Intelligent course arrangement method and system based on multi-objective optimization and deep learning

Abstract

The invention discloses an intelligent course arrangement method based on multi-objective optimization and deep learning, and relates to the technical field of intelligent course arrangement, and the method comprises the steps: collecting multi-source data, and constructing a spatial-temporal feature matrix; mining a historical class schedule conflict mode by using a knowledge graph and constructing an association relationship model; newly-added course constraints are converted into vectors, and a comprehensive constraint model is formed through BERT model analysis. A graph neural network is utilized to capture association of courses, teachers, classrooms and time periods, a comprehensive constraint model is combined to generate candidate schemes meeting hard constraints such as teacher time conflicts and classroom capacity and partial soft constraints such as pre-repair relations and student course selection conflicts, and an optimal course arrangement scheme is obtained through optimization. The problems that an existing system is insufficient in multi-source data application, difficult in association construction, inaccurate in constraint analysis, difficult in complex contact capture, insufficient in multi-target consideration and the like are solved.

Description

Technical Field

[0001] This application relates to the technical field of intelligent course scheduling, and particularly to an intelligent course scheduling method and system based on multi-objective optimization and deep learning. Background Art

[0002] Traditional course scheduling methods often rely on manual experience and it is difficult to efficiently integrate and utilize massive and diverse information. With the expansion of the scale of education, the increase in the types of courses, the growth in the number of teachers and students, the allocation of classroom resources has become increasingly complex. Existing course scheduling systems have frequent problems in mining the correlation relationships among courses, teachers, and classrooms, handling new course constraints, capturing the complex connections among them, and taking into account multi-objective optimizations such as teacher burden and classroom utilization rate, and it is difficult to meet the requirements of modern education for the rationality and feasibility of course scheduling. Summary of the Invention

[0003] The purpose of this application is to provide an intelligent course scheduling method and system based on multi-objective optimization and deep learning, so as to solve problems such as insufficient utilization of multi-source data in existing systems, difficult construction of associations, inaccurate parsing of constraints, difficult capture of complex connections, and insufficient consideration of multiple objectives.

[0004] In view of the above technical problems, this application provides an intelligent course scheduling method and system based on multi-objective optimization and deep learning.

[0005] In the first aspect of the embodiments of this application, an intelligent course scheduling method based on multi-objective optimization and deep learning is provided. The method includes:

[0006] Collect historical course scheduling data, teacher and student preferences, classroom resources, and course attributes, and construct a spatio-temporal feature matrix. The classroom resources include capacity and equipment data, and the course attributes include course type, duration, and number of students;

[0007] Mine conflict patterns in the historical course schedule, and use a knowledge graph to construct an association relationship model of course - teacher - classroom;

[0008] Based on the association relationship model, convert the constraints of the new course into vector form, and parse the constraint information described in natural language through a pre-trained BERT model to obtain a comprehensive constraint model;

[0009] Use a graph neural network to capture the complex associations among courses, teachers, classrooms, and time periods, and at the same time use the comprehensive constraint model to generate candidate course scheduling plans. The candidate course scheduling plans satisfy hard constraints and soft constraints. The hard constraints include teacher time conflicts and classroom capacity, and the soft constraints include prerequisite relationships and student course selection conflicts. The prerequisite relationship is a prerequisite course that a student must complete before taking a subsequent course;

[0010] Based on the candidate class scheduling plan, further optimize multiple objectives and screen out the Pareto optimal solutions that meet all constraint conditions to obtain the optimal class scheduling plan. The optimization of multiple objectives includes balanced teacher burden, classroom utilization rate, and reduction of students' cross-campus movement.

[0011] In the second aspect of the embodiments of the present application, an intelligent class scheduling system based on multi-objective optimization and deep learning is provided. The system includes:

[0012] A data collection module, which collects historical class scheduling data, teacher and student preferences, classroom resources, and course attributes, and constructs a spatio-temporal feature matrix. The classroom resources include capacity and equipment data, and the course attributes include course type, duration, and number of students.

[0013] An association relationship model construction module, which mines conflict patterns from historical class schedules and constructs an association relationship model of courses - teachers - classrooms using a knowledge graph.

[0014] A comprehensive constraint model construction module, which, based on the association relationship model, converts the constraints of new courses into vector forms, and parses the constraint information described in natural language through a pre-trained BERT model to obtain a comprehensive constraint model.

[0015] A candidate class scheduling plan generation module, which uses a graph neural network to capture the complex associations between courses, teachers, classrooms, and time periods, and at the same time uses the comprehensive constraint model to generate candidate class scheduling plans that meet hard constraints and soft constraints. The hard constraints include teacher time conflicts and classroom capacity, and the soft constraints include prerequisite relationships and student course selection conflicts. The prerequisite relationship is the prerequisite course that students must complete before taking subsequent courses.

[0016] An optimal class scheduling plan obtaining module, which, based on the candidate class scheduling plan, further optimizes multiple objectives and screens out the Pareto optimal solutions that meet all constraint conditions to obtain the optimal class scheduling plan. The optimization of multiple objectives includes balanced teacher burden, classroom utilization rate, and reduction of students' cross-campus movement.

[0017] One or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0018] Collect historical course scheduling data, teacher and student preferences, classroom resources, and course attributes to construct a spatio-temporal feature matrix. The classroom resources include capacity and equipment data, and the course attributes include course type, duration, and number of students. Mine conflict patterns from historical course schedules and use a knowledge graph to construct an association relationship model of courses - teachers - classrooms. Based on the association relationship model, convert the constraints of new courses into vector form, and parse the constraint information described in natural language through a pre-trained BERT model to obtain a comprehensive constraint model. Use a graph neural network to capture the complex associations between courses, teachers, classrooms, and time periods, and at the same time use the comprehensive constraint model to generate candidate course scheduling plans that meet hard constraints and soft constraints. The hard constraints include teacher time conflicts and classroom capacity, and the soft constraints include prerequisite relationships and student course selection conflicts. The prerequisite relationship is the prerequisite course that students must complete before taking subsequent courses. Based on the candidate course scheduling plan, further optimize multiple objectives and screen out the Pareto optimal solutions that meet all constraint conditions to obtain the optimal course scheduling plan. The optimization of multiple objectives includes balanced teacher workload, classroom utilization rate, and reduction of student cross-campus movement. It solves the problems of insufficient utilization of multi-source data, difficult association construction, inaccurate constraint parsing, difficult capture of complex connections, and insufficient consideration of multiple objectives in existing systems.

