Machine room learning result evaluation system based on real name binding and intelligent acquisition

Abstract

The invention relates to the technical field of informatization of knowledge teaching and skill cultivation, and discloses a machine room learning result evaluation system based on real-name binding and intelligent acquisition. The system comprises a real-name binding module, an automatic data acquisition module, a dynamic data matrix modeling module, a dynamic weight optimization module, a comprehensive score output module, a hybrid network adaptation module and a federated learning coordinator, wherein the real-name binding module is used for carrying out uniqueness binding on learner identity information and a terminal equipment hardware feature code; and the data automatic acquisition module is used for acquiring a screen operation log, a code compiling result and training activity space interactive behavior data of the learner end in real time. Through dynamic weight optimization, multi-dimensional data fusion modeling, intelligent network adaptation and visualization technologies, the defects of scoring subjectivity, interaction delay and strategy general of a traditional system are overcome, and technical improvement of evaluation objectivity, transmission stability and decision accuracy is achieved.

Description

Technical Field

[0001] The present invention relates to the field of information technology for knowledge imparting and skill training, and in particular to a computer room learning achievement evaluation system based on real-name binding and intelligent collection. Background Art

[0002] With the widespread adoption of information technology in knowledge and skills development, training evaluation systems are gradually shifting from single-score assessments to multi-dimensional data fusion and analysis. However, the diverse nature of knowledge and skills development scenarios (e.g., programming classes focus on code quality, theory classes emphasize test scores), the dynamic evolution of learner performance (e.g., long-term learning curve fluctuations), and the complexity of network transmission environments (e.g., unstable signals in remote areas) place higher demands on the objectivity, adaptability, and stability of evaluation systems.

[0003] Traditional knowledge and skills evaluation systems typically use a fixed-weight approach, whereby knowledge and skills assessment experts pre-set the weight ratios for each dimension (e.g., attendance, testing, and interaction) based on their experience, and generate a comprehensive score through linear weighting. At the data collection level, attendance records, test scores, and the number of interactions within training activity spaces are independently stored in different subsystems. Some systems attempt to introduce a time dimension to record the performance of a single training activity space. At the network transmission level, most systems rely on a single network protocol (e.g., Wi-Fi) to transmit interaction data. Some solutions utilize a dual-link backup mechanism, but the switching logic is triggered by a fixed threshold. Visualization modules often use static charts to display total scores or individual scores.

[0004] Existing technical solutions still have some shortcomings. First, fixed weights cannot capture the differences in data distribution across different knowledge modules. For example, code diversity in programming classes is strongly negatively correlated with test scores, but traditional models do not dynamically adjust the weights of the two, causing scores to deviate from actual ability. The present invention quantifies the statistical correlation between dimensions through covariance analysis and combines it with constrained optimization to generate dynamic weights, allowing the scoring model to adapt to the characteristics of different knowledge modules. Secondly, heterogeneous data stored in a distributed manner is separated from the time scale due to dimensional differences (such as attendance being binary and code diversity being continuous), making it difficult to construct a spatiotemporal evolution model of learner performance. The present invention integrates multi-source data into a structured matrix across the training activity space through normalization processing and time series matrix modeling, supporting long-term trend analysis. In addition, single-protocol transmission lacks adaptive switching capabilities when signals fluctuate, resulting in delays or loss of real-time interactive data (such as online assessment results). The present invention uses fuzzy logic to evaluate network status, dynamically switching protocols or enabling dual-link redundant transmission, and combines erasure coding technology to improve packet loss resistance. Finally, static charts cannot reveal the hidden performance patterns of learner groups (such as the "high interaction, low coding ability" group). This invention uses cluster analysis to identify cluster characteristics and, combined with heat maps, maps multidimensional data into spatiotemporal color codes, visually presenting individual weaknesses and group evolution patterns. Finally, traditional systems fail to effectively address device tampering or fraudulent use (such as false attendance), which can lead to data contamination. This invention uses real-name binding and an attenuation coefficient to dynamically suppress the impact of abnormal data, improving scoring credibility. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a computer room learning achievement evaluation system based on real-name binding and intelligent collection, which solves the problems of subjective scoring deviation, data integration difficulty, poor real-time interaction stability and weak targeted knowledge and skills cultivation strategy caused by fixed weight distribution, heterogeneous data fragmentation, network transmission rigidity and insufficient group pattern mining in the existing knowledge and skills evaluation system.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a computer room learning achievement evaluation system based on real-name binding and intelligent collection, the system comprising a real-name binding module, an automatic data collection module, a dynamic data matrix modeling module, a dynamic weight optimization module, a comprehensive scoring output module, a hybrid network adaptation module, and a federated learning coordinator:

[0007] The real-name binding module is used to uniquely bind the learner's identity information with the terminal device's hardware feature code, generate a device identifier, and encrypt and store it;

[0008] An automatic data collection module, which is in communication with the real-name binding module and is used to collect the learner's screen operation log, code compilation results and training activity space interactive behavior data in real time after verifying the learner's identity based on the device identifier;

[0009] The dynamic data matrix modeling module receives data from the automatic data acquisition module, uses a generative adversarial network to generate a reference code variant library, uses a discriminator to compare the learner's code logic and structural defects, and outputs a dynamic difference score. It also constructs attendance status, test scores, code difference, and the number of spatial interactions in training activities into a multidimensional data matrix in time series.

