Simulation training personalized recommendation analysis method, system, equipment and medium

Abstract

The invention discloses a simulation training personalized recommendation analysis method, system and device and a medium, and relates to the technical field of computer platform load balancing, and the method comprises the steps: collecting user behavior data, carrying out the preprocessing, and carrying out the feature marking and grading of a user; quantizing user behavior characteristics by using a distributed machine learning platform, and dynamically adjusting user groups through a clustering algorithm; and generating personalized recommendation by adopting collaborative filtering and reinforcement learning algorithms, and dynamically adjusting task difficulty based on real-time feedback to optimize recommendation weight. According to the method, multi-dimensional behavior characteristics are clustered through a K-means + + clustering algorithm, and dynamic user grouping is realized in combination with a dynamic threshold adjustment mechanism; by designing a recommendation algorithm architecture, not only can the new user be ensured to quickly adapt, but also individuation and dynamic adaptation of the recommended content are realized, and the classification error rate is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of pipeline detection, and in particular to a method, system, equipment and medium for personalized recommendation analysis of simulation training. Background Art

[0002] With the innovation and development of science and technology, the work of power grid dispatching requires higher efficiency and accuracy, and therefore the standards and requirements for dispatchers are also higher. Traditional training methods use fixed courses and require a lot of manpower support. The time is not flexible, and it is difficult for dispatchers to learn independently. In addition, the results of previous training need to be manually analyzed, which is time-consuming and ineffective. It is impossible to optimize the training content, resulting in delayed training feedback. The teaching standards of different coaches vary, which may lead to inconsistent terminology and communication misunderstandings in actual work. It takes a lot of time from the new employees joining the company to training and taking up their posts, which invisibly adds a burden to the already heavy dispatching department.

[0003] However, existing user behavior data analysis technologies have the following major flaws. First, user classification standards are rigid. They divide users into levels based on a single static indicator without integrating multi-dimensional behavioral characteristics, resulting in insufficient granularity in user portraits and an inability to accurately identify potential capability differences. Second, recommendation algorithms lack a dynamic adaptation mechanism. Most systems use fixed thresholds or predefined scenario libraries and cannot respond to capability changes in real time. Finally, existing technologies do not achieve closed-loop optimization. Most recommendation models rely on offline data updates and cannot adjust recommendation weights through real-time feedback.

[0004] To address these issues, this paper proposes a personalized training recommendation analysis method based on user behavior data analysis. By analyzing user behavior, leveraging machine learning and reinforcement learning techniques, this method accurately recommends training content to users, shortening training cycles, improving training quality, and enabling adaptive training. This increases accuracy in production environments and fosters a personalized training mechanism tailored to each student's needs. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the present invention provides a personalized recommendation analysis method for simulation training to solve the problems of the existing simulation training recommendation method, such as the single user classification standard, static recommendation algorithm, insufficient scenario generation capability, and how to dynamically optimize the adaptation of training content through real-time feedback.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In the first aspect, the present invention provides a personalized recommendation analysis method for simulated training, including: collecting user behavior data and preprocessing it, and labeling and grading users with features; using a distributed machine learning platform to quantify user behavior characteristics, and dynamically adjusting user groupings through a clustering algorithm; using collaborative filtering and reinforcement learning algorithms to generate personalized recommendations, and dynamically adjusting task difficulty based on real-time feedback to optimize recommendation weights.

[0009] As a preferred solution of the personalized recommendation analysis method for simulated training described in the present invention, the user behavior data is collected and pre-processed, including uploading the operation log to the server in real time through the user terminal; using SQL to clean the data, process missing values, and delete records with a missing rate of more than 50%; filling some missing fields with the average value of similar users; deduplication based on user ID, task ID, and timestamp, and filtering outliers; storing valid data in a distributed database to provide standardized input for subsequent feature annotation

[0010] As a preferred solution of the personalized recommendation analysis method for simulation training described in the present invention, the user feature labeling and grading includes dividing the user level according to the ratio of user error rate to completion time; after feature labeling, the user level label is associated with operation efficiency and knowledge blind spot index; the user level includes novice users, skilled users and expert users.

