Blockchain-based method for technician training and project dynamic authorization engine

Abstract

The present application proposes a blockchain-based technical personnel training and project dynamic authorization engine method, which comprises: obtaining the original training event sequence of the technical personnel; constructing a decision vector according to the training data chain, calculating an authorization adaptation score through a smart contract, obtaining a permission ID and a record change log chain; real-time monitoring of the permission change log, risk assessment through a permission freezing decision model, calculation of a freezing adaptation score, triggering a smart contract to freeze the permission when the freezing adaptation score exceeds a threshold, generating a warning notice and recording on the chain; constructing a feedback vector based on the permission freezing record, and calculating an optimization adaptation score through a certificate optimization decision model, and issuing a certificate by a smart contract after reaching the standard, and storing the certificate as an NFT, and synchronously adjusting the task role of the technical personnel and storing it on the chain. The present application improves the flexibility, efficiency and transparency of the system, and can meet the needs of modern enterprises for efficient, accurate and traceable management.

Description

Technical Field

[0001] This invention belongs to the field of blockchain, and in particular relates to a blockchain-based method for training technical personnel and dynamic authorization engine for projects. Background Technology

[0002] With the rapid development of information technology and artificial intelligence, technical personnel training and project authorization management have become critical issues for many enterprises and institutions. Existing technical personnel training systems are typically based on traditional offline or online training platforms, employing fixed curriculum structures and task allocation methods. These systems lack flexibility and personalization, making it difficult to plan precise learning paths based on the actual needs and capabilities of technical personnel. Traditional project authorization management systems are also often static; once permissions are set, they are difficult to adjust dynamically. This results in technical personnel not receiving tasks and project resources that match their abilities in real time, impacting not only their work efficiency but also reducing the overall project execution effect. Furthermore, existing systems have certain deficiencies in data security, access control, and task traceability. Most data records and task authorizations lack transparency and traceability, making them susceptible to human intervention and tampering, thus affecting trust and system fairness. Blockchain technology, due to its decentralized, immutable, transparent, and traceable characteristics, is gradually being applied to data management and access control. However, its practical application in technical personnel training and project authorization management still faces many challenges. Existing blockchain technology applications mostly remain at the level of data recording and transaction verification. They have not yet effectively solved the problem of dynamic authorization of technical personnel and personalized training path design, nor have they made full use of the existing smart contracts and decentralized features to achieve real-time and intelligent training and authorization decisions.

[0003] Therefore, how to combine emerging technologies, especially the combination of blockchain and artificial intelligence, to achieve intelligent and personalized training for technical personnel and dynamically adjust project authorization has become an urgent problem to be solved. Summary of the Invention

[0004] The purpose of this invention is to propose a blockchain-based method for dynamic authorization of technical personnel training and projects. This method not only solves the shortcomings of traditional systems in terms of permission management, customized training, and dynamic adjustment, but also greatly improves the flexibility, efficiency, and transparency of the system, meeting the needs of modern enterprises for efficient, accurate, and traceable management.

[0005] To achieve the above objectives, a first aspect of the present invention provides a blockchain-based method for technical personnel training and dynamic project authorization engine, the method comprising the following steps:

[0006] S1. Obtain the original training event sequence of technical personnel, process dynamic interference through attention mechanism and LSTM model, generate refined vectors, associate them with technical personnel identifiers and upload them to the blockchain for notarization, form a training data chain, and output the assessment results.

[0007] S2. Construct a decision vector based on the training data chain, calculate the authorization adaptation score through a smart contract, generate an authorization token when the authorization adaptation score reaches a threshold, update the access permissions and store them on the chain, and obtain the authorization ID and record the change log chain.

[0008] S3. Monitor permission change logs in real time, assess risks through permission freeze decision model, calculate freeze adaptation score, and trigger smart contract freeze permission when the freeze adaptation score exceeds the threshold, generate early warning notification and record it on the chain to obtain risk ID for identifying freeze instance.

[0009] S4. Construct a feedback vector based on the permission freeze record, and calculate the optimization adaptation score through the certificate optimization decision model. After the target is met, the smart contract issues a certificate with a unique identifier for NFT storage, and the task roles of technical personnel are adjusted and stored on the blockchain simultaneously; wherein, the feedback vector is the risk ID and early warning notification.

[0010] Furthermore, the original training event sequence includes a unique identifier for the technical personnel, a sequence of interaction intensity vectors for course modules, and assessment response records;

[0011] Specifically, the process of handling dynamic interference through an attention mechanism and an LSTM model to generate refined vectors, which are then associated with technical personnel identifiers and uploaded to the blockchain for evidence storage, forming a training data chain, is as follows:

[0012] The original training event sequence is integrated into a unified input vector;

[0013] The input vector is fed into the attention mechanism and LSTM model to generate a refined vector; wherein, the attention mechanism and LSTM model are specifically:

[0014] In the multi-head attention layer, three attention heads jointly calculate the attention score of a node to obtain the attention score of the corresponding node; the attention weighted sequence is input into the interference-adaptive LSTM unit for state update to obtain the hidden state of the current time step;

[0015] If the norm threshold of the hidden state at the current time step is less than a preset threshold, the hidden state at the current time step is used as a refined vector and associated with the unique identifier of the technician to form a structured record, which is then uploaded to the blockchain for evidence storage, forming a training data chain.