[0010] A dynamic weight optimization module calculates a covariance matrix based on the multidimensional data matrix and generates a scoring weight vector through principal component analysis and a constrained optimization algorithm;

[0011] A comprehensive scoring output module performs weighted fusion on the multidimensional data matrix according to the scoring weight vector, outputs the learner's comprehensive score, and generates knowledge and skills development strategy recommendations based on the cluster analysis results;

[0012] The hybrid network adapter module communicates with all modules and is used to coordinate the transmission priorities of real-time control instructions, batch data, and abnormal events in wireless or wired hybrid networks to ensure data synchronization between modules;

[0013] Through the blockchain smart contract executor, the scoring rules and policy generation logic are written into the chain code, triggering an unalterable execution process;

[0014] The federated learning coordinator, embedded in the hybrid network adaptation module, generates a federated scoring model by aggregating local model parameters across institutions, supporting unified evaluation standard mapping under multi-institutional data isolation.

[0015] The present invention provides a computer room learning achievement evaluation system based on real-name binding and intelligent collection. It has the following beneficial effects:

[0016] 1. The present invention automatically generates weights based on covariance analysis and constrained optimization algorithm through a dynamic weight optimization module, completely eliminating the subjective bias of manually setting weights. At the same time, the weight distribution is adjusted in real time according to the dynamic distribution characteristics of the training activity space data, so that the scoring model can adapt to the differentiated needs of different knowledge and skills training activity scenarios, ensuring that the scoring results are more scientific and reasonable.

[0017] 2. Through the dynamic data matrix modeling module, multi-dimensional heterogeneous data such as attendance, testing, interaction, and code differences are normalized and organized into time series, solving the integration difficulties caused by inconsistent data dimensions and missing time dimensions in traditional scoring systems, and providing a structured and scalable data foundation for subsequent analysis.

[0018] 3. The hybrid network adaptation module adopts multi-protocol adaptive switching and erasure code redundant transmission technology. It automatically switches to the optimal transmission protocol or enables dual-link parallel transmission in environments with network signal fluctuations or high packet loss rates, significantly reducing the delay and interruption risk of real-time interaction in training activity spaces. It is especially suitable for knowledge and skills training activities in remote areas with unstable network conditions.

[0019] 4. The comprehensive scoring output module divides learner groups into excellent / medium / poor levels through cluster analysis, and generates targeted strategic recommendations based on cluster characteristics and the degree of deviation from preset knowledge and skills development goals, helping knowledge transmitters quickly identify weak links in knowledge and skills development and achieve data-driven, precise knowledge and skills development improvements. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a system structure diagram of the present invention;

[0021] Figure 2 This is an architectural diagram of the real-name binding module of the present invention;

[0022] Figure 3 This is an architectural diagram of the automatic data acquisition module of the present invention;

[0023] Figure 4 This is an architectural diagram of the dynamic data matrix modeling module of the present invention. DETAILED DESCRIPTION

[0024] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the specification of the present invention. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0025] Please see the attached Figure 1 -Attached Figure 4 The embodiment of the present invention provides a computer room learning achievement evaluation system based on real-name binding and intelligent collection. The system includes a real-name binding module, an automatic data collection module, a dynamic data matrix modeling module, a dynamic weight optimization module, a comprehensive score output module, and a hybrid network adaptation module:

[0026] The real-name binding module is used to uniquely bind the learner's identity information with the terminal device's hardware feature code, generate a device identifier, and encrypt and store it;

[0027] The real-name binding module uses multi-level hardware feature extraction and encryption binding technology to establish a unique association between the learner's identity and the terminal device, ensuring the non-repudiation of knowledge and skills training data collection and the legitimacy of the device. The module includes a hardware feature code generation unit, an encryption binding unit, and an anomaly detection unit. The units work together as follows:

[0028] The hardware signature generation unit is used to extract unalterable hardware identification information from the terminal device and generate a unique hash value. The unit first collects the device's physical network address (MAC address) through the application program interface provided by the operating system. Preferably, the physical network address is a globally unique identifier hardened to the device's network card at the factory, such as the MAC address of an Ethernet or wireless network card.

[0029] At the same time, the unit reads the serial number of the central processing unit (CPU) through the underlying instruction set. Preferably, the serial number is an unmodifiable identification code written into the chip by the CPU manufacturer. To avoid the risk of privacy leakage caused by directly storing the original hardware information, the unit hashes the above hardware identification information: first, a hash value of the CPU serial number is generated, and its calculation formula is:

[0030] H cpu =SHA-256(CPUSerial);

[0031] SHA-256(·) is the SHA-256 hash function, which outputs a 256-bit hexadecimal string. cpu is the hash value of the CPU serial number; CPUSerial is the original CPU serial number string.

[0032] Furthermore, the MAC address and the CPU serial number hash value are concatenated and hashed twice to generate a unique device identifier:

[0033] H dev =SHA-256(M mac ∥H cpu );

[0034] Among them, M mac is the MAC address of the learner's terminal device; H cpu is the hash value of the CPU serial number; ∥ is the string concatenation operator; H dev is the unique device identifier used for learner-device binding; SHA-256(·) is the SHA-256 hash function, which outputs a 256-bit hexadecimal string.

[0035] The encryption binding unit is used to store the learner's student ID and device identifier in irreversible encryption. The unit receives the learner's student ID S provided by the educational system. id , and compare it with the device's unique identifier Hdev Encrypted using the Advanced Encryption Standard (AES) algorithm to generate ciphertext data:

[0036] E=AES-256(S id ∥H dev ,Key);

[0037] Among them, Key is the 256-bit encryption key; H dev is the bound device identifier stored in the database; E is the encrypted binding relationship data, which is stored in the database and cannot be restored to the original student number or device identifier through reverse engineering.