[0011] As a preferred solution of the simulated training personalized recommendation analysis method described in the present invention, the use of a distributed machine learning platform to quantify user behavior characteristics includes counting the user's average timeliness rate and average accuracy rate through a distributed computing platform to extract the user's behavioral preference characteristics.

[0012] As a preferred solution of the personalized recommendation analysis method for simulation training described in the present invention, the dynamic adjustment of user grouping through the clustering algorithm includes generating role labels, using the K-means++ clustering algorithm, pre-defining the initial clustering center based on historical excellent user data, and inputting the standardized error rate, task time, and knowledge section weight characteristics for clustering grouping; updating once a quarter through the dynamic threshold adjustment mechanism, and recalibrating the threshold based on the performance distribution of all users; for boundary value users, introducing weight coefficients α and β for comprehensive scoring through multi-indicator weighted judgment.

[0013] As a preferred solution of the personalized recommendation analysis method for simulation training described in the present invention, the use of collaborative filtering and reinforcement learning algorithms to generate personalized recommendations includes calculating user cosine similarity through a collaborative filtering mechanism, recommending scenarios with the highest completion rate among similar users, and recommending high-frequency general scenarios to new users by default; dynamically adjusting the scenario difficulty coefficient based on the user's current level, knowledge blind spots, and recent task performance through a reinforcement learning algorithm; presetting static standard training scenarios based on expert users; after the user completes the task, the operation log is written to the Kafka message queue in real time, the user similarity matrix is ​​fully updated daily, and the Q-table is fine-tuned every hour to optimize the recommendation strategy.

[0014] As a preferred solution of the personalized recommendation analysis method for simulation training described in the present invention, the method of dynamically adjusting task difficulty based on real-time feedback to optimize recommendation weights includes: the system records the user's task completion rate, and imports new data into user portraits by time; recalculates the user level based on the new task completion time and error rate, triggering dynamic difficulty adjustment.

[0015] In the second aspect, the present invention provides a simulated training personalized recommendation analysis system, including: a level division module, a user grouping module, and a personalized recommendation module; the level division module is used to collect user behavior data and perform preprocessing, and to perform feature labeling and level division on users; the user grouping module is used to use a distributed machine learning platform to quantify user behavior characteristics, and dynamically adjust user groups through a clustering algorithm; the personalized recommendation module is used to generate personalized recommendations using collaborative filtering and reinforcement learning algorithms, and dynamically adjust the task difficulty based on real-time feedback to optimize the recommendation weight.

[0016] In a third aspect, the present invention provides an electronic device, comprising:

[0017] memory and processor;

[0018] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the simulation training personalized recommendation analysis method are implemented.

[0019] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the simulated training personalized recommendation analysis method.

[0020] Compared with the existing technology, the present invention has the following beneficial effects: the present invention clusters multi-dimensional behavioral features through the K-means++ clustering algorithm, and combines it with a dynamic threshold adjustment mechanism to achieve dynamic user grouping, solving the problem of the single traditional static classification standard; collaborative filtering calculates user cosine similarity to recommend high-completion scenarios of similar users, ensuring that new users can adapt quickly; reinforcement learning dynamically adjusts the difficulty coefficient based on real-time task performance, and optimizes the reward function strategy to achieve personalized and dynamic adaptation of recommended content, thereby reducing the classification error rate; user operation logs are written into the Kafka message queue in real time, the user similarity matrix is ​​updated daily, and the Q-table is fine-tuned every hour to form a closed-loop optimization and real-time feedback, thereby improving training efficiency and rational resource allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0022] Figure 1 The figure is a schematic diagram of the overall process of the personalized recommendation analysis method for simulated training according to one embodiment of the present invention.