[0016] Furthermore, S1 also includes:

[0017] The assessment response record is input into the LSTM model, and the temporal feature vector is extracted through cell state updates. The assessment response record and the temporal feature vector are combined by multiplication and fusion, and the corrected feature vector is obtained by correction through interference bias. The corrected feature vector is input into the fully connected layer, mapped to the assessment score, and the assessment result is output in combination with preset rules.

[0018] Furthermore, S2 also includes:

[0019] The training data chain and assessment results are integrated into a decision vector;

[0020] The decision vector is input into the smart contract decision model to obtain the authorization decision vector;

[0021] Based on the authorization decision vector combined with the training data chain and assessment results, an authorization adaptation score is calculated to evaluate the appropriateness of the current decision.

[0022] When the authorization adaptation score reaches a preset threshold, the smart contract function is activated; an authorization token is generated using the authorization adaptation score as an input vector as a parameter, and the token is updated in real time with the project management system via the oracle API to update the corresponding access entries; at the same time, all changes are hashed and stored in the blockchain consensus chain to ensure the immutability and traceability of the data.

[0023] Furthermore, the smart contract decision-making model includes a threshold activation layer and an adaptive weight feedforward unit; wherein,

[0024] The threshold activation layer uses the sigmoid function to process the pass / fail judgment, converting it to a value close to 1 for pass and a value close to 0 for fail.

[0025] The adaptive weighted feedforward unit includes:

[0026] The first layer is a linear layer; the weight matrix is ​​initialized with a uniform distribution.

[0027] The second layer is the ReLU activation layer, used for nonlinear transformation.

[0028] The third layer is the output layer, which uses the sigmoid function to compress the values ​​to the [0,1] interval, forming the final authorization decision vector.

[0029] Furthermore, S3 also includes:

[0030] Real-time monitoring of permission IDs and recording of change log chains are integrated into a monitoring vector;

[0031] The monitoring vector is input into the permission freeze decision model; wherein the permission freeze decision model includes a temporal embedding layer and a risk-adaptive GRU model, and outputs a risk assessment vector.

[0032] The freeze adaptation score is calculated based on the risk assessment vector, monitoring permission ID, and record change log chain. When the freeze adaptation score exceeds a preset threshold, the smart contract freeze function is activated. The freeze function generates a freeze token with the risk assessment vector as a parameter, updates the permission status on the blockchain to frozen, and issues a warning notification through the event log to alert potential abuse risks.

[0033] Furthermore, the temporal embedding layer processes the record change log chain through a positional encoding function, wherein the embedding dimension of each log node matches the change dimension; the risk-adaptive GRU model is a three-layer GRU network, the structure of each layer of which is as follows:

[0034] The first layer is a GRU layer, which uses a gating mechanism and a sigmoid function to process the input gate and forget gate, capturing historical information.

[0035] The second layer is the attention enhancement layer, which applies a self-attention mechanism to each node to highlight key changes;

[0036] The third layer is the output layer, which obtains the final risk vector through linear transformation and generates the output result through ReLU activation.

[0037] Furthermore, S4 specifically includes:

[0038] Obtain the feedback vector, use the risk ID as a risk reference point, extract the event chain of the early warning information, and form multi-dimensional historical data;

[0039] The feedback vector is input into the certificate optimization decision model, and the final decision vector is output; wherein, the certificate optimization decision model includes an event feedback embedding layer and an adaptive Transformer encoding unit;

[0040] The optimization adaptation score is calculated based on the final decision vector, risk ID, and early warning information; when the optimization adaptation score reaches a threshold, the smart contract optimization function is activated; the smart contract optimization function takes the final decision vector as the input vector, generates an electronic certificate as NFT proof, and updates the task role allocation;

[0041] All changes will be fed back through the blockchain event log to ensure that certificate issuance and role adjustment are synchronized and agreed upon, without delay and without tampering.

[0042] Furthermore, the event feedback embedding layer processes the event chain in the warning information through a positional encoding function, and the embedding dimension of each event node matches the notification attribute; the adaptive Transformer encoding unit includes a three-layer structure:

[0043] The first layer is a multi-head self-attention layer, which is used to capture the dependencies between events. It calculates the weighted influence of nodes by scaling the dot product of the query-key-value vector and adding positional bias.

[0044] The second layer is the normalization layer, which will be used to prevent gradient explosion;

[0045] The third layer is a feedforward network layer, used for nonlinear projection to optimize the final decision vector.

[0046] Furthermore, S1 also includes:

[0047] After the training data chain is formed, the attention mechanism, the attention score output by the LSTM model, and the LSTM state update results are combined to generate a hash chain using SHA-256 and stored through a smart contract consensus mechanism to form an immutable on-chain path.