[0038] Preferably, the key is stored in a segmented manner, with some keys stored in a secure hardware module on the server side and others injected by the system administrator during initialization. This encryption process ensures that even if the database is compromised, attackers cannot directly obtain the binding between the learner's ID and the device.

[0039] The anomaly detection unit is used to verify the legitimacy of the terminal device in real time. When the learner logs into the system, the unit re-collects the MAC address and CPU serial number of the current device and generates a real-time device identifier H′ according to the above hash process. dev By comparing the real-time identifier H′ dev H stored in the database dev , determine whether the device has changed:

[0040] H′ dev ≠H dev ;

[0041] Among them, H′ dev The real-time calculated identifier of the current device; H dev The bound device identifier stored in the database.

[0042] Preferably, the abnormal event record includes a timestamp, an original device identifier, a real-time identifier, and a network IP address, and is reported to the knowledge imparter management terminal in real time through an independent communication channel.

[0043] Furthermore, when the unit detects a device anomaly, it will simultaneously perform the following operations: temporarily freeze the data upload permission of the current terminal and push a secondary authentication request to the learner terminal. Preferably, the secondary authentication uses a dynamic password or biometric recognition (such as camera face comparison) to distinguish between device loss and malicious use scenarios.

[0044] The hardware signature generation unit solves the problem of single hardware identifiers being easily tampered with through double hashing and concatenation operations, ensuring the uniqueness and anti-collision of device identification.

[0045] The encryption binding unit uses the AES-256 algorithm and key segmentation management to prevent the learner-device binding relationship from being reverse cracked or leaked in batches;

[0046] The anomaly detection unit uses real-time hash comparison and dynamic verification mechanisms to quickly identify device change behaviors and isolate risks, providing a basic guarantee for the reliability of subsequent data collection.

[0047] The automatic data collection module is connected to the real-name binding module and is used to collect the learner's screen operation log, code compilation results and training activity space interactive behavior data in real time after verifying the learner's identity based on the device identifier;

[0048] The automatic data collection module uses event triggering mechanism and code semantic analysis technology to capture learner terminal operation data in real time and quantify their programming behavior characteristics. The module includes event triggering unit, code analysis unit and difference calculation unit. The technical implementation method is as follows:

[0049] The event triggering unit is used to drive screenshot capture and data collection according to preset rules. The unit monitors code compilation completion events on the learner's terminal through the underlying operating system interface. Preferably, the compilation completion event is a compilation success signal returned by the integrated development environment (IDE) or an executable file generation operation. When such an event is detected, the current screen image is immediately captured and saved as a lossless compressed image file.

[0050] The unit also periodically triggers screenshots, preferably at intervals that avoid conflicts with learners' active actions. The resolution of the captured images matches the native screen resolution, ensuring clear recognition of code text. The screenshot operation is combined with event-driven data collection to ensure both real-time performance and data integrity.

[0051] The code analysis unit is used to extract code text from the screen image and parse it into a structured syntax tree. The unit first uses an optical character recognition (OCR) engine to locate and convert the code area in the screenshot. Preferably, the OCR engine uses an end-to-end recognition model based on convolutional neural networks and supports multiple programming language character sets (such as C / C++ and Python special symbols). The recognition results are output as a plain text code file encoded in UTF-8, preserving the original indentation and syntax structure.

[0052] Furthermore, the unit converts the text code into an abstract syntax tree (AST) through a syntax parser. Preferably, the parser generates a tree structure based on the grammatical rules of the target programming language, where each node represents a code element (such as a variable declaration, a function call, or a control statement). During the parsing process, the unit records the following key node information:

[0053] The name, data type and scope of the variable declaration node;

[0054] The name, parameter list, and return value type of the function call node;

[0055] Conditional expressions and branch paths of control flow nodes.

[0056] The difference calculation unit is used to quantify the structural and logical differences between the learner code and the reference code. The unit receives the learner code AST and the reference code AST preset by the knowledge imparter, and calculates the difference index using the following formula:

[0057] Structural difference:

[0058]

[0059] Among them, AST stu Abstract syntax tree for learner code; AST ref The abstract syntax tree of the reference code provided to the knowledge imparter; N(·) is the total number of nodes in the abstract syntax tree; D struct It is the structural difference, reflecting the difference in code form.

[0060] Logical difference:

[0061]

[0062] Where CF(·) is the set of control flow paths of the code; ∩ is the set intersection operator; N paths is the total number of control flow paths in the reference code; D logic is the logical difference.

[0063] The event trigger unit avoids the resource waste of traditional polling collection by coordinating compilation events with periodic screenshots, ensuring the integrity of the code submission stage.

[0064] The code analysis unit uses cascade processing of OCR and syntax parser to solve the problem that screenshots cannot directly obtain code text, while retaining code semantic information through AST parsing;

[0065] The difference calculation unit overcomes the defect of traditional row comparison method in being unable to identify code logic errors through multi-dimensional quantification of structural difference and logical difference, providing reliable input for subsequent dynamic weight scoring.

[0066] The dynamic data matrix modeling module receives data from the automatic data acquisition module, uses a generative adversarial network to generate a reference code variant library, uses a discriminator to compare the learner's code logic and structural defects, and outputs a dynamic difference score. It also constructs attendance status, test scores, code difference, and the number of spatial interactions in training activities into a multidimensional data matrix in time series.