[0019] The above description is only an overview of the technical solution of this application. In order to be able to more clearly clarify the technical means of this application, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features, and advantages of this application more obvious and understandable, the specific embodiments of this application are specifically given below. Brief Description of the Drawings

[0020] To more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings of the embodiments of the present disclosure will be briefly introduced below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of this application. It should be understood that the operations in the front or below do not necessarily need to be executed precisely in sequence. On the contrary, according to needs, they can be executed in reverse order or simultaneously. At the same time, other operations can also be added to these processes, or one or several operations can be removed from these processes.

[0021] Figure 1 It is a schematic flowchart of an intelligent course scheduling method based on multi-objective optimization and deep learning provided by an embodiment of this application;

[0022] Figure 2 It is a schematic structural diagram of an intelligent course scheduling system based on multi-objective optimization and deep learning provided by an embodiment of this application.

[0023] Description of the drawing reference numerals: data acquisition module 10, correlation relationship model construction module 20, comprehensive constraint model construction module 30, candidate class scheduling plan generation module 40, optimal class scheduling plan obtaining module 50. Specific implementation manners

[0024] The present application provides an intelligent class scheduling method and system based on multi-objective optimization and deep learning, and solves the problems of insufficient utilization of multi-source data, difficult correlation construction, inaccurate constraint parsing, difficult capture of complex relationships, and insufficient consideration of multiple objectives in the existing system.

[0025] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts shall fall within the protection scope of the present application.

[0026] It should be noted that the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server including a series of steps or units is not necessarily limited to those clearly listed steps or units, but may include other steps or modules not clearly listed or inherent to these processes, methods, products or devices.

[0027] Embodiment 1, as Figure 1 shown, the present application provides an intelligent class scheduling method based on multi-objective optimization and deep learning, wherein the method includes:

[0028] Collect historical class scheduling data, teacher and student preferences, classroom resources, and course attributes, and construct a spatio-temporal feature matrix. The classroom resources include capacity and equipment data, and the course attributes include course type, duration, and number of students;

[0029] Specifically, to build the spatio-temporal feature matrix required for intelligent course scheduling, it is necessary to comprehensively collect multi-source key data. First, for historical course scheduling data, export the course scheduling details of multiple past semesters or academic years from the school's educational administration management system. These data cover various subject courses, and specifically record their specific time arrangements for each class session from Monday to Friday, as well as their allocations to different numbered classrooms. At the same time, the relevant information of the teaching teachers is clarified. Second, when collecting teachers' preferences, use a specially designed questionnaire to accurately ask about the teaching time periods that teachers expect, such as morning, afternoon, or evening. At the same time, encourage teachers to elaborate on their needs for specific classroom equipment, such as the requirements for projectors in multimedia courses and various experimental equipment required for experimental courses. Third, to understand students' preferences, an online class survey can be adopted, focusing on mastering which time periods the student group prefers to centrally arrange different course types, such as theory courses and practical courses, in order to achieve efficient knowledge learning and centralized skill training. Fourth, when collecting classroom resource information, conduct on-site inspections of each classroom, specifically record its maximum capacity, and comprehensively register the equipment in the classroom, such as the projectors and their specifications, the types, quantities, and applicable course ranges of various experimental equipment. Fifth, for course attribute and priority data, obtain them from the course syllabus and teaching plan documents provided by the school's teaching management department, and clarify whether the course belongs to a theory course, an experimental course, or a practical course. And based on the key points of the teaching plan and credit settings, determine the priority of each course in the teaching system. Sixth, after obtaining the above multi-source data, integrate them into the spatio-temporal feature matrix. It can intuitively show the allocation possibilities of each course at different times and in different classrooms, providing a structured data form for using data mining algorithms to explore course scheduling rules and running course scheduling algorithms to generate scientific and reasonable course scheduling plans.

[0030] Further construction of the spatio-temporal feature matrix includes:

[0031] Define C as the set of courses to be scheduled, C = {C1, C2,..., C n}, where n is the total number of courses;

[0032] Define T as the set of each time period, T = {T1, T2,..., T m}, where m is the total number of time periods;

[0033] Define R as the set of available classrooms, R = {R1, R2,..., R k}, where k is the total number of classrooms;

[0034] The dimension of the spatio-temporal feature matrix X is:

[0035] X = C × T × R;

[0036] Among them, C is the set of courses to be scheduled, T is the set of each time period, and R is the set of available classrooms;

[0037] The size of the spatio-temporal feature matrix X is n×m×k, that is, a combination of n courses, m time periods, and k classrooms.

[0038] Mine the conflict patterns from the historical class schedules and use a knowledge graph to construct an association relationship model of courses - teachers - classrooms;

[0039] Specifically, by analyzing the historical class scheduling data, discover the rules of class scheduling conflicts, provide a basis for future optimization of class scheduling, and through knowledge graph technology, establish associations among elements such as courses, teachers, and classrooms to support intelligent class scheduling analysis.

[0040] Furthermore, mine the conflict patterns from the historical class schedules and use a knowledge graph to construct an association relationship model of courses, teachers, and classrooms, including:

[0041] Extract relevant class scheduling information from the historical class schedules. By analyzing the historical class scheduling data, identify possible conflict patterns among different courses, teachers, and classrooms. The conflict patterns include time conflicts, teacher conflicts, classroom conflicts, and student conflicts;

[0042] Use a frequent pattern mining algorithm to discover the common conflict patterns in the historical class schedules, record the conflict patterns and conflict types, and use the conflict patterns and conflict types as conflict constraint rules to form a conflict pattern library and obtain a conflict pattern graph;

[0043] Based on the historical class scheduling data, construct a knowledge graph of courses, teachers, and classrooms, with courses, teachers, and classrooms as the three main entities in the knowledge graph;

[0044] Establish the associations between courses and teachers, courses and classrooms, and teachers and classrooms to obtain an entity relationship model.