[0048] The beneficial technical effects of the present invention are at least as follows:

[0049] The main innovation of this invention lies in its novel approach to addressing the shortcomings of traditional systems by combining blockchain technology with artificial intelligence algorithms, particularly in the practical application of technical personnel training and project authorization management. By introducing blockchain technology, this invention provides a decentralized, transparent, and tamper-proof management method for the training process and project authorization, ensuring clear traceability of all training records, authorization decisions, and task assignments, effectively preventing human tampering and data misuse. Simultaneously, leveraging artificial intelligence algorithms, especially deep learning and reinforcement learning, this invention enables real-time assessment and dynamic adjustment of technical personnel's capabilities. Based on their learning progress, task completion, and project performance, it automatically generates personalized training paths and task authorizations, ensuring that technical personnel always participate in projects or tasks best suited to their abilities and development stages. This innovative combination not only solves the deficiencies of traditional systems in terms of access control, customized training, and dynamic adjustment, but also significantly improves the system's flexibility, efficiency, and transparency, meeting the needs of modern enterprises for efficient, accurate, and traceable management. Attached Figure Description

[0050] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0051] Figure 1 This is a flowchart of a blockchain-based technical personnel training and project dynamic authorization engine method disclosed in an embodiment of the present invention. Detailed Implementation

[0052] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0053] like Figure 1 As shown in the embodiment of the present invention, a blockchain-based technical personnel training and project dynamic authorization engine method is provided, the method comprising:

[0054] S1. Obtain the original training event sequence of technical personnel, process dynamic interference through attention mechanism and LSTM model, generate refined vectors, associate them with technical personnel identifiers and upload them to the blockchain for evidence storage, form a training data chain, and output the assessment results.

[0055] Specifically, this step is the initial stage of this invention, responsible for collecting training and assessment data of technical personnel, and providing a computational foundation for subsequent data processing and automated contract execution through innovative sequence models and blockchain notarization technology. This solution addresses the unique characteristics of training data, particularly the close relationship between temporal dependencies and dynamic project interference in the sequence module (e.g., real-time task changes may cause interaction interruptions, and traditional static processing methods cannot meet the immediate needs of the authorization engine). It designs a dedicated refinement process to ensure the logical smoothness of data input into the optimization loop and directly drive subsequent steps such as contract automation, thereby improving the overall responsiveness and accuracy of authorization.

[0056] As the initial step of the method, the input to this step includes the raw training event sequence collected by an external system. This data includes technician identifiers (e.g., user_id, used to uniquely identify participants), course module interaction paths (e.g., interaction_sequence, a vector sequence representing the interaction intensity values ​​of each node), and assessment response records (e.g., response_values, recording the response data at the end of the sequence). This data constitutes the initial vector chain, providing the foundation for subsequent model calculations.

[0057] The system first integrates the original training event sequences into a unified input vector. Specifically, the technician identifier `user_id` serves as the global index, `interaction_sequence` serves as the main time-series data (each node corresponds to one dimension, such as n modules corresponding to an n-dimensional vector), and `response_values` is directly appended to the end of the sequence as the target computation node. Addressing the specific problem of inter-module dependencies being affected by project dynamics (e.g., interaction interruptions increase authorization errors), this invention designs a hybrid attention mechanism and an interference-adaptive sequence model to refine the data. This model consists of a simplified multi-head attention layer and an interference-adaptive LSTM unit.

[0058] In a multi-head attention layer, three attention heads jointly calculate the attention score of a node, using the following formula:

[0059]

[0060] Among them, a i is the attention score of the i-th node (ranging from 0 to 1, used for weighted sequence nodes), and q is the query vector, derived from the input vector of the current node (dimension d). k =48); k is the key vector, derived from the input vector associated with the current node (both have the same dimension d). k =48); d k is the vector dimension (fixed at 48, used to scale the dot product), φ is the adjustment coefficient (fixed value of 0.15), and ι is the interference adjustment term (calculated from the Euclidean norm of adjacent nodes, with a value from 0 to positive infinity, serving to compensate for dynamic interference). The softmax function is used to normalize the attention score. The innovation of this mechanism lies in ι, which calculates the bias of the input main time series data interaction_sequence (i.e., ∥n). j -n j-1 ∥, where n j (where the vector of the j-th node is used to adjust attention, supporting nonlinear compensation for interference, thereby reducing the response latency of the licensing engine.)