[0067] In this embodiment, this module receives the original operation logs, voice interaction data, and code submission records from the automatic data acquisition module, and generates a standardized spatiotemporal data matrix by combining a generative adversarial network with statistical analysis. Its technical implementation and collaborative logic are as follows:

[0068] Input preparation: Extract the reference code of the target function (such as the sorting algorithm implementation) from the course knowledge graph, parse its abstract syntax tree (AST) and control flow graph (CFG) as the input benchmark for the generative adversarial network.

[0069] Generator operation: The generator based on the Transformer architecture performs syntax-level mutation on the reference code to generate candidate codes that are logically equivalent but structurally different;

[0070] Perform syntax parsing and unit testing on the generated code, and only retain variant codes that pass compilation and are functionally equivalent and store them in the variant library;

[0071] Input the learner code and the reference code into the symbolic execution engine to generate a set of path constraints Φ 学习 and Φ 参考 , calculate the deviation of the constraint coverage of the two:

[0072]

[0073] Among them, Φ 参考 is the path constraint set of the reference code; Φ 学习 is the path constraint set of the learner code; |Φ 参考 | is the cardinality of the reference code path constraint set, representing the number of all logical paths that the reference code should cover ideally; |Φ 学习 ∩Φ 参考 | is the intersection cardinality of the learner code and the reference code path constraint set, indicating the number of expected logical paths of the reference code that the learner code correctly covers; S 逻辑 Score logical flaws.

[0074] AST difference calculation: extract the AST of the learner code, compare it with the AST of each reference variant in the variant library by tree edit distance, and take the minimum similarity as the structural defect score;

[0075] Dynamically adjust the weight ratio of logical and structural defects according to the course objectives to generate a comprehensive difference score:

[0076] S 差异 =α·S 逻辑 +β·S 结构 ;

[0077] Among them, S 差异is the comprehensive difference score between the learner code and the reference code variant library; α is the logical defect score weight coefficient; S 逻辑 is the logic defect score; β is the structural defect score weight coefficient; S 结构 Score structural defects.

[0078] If the learner code has serious errors (such as dead loops, unhandled exceptions) during symbolic execution or test case verification, S is directly assigned. 差异 =1 (the highest defect level), and trigger a real-time alarm to the teacher's end.

[0079] The data normalization unit is used to map heterogeneous data to a unified dimension range, eliminating the impact of dimensional differences on scoring. The unit receives the original data stream verified by the real-name binding module, including attendance status, test scores, number of interactions in the training activity space, and code difference indicators, and normalizes them according to the following rules:

[0080] Attendance status: binary value conversion, absence is marked as 0, attendance is marked as 1. Preferably, absence is determined based on the number of consecutive unresponsive heartbeat packets in the device login log exceeding a preset threshold;

[0081] Test score: linearly normalized to the range of 0,1, the calculation formula is:

[0082]

[0083] Among them, S raw It is the learner’s original score in a single training activity space test, usually in percentage or fixed score. min The lowest score of all learners in the class on the same test, dynamically calculated within the current training activity space data; S max The highest score of all learners in the class on the same test, used to calibrate the normalized benchmark; S norm It is the normalized test score, mapped to the range of 0 and 1.

[0084] Preferably, the score range is dynamically calculated based on a single training activity space test to avoid interference from historical data;

[0085] Interaction times in the training activity space: The hyperbolic tangent function is used to suppress abnormally high activity behaviors. The calculation formula is:

[0086] a interact =tanh(N raise +N ask );

[0087] Among them, N raise N is the number of times a learner raises his hand in a single training activity space; askis the number of questions asked by the learner in a single training activity space; tanh(·) is the hyperbolic tangent function; a interact is the normalized spatial interaction index of training activities.

[0088] Preferably, this design avoids a small number of learners artificially increasing their scores through frequent ineffective interactions;

[0089] Code difference: directly reference the structural difference D output by the data automatic acquisition module struct Logical difference D logic , retaining its original value range 0,1.

[0090] The matrix expansion unit is used to organize data in time series, constructing a dynamically growing multidimensional matrix. After each lesson, a new row of data is added to the unit, with fixed column dimensions: [Attendance Status, Test Score, Structural Difference, Logical Difference, Training Activity Space Interaction Value]. The data matrix for a single learner is represented as:

[0091]

[0092] Among them, k2 is the learner number; n2 is the number of times the class has been attended.

[0093] Preferably, the matrix adopts a sparse storage format, retaining only non-zero elements and their coordinates to save storage space.

[0094] The abnormal marking unit is used to identify and correct data deviations caused by device abnormalities. dev ≠H dev ), the unit applies an attenuation coefficient to the data row corresponding to the abnormal knowledge module:

[0095]

[0096] Where λ is the attenuation factor; is the data vector of all columns in the i-th row of the data matrix of the k-th learner;

[0097] Preferably, the factor is set to a constant less than 1 to reduce the weight contribution of abnormal data to the score. Preferably, the decay operation is performed before the data is written into the matrix to ensure that the subsequent processing flow is not disturbed.

[0098] The data normalization unit solves the heterogeneity problem of attendance, test, interaction and other data through multi-strategy mapping to ensure the comparability of data within the matrix; the matrix expansion unit adopts a time series organization method to retain the dynamic evolution characteristics of learners' performance, providing a time dimension analysis basis for subsequent covariance-based weight optimization; the anomaly marking unit isolates the interference of abnormal device data through the attenuation coefficient to improve the robustness of the scoring system. Its technical implementation relies on the real-time verification results of the real-name binding module.