[0045] Specifically, in the process of constructing an intelligent course scheduling system, the first step is to comprehensively extract relevant course scheduling information from historical course schedules. By deeply analyzing these historical course scheduling data, accurately identify possible conflict patterns at different levels of courses, teachers, classrooms, and students. To further explore potential common conflict patterns in the course schedule, a frequent pattern mining algorithm is adopted. This algorithm records these conflict patterns and their corresponding conflict types, and organizes them into conflict constraint rules, and then constructs a conflict pattern library, which is finally visually presented in the form of a graph to form a conflict pattern graph. At the same time, a knowledge graph is constructed based on historical course scheduling data, with courses, teachers, and classrooms set as the three main entities of the knowledge graph. On this basis, the teaching association between courses and teachers, the arrangement association between courses and classrooms, and the indirect association between teachers and classrooms formed through courses are established, so as to obtain an entity relationship model that comprehensively reflects the relationship among the three, providing a rich and structured data basis for subsequent course scheduling decisions and helping to generate a more scientific and reasonable course scheduling plan.

[0046] Furthermore, using the frequent item mining algorithm to discover common conflict patterns in the course schedule, recording the conflict patterns and conflict types, taking the conflict patterns and conflict types as conflict constraint rules, forming a conflict pattern library, and obtaining a conflict pattern graph, further includes:

[0047] Adopt the frequent item mining algorithm, determine the lower limit of support according to historical course scheduling data, and filter out low-frequency conflict patterns;

[0048] Analyze the combinations of courses, teachers, and classrooms that often appear together within the same time period to form frequent item sets;

[0049] Analyze each frequent item set to identify its corresponding conflict type. The conflict types include teacher time conflict, classroom resource conflict, and student course selection conflict. The teacher time conflict means that the same teacher arranges multiple courses at the same time. The classroom resource conflict means that within the same time period, multiple courses compete for the same classroom. The student course selection conflict means that some students simultaneously select courses with overlapping time;

[0050] Summarize the conflict patterns and their conflict types to form a preliminary set of conflict constraint rules;

[0051] Organize and store all conflict patterns, corresponding conflict types, and constraint rules to construct a conflict pattern library;

[0052] Utilize the conflict pattern library to construct a conflict pattern graph by taking each conflict pattern as the attributes of nodes and edges in the graph. The nodes represent courses, teachers, and classrooms, and the edges represent conflict relationships.

[0053] Specifically, when optimizing the course scheduling system, a frequent pattern mining algorithm is used to sort out the common conflict patterns in the course schedule. This algorithm first conducts a comprehensive analysis of the historical course scheduling data, screening out those occasional and rarely occurring conflict patterns, and only retaining the conflict situations that have repeatedly occurred in past course scheduling, making the subsequent analysis more valuable. Then, the algorithm focuses on analyzing the combinations of courses, teachers, and classrooms that frequently co-occur within the same time period. These combinations often hide conflicts. For each frequently occurring item set, the algorithm will carefully identify its corresponding conflict types, including teacher time conflicts (where the same teacher is scheduled for multiple teaching tasks at the same time, making it impossible for the teacher to attend to all), classroom resource conflicts (where multiple courses compete for the same classroom within the same time period, resulting in classroom usage contradictions), and student course selection conflicts (where the course times selected by some students overlap, causing students to be unable to participate simultaneously). After identifying all conflict patterns and their conflict types, this information is summarized to form a preliminary set of conflict constraint rules. Subsequently, the mined conflict patterns are transformed into practical constraint rules, and all conflict patterns, corresponding conflict types, and constraint rules are sorted out and stored to build a conflict pattern library. Finally, with the help of the conflict pattern library, using courses, teachers, and classrooms as nodes and conflict relationships as edges, and taking each conflict pattern as an attribute of the nodes and edges, a conflict pattern graph is constructed. This graph can visually present various conflict relationships in course scheduling, providing a clear reference for course schedulers and helping to formulate a more reasonable course scheduling plan.

[0054] Furthermore, an intelligent course scheduling method based on multi-objective optimization and deep learning is characterized in that, based on the historical course scheduling data, a knowledge graph of courses, teachers, and classrooms is constructed, taking courses, teachers, and classrooms as the three main entities in the knowledge graph, and further including:

[0055] Obtain historical course scheduling data, where the historical course scheduling data includes: course information, teacher information, and classroom information. The course information includes course name, course number, course type, class hours, prerequisite relationship, and the number of students selecting the course. The teacher information includes teacher name, teacher number, major, courses that can be taught, available time, and affiliated department. The classroom information includes classroom number, capacity, equipment type, and affiliated building;

[0056] Define the main entities and their attributes in the knowledge graph, where the entities include courses, teachers, and classrooms;

[0057] According to the historical course scheduling data and domain knowledge, determine the association relationships between entities and construct triples. The association relationships include the relationship between courses and teachers, the relationship between courses and classrooms, the indirect association between teachers and classrooms, the course prerequisite relationship, and the teacher available time constraint;

[0058] Extract entities such as courses, teachers, and classrooms and their attributes using the preprocessed data;

[0059] Extract the relationships between entities based on the course scheduling records and domain knowledge to form a set of triples.

[0060] Import all entities and relationships into a graph database to construct a knowledge graph.

[0061] Use data rules to verify the correctness of entities and relationships in the knowledge graph.