[0061] Next, the attention-weighted sequence is input into the interference-adaptive LSTM unit for state updates. The hidden dimension of the LSTM unit is 96, and the state update formula is:

[0062] h t =o t ⊙tanh(c t )

[0063] Among them, h t This represents the hidden state at time step t (a vector with a dimension of 96, responsible for carrying sequence information to the next time step), ot It is the output gate (computed using the standard sigmoid function, with values ​​ranging from 0 to 1), tanh is the hyperbolic tangent activation function (with values ​​ranging from -1 to 1), and c t It refers to the cell state, and its calculation formula is:

[0064]

[0065] Among them, f t For the forget gate (computed via sigmoid), c t-1 Given the cell state at the previous time step (dimension 96), i t For the input gate (computed via sigmoid), Here, ψ represents the candidate cell state (calculated using tanh), δ is the adjustment coefficient (fixed at 0.3, used to adjust the regularization term weight), δ is the interference bias vector (extracted from the assessment response records `response_values`), and n is the number of modules (e.g., 10 modules). The regularization term in this formula... It is specifically designed to handle interference deviations in assessment responses, ensuring that the generated refined vector has the smallest error, thereby providing an accurate basis for subsequent dynamic authorization.

[0066] Once the verification passes (based on the final h) t The vector (norm threshold) is used to refine the data. The refined vector is then packaged with the user_id to form a structured record, which is then uploaded to the blockchain for notarization. After serialization, this data includes the attention score and LSTM state update results. A hash chain is generated using SHA-256, and the data is stored through a smart contract consensus mechanism, forming an immutable on-chain path.

[0067] The final record includes:

[0068] training_module_record: The training data chain, which uses an attention-weighted interaction_sequence and updates it via LSTM. It extracts the course_id and calculates the completion_metric (i.e., the average attention score, calculated as ∑a). i / n).

[0069] assessment_result: The assessment result is calculated based on the product of cell state update and response_values ​​in the LSTM formula, combined with the subtraction operation of interference bias.

[0070] S2. Construct a decision vector based on the training data chain, calculate the authorization adaptation score through a smart contract, generate an authorization token when the authorization adaptation score reaches a threshold, update the access permissions and store them on the chain, obtain the authorization ID and record the change log chain.

[0071] Specifically, this step is responsible for automating smart contract authorization based on the aforementioned training data and linking it with project management systems (such as ERP systems) to form a core chain for dynamic permission allocation. This solution optimizes the authorization linkage problem in the scenario of this invention—the training records in step 1 need to be mapped to project tasks in real time, but traditional static systems are easily affected by data latency and inconsistency issues, causing authorization decisions to fail to adapt to the dynamic needs of the project. To address this, an innovative contract decision-making process is designed, building a crucial bridge in enterprise project management through an end-to-end mechanism from data foundation to permission optimization. This ensures that authorization can quickly respond to changes in training, thereby improving the overall engine's transparency and response efficiency.

[0072] The input to this step comes from the output of step 1, including the training data chain `training_module_record` (containing course ID and completion metric, used to identify courses and quantify completion) and the assessment results `assessment_result` (containing assessment score and qualified status, used for assessment and qualification). These inputs serve as the starting vector for this step, directly driving contract decisions and ensuring the continuous linkage of the entire process.

[0073] The operational details of the steps are as follows: The system first integrates the training data chain training_module_record and the assessment result assessment_result into a unified decision vector, in which course_id serves as the matching anchor, course completion_metric and assessment_score are merged into a quantitative component, and qualified_status serves as the activation switch.

[0074] To address the dynamic inconsistency problem in the scenario described in this invention (e.g., a high completion_metric but a low qualified_status may indicate potential project risk), this invention designs an innovative smart contract decision-making model, which includes a threshold activation layer and an adaptive weight feedforward unit. Specifically, the threshold activation layer uses the sigmoid function to process the qualified_status, converting it to a value close to 1 for passing and close to 0 for failing. Subsequently, the adaptive weight feedforward unit consists of three layers:

[0075] The first layer is a linear layer (with an input dimension of 4, matched with course_id encoding, completion_metric, assessment_score, and qualified_status activation respectively), and the weight matrix is ​​initialized with a uniform distribution.

[0076] The second layer is a ReLU activation layer (with a hidden layer dimension of 32), used for non-linear transformation.

[0077] The third layer is the output layer, which uses the sigmoid function to compress the values ​​to the [0,1] interval, forming the final authorization decision vector.

[0078] The design of this decision model enhances the system's adaptability to training data, especially when dealing with inconsistencies between training and project dynamics. It prioritizes projects with high matching records, supporting fast and low-latency decision-making.

[0079] To further refine the granularity of authorization and meet the needs of enterprises in distinguishing between "authorized content" and "authorization level" in actual operations, this step expands upon the original decision vector structure by introducing three types of information: capability table parameters, authorization level classification, and authorization basis tags. This data is jointly provided by the organizational personnel system and training record system and injected in real-time via an interface. It includes: capability item vectors (e.g., a "skill matrix," used to represent standardized scoring sequences of various equipment or technical capabilities), authorization level request tags (e.g., enumeration types such as "equipment operation," "inspection and review," and "report issuance"), and qualification metadata (e.g., certification records, educational information, and work experience). These extended dimensions are uniformly mapped to an additional structure of the input vector as multi-dimensional supplementary input.