[0099] Dynamic weight optimization module calculates the covariance matrix based on the multidimensional data matrix and generates the scoring weight vector through principal component analysis and constrained optimization algorithm;

[0100] The dynamic weight optimization module dynamically generates scoring weights through covariance analysis and constrained optimization algorithms, eliminating the subjective bias of manually set weights and ensuring that the scoring model adapts to the data distribution characteristics of different training activity spaces. The module includes a covariance calculation unit, a principal component analysis unit, and a constrained optimization unit. Its technical implementation and data interaction logic are as follows:

[0101] The covariance calculation unit is used to quantify the statistical correlation between the data of each dimension and provide a mathematical basis for weight optimization. The unit receives the matrix set of all learners in the class {M (1) ,M (2) ,…,M (K)}, where each learner matrix Contains five columns of data: attendance, test scores, structural differences, logical differences, and training activity space interaction values.

[0102] The unit first calculates the global mean vector μ = [μ1, μ2, μ3, μ4, μ5] for each column of data.

[0103] The unit then computes the covariance matrix across learners and across the training activity space Its element C pq Defined as:

[0104]

[0105] Among them, C pq is the element in the pth row and qth column of the covariance matrix; is the normalized value of the i-th row and p-th column in the dynamic data matrix of the k-th learner; μ p and μ q represent the means of the pth and qth dimensions respectively; K is the total number of learners; n is the number of rows of training activity space data for a single learner; and K·n-1 is the unbiased estimate correction term for covariance calculation.

[0106] Preferably, the covariance matrix reflects the correlation between the data in each dimension, such as the negative correlation between test scores and code diversity (high-scoring learners usually have low code diversity).

[0107] The principal component analysis unit is used to extract the main direction of change in the data and generate the initial weight vector. The unit performs eigenvalue decomposition on the covariance matrix C, solving its eigenvalues λ1≥λ2≥…≥λ5 and the corresponding eigenvectors v1,v2,…,v5. The eigenvector v1 corresponding to the largest eigenvalue is selected as the initial weight;

[0108] Preferably, the principal component direction represents the projection direction where the data variance is the largest, so that the scoring model can capture the main differences in learner performance.

[0109] The constrained optimization unit is used to impose reasonable constraints on the initial weights to generate the final optimized weights. The objective function of the constrained optimization unit is:

[0110]

[0111] Among them, w j is the weight of the j-th dimension; is the normalized value of the jth dimension of the i-th lesson of the k-th learner; K is the total number of learners; and n is the number of rows of training activity space data for a single learner.

[0112] The covariance calculation unit identifies key influencing dimensions (such as the strong correlation between code differences and test scores) by statistically analyzing the correlation of class data, providing a data-driven basis for weight allocation.

[0113] The comprehensive scoring output module performs weighted fusion on the multidimensional data matrix according to the scoring weight vector, outputs the learner's comprehensive score, and generates knowledge and skills development strategy recommendations based on the cluster analysis results;

[0114] The comprehensive scoring output module uses multi-dimensional weighted fusion and cluster analysis technology to generate a comprehensive score of learners' training activity space performance and knowledge and skills development strategy recommendations, providing data-driven decision support for knowledge imparters. The module includes a weighted summation unit, a cluster analysis unit, a strategy recommendation unit, and a heat map visualization unit. Its technical implementation and data interaction logic are as follows:

[0115] The weighted summation unit is used to integrate the weight vector generated by the dynamic weight optimization module with the learner's dynamic data matrix to calculate the comprehensive score. The unit receives the optimal weight vector W output by the dynamic weight optimization module. opt =[w1,w2,w3,w4,w5], where w1 to w5 correspond to the weights of attendance status, test score, structural difference, logical difference, and training activity space interaction value, respectively.

[0116] For the K-th learner, the comprehensive score calculation formula is:

[0117]

[0118] Among them, Score is the learner's comprehensive score, reflecting the weighted performance of multi-dimensional data under dynamic weights; K is the total number of learners; n is the number of rows of training activity space data for a single learner; W opt,j is the optimal weight of the j-th dimension; is the normalized value of the i-th row and j-th column in the data matrix of the k-th learner.

[0119] The cluster analysis unit is used to cluster the performance of learners and identify the direction of knowledge and skills cultivation and improvement. The unit collects the dynamic data matrix of all learners in the class {M (1) ,M (2) ,…,M (K) Input the clustering algorithm. Preferably, the K-means algorithm is used with the preset number of categories being 3 to calculate the category label of each learner and the coordinates of the center of each cluster.

[0120] During clustering, the unit calculates the Euclidean distance between the learner matrix and the cluster center:

[0121]

[0122] in, represents the value of the i-th row and j-th column of the center matrix of the c-th cluster; d (k,c) is the Euclidean distance between learner k and cluster c, which is used to determine the performance category to which the learner belongs; i is the index of the training activity space; j is the dimension index.

[0123] The strategy recommendation unit generates knowledge and skills training adjustment suggestions based on the clustering results. The unit calculates the Euclidean distance between each cluster center and the preset reference center (such as the center of the top students cluster). like This triggers the generation of knowledge and skills cultivation strategy recommendations. Preferably, the reference center is trained by historical excellent learner data, and the threshold θ is dynamically set according to the knowledge and skills cultivation activity scenario.

[0124] The heat map visualization unit maps the learner matrix data to color intensity. The rendering formula is:

[0125] Color(a ij )=(255(1-a ij ),255a ij ,0);

[0126] Among them, Color(a ij ) is the output of the entire color mapping function; a ijis the normalized value of row i and column j in the learner data matrix; 255(1-a ij ) is the intensity value of the red channel; 255a ij The intensity value of the green channel; 0 means the blue channel is fixed to 0 and does not participate in color rendering.

