and the like of the student examination results through the visualization tool, and timely discovering and processing abnormal conditions.

In one embodiment of the present specification, step S3 is specifically:

specifically, for example, a high-dimensional data is subjected to a dimension reduction process, for example, in an image process, a dimension reduction process is performed using a method such as Principal Component Analysis (PCA).

Specifically, for example, for time series data, time series activity features such as average value, variance, kurtosis are extracted therein. For example, in student homework submission situation analysis, the time series activity characteristics of the average, variance, maximum value of the number of homework submissions per day of the student can be extracted. And extracting time sequence position characteristics, such as time intervals, accumulated time and the like, of the time sequence data according to the position of the time sequence data on the time axis. For example, in student attendance analysis, time sequence position features such as the number of days of continuous attendance, the number of times of absence and leave, and the like of students can be extracted.

Specifically, for example, the time series data is used to predict future academic performance of the student. This may be done by monitoring the student's activities, such as course access records, job submission times, online test scores, etc. in the learning management system, as well as other relevant information such as social media activities, sleep records, etc. to determine the student's learning behavior pattern. Machine learning algorithms can then be used to build a predictive model that will predict students' future academic performance based on their historical behavior and other factors, and identify risky students by comparing the predicted results to actual results.

In particular, the time-lapse positioning features and time-lapse activity features are coupled, for example, by weighting based on student educational data, such as using a weighted coupling method to analyze the student's behavior and performance on an online learning platform. For example, interaction data of students with an online learning platform, such as access times, learning time, job submission conditions, etc., can be collected, and the student behavior can be modeled in combination with location information, usage of input devices (such as a mouse, a keyboard), and other related data, and then a weighted coupling method can be used to combine time-series location features with time-series activity features, and assign a different importance weight to each feature.

In particular, the real-time activity information of the student is noted, for example, using the risk level information. For example, on an online learning platform, students 'course access, job submission times, etc. may be monitored in real-time and these activities may be categorized into different risk levels based on historical data and education institutions' specifications. These levels may reflect the learning status, interests and needs of the students.

In particular, behavioral models are built, for example, using student risk level real-time activity data, to help educational institutions better understand students and provide personalized support and guidance. For example, on an online learning platform, historical behavioral data and risk level real-time activity data of a student may be collected and combined to construct a behavioral model of the student. In this model, each student is assigned a set of features including their location, input device usage, number of visits, learning time, job submission, etc. Machine learning algorithms can then be used to analyze these features and extract therefrom several features that are most important to student learning outcome. Next, it is contemplated that students may be separated into different groups, such as excellent, normal, problematic, risk-free, high risk-level, and personalized support and instruction may be provided to each student based on the characteristics and behavioral models of these groups.

In one embodiment of the present specification, step S32 is specifically:

specifically, for example, on an online learning platform, data such as access logs of students, job submission records, test scores, and the like may be collected and converted into time series data such as a time series or an event series.

Specifically, for example, using a time series analysis method such as ARIMA (autoregressive integrated moving average model) or LSTM (long short term memory model), model parameters are extracted from time series data as features, or basic statistical features including statistics of maximum, minimum, mean, median, variance, standard deviation, kurtosis, skewness, and the like.

Specifically, for example, shot image position correction is performed using position information in the time-series activity recording data to obtain a time-series position feature. For example, in educational institutions, cameras or other location-aware devices may be installed to collect student location information in a classroom and combine that information with time-series recorded data. Then, the captured image may be subjected to positional correction using image processing techniques such as computer vision and deep learning.

specifically, for example, student position information in the time-series activity record data is used to control the corresponding cameras to perform image acquisition operation so as to obtain student image information. For example, in an educational institution, multiple cameras may be installed in a classroom and used to capture images of students in different areas.

specifically, student image recognition is performed using, for example, student image information to obtain student image recognition information. For example, in educational institutions, information such as facial expressions, body gestures, and gestures of students may be captured by cameras and analyzed and identified by techniques such as computer vision and deep learning.

Specifically, for example, when it is determined that the student image identification information is to contain the true student image identification information, time-series position feature extraction may be performed according to the position information of the student in the teaching room to obtain the time-series position feature.