[0023] Figure 2 The figure is a schematic diagram of the overall process of the personalized recommendation analysis system for simulated training according to one embodiment of the present invention. DETAILED DESCRIPTION

[0024] To make the above-mentioned objects, features, and advantages of the present invention more clearly understood, the following detailed description of the specific embodiments of the present invention is given in conjunction with the accompanying drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without creative work should fall within the scope of protection of the present invention.

[0025] Example 1, with reference to Figure 1 , as one embodiment of the present invention, provides a personalized recommendation analysis method for simulated training, comprising:

[0026] S1: Collect and pre-process user behavior data, and then label and classify users;

[0027] S2: Utilizes a distributed machine learning platform to quantify user behavior characteristics and dynamically adjust user groups through clustering algorithms;

[0028] S3: Use collaborative filtering and reinforcement learning algorithms to generate personalized recommendations, and dynamically adjust task difficulty based on real-time feedback to optimize recommendation weights.

[0029] It should be noted that power grid dispatching involves high-risk scenarios. Training can help dispatchers master standardized operating procedures, unify terminology, ensure quick and accurate decision-making in complex environments, and reduce the risk of human operational errors. Through systematic training, new employees can quickly become familiar with business logic and operating specifications, reduce the time cost from joining the company to taking up the post, and alleviate the manpower burden of the dispatching department. Traditional training relies on fixed courses and is difficult to solve individual weaknesses in a targeted manner. Simulation training dynamically identifies user error types, pushes special training, and systematically fills skill gaps. At the same time, standardized scoring models and real-time data recording avoid the subjective bias of manual scoring, ensuring that the training results are objective and scientific, and directly mapped to the operational efficiency and accuracy in actual work. Therefore, simulation training for user behavior data analysis is very important.

[0030] Therefore, in response to the above-mentioned simulation training problems, through steps S1-S3, user behavior data is collected and preprocessed, and users are labeled and graded with characteristics, realizing standardized data collection and cleaning of user behavior data, ensuring the data quality of subsequent analysis, avoiding noise interference in model training, and realizing user feature labeling and grade division, providing an initial user stratification basis for subsequent recommendations, and solving the "one-size-fits-all" drawbacks of traditional training; using a distributed machine learning platform to quantify user behavior characteristics, and dynamically adjusting user groups through clustering algorithms, realizing dynamic and accurate adaptation of user portrait construction and user grouping, avoiding model rigidity; combining collaborative filtering and reinforcement learning, building a real-time optimized personalized recommendation engine, realizing the maximization of long-term skill gain.

[0031] Example 2, reference Figure 1 , which is an embodiment of the present invention, provides a personalized recommendation analysis method for simulated training based on the above embodiment.

[0032] In the embodiment of the present application, collecting user behavior data and preprocessing it in step S1 includes the following steps A1-A3:

[0033] A1: Upload operation logs to the server in real time through the user terminal;

[0034] A2: Use SQL to clean the data, handle missing values, and delete records with a missing value rate exceeding 50%. Fill in missing fields with the average value of similar users. De-duplicate the user ID, task ID, and timestamp, and filter outliers.

[0035] A3: Store valid data in a distributed database to provide standardized input for subsequent feature annotation.

[0036] Specifically, in step A1, the operation log includes structured data and unstructured data;

[0037] Structured data includes training duration, assessment scores, and task completion time;

[0038] Unstructured data includes user feedback text;

[0039] In step A2, outliers are defined as records where the training duration exceeds 3 times the standard deviation or is negative;

[0040] Filtering outliers includes marking abnormal records and storing them in logs for manual review.

[0041] In the embodiment of the present application, the user feature labeling and classification in step S1 includes the following steps B1-B2:

[0042] B1: Divide users into different levels based on the ratio of user error rate to completion time;

[0043] B2: After feature annotation, associate the user level label with the operation efficiency and knowledge blind spot index.