[0127] The hybrid network adapter module communicates with all modules and is used to coordinate the transmission priorities of real-time control instructions, batch data, and abnormal events in wireless or wired hybrid networks to ensure data synchronization between modules;

[0128] The hybrid network adaptation module uses multi-protocol adaptive switching and transmission path optimization technology to address latency, jitter, and packet loss issues in data transmission for knowledge and skills training terminals in heterogeneous network environments, ensuring the stability of real-time training activity space interactions. The module includes a network status monitoring unit, a protocol switching decision unit, and a transmission optimization unit. Its technical implementation and collaborative logic are as follows:

[0129] The network status monitoring unit is used to collect network environment parameters of terminal devices in real time, providing data support for protocol switching and path selection. The unit periodically detects the link quality indicators of the currently connected wireless network (such as Wi-Fi, 5G) and wired network, including:

[0130] Signal Strength (RSSI): The received signal strength indicator value is obtained through the physical layer driver interface. Preferably, the value is smoothed by Kalman filtering to eliminate instantaneous fluctuations;

[0131] End-to-end delay: The round-trip time (RTT) is measured by sending ICMP probe packets, and the one-way delay τ = RTT2 is calculated;

[0132] Packet loss rate: The ratio of TCP retransmission times to total transmission times within a statistical unit of time. Preferably, the statistical window is set to dynamically adjust to adapt to network fluctuations.

[0133] The unit encapsulates the above parameters into a network state vector Q = [RSSI, τ, p loss ], where p loss is the packet loss rate and is pushed to the protocol switching decision unit.

[0134] The protocol switching decision unit triggers transmission protocol switching or dual-link parallel transmission based on the network status assessment results and preset thresholds. The unit uses a fuzzy logic evaluation model, defining the input variable as Q and the output variable as the protocol switching action A∈{maintain, switch to backup protocol, dual-link transmission}.

[0135] The core of the evaluation model is the membership function and rule base:

[0136] Membership function: Defines three fuzzy sets for each network parameter: "Excellent," "Medium," and "Poor." Fuzzy rule base: Contains combination rules such as "If RSSI is excellent and latency is excellent, maintain the current protocol." Rule weights are adjusted based on historical handover success rate feedback.

[0137] The transmission optimization unit is used to optimize the data transmission path and redundancy strategy under the selected protocol. When dual-link transmission is enabled, the unit divides the data stream into multiple sub-blocks and transmits them in parallel through the main link (such as Wi-Fi) and the backup link (such as 5G). The receiving end reassembles the data based on the erasure code. Preferably, the factor is dynamically adjusted according to the real-time packet loss rate.

[0138] Define transfer priority rules:

[0139]

[0140] Among them, Priority is the transmission priority of the data packet; 0 is the multidimensional training activity space data generated by the dynamic data matrix modeling module; 1 is the device identifier mismatch or hardware tampering event; 2 is the real-time control instruction sent by the knowledge transmitter.

[0141] The continuous changes of network status are quantified by fuzzy sets to avoid the hard switching jitter problem of traditional threshold methods. It relies on the real-time evaluation of the network state vector Q and forms a closed-loop feedback with the status monitoring unit to ensure the timeliness of the switching action.

[0142] Through the blockchain smart contract executor, the scoring rules and policy generation logic are written into the chain code, triggering an unalterable execution process;

[0143] In this embodiment, the blockchain smart contract executor implements tamper-proof execution of scoring rules and policies through chaincode logic encapsulation and distributed ledger technology. Specifically, the executor receives the scoring weight vector generated by the dynamic weight optimization module and the policy generation conditions from the comprehensive scoring output module, converts them into verifiable smart contract code, and deploys them to the blockchain network, forming an automated execution system under the full-node consensus mechanism.

[0144] The chain code logic injection unit uses modular programming to convert the scoring rules into chain code executable logic. Scoring weight vector W=(w1,w2,...,w d ) is written into the chaincode state database through serialization, where the weight coefficients of each dimension strictly correspond to the column dimensions of the dynamic data matrix. The policy generation logic is encoded as the chaincode transaction trigger condition. When a new data row or device identifier anomaly in the dynamic data matrix is detected, the scoring calculation function is automatically called:

[0145]

[0146] Among them, w i represents the normalized indicator value of the i-th dimension in the dynamic data matrix, including attendance status, test score, logic defect score, structural defect score and interaction index; x i is the normalized index value of the i-th dimension; d is the total number of column dimensions of the multidimensional data matrix.

[0147] Preferably, the chain code imposes normalization constraints on the weight coefficients to ensure The mathematical conditions of are automatically enforced at every weight update.

[0148] The on-chain and off-chain storage units use a hierarchical storage architecture to achieve verifiability and scalability of scoring rules. The chain code only stores the scoring weight vector hash value H(W) = SHA256(W) and the bytecode fingerprint of the policy triggering logic. The complete weight data and policy rules are processed by the encryption algorithm and stored in the decentralized file system. Preferably, a content-addressable identifier is used to establish an index association between the on-chain hash value and the off-chain data. The specific generation method is as follows:

[0149] CID=Multihash(Encrypt(W,K pub ));

[0150] Among them, K pub The public key assigned to the federated learning coordinator ensures the controllability of cross-institutional data calls; W is the scoring weight vector generated by the dynamic weight optimization module; Encrypt(W,K pub ) is a public key encryption function; Multihash(·) is a multiple hash function.