[0080] In terms of model structure, the original feedforward authorization decision network adds a "conditional partial derivative branch." This branch applies conditional constraints based on the authorization level label, controlling the activation path of the main output layer so that different authorization levels only respond to corresponding input combinations. For example, if the applicant has "report issuance" authorization, a higher assessment score and more complete certification qualifications are required. This structure is implemented internally through gating logic, with the authorization level signal controlling the output layer state to ensure that the smart contract's output authorization type matches the application objective. Simultaneously, the skill matrix participates in intermediate layer calculations to determine whether the applicant possesses the operational capabilities of the equipment required for the requested task, thereby improving the business consistency and interpretability of the authorization logic.

[0081] Ultimately, the smart contract's output includes not only a authorization flag but also the specific authorized object and permission level. For example, it generates a structured result: authorization passed, level "inspection and review," with the authorized scope including equipment such as "high-pressure detector" and "temperature control furnace." This structure is further encapsulated into a multi-field permission token within the contract chain and bound and updated to the role permission fields of the enterprise project system through an interface. For instance, the "inspection and review" permission can be injected into the review interface of the project task module.

[0082] Next, this invention calculates an authorization fitness score based on the generated decision vector, which evaluates the suitability of the current decision in terms of linkage. The formula is as follows:

[0083]

[0084] Where ν is the authorization adaptation score (range 0 to 1, used to quantify the degree of linkage in authorization decisions); ξ is the completion_metric (input from training_module_record, representing course completion, e.g., 0.85); ζ is the assessment_score (input from assessment_result, representing assessment score, e.g., 92); ω is the qualified_status activation value (calculated through a threshold layer, 1 for qualified, 0 for unqualified); π is the pre-encoded weight of course_id (blockchain mapping, e.g., 1.1); κ is the adjustment coefficient (fixed at 0.4, used to control the strength of the regularization term); ρ is the inconsistency regularization term (calculated as...). (used to penalize deviations); This represents the square root function, used for L2-like nonlinear penalties. The unique aspect of this formula is... The regularization term is specifically designed to penalize high-risk records, especially when inconsistencies arise between completion and score (e.g., project interruptions leading to accumulated bias). This design ensures that low-inconsistency scores are prioritized, helping to maintain the stability and accuracy of the entire system.

[0085] Once the authorization adaptation score ν reaches a preset threshold (e.g., 0.75), the system will activate a smart contract function. This function generates an authorization token using the input vector as a parameter and updates the corresponding access entries in real time with the project management system via an oracle API. Simultaneously, all changes are hashed and stored in the blockchain consensus chain, ensuring data immutability and traceability. Through this mechanism, the authorization path is fully recorded and linked to subsequent steps, such as the freeze check in step 3, ensuring seamless and synchronous dynamic adjustments both internally and externally within the system.

[0086] S3. Monitor permission change logs in real time, assess risks through the permission freeze decision model, calculate the freeze adaptation score, and trigger the smart contract to freeze permissions when the freeze adaptation score exceeds the threshold, generate an early warning notification and record it on the chain, and obtain a risk ID to identify the freeze instance.

[0087] Specifically, this step implements dynamic management and automated freezing of permissions based on the aforementioned authorization data, forming a core component of project risk control. This solution is specifically optimized for the unique needs of permission freezing in the scenario described in this invention—the authorization records in step 2 require real-time monitoring of project dynamics (e.g., change logs showing expiration or fluctuations, leading to execution risks; traditional manual systems easily overlook these disturbances and cannot adapt to immediate adjustments in the training-authorization chain). To address this, an innovative freeze decision-making process is designed, supporting a full-chain mechanism from authorization to certificate optimization, enhancing the protection capabilities and operational continuity of the permission management system, ensuring freeze responses to authorization changes, and improving system automation and security.

[0088] The input for this step comes from the output of step 2, including authorization_id (permission ID, used to uniquely identify the authorized instance) and permission changes. r ecord (a chain of change logs containing timestamps and state change information used to track authorized state changes). These inputs serve as the starting vector for this step, driving subsequent freeze evaluations to ensure the continuity of the management chain.

[0089] In terms of operation, the system will change the authorization_id and permissions. r ecord is integrated into a unified monitoring vector: authorization_id serves as a global index for permission changes. r Ecord extracts timestamp sequences and status metrics from each item in the log chain to form complete time-series historical data. This addresses a problem in the scenario described in this invention—the need for access control to handle the cumulative risks of authorized data (e.g., changes). rEcord data shows that frequent fluctuations may indicate project instability, requiring automated freezing supported by non-linear expiration compensation, while the blockchain consensus mechanism needs to integrate early warning functions to prevent abuse—this employs an innovative permission freezing decision model. This model includes a time-series embedding layer and risk-adaptive GRU units.

[0090] Specifically, the temporal embedding layer handles permission changes through a positional encoding function. r The ecord log chain, where each log node's embedding dimension matches its change dimension (e.g., 6 dimensions including timestamp, status code, ID association, change magnitude, interval difference, and cumulative count). Then, the risk-adaptive GRU model is a three-layer GRU network, with the following layer structure:

[0091] The first layer is a GRU layer (with a hidden dimension of 48), which uses a gating mechanism and a sigmoid function to process the input gate and forget gate, capturing historical information.