[0044] Specifically, in step B1, the cleaned data is feature labeled;

[0045] The annotated features include user business level and high-frequency error types;

[0046] User service level includes user error rate and completion time ratio.

[0047] And divide users into levels according to their business level;

[0048] Among them, user levels include novice users, skilled users, and expert users;

[0049] Novice users are those who have just started training, have a low level of professional skills, and need basic training. Their data characteristics include completing less training content, having lower scores, making mistakes concentrated on basic knowledge points, and working slowly.

[0050] Skilled users are those who have mastered the basics and possess certain business capabilities, but still have room for improvement. They can complete most of the training content but still have difficulty solving some complex problems, and their errors are concentrated on advanced knowledge points.

[0051] Expert users are users with high professional skills and rich experience. They have high training scores, make fewer mistakes, are efficient in answering questions, and can correctly handle complex problems.

[0052] At the user service level, the user error rate is expressed as:

[0053]

[0054] Where E represents the number of incorrect operations or incorrect answers in the user's task, and N represents the total number of tasks completed by the user in a specific time period;

[0055] Completion time ratio, expressed as:

[0056]

[0057] Among them, T actual Indicates the time it actually takes for the user to complete the task, T standard Indicates the preset standard completion time of the task;

[0058] Among them, the novice user classification criteria include R error ≥60% or R time ≥150%;

[0059] Skilled user classification criteria include 30% ≤ R error <60% and 120%≤R time <150%;

[0060] Expert user classification criteria include R error <30% and R time ≤120%.

[0061] In the embodiment of the present application, step S2 utilizes a distributed machine learning platform to quantify user behavior characteristics, including using a distributed computing platform to calculate the average timeliness rate and average accuracy rate of users and extract the user's behavior preference characteristics;

[0062] Specifically, the labeled feature data is input, and the average timeliness and accuracy of each user are calculated. The operation efficiency and knowledge blind spot index of each user are quantified, and the user's behavioral preference characteristics are extracted.

[0063] The user's operating efficiency is expressed as:

[0064]

[0065] Among them, Q task Indicates the total number of tasks completed by the user, t task Indicates the time it takes for the user to complete the total number of tasks;

[0066] The knowledge blind spot index is expressed as:

[0067]

[0068] Among them, U represents the error rate of the user's i-th category of knowledge, and F represents the average error rate of all users.

[0069] The average timeliness rate is expressed as:

[0070]

[0071] Among them, M represents the total number of students, C i represents the on-time completion rate of the i-th student;

[0072] The average accuracy is expressed as:

[0073]

[0074] Among them, A i represents the training accuracy of the i-th student;

[0075] User behavior preference characteristics include the time spent on training for a specific business, the courses they have recently taken multiple times, and the business knowledge points they focus on training;

[0076] We directly extract data on training duration, accuracy, timeout rate, average accuracy over the past seven days, and task completion time fluctuation coefficients. We perform one-hot encoding on error types and statistically analyze time series features. We use linear regression slopes to analyze the accuracy trends and periodicity of users' last five tasks.

[0077] Periodicity characteristics include differences in training efficiency during daily active time periods (morning / noon / evening).

[0078] In an optional implementation, statistics on the average timeliness and average accuracy of users can also be written into the Kafka message queue architecture in real time through operation logs, using the Apache Flink stream processing engine to calculate the average timeliness and average accuracy of users in real time.

[0079] In another optional implementation, the average timeliness and average accuracy of users can also be calculated based on the data stored in a distributed database, using the continuous query function of the InfiuxDB time series database to calculate the average timeliness and average accuracy of users.

[0080] In the embodiment of the present application, step S2 dynamically adjusts user groups by clustering algorithm, including the following steps C1-C3:

[0081] C1: Generate role labels using the K-means++ clustering algorithm. Initial cluster centers are predefined based on historical excellent user data. Normalized error rates, task duration, and knowledge section weights are input for clustering and grouping.