[0151] When the integrity of the scoring rules needs to be verified, the off-chain data is retrieved through the CID and decrypted, and the hash value is recalculated to match the on-chain record.

[0152] The cross-chain verification unit implements multi-chain data consistency verification through a lightweight node protocol. When the federated learning coordinator aggregates the scoring models of each participant, the smart contract executor submits the local chaincode rule hash list to the cross-chain relay, where Merkle tree proofs are used to verify the consistency of each node's chaincode logic with the global constraints. Preferably, a relay chain model is used to achieve secondary consensus on the verification results of heterogeneous blockchain networks to prevent model bias caused by data tampering on a single chain.

[0153] In this embodiment, the smart contract executor and the hybrid network adapter module form a closed-loop exception handling system. When a device identifier change event is detected, the chaincode automatically triggers the following transaction flow: first, the scoring output channel of the associated learner is frozen; second, the abnormal event characteristics (including the device hardware fingerprint change record and the threshold-exceeding difference between the logical and structural defect scores) are written to the blockchain's immutable ledger; and finally, a manual review request is pushed to the knowledge imparter. The freeze persists until the federated learning coordinator verifies the authenticity of the node's historical data, at which point a majority of nodes vote to lift the freeze.

[0154] The interaction between the executor and the dynamic data matrix modeling module is implemented through an event-driven mechanism. Event listeners for newly added data rows in the matrix trigger chaincode transaction invocations in real time. After the score calculation is completed within the blockchain network, the resulting hash value is written back to the matrix metadata field. Preferably, an asynchronous callback mechanism is used to avoid chaincode execution blocking caused by high-concurrency data transmission. Timestamp signatures are used to ensure temporal consistency between data rows and score results.

[0155] The federated learning coordinator, embedded in the hybrid network adaptation module, generates a federated scoring model by aggregating local model parameters across institutions, supporting unified evaluation standard mapping under multi-institutional data isolation.

[0156] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. The computer room learning achievement evaluation system based on real-name binding and intelligent collection is characterized by: The system includes a real-name binding module, an automatic data collection module, a dynamic data matrix modeling module, a dynamic weight optimization module, a comprehensive scoring output module, a hybrid network adaptation module, and a federated learning coordinator: The real-name binding module is used to uniquely bind the learner's identity information with the terminal device's hardware feature code, generate a device identifier, and encrypt and store it; An automatic data collection module, which is in communication with the real-name binding module and is used to collect the learner's screen operation log, code compilation results and training activity space interactive behavior data in real time after verifying the learner's identity based on the device identifier; The dynamic data matrix modeling module receives data from the automatic data acquisition module, uses the generative adversarial network to generate a reference code variant library, compares the learner's code logic and structural defects through the discriminator, and outputs a dynamic difference score; The attendance status, test scores, code difference and number of spatial interactions of training activities were constructed into a multidimensional data matrix according to time series; A dynamic weight optimization module calculates a covariance matrix based on the multidimensional data matrix and generates a scoring weight vector through principal component analysis and a constrained optimization algorithm; A comprehensive scoring output module performs weighted fusion on the multidimensional data matrix according to the scoring weight vector, outputs the learner's comprehensive score, and generates knowledge and skills development strategy recommendations based on the cluster analysis results; The hybrid network adapter module communicates with all modules and is used to coordinate the transmission priorities of real-time control instructions, batch data, and abnormal events in wireless or wired hybrid networks to ensure data synchronization between modules; Through the blockchain smart contract executor, the scoring rules and policy generation logic are written into the chain code, triggering an unalterable execution process; The federated learning coordinator, embedded in the hybrid network adaptation module, generates a federated scoring model by aggregating local model parameters across institutions, supporting unified evaluation standard mapping under multi-institutional data isolation.

2. The computer room learning achievement evaluation system based on real-name binding and intelligent collection according to claim 1 is characterized in that: The real-name binding module includes: Hardware signature generation unit, used to collect the MAC address and CPU serial number of the device and generate a hash value H dev =SHA-256(M mac ∥H cpu ); Among them, M mac is the MAC address of the learner's terminal device; H cpu is the hash value of the CPU serial number; ∥ is the string concatenation operator; H dev is the unique device identifier used for learner-device binding; SHA-256(·) is the SHA-256 hash function, which outputs a 256-bit hexadecimal string; Encrypted binding unit, learner ID S id With H dev Encrypted storage using the AES algorithm; The anomaly detection unit detects the current device hash value H′ dev ≠H dev When , the abnormal event record is triggered, where H′ dev The real-time calculated identifier of the current device; H dev The bound device identifier stored in the database.

3. The computer room learning achievement evaluation system based on real-name binding and intelligent collection according to claim 1 is characterized in that: The automatic data acquisition module includes: Event trigger unit, triggering screenshot at preset time interval ΔT or code compilation completion event; The code analysis unit uses OCR to recognize the code text in the screenshot and parse it into an abstract syntax tree; Difference calculation unit, calculates the structural difference D between the learner code and the reference code struct and logical difference D logic .

4. The computer room learning achievement evaluation system based on real-name binding and intelligent collection according to claim 3 is characterized in that: In the difference calculation unit: Structural difference D struct is the normalized value of the difference in the number of AST nodes: Among them, AST stu Abstract syntax tree for learner code; AST ref The abstract syntax tree of the reference code provided to the knowledge imparter; N(·) is the total number of nodes in the abstract syntax tree; D struct is the structural difference, reflecting the difference in code form; Logical difference D logic is the complement of the control flow path matching rate: Where CF(·) is the set of control flow paths of the code; ∩ is the set intersection operator; N paths is the total number of control flow paths in the reference code; D logic is the logical difference.