[0092] The second layer is the attention enhancement layer (with 2 heads), which applies a self-attention mechanism to each node to highlight key changes;

[0093] The third layer is the output layer, which obtains the final risk vector through linear transformation and generates the output result through ReLU activation.

[0094] The model processes each log node sequentially from the beginning of the input vector, applies temporal embedding, and then iteratively updates it through a GRU layer and attention mechanism to generate an intermediate risk assessment vector. This design optimizes the dynamic interference in the authorization-to-freeze process, ensuring that high-risk authorization records are identified and frozen first to support real-time risk decision-making.

[0095] Next, a freeze suitability score is calculated based on the risk assessment vector using the following formula. This score is used to assess the suitability of the freeze:

[0096]

[0097] Where o represents the freeze adaptation score (value range 0 to 1, quantifying the degree of freeze decision); χ is the stability weight of authorization_id (derived from id hash, e.g., 0.95 indicates high stability); ψ is the permission change... r In `ecord`, the timestamp interval variance (calculated via the log chain, e.g., variance value is 200); β is the adjustment coefficient (fixed at 0.5, used to control the strength of the logarithm); γ is the fluctuation regularization term (calculated as (∑(state change amplitude)). 2The formula uses a regularization term of log(1+|γ|) to effectively suppress volatility, ensuring the system can respond promptly to permission records with significant changes and avoiding potential risks. This is especially useful when permission data fluctuates frequently or is implicitly expired (e.g., due to accumulated log deviations caused by project cycles). This regularization term smooths highly volatile data using a logarithmic function, helping to determine whether to trigger a freeze.

[0098] When the freeze adaptation score exceeds a preset threshold (e.g., 0.85), the system activates the smart contract freeze function. This function takes the input vector as a parameter, generates a freeze token, and updates the permission status on the blockchain to frozen. Simultaneously, the system issues a warning notification via event logs, alerting to potential abuse risks. The entire process begins with vector hashing, with the contract recording events to ensure seamless freeze synchronization with the consensus chain. The freeze path is recorded on the blockchain, ensuring the transparency and traceability of each freeze decision and directly connecting logically to subsequent steps (such as the certificate optimization decision in step 4).

[0099] The output of the step includes a permission freeze record, which contains:

[0100] freeze_id (generated by combining the freeze formula o with the hash of authorization_id, such as SHA-256 combination, obtained by integrating GRU unit and regular expression calculation);

[0101] Warning information (based on a blockchain event chain generated by contract execution, including notification timestamp and reason for freezing, extracted directly from the freezing event).

[0102] S4. Construct a feedback vector based on the permission freeze record, and calculate the optimization adaptation score through the certificate optimization decision model. After the target is met, the smart contract issues a certificate with a unique identifier for NFT storage, and the task roles of technical personnel are adjusted and stored on the blockchain simultaneously; wherein, the feedback vector is the risk ID and early warning notification.

[0103] Specifically, this step, based on the aforementioned frozen data, automates the issuance of electronic certificates and optimizes task roles, forming a feedback loop for the method of this invention. This solution is optimized for the specific characteristics of the optimization data in the scenario of this invention—the frozen records in step 3 need to be transformed into positive feedback (for example, tasks can be upgraded in cases with low freeze_id risk, while traditional fixed optimization methods easily overlook chain accumulation, leading to a mismatch between certificates and project tasks). To address this, this invention designs an innovative optimization decision-making process, supporting continuous personalized optimization from risk management through a full-chain mechanism, achieving seamless connection from freeze control to task adjustment, and improving the overall engine's adaptability and long-term efficiency.

[0104] The input comes from the output of step 3, including freeze_id (a risk ID used to identify frozen instances) and alerts. n otification (recording the chain of notification events, including timestamps, reasons, severity, etc.). These inputs serve as the starting vector for this step, directly driving the optimization evaluation to ensure the closed-loop continuity of optimization decisions.

[0105] First, the system will input the freeze_id and the warning. n OTATION is integrated into a unified feedback vector. `freeze_id` serves as a risk reference point for early warning. n The event chain of otification extracts information such as timestamp sequence, cause classification, and severity item by item to form multidimensional historical data.

[0106] To address the problem in the scenario of this invention—task optimization requires handling the reverse loop of frozen data (e.g., n Accumulated causes in otification may lead to high risk, requiring demotion of task roles; while low risk may trigger certificate issuance to incentivize personnel. The system needs to support nonlinear reversal compensation for automated personalized optimization. This invention adopts an innovative certificate optimization decision model, which includes an event feedback embedding layer and an adaptive Transformer coding unit.

[0107] The event feedback embedding layer processes warnings through a position encoding function. n In otification, the event chain is such that the embedding dimensions of each event node match notification attributes (e.g., 8 dimensions including timestamp, reason code, ID association, severity, impact score, chain position, cumulative risk, and reversal potential). Subsequently, the adaptive Transformer encoding unit comprises a three-layer structure:

[0108] The first layer is a multi-head self-attention layer (3 heads, 72 dimensions), which is used to capture the dependencies between events. The weighted influence of nodes is calculated by scaling the dot product of the query-key-value vector and adding positional bias.