[0062] Specifically, in the intelligent course scheduling method system based on multi-objective optimization and deep learning, first, a large amount of comprehensive historical course scheduling data needs to be obtained from multiple channels such as the school's educational administration system and teaching archives. In terms of course information, the course name is the intuitive identifier of the course, such as "Advanced Mathematics"; the course number is the unique code of the course for precise system identification; the course type is divided into theoretical courses, experimental courses, practical courses, etc., and different types of courses have different requirements for teaching environments and time arrangements; the class hours specify the teaching duration of the course, for example, "Advanced Mathematics" has 64 class hours; the prerequisite relationship reflects the logical order between courses, like learning "Advanced Mathematics" is a prerequisite for learning "Probability Theory and Mathematical Statistics"; the number of students selecting a course reflects the elective situation of students for the course, which affects the selection of classroom capacity. In teacher information, the teacher's name is convenient for identifying individuals, and the teacher number is the unique identifier of the teacher in the system; the major indicates the teacher's subject area, such as the physics major; the courses that can be taught list the list of courses that the teacher is capable of teaching; the available time details the time periods when the teacher can teach, for example, Monday morning, Wednesday afternoon, etc.; the affiliated department clarifies the teacher's affiliated department. In classroom information, the classroom number is the unique identifier of each classroom, like "Classroom 302 in Teaching Building 1"; the capacity determines the number of students that the classroom can accommodate; the equipment type covers projectors, experimental equipment, etc., which affects the suitability of the course; the affiliated building determines the geographical location of the classroom. Next, define the main entities of the knowledge graph as courses, teachers, and classrooms, and precisely define their respective attributes. In addition to the above information, course attributes may also include credits, course difficulty, etc.; teacher attributes may also involve professional titles, years of teaching experience, etc.; classroom attributes may also include whether it has multimedia equipment, whether it is a special function classroom, etc. Then, deeply study the historical course scheduling data and combine the professional knowledge in the education field to determine the complex and diverse association relationships between entities. There is a clear teaching relationship between courses and teachers, that is, teachers are responsible for teaching specific courses; the relationship between courses and classrooms is an arrangement relationship, and courses need to be arranged in suitable classrooms for teaching; teachers and classrooms are indirectly associated through courses because teachers are related to the classrooms where they teach through the courses they teach; the prerequisite relationship of courses clarifies the sequence of course learning; the teacher's available time constraint limits that teachers can only teach within a specific time period. Construct these relationships as triples, such as (Teacher A, teaches, Course X), (Course X, is arranged in, Classroom 101), etc.

[0063] Using data mining techniques, entities such as courses, teachers, and classrooms and their attributes are extracted from the processed data. At the same time, based on detailed course scheduling records and established rules in the education field, such as that course schedules cannot conflict in time and teachers cannot teach different classes at the same time, the relationships between various entities are extracted, and then a set of triples is formed. Finally, all the extracted entities and relationships are imported into the graph database. During the import process, according to the characteristics of the graph database and the structural requirements of the knowledge graph, the data is reasonably stored and organized, thus successfully constructing the knowledge graph. After the construction is completed, using preset data rules, such as checking whether the prerequisite relationship of courses is logical and whether the teaching arrangements of teachers are within their available time, etc., the correctness of the entities and relationships in the knowledge graph is comprehensively verified to ensure that the knowledge graph can accurately and reliably reflect the real relationships among courses, teachers, and classrooms, providing a solid and accurate data support for subsequent intelligent course scheduling with the help of multi-objective optimization and deep learning techniques.

[0064] Based on the association relationship model, the constraints of the new course are transformed into vector form, and the pre-trained BERT model is used to parse the constraint information described in natural language to obtain a comprehensive constraint model;

[0065] Specifically, based on the previously constructed association relationship model of courses, teachers, and classrooms, when facing a new course, its relevant constraint conditions need to be transformed. Constraints such as class time requirements, classroom equipment requirements, and teaching teacher qualification requirements for the new course are transformed into vector forms that can be understood and processed by a computer. Then, with the help of the pre-trained BERT model, which is good at understanding and analyzing natural language, the constraint information of the new course described in natural language is parsed. Through operations such as semantic understanding and feature extraction of natural language, the BERT model combines this information with the association relationship model, and finally obtains a comprehensive constraint model that comprehensively considers various constraints and association relationships, providing a comprehensive and accurate constraint basis for subsequent intelligent course scheduling.

[0066] Furthermore, based on the association relationship model, transforming the constraints of the new course into vector form and using the pre-trained BERT model to parse the constraint information described in natural language to obtain a comprehensive constraint model also includes:

[0067] Vectorize the attributes of courses, teachers, and classrooms and the association relationships between them, and the vectorization includes course vectorization, teacher vectorization, and classroom vectorization;

[0068] The course vectorization means that each course can be represented as a vector through course attributes, and the course attributes include course type, duration, and number of students;

[0069] The teacher is vectorized, and each teacher is represented as a vector through their teaching ability and available time period attributes;

[0070] The classroom is vectorized, and the attributes of each classroom can be transformed into a vector. The attributes of each classroom include classroom capacity, equipment requirements, and time period availability;

[0071] Obtain the constraint information of the natural language description of the new course, and use the pre-trained BERT model to parse the constraint information to obtain the context representation of each word;

[0072] Combine the natural language constraint information parsed by BERT with the existing constraint conditions in the course scheduling system to form a comprehensive constraint model.