5. The computer room learning achievement evaluation system based on real-name binding and intelligent collection according to claim 1 is characterized in that: The dynamic data matrix modeling module includes: Based on the reference code in the course knowledge graph, a generative adversarial network is used to generate a library of logically equivalent but syntactically different code variants. The library covers three typical programming features: loop nesting optimization, exception handling completeness, and code reuse rate. The following comparative analysis is performed on the code submitted by the learners: Logic defect detection: Calculate the deviation between the learner code and the reference code in the control flow graph path coverage; Structural defect detection: Compare the abstract syntax tree similarity between the learner code and the generator variant library to identify redundant or inefficient structures; Output dynamic difference scores, where the weight of logical defects is 0.6 and the weight of structural defects is 0.4; The data normalization unit maps attendance status, test scores, code differences, and the number of training activity space interactions to the interval [0,1]. The number of training activity space interactions is suppressed by the hyperbolic tangent function: a interact =tanh(N raise +N ask ); Among them, N raise N is the number of times a learner raises his hand in a single training activity space; ask is the number of questions asked by the learner in a single training activity space; tanh(·) is the hyperbolic tangent function; a interact is the normalized spatial interaction index of training activities; Matrix expansion unit, which adds a row of data to the learner matrix for each class, with column dimensions including attendance, test, structural difference, logical difference, and interaction; The abnormality marking unit adds a decay coefficient λ = 0.5 to the training activity space data of equipment changes.

6. The computer room learning achievement evaluation system based on real-name binding and intelligent collection according to claim 1 is characterized in that: The dynamic weight optimization module includes: Covariance calculation unit, calculates the covariance matrix based on the matrix set of all learners in the class Its elements are: Among them, C pq is the element in the pth row and qth column of the covariance matrix; is the normalized value of the i-th row and p-th column in the dynamic data matrix of the k-th learner; μ p and μ q Represent the means of the pth and qth dimensions respectively; K is the total number of learners; n is the number of rows of training activity space data for a single learner; K·n-1 is the unbiased estimate correction term for covariance calculation; The principal component analysis unit extracts the eigenvector V1 corresponding to the maximum eigenvalue of the covariance matrix as the initial weight W pca ; Constrained optimization unit, constructs the Lagrangian function and solves the optimal weight W to maximize the objective function opt .

7. The computer room learning achievement evaluation system based on real-name binding and intelligent collection according to claim 6 is characterized in that: The objective function of the constraint optimization unit is: Among them, w j is the weight of the j-th dimension; is the normalized value of the jth dimension of the i-th class of the k-th learner; K is the total number of learners; n is the number of rows of training activity space data for a single learner.

8. The computer room learning achievement evaluation system based on real-name binding and intelligent collection according to claim 1 is characterized in that: The comprehensive score output module includes: The weighted summation unit performs weighted accumulation of matrix row data lesson by lesson according to the optimal weight: Among them, Score is the learner's comprehensive score, reflecting the weighted performance of multi-dimensional data under dynamic weights; K is the total number of learners; n is the number of rows of training activity space data for a single learner; W opt,j is the optimal weight of the j-th dimension; is the normalized value of the i-th row and j-th column in the data matrix of the k-th learner; Cluster analysis unit, identifying abnormal learner groups through K-means algorithm, with the number of clusters k=3, where k=3 is the number of clusters of K-means algorithm; The strategy recommendation unit outputs knowledge and skill cultivation adjustment suggestions when the Euclidean distance between the class center and the reference center exceeds the threshold θ, where θ is the Euclidean distance threshold between the class center and the reference center.

9. The computer room learning achievement evaluation system based on real-name binding and intelligent collection according to claim 1 is characterized in that: The hybrid network adaptation module includes: Protocol selection unit, real-time control instructions are transmitted via UDP broadcast, and matrix incremental data are transmitted via TCP fragment compression; Priority management unit, defines the transmission priority rules: Among them, Priority is the transmission priority of the data packet; 0 is the multidimensional training activity space data generated by the dynamic data matrix modeling module; 1 is the device identifier mismatch or hardware tampering event; 2 is the real-time control instruction sent by the knowledge transmitter.

10. The computer room learning achievement evaluation system based on real-name binding and intelligent collection according to claim 1 is characterized in that: The triggering of the tamper-proof execution process includes: The chain code logic injection unit is used to write the scoring weight vector W generated by the dynamic weight optimization module and the policy generation conditions of the comprehensive scoring output module into the blockchain chain code, and define the following tamper-proof triggering rules: When a new data row is added to the dynamic data matrix modeling module, the chain code is triggered to perform the scoring calculation And store the resulting hash value on the chain, where w i The weight coefficient of the i-th dimension generated by the dynamic weight optimization module; x i is the normalized index value of the i-th dimension in the dynamic data matrix modeling module; d is the total number of column dimensions of the multidimensional data matrix; When the hybrid network adapter module detects an abnormal event in the device identifier, the chain code automatically freezes the scoring output of the associated learner and generates a manual review request event; The on-chain-off-chain storage unit is used to encrypt and store the original scoring rule data in IPFS. Only the IPFS content identifier and the rule hash value are retained in the chain code. The off-chain data can be verified and queried through the rule hash value index; Cross-chain verification unit: When the federated learning coordinator aggregates cross-institutional scoring models, it verifies the consistency of the chaincode rules of each institution through the cross-chain protocol to ensure that the generation of federated model weights meets the preset constraints.

Images