Specifically, for example, when the student image identification information is determined to include false student image identification information, a camera in the campus may be controlled to perform image acquisition and student image identification. And after the student image identification information is iteratively confirmed to contain the real student image identification information, performing relative position calibration on the time sequence activity record data by utilizing the current student position information, and performing time sequence position feature extraction to obtain time sequence position features. A plurality of cameras are installed in a campus, and information such as identity, facial expression, body posture and the like of students is identified through computer vision technology and a deep learning algorithm. If it is determined that the student image identification information is the student image identification information containing the false, the cameras are required to be controlled to perform image acquisition and student image identification until it is determined that the student image identification information is the student image identification information containing the true.

specifically, for example, a time series activity feature is extracted. These characteristics may include information in terms of time, place, type, etc., such as: time characteristics: including lesson time, rest time between lessons, lunch time, school time, etc. Location feature: including classrooms, laboratories, libraries, playgrounds, etc. Type characteristics: including listening and speaking, discussion, experimentation, self-study, etc.

Specifically, for example, a suitable denoising method, such as a denoising algorithm based on total variation, is selected according to spatial and frequency domain features of an image, such as local variance, gradient, wavelet coefficient, and the like.

Specifically, for example, for certain time series data with periodicity, fourier transform or wavelet transform may be used to extract its frequency domain features and reduce it to fewer dimensions by PCA or other dimension reduction algorithm. The resulting dimension-reduced features are then normalized, e.g. scaling all feature values to between 0 and 1.

In particular, for example, in a time series classification task, all time series data may be divided into a training set and a test set. For each time series sample, it may be divided into fixed length sub-sequences using a sliding window technique. The normalized features of each sub-sequence are then taken as input features and assigned to a training set or test set.

In particular, for example in a time series classification task, a conditional feature selection method may be used to determine which time series activity features are best able to distinguish between different categories, such as feature selection algorithms based on information gain or information gain ratio, e.g. the C4.5 algorithm. The algorithm divides the training set sample into different nodes, calculates the information entropy of each node, selects the characteristic with the maximum information gain or the information gain ratio as the basis of node splitting, and adds the characteristic into the decision tree.

Specifically, using algorithms such as C4.5 or ID3 to construct a time series activity feature decision tree model requires extracting a set of conditional time series activity features for each time series sample and training the model using these features.

specifically, rule-based pruning methods are applied, for example, to reduce model complexity, such as removing leaf nodes with only a small number of samples or merging similar leaf nodes at a particular node, such as merging or deleting the gender of students as a teaching progress of non-physical lessons follows.

Specifically, for example, new data in the test set is input into the model and errors are calculated from the model output, such as cross-checking or recall, the errors can be used to adjust the model parameters and the model trained again, repeating the process until the error reaches a desired level.

In one embodiment of the present disclosure, step S34 is specifically:

specifically, for example, the time series data is divided by using a sliding window or a time slice, and then the time series position feature and the time series activity feature in each time period are extracted and connected in series according to the time sequence to form a time series feature.

Specifically, the features of each time step are evaluated, for example, using a statistical-based feature selection method, such as analysis of variance (ANOVA) or chi-square test, and the most discriminant features are selected.

In one embodiment of the present specification, step S36 is specifically:

specifically, for example, all data is randomly divided into a training set and a test set in a certain proportion, for example, 80% of data is divided into the training set and 20% is divided into the test set.

Specifically, for example, using cost-sensitive learning methods, different cost weights are applied to the misclassified samples for different error types (e.g., false positives and false negatives), and this weighted loss function is optimized during classifier training.

Specifically, a decision tree model is built on conditional risk level real-time activity data, e.g. based on ID3, C4.5 or CART algorithm.

In particular, the model is evaluated and optimized, for example, using cross-validation or recall calculation methods.

At least one processor; and

A memory communicatively coupled to the at least one processor; wherein,

The present embodiments are, therefore, to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.

The foregoing is only a specific embodiment of the invention to enable those skilled in the art to understand or practice the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. The intelligent campus education big data fusion method is characterized by comprising the following steps of:

Step S1: acquiring student educational activity data;

Step S3: fusion construction is carried out according to standard real-time data, so that a student behavior model is constructed; the step S3 specifically comprises the following steps:

step S33: the student activity risk identification model is used for identifying according to the time sequence activity characteristics, so that student activity risk information is obtained, and the construction steps of the student activity risk identification model in the step S33 are specifically as follows:

Acquiring standard time sequence activity data, wherein the standard time sequence activity data comprises standard activity time data, standard activity place data and standard activity type data, and the standard time sequence activity data comprises legal time sequence activity data and abnormal time sequence activity data;