[0082] C2: Updated quarterly through a dynamic threshold adjustment mechanism, recalibrating the threshold based on the performance distribution of all users;

[0083] C3: For users with borderline values, weight coefficients α and β are introduced through multi-index weighted judgment to perform comprehensive scoring.

[0084] Specifically, in step C1, user behavior is quantified and statistically modeled, and the Poisson distribution is used to model user task completion time and identify abnormally delayed tasks;

[0085] Generate role tags, where the role tags include static tags and skill level tags;

[0086] Among them, static tags include job roles and initial grades based on entry assessments;

[0087] The skill level label includes dynamic updates based on three consecutive task performances. If a user is at the expert level but fails to perform well in three consecutive tasks, the user will be downgraded from expert to skilled.

[0088] By using the K-means++ algorithm, we pre-define and optimize the initial cluster centers based on historical excellent user data, and input standardized weighted features such as error rate, task duration, and knowledge section;

[0089] Use clustering algorithms to pre-process user portrait data, design user behavior models based on job level requirements, and divide them into three user behavior groups: intensive training group, medium potential group, and efficient and skilled group;

[0090] Among them, the intensive training group is for users with high error rates, exceeding level requirements, and taking a long time to complete. They need to strengthen their basic business capabilities and receive comprehensive training;

[0091] The medium potential group has reached the error rate requirement, but the time required to complete the task is not up to the requirement. They need to undergo advanced and enhanced training to improve their efficiency.

[0092] The efficient and skilled group has an error rate within the requirements and a completion time within the average time. The system will provide students with more difficult training to cultivate their ability to solve complex problems and make decisions.

[0093] The user behavior model judgment rules include: if the user error rate R error > Job level requires an upper limit on error rate and a completion time ratio R time > If the job level requires a lower limit on the completion time, the user will be divided into the intensive training group;

[0094] If the user error rate R error ≤ upper limit of error rate required by job level, and completion time ratio R time > If the job level requires a lower limit on the completion time, the user will be classified as a medium potential group;

[0095] If the user error rate Rerror ≤ upper limit of error rate required by job level, and completion time ratio R time If the completion time is less than or equal to the lower limit of the job level requirement, the user will be classified into the efficient and skilled group.

[0096] It should be noted that if the total number of tasks of a user is less than 5, classification will not be performed for the time being and will be performed after more data is accumulated.

[0097] In step C2, the dynamic threshold adjustment mechanism includes regular updates of job level thresholds based on historical data, and recalibration based on the performance distribution of all users once every quarter;

[0098] The job level threshold includes the upper limit of the job level requirement error rate and the lower limit of the job level requirement completion time.

[0099] In step C3, for users with boundary values, i.e., users close to the upper limit of the error rate or the lower limit of the completion time required by the job level, weight coefficients are introduced, and the α and β coefficients are set, so that α + β = 1;

[0100] Assume that the job level requires the upper limit of error rate to be R error_max The minimum completion time required for the job level is T time_min , and give a comprehensive score to the borderline users, expressed as:

[0101] S core =α·R error_max +β·T time_min

[0102] It should be noted that the values ​​of the α and β coefficients are set according to specific business requirements. For example, for positions that require precision, α>0.5; for positions that require efficiency, β>0.5.

[0103] In the embodiment of the present application, the use of collaborative filtering and reinforcement learning algorithms to generate personalized recommendations in step S3 includes the following steps D1-D4:

[0104] D1: Calculate user cosine similarity through collaborative filtering and recommend scenarios with the highest completion rate among similar users. For new users, high-frequency and common scenarios are recommended by default.

[0105] D2: Dynamically adjust the scenario difficulty coefficient based on the user's current level, knowledge gaps, and recent task performance through reinforcement learning algorithms;

[0106] D3: Based on expert users, pre-set static standard training scenarios;

[0107] D4: After the user completes the task, the operation log is written to the Kafka message queue in real time. The user similarity matrix is ​​fully updated daily, and the Q-table is fine-tuned every hour to optimize the recommendation strategy.