[0073] Specifically, in the intelligent course scheduling system, based on the established association relationship model of courses, teachers, and classrooms, a series of key steps are carried out when a new course is added. First, the attributes of courses, teachers, and classrooms and their association relationships need to be transformed into vector forms, which includes course vectorization, teacher vectorization, and classroom vectorization. Course vectorization means representing each course as a vector according to its course attributes. These course attributes cover the course type, such as whether it is a theory course, an experiment course, or a practical course; the duration, which clarifies the total teaching duration of the course; and the number of students, that is, the number of students who choose this course. Through the quantitative combination of these attributes, each course has a unique representation in the vector space. Teacher vectorization is to represent each teacher as a vector through their teaching ability, such as which subject areas they are good at teaching courses in, and the available time period attributes. This enables the key information of teachers in the course scheduling system to be presented in vector form, facilitating subsequent operations and matching. Classroom vectorization is to transform the attributes of each classroom into a vector. Classroom attributes include classroom capacity, that is, the maximum number of students that the classroom can accommodate; equipment requirements, such as whether special equipment like projectors and experimental equipment needs to be equipped; and time period availability, indicating which time periods in a week the classroom is available. Then, obtain the constraint information described in natural language for the new course, such as "This course needs to be taught in a large classroom equipped with a multimedia projector and is preferably scheduled on Tuesday or Thursday afternoons". At this time, use the BERT model pre-trained with a large amount of data to parse this constraint information. The BERT model can deeply understand the semantics of natural language. Through complex neural network operations, it obtains the context representation of each word in the context of the entire sentence, thereby accurately grasping the connotation of the constraint information. Finally, combine the natural language constraint information parsed by the BERT model with the existing constraint conditions in the course scheduling system, such as teacher time conflict restrictions and classroom usage rules, and comprehensively consider various factors, and then form a comprehensive and perfect comprehensive constraint model. This model integrates the specific requirements of the new course and the original constraint rules of the system, providing an accurate and comprehensive constraint basis for the subsequent intelligent course scheduling algorithm, and helping to generate a more scientific and reasonable course scheduling plan.

[0074] Use a graph neural network to capture the complex associations among courses, teachers, classrooms, and time periods, and at the same time use the comprehensive constraint model to generate candidate course scheduling plans. The candidate course scheduling plans satisfy hard constraints and soft constraints. The hard constraints include teacher time conflicts and classroom capacity, and the soft constraints include prerequisite relationships and student course selection conflicts. The prerequisite relationship is the prerequisite course that a student must complete before taking a subsequent course.

[0075] Furthermore, using a graph neural network to capture the complex associations among courses, teachers, classrooms, and time periods, and at the same time using the comprehensive constraint model to generate candidate course scheduling plans, includes:

[0076] Taking courses, teachers, and classrooms as nodes, establish edges between the nodes;

[0077] Construct a feature vector for each node;

[0078] Adopt a GNN model, aggregate information through the adjacency matrix, learn the high-dimensional representation of nodes, and capture local and global correlation relationships;

[0079] Utilize the class schedule after GNN learning, combine the time period and classroom features, calculate the allocation probabilities of courses at different times and in different classrooms through a multi-layer perceptron, and perform normalization to obtain the probability matrix of the candidate class scheduling scheme, and then generate multiple preliminary class scheduling schemes through Top-K sampling.

[0080] Specifically, in the intelligent course scheduling system, our goal is to formulate a reasonable course arrangement plan. To achieve this goal, two important tools and methods are mainly adopted, namely graph neural network and comprehensive constraint model. First, we regard courses, teachers, and classrooms as individual "points" (nodes). Then, "lines" (i.e., edges) are established based on their actual connections. For example, if a teacher is to teach a certain course, a line is connected between the node representing this teacher and the node representing this course; if a certain course is scheduled to be taught in a certain classroom, the node representing this course and the node representing this classroom are connected. In this way, we construct a structure similar to a network. Next, we need to assign some feature information to each "point" (node) to turn them into feature vectors. For the course node, we will include information such as the type of the course (e.g., theoretical course or practical course), the duration of the course (how many hours in total), the number of students (how many students choose this course), etc. For the teacher node, there will be information such as the teaching ability of the teacher (e.g., which subjects the teacher is good at teaching), the available time periods (when the teacher can teach), etc. And the classroom node includes information such as the capacity of the classroom (how many students it can accommodate), equipment requirements (whether there are projectors, experimental equipment, etc.), the availability of time periods (when this classroom can be used), etc. After obtaining these more comprehensive representations of the courses through the graph neural network (GNN), combined with the information of time periods (such as Monday morning, Tuesday afternoon, etc.) and the feature information of the classrooms, they are all input into a multi-layer perceptron. This multi-layer perceptron will perform calculations to calculate the possibility (i.e., probability) of arranging this course at different times and in different classrooms. Then, these probabilities are normalized to keep them within a reasonable range. In this way, we obtain a probability matrix of candidate course scheduling plans, which contains the likelihood of various courses being arranged at different times and in different classrooms. Finally, through the method of Top-K sampling, several combinations with higher possibilities are selected from this probability matrix to generate multiple preliminary course scheduling plans. And these preliminary course scheduling plans need to meet certain conditions. Among them, hard constraints must be strictly adhered to, such as there should be no time conflict where a teacher has to teach two courses at the same time, and the capacity of the classroom should be able to accommodate the students who choose this course. While soft constraints, although not absolutely cannot be violated, should be satisfied as much as possible, such as the prerequisite relationship between courses (one course must be learned before another course can be learned), and reducing the course selection conflict of students (students cannot choose two courses that are taught at the same time).

[0081] For the candidate course scheduling plans, further optimize multiple objectives and screen out the Pareto optimal solutions that meet all constraint conditions to obtain the optimal course scheduling plan.

[0082] Furthermore, it also includes:

[0083] Use a graph neural network to capture the correlations among courses, teachers, and classrooms, and combine with a comprehensive constraint model to generate multiple candidate course scheduling plans that meet the basic constraint conditions;

[0084] Take the candidate course scheduling plans as the initial population and input them into a multi-objective optimization algorithm to obtain an updated population;

[0085] Define multiple objective functions for each plan, and at the same time embed the comprehensive constraints as evaluation indicators. The multiple objective functions include teacher burden, classroom utilization rate, and student cross-campus moving distance. The comprehensive constraints include hard constraints and soft constraints;

[0086] Adopt an improved NSGA-III multi-objective optimization method combined with a comprehensive constraint model and a candidate plan generation and constraint correction mechanism to obtain a set of Pareto optimal candidate course scheduling plans that meet the requirements of hard constraints and soft constraints during the crossover and mutation processes.