[0109] The second layer is the normalization layer (LayerNorm), which will be used to prevent gradient explosion;

[0110] The third layer is a feedforward network layer (192 hidden dimensions, using the GELU activation function), used for non-linear projection to optimize the final decision vector.

[0111] This model processes each event node sequentially, first applying feedback embedding, then iteratively constructing an optimized decision vector through Transformer layers involving attention calculation, normalization adjustment, and feedforward projection. This innovative design specifically addresses feedback interference during the optimization process, ensuring that low-risk frozen records are prioritized for certificate issuance and task role upgrades, supporting the engine's continuous personalized decision-making. In enterprise project management, such as technology projects, technical personnel roles can be quickly adjusted based on historical frozen records, reducing retraining costs and improving personnel matching accuracy.

[0112] Next, the optimization fitness score is calculated based on the optimization decision vector using the following formula to assess certificate issuance and task suitability:

[0113]

[0114] Where ρ represents the fitness score (values ​​range from 0 to 1, used to quantify the degree of optimization); α is the inverse risk weight of freeze_id (derived from the hash of the input id, e.g., 0.92 is low risk); β is the warning value. n In otification, the strength of the most recent event (calculated at the end of the chain, strength value such as 0.8); γ is the adjustment coefficient (fixed value of 0.65, used to control the strength of sinh); δ is the cumulative mean of causes (average of the output of the third layer of Transformer, for example 0.4); ∈ is the feedback inverse regularization term (calculated as...). (Used to reward low-biased cycles); sinh is a hyperbolic sine function (used for nonlinear amplification reversal). The design of this formula ensures that high-risk frozen records are properly adjusted by embedding the denominator through the subtraction of the inverse regularization term of sinh(δ·∈), while low-biased cycles are prioritized for certificate issuance and task role promotion.

[0115] For example, when α = 0.92, β = 0.75, δ = 0.3, and ∈ = 0.9, the optimal fitness score is calculated as follows:

[0116]

[0117] When the optimization score ρ reaches a threshold (e.g., 0.88), the system activates the smart contract optimization function. This contract, using the input vector as a parameter, generates a digital certificate cert_id as an NFT for storage and updates task role assignments. All changes are reflected through the blockchain event log, ensuring synchronized consensus, zero delay, and immutability in certificate issuance and role adjustments.

[0118] The output of the step includes:

[0119] The certificate's unique identifier, cert_id, is generated by an optimized formula ρ and a hash of freeze_id, for example, using a combination of SHA-256, based on Transformer units and inverse regularization calculations.

[0120] Task optimization a Djust (a blockchain adjustment chain based on contract execution, recording new task roles and allocation details, extracted directly from optimization events)

[0121] The foregoing has described specific embodiments of this specification; other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than those shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily have to follow the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0122] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0123] For ease of description, the above devices are described in terms of function, divided into various units. Of course, in implementing this specification, the functions of each unit can be implemented in one or more software and / or hardware components.

[0124] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, the embodiments of this specification can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the embodiments of this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0125] This specification is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this specification. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0126] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0127] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0128] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0129] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0130] Computer-readable media include both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0131] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0132] This specification can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This specification can also be practiced in distributed computing environments, where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0133] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0134] Finally, it should be noted that the notarized lottery platform based on national cryptographic algorithms disclosed in this embodiment of the invention is merely a preferred embodiment of the invention and is only used to illustrate the technical solutions of the invention, not to limit it. Although the invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the invention.