[0108] Specifically, design the recommendation algorithm architecture, which includes collaborative filtering mechanism and reinforcement learning algorithm;

[0109] In step D2, the reward function is expressed as:

[0110] R=α·S+β·ΔE

[0111] Among them, S represents the number of consecutive successes of the user, and ΔE represents the improvement in user efficiency.

[0112] In step D3, static standard training scenarios are preset for expert users;

[0113] Static standard training scenarios are preset. Specifically, by accessing real-time power grid data, multi-fault concurrent scenarios are generated, and GAN is used to generate adversarial networks to simulate complex scenarios to ensure compliance with real physical constraints.

[0114] It should be noted that for novice users, correction training for high-frequency error scenarios should be recommended based on the knowledge blind spot labels in the user portrait; for experienced users, advanced tasks with strong correlation are recommended based on user behavior preferences, such as common instruction types and high-frequency error instructions; for expert users, the system will combine real-time and real power grid data to dynamically generate high-risk scenarios and improve user processing capabilities.

[0115] It should also be noted that the system will recommend the most completed training scenarios among similar user groups based on the collected data, such as the number of consecutive successes, task completion time, error type distribution, etc., and maximize the long-term benefits of user skill improvement by dynamically adjusting the task difficulty.

[0116] In the implementation manner of the present application, in step S3, dynamically adjusting the task difficulty based on real-time feedback to optimize the recommendation weight includes the system recording the user's task completion rate, and importing new data into the user portrait by time; recalculating the user level based on the new task completion time and error rate, and triggering dynamic difficulty adjustment.

[0117] Specifically, dynamic difficulty adjustment includes increasing the complexity of the task if the user succeeds continuously and the efficiency improves; if the error rate exceeds the threshold or the task times out, it will fall back to a lower difficulty scenario;

[0118] After the user profile is updated, the system optimizes the recommendation model through incremental learning to ensure that the training content is dynamically matched with user capabilities.

[0119] In an optional implementation, the recommendation weight optimization rule is to increase the weight of high completion rate scenarios by Δw, which is expressed as:

[0120] Δw=C·0.1+ΔE·0.2

[0121] Where C represents the user task completion rate.

[0122] Example 3, reference Figure 2 The above is a schematic diagram of a method for personalized recommendation analysis of simulated training. It should be noted that the technical solution of this system for personalized recommendation analysis of simulated training is based on the same concept as the technical solution of the aforementioned method for personalized recommendation analysis of simulated training. For details not described in detail in the technical solution of the system for personalized recommendation analysis of simulated training in this embodiment, please refer to the description of the technical solution of the aforementioned method for personalized recommendation analysis of simulated training.

[0123] This embodiment also provides a simulated training personalized recommendation analysis system, which includes: a level classification module, a user grouping module, and a personalized recommendation module.

[0124] Among them, the grading module is used to collect and pre-process user behavior data, and to label and grade users; the user grouping module is used to quantify user behavior characteristics using a distributed machine learning platform, and dynamically adjust user groups through clustering algorithms; the personalized recommendation module is used to generate personalized recommendations using collaborative filtering and reinforcement learning algorithms, and dynamically adjust task difficulty based on real-time feedback to optimize recommendation weights.

[0125] This embodiment also provides an electronic device suitable for personalized recommendation analysis of simulated training, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the personalized recommendation analysis method for simulated training proposed in the above embodiment.

[0126] This embodiment further provides a storage medium storing a computer program, which, when executed by a processor, implements the personalized recommendation analysis method for simulated training as proposed in the above embodiment.

[0127] The storage medium proposed in this embodiment and the method for implementing personalized recommendation analysis for simulated training proposed in the above embodiment belong to the same inventive concept. For technical details not fully described in this embodiment, please refer to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0128] Through the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented with the help of software and necessary general hardware, and of course can also be implemented by hardware. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods of various embodiments of the present invention.