[0087] Specifically, in the intelligent course scheduling process, first, use a graph neural network to deeply capture the complex correlations among courses, teachers, and classrooms, and at the same time combine with a comprehensive constraint model to generate multiple candidate course scheduling plans that meet basic constraint conditions such as teacher time conflict restrictions and classroom capacity adaptation. Then, based on these candidate plans, further optimization work is carried out to balance multiple important objectives. For teacher burden balance, ensure that the teaching tasks of each teacher are reasonably distributed, avoiding some teachers having too heavy teaching loads and some having too light. In terms of classroom utilization rate, strive to make the classrooms be fully and reasonably used in each time period, reducing idle waste. Reducing student cross-campus movement is from the perspective of students, reducing the situation of running around between different campuses due to course arrangements, saving students' time and energy. Use an improved NSGA-III multi-objective optimization method, combined with a comprehensive constraint model and a candidate plan generation and constraint correction mechanism, to screen out a set of Pareto optimal candidate course scheduling plans that meet the requirements of hard constraints and soft constraints in the crossover and mutation links, and then determine the optimal course scheduling plan from them to achieve scientific and efficient intelligent course scheduling.

[0088] Furthermore, taking the candidate course scheduling plans as the initial population and inputting them into the multi-objective optimization algorithm also includes:

[0089] Take the candidate course scheduling plans generated by the graph neural network and the comprehensive constraint model as the initial population;

[0090] Calculate the fitness values of multiple objectives for each candidate plan. The fitness values include teacher burden, classroom utilization rate, student movement cost, and conflict score;

[0091] Use the non-dominated sorting algorithm to layer the candidate plans and screen out the solutions on each Pareto front;

[0092] In the crossover operation, new candidate solutions are generated by swapping partial genes of two solutions, where the partial genes are time arrangements, teacher assignments, and classroom assignments;

[0093] The mutation operation fine-tunes some of the course scheduling decisions;

[0094] The new solutions generated by crossover and mutation are combined with the existing population, and non-dominated sorting and crowding distance calculation are performed again to update the population.

[0095] Specifically, in the optimization process of intelligent course scheduling, first, the candidate course scheduling solutions generated by the graph neural network and the comprehensive constraint model are regarded as the initial population and input into the multi-objective optimization algorithm. Subsequently, the fitness values of multiple objectives of each candidate solution are calculated. Among them, the teacher burden measures the balance degree of teacher teaching task allocation, the classroom utilization rate reflects the usage efficiency of classroom resources, the student movement cost reflects the cost generated by students' cross-campus movement due to course arrangements, and the conflict score is used to judge whether the solution violates various hard constraints and soft constraints. Then, the non-dominated sorting algorithm is used to stratify according to the superiority and inferiority relationship of each candidate solution on multiple objectives, so as to screen out the solutions on each Pareto front. These solutions represent a balanced state in multi-objective optimization, that is, without reducing the performance of other objectives, it is impossible to further improve the performance of a certain objective. In the crossover operation stage, some key information (i.e., partial genes) such as time arrangements, teacher assignments, and classroom assignments are selected from two different candidate solutions for exchange, thereby generating new candidate solutions, hoping to integrate the advantages of different solutions in this way. The mutation operation is to make fine adjustments to some of the course scheduling decisions, introducing new change factors into the population to prevent the algorithm from falling into a local optimal solution. Finally, the new solutions generated by the crossover and mutation operations are combined with the existing population, and the non-dominated sorting algorithm is executed again to re-determine the position relationship of each solution in the multi-objective space, and the crowding distance is calculated. This distance is used to measure the density between solutions on the same Pareto front. The larger the distance, the sparser the distribution of solutions around the solution, and the better the diversity of the solution. Through this series of operations, the population is updated, and then gradually better course scheduling solutions are screened out.

[0096] In summary, the embodiments of the present application have at least the following technical effects:

[0097] Collect historical course scheduling data, teacher and student preferences, classroom resources, and course attributes to construct a spatio-temporal feature matrix. The classroom resources include capacity and equipment data, and the course attributes include course type, duration, and number of students. Mine conflict patterns from historical course schedules and use a knowledge graph to construct an association relationship model of courses - teachers - classrooms. Based on the association relationship model, transform the constraints of new courses into vector form, and parse the constraint information described in natural language through a pre-trained BERT model to obtain a comprehensive constraint model. Use a graph neural network to capture the complex associations between courses, teachers, classrooms, and time periods, and at the same time use the comprehensive constraint model to generate candidate course scheduling plans. The candidate course scheduling plans satisfy hard constraints and soft constraints. The hard constraints include teacher time conflicts and classroom capacity, and the soft constraints include prerequisite relationships and student course selection conflicts. The prerequisite relationship is the prerequisite course that a student must complete before taking a subsequent course. Based on the candidate plans, further optimize multiple objectives and screen out the Pareto optimal solutions that meet all constraint conditions to obtain the optimal course scheduling plan. The optimization of multiple objectives includes balanced teacher workload, classroom utilization rate, and reduction of student cross-campus movement.

[0098] Embodiment 2. Based on the same inventive concept as the intelligent course scheduling method based on multi-objective optimization and deep learning in the foregoing embodiment, as Figure 2 shown, the present application provides an intelligent course scheduling system based on multi-objective optimization and deep learning. The system in the embodiments of the present application and the method embodiments are based on the same inventive concept. Among them, the system includes:

[0099] A data collection module 10, which is configured to collect historical course scheduling data, teacher and student preferences, classroom resources, and course attributes to construct a spatio-temporal feature matrix. The classroom resources include capacity and equipment data, and the course attributes include course type, duration, and number of students;

[0100] An association relationship model construction module 20, which is configured to mine conflict patterns from historical course schedules and use a knowledge graph to construct an association relationship model of courses - teachers - classrooms;

[0101] A comprehensive constraint model construction module 30, which is configured to, based on the association relationship model, transform the constraints of new courses into vector form, and parse the constraint information described in natural language through a pre-trained BERT model to obtain a comprehensive constraint model;

[0102] Candidate course scheduling plan generation module 40. The candidate course scheduling plan generation module 40 captures the complex associations among courses, teachers, classrooms, and time periods using a graph neural network, and at the same time generates candidate course scheduling plans using a comprehensive constraint model. The candidate course scheduling plans satisfy hard constraints and soft constraints. The hard constraints include teacher time conflicts and classroom capacities. The soft constraints include prerequisite relationships and student course selection conflicts. The prerequisite relationship is a prerequisite course that a student must complete before taking subsequent courses.