Claims

1. A blockchain-based method for technical personnel training and dynamic project authorization engine, characterized in that: The method includes the following steps: S1. Obtain the original training event sequence of technical personnel, process dynamic interference through attention mechanism and LSTM model, generate refined vectors, associate them with technical personnel identifiers and upload them to the blockchain for notarization, form a training data chain, and output the assessment results. S2. Construct a decision vector based on the training data chain, calculate the authorization adaptation score through a smart contract, generate an authorization token when the authorization adaptation score reaches a threshold, update the access permissions and store them on the chain, and obtain the authorization ID and record the change log chain. S3. Monitor permission change logs in real time, assess risks through permission freeze decision model, calculate freeze adaptation score, and trigger smart contract freeze permission when the freeze adaptation score exceeds the threshold, generate early warning notification and record it on the chain to obtain risk ID for identifying freeze instance. S4. Construct a feedback vector based on the permission freeze record, and calculate the optimization adaptation score through the certificate optimization decision model. After the target is met, the smart contract issues a certificate with a unique identifier for NFT storage, and the task roles of technical personnel are adjusted and stored on the blockchain simultaneously; wherein, the feedback vector is a risk ID and a warning notification; The original training event sequence includes a unique identifier for technical personnel, a sequence of interaction intensity vectors for course modules, and assessment response records. Specifically, the process of handling dynamic interference through an attention mechanism and an LSTM model to generate refined vectors, which are then associated with technical personnel identifiers and uploaded to the blockchain for evidence storage, forming a training data chain, is as follows: The original training event sequence is integrated into a unified input vector; The input vector is fed into the attention mechanism and LSTM model to generate a refined vector; wherein, the attention mechanism and LSTM model are specifically: In the multi-head attention layer, three attention heads jointly calculate the attention score of a node to obtain the attention score of the corresponding node; the attention weighted sequence is input into the interference-adaptive LSTM unit for state update to obtain the hidden state of the current time step; If the norm threshold of the hidden state at the current time step is less than a preset threshold, the hidden state at the current time step is used as a refined vector and associated with the unique identifier of the technician to form a structured record, which is then uploaded to the blockchain for evidence storage to form a training data chain. The S1 further includes: The assessment response record is input into the LSTM model, and the temporal feature vector is extracted through cell state update. The assessment response record and the temporal feature vector are combined by multiplication and fusion, and the correction is obtained by interference bias revision. The correction feature vector is input into the fully connected layer, mapped to the assessment score, and the assessment result is output according to the preset rules. The S2 further includes: The training data chain and assessment results are integrated into a decision vector; The decision vector is input into the smart contract decision model to obtain the authorization decision vector; Based on the authorization decision vector combined with the training data chain and assessment results, an authorization adaptation score is calculated to evaluate the appropriateness of the current decision. When the authorization adaptation score reaches a preset threshold, the smart contract function is activated; the authorization adaptation score is used as the input vector parameter to generate an authorization token, and the token is updated in real time with the project management system via the oracle API; at the same time, all changes are hashed and stored in the blockchain consensus chain to ensure the immutability and traceability of the data. The smart contract decision model includes a threshold activation layer and an adaptive weight feedforward unit; wherein... The threshold activation layer uses the sigmoid function to process the pass / fail judgment, converting it to a value close to 1 for pass and a value close to 0 for fail. The adaptive weighted feedforward unit includes: The first layer is a linear layer; the weight matrix is ​​initialized with a uniform distribution. The second layer is the ReLU activation layer, used for nonlinear transformation; The third layer is the output layer, which uses the sigmoid function to compress the values ​​to the [0,1] interval, forming the final authorization decision vector; The S3 further includes: Real-time monitoring of permission IDs and recording of change log chains are integrated into a monitoring vector; The monitoring vector is input into the permission freeze decision model; wherein the permission freeze decision model includes a temporal embedding layer and a risk-adaptive GRU model, and outputs a risk assessment vector. The freeze adaptation score is calculated based on the risk assessment vector, monitoring permission ID, and record change log chain. When the freeze adaptation score exceeds a preset threshold, the smart contract freeze function is activated. The freeze function generates a freeze token with the risk assessment vector as a parameter, updates the permission status on the blockchain to frozen, and issues a warning notification through the event log to alert potential abuse risks. The temporal embedding layer processes the record change log chain through a positional encoding function, wherein the embedding dimension of each log node matches the change dimension; the risk-adaptive GRU model is a three-layer GRU network, and its layer structure is as follows: The first layer is a GRU layer, which uses a gating mechanism and a sigmoid function to process the input gate and forget gate, capturing historical information. The second layer is the attention enhancement layer, which applies a self-attention mechanism to each node to highlight key changes; The third layer is the output layer, which obtains the final risk vector through linear transformation and generates the output result through ReLU activation. S4 specifically includes: Obtain the feedback vector, use the risk ID as a risk reference point, extract the event chain of the early warning information, and form multi-dimensional historical data; The feedback vector is input into the certificate optimization decision model, and the final decision vector is output; wherein, the certificate optimization decision model includes an event feedback embedding layer and an adaptive Transformer encoding unit; The optimization adaptation score is calculated based on the final decision vector, risk ID, and early warning information; when the optimization adaptation score reaches a threshold, the smart contract optimization function is activated; the smart contract optimization function takes the final decision vector as the input vector, generates an electronic certificate as NFT proof, and updates the task role allocation; All changes will be fed back through the blockchain event log to ensure that certificate issuance and role adjustment are synchronized and agreed upon, without delay and without tampering; The event feedback embedding layer processes the event chain in the warning information through a positional encoding function, and the embedding dimension of each event node matches the notification attribute; the adaptive Transformer encoding unit includes a three-layer structure: The first layer is a multi-head self-attention layer, which is used to capture the dependencies between events. It calculates the weighted influence of nodes by scaling the dot product of the query-key-value vector and adding positional bias. The second layer is the normalization layer, which will be used to prevent gradient explosion; The third layer is a feedforward network layer, used for nonlinear projection to optimize the final decision vector.

2. The blockchain-based technical personnel training and project dynamic authorization engine method according to claim 1, characterized in that, S1 further includes: After the training data chain is formed, the attention mechanism, the attention score output by the LSTM model, and the LSTM state update results are combined to generate a hash chain using SHA-256 and stored through a smart contract consensus mechanism to form an immutable on-chain path.

Images