[0129] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A personalized recommendation analysis method for simulated training, characterized in that: include: Collect user behavior data and pre-process it, and then label and grade users according to their characteristics; Utilize a distributed machine learning platform to quantify user behavior characteristics and dynamically adjust user groups through clustering algorithms; Collaborative filtering and reinforcement learning algorithms are used to generate personalized recommendations, and the task difficulty is dynamically adjusted based on real-time feedback to optimize the recommendation weight.

2. The personalized recommendation analysis method for simulation training according to claim 1, characterized in that: The collecting and pre-processing of user behavior data includes uploading operation logs to the server in real time via the user terminal; Use SQL to clean the data, handle missing values, and delete records with a missing value rate exceeding 50%; Some missing fields are filled with the mean value of similar users; Combined deduplication based on user ID, task ID, and timestamp to filter outliers; Valid data is stored in a distributed database to provide standardized input for subsequent feature annotation.

3. The personalized recommendation analysis method for simulation training according to claim 2, characterized in that: The user feature labeling and grading includes grading the user based on the ratio of the user error rate to the completion time; After feature annotation, the user level label is associated with the operation efficiency and knowledge blind spot index; The user levels include novice users, experienced users, and expert users.

4. The personalized recommendation analysis method for simulation training according to claim 3, characterized in that: The use of a distributed machine learning platform to quantify user behavior characteristics includes using a distributed computing platform to count users' average timeliness and average accuracy to extract users' behavior preference characteristics.

5. The personalized recommendation analysis method for simulation training according to claim 4, characterized in that: The dynamic adjustment of user grouping by clustering algorithm includes generating role labels, using K-means++ clustering algorithm, pre-defining initial cluster centers based on historical excellent user data, and inputting standardized error rate, task time, and knowledge section weight features for clustering grouping; The threshold is updated quarterly through a dynamic threshold adjustment mechanism, which recalibrates the threshold based on the performance distribution of all users; For users with borderline values, weight coefficients α and β are introduced through multi-index weighted judgment to perform comprehensive scoring.

6. The personalized recommendation analysis method for simulation training according to claim 5, characterized in that: The use of collaborative filtering and reinforcement learning algorithms to generate personalized recommendations includes calculating user cosine similarity through the collaborative filtering mechanism, recommending the scenarios with the highest completion rate among similar users, and recommending high-frequency common scenarios by default to new users; Dynamically adjust the scene difficulty coefficient based on the user's current level, knowledge blind spots, and recent task performance through reinforcement learning algorithms; Preset static standard training scenarios based on expert users; After the user completes the task, the operation log is written to the Kafka message queue in real time, the user similarity matrix is ​​fully updated daily, and the Q-table is fine-tuned every hour to optimize the recommendation strategy.

7. The personalized recommendation analysis method for simulation training according to claim 6, characterized in that: The dynamic adjustment of task difficulty based on real-time feedback to optimize recommendation weights includes the system recording the user's task completion rate and importing new data into user profiles by time; Recalculate user levels based on new task completion times and error rates, triggering dynamic difficulty adjustments.

8. A personalized recommendation analysis system for simulated training, applying the method according to any one of claims 1 to 7, characterized in that: include: Level classification module, user grouping module, and personalized recommendation module; The grading module is used to collect and pre-process user behavior data, and to perform feature annotation and grading on users; The user grouping module is used to quantify user behavior characteristics using a distributed machine learning platform and dynamically adjust user groups through a clustering algorithm; The personalized recommendation module is used to generate personalized recommendations using collaborative filtering and reinforcement learning algorithms, and dynamically adjust task difficulty based on real-time feedback to optimize recommendation weights.

9. An electronic device comprising: memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the simulation training personalized recommendation analysis method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the simulation training personalized recommendation analysis method according to any one of claims 1 to 7.

Images