[0103] Optimal course scheduling plan obtaining module 50. The optimal course scheduling plan obtaining module 50 further optimizes multiple objectives based on the candidate plans and selects the Pareto optimal solutions that satisfy all constraint conditions to obtain the optimal course scheduling plan. The optimization of multiple objectives includes balanced teacher workload, classroom utilization rate, and reduction of student cross-campus movement.

[0104] It should be noted that the above order of the embodiments of the present application is only for description and does not represent the superiority or inferiority of the embodiments. And the above description of specific embodiments of this specification has been made. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be executed in a different order than in the embodiments and still achieve the desired results. Additionally, the processes depicted in the figures do not necessarily require the specific order or sequential order shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0105] The above are only the preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included within the protection scope of the present application.

[0106] This specification and the drawings are only exemplary descriptions of the present application and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the present application. Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalent technologies, the present application is intended to include these changes and modifications.

Claims

1. An intelligent course scheduling method based on multi-objective optimization and deep learning, characterized in that, Including: Collect historical course scheduling data, teachers' and students' preferences, classroom resources, and course attributes to construct a spatio-temporal feature matrix. The classroom resources include capacity and equipment data, and the course attributes include course type, duration, and number of students; Mine conflict patterns from historical course schedules and use a knowledge graph to construct an association relationship model of courses - teachers - classrooms; Based on the association relationship model, convert the constraints of new courses into vector form, and parse the constraint information described in natural language through a pre-trained BERT model to obtain a comprehensive constraint model; Use a graph neural network to capture the complex associations among courses, teachers, classrooms, and time periods. At the same time, use the comprehensive constraint model to generate candidate course scheduling plans that meet hard constraints and soft constraints. The hard constraints include teacher time conflicts and classroom capacity, and the soft constraints include prerequisite relationships and student course selection conflicts. The prerequisite relationship is the prerequisite course that students must complete before taking subsequent courses; Based on the candidate course scheduling plan, further optimize multiple objectives and screen out the Pareto optimal solutions that meet all constraint conditions to obtain the optimal course scheduling plan. The optimization of multiple objectives includes balanced teacher workload, classroom utilization rate, and reduction of students' cross-campus movement.

2. The intelligent course scheduling method based on multi-objective optimization and deep learning according to claim 1, wherein the construction of the spatio-temporal feature matrix is characterized in that Including: Define \(C\) as the set of courses to be sorted, \(C=\{C_1,C_2,\cdots,C\ n \}\), where \(n\) is the total number of courses; Define T as the set of each time period, T = {T1, T2,..., T m}, where m is the total number of time periods; Define R as the set of available classrooms, R = {R1, R2,..., R k}, where k is the total number of classrooms; The dimension of the spatio-temporal feature matrix X is: X = C × T × R; where C is the set of courses to be scheduled, T is the set of each time period, and R is the set of available classrooms; The size of the spatio-temporal feature matrix X is n × m × k, that is, a combination of n courses, m time periods, and k classrooms.

3. The intelligent course scheduling method based on multi-objective optimization and deep learning according to claim 1, wherein the mining of conflict patterns from historical course schedules and using a knowledge graph to construct an association relationship model of courses, teachers, and classrooms includes: Extract relevant course scheduling information from historical course schedules. By analyzing historical course scheduling data, identify possible conflict patterns among different courses, teachers, and classrooms. The conflict patterns include time conflicts, teacher conflicts, classroom conflicts, and student conflicts; Use a frequent pattern mining algorithm to discover common conflict patterns in historical course schedules, record the conflict patterns and conflict types, and use the conflict patterns and conflict types as conflict constraint rules to form a conflict pattern library and obtain a conflict pattern graph; Based on historical course scheduling data, construct a knowledge graph of courses, teachers, and classrooms, with courses, teachers, and classrooms as the three main entities in the knowledge graph; Establish the associations between courses and teachers, courses and classrooms, and teachers and classrooms to obtain an entity relationship model.

4. The intelligent course scheduling method based on multi-objective optimization and deep learning according to claim 3, wherein The use of the frequent item mining algorithm to discover common conflict patterns in the course schedule, record the conflict patterns and conflict types, and use the conflict patterns and conflict types as conflict constraint rules to form a conflict pattern library and obtain a conflict pattern graph further includes: Adopt a frequent item mining algorithm to determine the lower limit of support according to historical course scheduling data and filter out low-frequency conflict patterns; Analyze the combinations of courses, teachers, and classrooms that often appear together within the same time period to form frequent item sets; Analyze each frequent item set to identify its corresponding conflict type, where the conflict types include teacher time conflict, classroom resource conflict, and student course selection conflict. The teacher time conflict means that the same teacher has scheduled multiple courses at the same time. The classroom resource conflict means that multiple courses compete for the same classroom within the same time period. The student course selection conflict means that some students have selected courses with overlapping time slots at the same time; Summarize the conflict patterns and their conflict types to form a preliminary set of conflict constraint rules; Organize and store all conflict patterns, corresponding conflict types, and constraint rules to build a conflict pattern library; Using the conflict pattern library, construct a conflict pattern graph by taking each conflict pattern as the attributes of nodes and edges in the graph. The nodes represent courses, teachers, and classrooms, and the edges represent conflict relationships.