Extracting time sequence activity characteristics according to the standard time sequence activity data, so as to generate standard time sequence activity characteristics;

optimizing denoising according to the standard time sequence activity characteristic, so as to obtain a denoising time sequence activity characteristic;

Performing dimension reduction normalization according to the denoising time sequence activity feature so as to obtain a normalized time sequence activity feature;

Dividing according to the normalized time sequence activity characteristics so as to generate training time sequence activity characteristics and testing time sequence activity characteristics;

Selecting the condition characteristic according to the training time sequence activity characteristic, thereby obtaining the condition time sequence activity characteristic;

generating an optimal characteristic node according to the conditional time sequence activity characteristics, so as to construct a time sequence activity characteristic decision tree model;

Pruning is carried out after optimization according to the time sequence activity characteristic decision tree model, so that an optimized time sequence activity characteristic decision tree model is obtained;

Performing error iteration on the optimized time sequence activity feature decision tree model according to the test time sequence activity features, so as to obtain a student activity risk identification model;

Step S36: fusing and constructing the real-time activity data according to the risk level, so as to construct a student behavior model;

2. The method according to claim 1, wherein the standard real-time data comprises a first standard real-time data and a second standard real-time data, and step S2 is specifically:

data cleansing is carried out on student education activity data, so that cleansing activity data are generated;

Performing distributed calculation according to the cleaning activity data, thereby obtaining distributed activity data;

judging whether the distribution activity data is larger than or equal to preset distribution threshold data or not;

When the distribution activity data is determined to be greater than or equal to the preset distribution threshold data, performing first standardized calculation according to the cleaning activity data, so as to generate first standard real-time data;

And when the distribution activity data is determined to be smaller than the preset distribution threshold data, performing second standardized calculation according to the cleaning activity data, so as to generate second standard real-time data, wherein the first standardized calculation is different from the data standardized calculation adopted by the second standardized calculation.

3. The method according to claim 1, wherein step S32 is specifically:

Performing time sequence conversion according to the preprocessing data to obtain time sequence activity record data;

Extracting time sequence activity characteristics according to the time sequence activity record data, thereby obtaining time sequence activity characteristics;

and performing shot image position correction according to the position information in the time sequence activity record data, thereby obtaining time sequence position characteristics.

4. A method according to claim 3, wherein the step of correcting the position of the captured image is in particular:

the corresponding cameras are controlled to perform image acquisition operation through student position information in the time sequence activity record data, so that student image information is obtained;

Carrying out student image recognition according to the student image information so as to obtain student image recognition information;

when the student image identification information is determined to contain true student image identification information, extracting time sequence position features according to time sequence activity record data, so as to obtain time sequence position features;

When the student image identification information is determined to contain false student image identification information, controlling a camera in the campus to acquire images and conduct student image identification, iterating until the student image identification information is determined to contain true student image identification information, conducting relative position calibration on time sequence activity record data by utilizing current student position information and conducting time sequence position feature extraction, and therefore obtaining time sequence position features.

5. The method according to claim 1, wherein step S34 is specifically:

6. The method according to claim 1, wherein step S36 is specifically:

Carrying out data division according to the risk level real-time activity data so as to obtain training risk level real-time activity data and testing risk level real-time activity data;

Selecting minimum cost conditions according to the training risk level real-time activity data, so as to obtain the condition risk level real-time activity data;

constructing a decision tree according to the real-time activity data of the conditional risk level, thereby constructing a real-time activity decision tree model of the risk level;

And carrying out iterative model evaluation on the risk level real-time activity decision tree model according to the test risk level real-time activity data, thereby determining a student behavior model.

7. The method of claim 6, wherein the iterative model evaluation uses a historical corrected student activity evaluation calculation formula for model evaluation, wherein the historical corrected student activity evaluation calculation formula is specifically:

Assessment of index for student behavioral model,/> Is the initial item,/>Weight coefficient for student behavior model accuracy rate,/>For student behavior model accuracy rate,/>For the weight coefficient correction term,/>For student behavior model history accuracy rate,/>For the total accuracy of student behavior model,/>Is the weight coefficient of student behavior risk coefficient,/>For student behavior risk coefficient,/>For the real example number,/>Is the number of false negative cases,/>And evaluating the correction term of the index for the student behavior model.

8. An intelligent campus education big data fusion platform, which is characterized by comprising:

At least one processor; and

A memory communicatively coupled to the at least one processor; wherein,

The memory stores a computer program executable by the at least one processor to enable the at least one processor to perform a smart campus education big data fusion method as claimed in any one of claims 1 to 7.