5. The intelligent course scheduling method based on multi-objective optimization and deep learning according to claim 3, characterized in that, Based on historical course scheduling data, build a knowledge graph of courses, teachers, and classrooms, taking courses, teachers, and classrooms as the three main entities in the knowledge graph. It also includes: Obtain historical course scheduling data, which includes course information, teacher information, and classroom information. The course information includes course name, course number, course type, class hours, prerequisite relationship, and number of students selecting the course. The teacher information includes teacher name, teacher number, major, courses that can be taught, available time, and affiliated department. The classroom information includes classroom number, capacity, equipment type, and affiliated building; Define the main entities and their attributes in the knowledge graph. The entities include courses, teachers, and classrooms; Based on historical course scheduling data and domain knowledge, determine the association relationships between entities and construct triples. The association relationships include the relationship between courses and teachers, the relationship between courses and classrooms, the indirect association between teachers and classrooms, the course prerequisite relationship, and the teacher available time constraint; Extract entities such as courses, teachers, and classrooms and their attributes using the preprocessed data; Extract the relationships between entities based on course scheduling records and domain knowledge to form a triple set; Import all entities and relationships into a graph database to build a knowledge graph; Use data rules to verify the correctness of entities and relationships in the knowledge graph.

6. The intelligent course scheduling method based on multi-objective optimization and deep learning according to claim 1, characterized in that, Based on the association relationship model, convert the constraints of the new course into vector form, and parse the constraint information described in natural language through a pre-trained BERT model to obtain a comprehensive constraint model. It also includes: Vectorize the attributes of courses, teachers, and classrooms and their association relationships. The vectorization includes course vectorization, teacher vectorization, and classroom vectorization; The course vectorization means that each course can be represented as a vector through course attributes, and the course attributes include course type, duration, and number of students; The teacher vectorization means that each teacher is represented as a vector through their teaching ability and available time period attributes; The classroom vectorization means that the attributes of each classroom can be converted into a vector, and the attributes of each classroom include classroom capacity, equipment requirements, and time period availability; Obtain the constraint information described in natural language for the new course, and use a pre-trained BERT model to parse the constraint information to obtain the context representation of each word; Combine the natural language constraint information parsed by BERT with the existing constraint conditions in the course scheduling system to form a comprehensive constraint model.

7. The intelligent course scheduling method based on multi-objective optimization and deep learning according to claim 1, wherein The method of using a graph neural network to capture the complex associations between courses, teachers, classrooms, and time periods, and simultaneously using the comprehensive constraint model to generate candidate course scheduling plans includes: Take courses, teachers, and classrooms as nodes and establish edges between the nodes; Construct feature vectors for each node; Adopt a GNN model to aggregate information through an adjacency matrix, learn the high-dimensional representations of the nodes, and capture local and global association relationships; Use the course schedule after GNN learning, combine the time period and classroom features, calculate the allocation probabilities of courses at different times and in different classrooms through a multi-layer perceptron, and perform normalization to obtain the probability matrix of the candidate course scheduling plan, and then generate multiple preliminary course scheduling plans through Top-K sampling.

8. The intelligent course scheduling method based on multi-objective optimization and deep learning according to claim 1, characterized in that The candidate course scheduling plan is further optimized for multiple objectives and the Pareto optimal solutions that meet all constraint conditions are screened out to obtain the optimal course scheduling plan, which also includes: Use a graph neural network to capture the associations between courses, teachers, and classrooms, and combine the comprehensive constraint model to generate multiple candidate course scheduling plans that meet the basic constraint conditions; Take the candidate course scheduling plan as the initial population and input it into a multi-objective optimization algorithm to obtain an updated population; Define multiple objective functions for each plan, and at the same time embed the comprehensive constraint as an evaluation index. The multiple objective functions include teacher burden, classroom utilization rate, and student cross-campus movement distance. The comprehensive constraint includes hard constraints and soft constraints; Adopt an improved NSGA-III multi-objective optimization method combined with the comprehensive constraint model and the candidate solution generation and constraint correction mechanism to obtain a set of Pareto optimal candidate course scheduling plans that meet the requirements of hard constraints and soft constraints during the crossover and mutation processes.

9. The intelligent course scheduling method based on multi-objective optimization and deep learning according to claim 8, characterized in that The step of taking the candidate course scheduling plan as the initial population and inputting it into a multi-objective optimization algorithm also includes: Take the candidate course scheduling plans generated by the graph neural network and the comprehensive constraint model as the initial population; Calculate the fitness values of multiple objectives for each candidate plan. The fitness values include teacher burden, classroom utilization rate, student movement cost, and conflict score; Use the non-dominated sorting algorithm to stratify the candidate plans and screen out the solutions on each Pareto front; In the crossover operation, generate new candidate plans by exchanging some genes of two plans. The some genes are time arrangement, teacher assignment, and classroom assignment; The mutation operation fine-tunes some course scheduling decisions; Merge the new solutions generated by crossover and mutation with the existing population, perform non-dominated sorting and crowding distance calculation again, and update the population.

10. An intelligent course scheduling system based on multi-objective optimization and deep learning, characterized in that, The system for implementing the intelligent course scheduling method based on multi-objective optimization and deep learning according to any one of claims 1 to 9, the system includes: A data collection module, which collects historical course scheduling data, teacher and student preferences, classroom resources, and course attributes, constructs a spatio-temporal feature matrix. The classroom resources include capacity and equipment data, and the course attributes include course type, duration, and number of students; Association relationship model construction module, which mines conflict patterns from historical class schedules and constructs an association relationship model of courses - teachers - classrooms using a knowledge graph; Comprehensive constraint model construction module, which, based on the association relationship model, converts the constraints of new courses into vector forms, and parses the constraint information described in natural language through a pre-trained BERT model to obtain a comprehensive constraint model; Candidate class scheduling plan generation module, which uses a graph neural network to capture the complex associations between courses, teachers, classrooms, and time periods, and at the same time uses the comprehensive constraint model to generate candidate class scheduling plans. The candidate class scheduling plans satisfy hard constraints and soft constraints. The hard constraints include teacher time conflicts and classroom capacity, and the soft constraints include prerequisite relationships and student course selection conflicts. The prerequisite relationship is the prerequisite course that students must complete before taking subsequent courses; Optimal class scheduling plan obtaining module, which, based on the candidate class scheduling plans, further optimizes multiple objectives and screens out the Pareto optimal solutions that meet all constraint conditions to obtain the optimal class scheduling plan. The optimization of multiple objectives includes balanced teacher workload, classroom utilization rate, and reduction of student cross-campus movement.

Images