Machine learning-based power safety training intelligent examination and evaluation system

Abstract

This invention discloses an intelligent assessment and evaluation system for power safety training based on machine learning, comprising a data acquisition module, a behavior analysis module, a knowledge assessment module, a comprehensive evaluation module, and a feedback generation module. Through the collaborative work of multiple modules, this invention achieves intelligent management of the entire power safety training and assessment process. The system utilizes machine learning technology to conduct multi-dimensional analysis of trainees' theoretical knowledge and practical behavior, significantly improving the objectivity and accuracy of the assessment. By constructing a power safety knowledge graph and an operational behavior standard library, the system can comprehensively cover various safety scenarios and achieve accurate assessment of trainees' safety awareness and operational skills based on deep learning-based multimodal data fusion analysis. The personalized feedback module can generate targeted training suggestions based on trainees' weaknesses, effectively improving training outcomes. Simultaneously, the system supports remote online assessment, greatly reducing training costs and time investment.

Description

Technical Field

[0001] This invention relates to the field of power system technology, and specifically to a machine learning-based intelligent assessment and evaluation system for power safety training. Background Technology

[0002] Currently, safety training and assessment in the power industry mainly adopts a combination of theoretical examinations and manual practical assessments, which suffers from problems such as low assessment efficiency, high subjectivity, and limited coverage of scenarios. Traditional theoretical examinations cannot accurately reflect trainees' ability to apply safety knowledge, while practical assessments are limited by venues, equipment, and the personal experience of examiners, making it impossible to conduct detailed analysis of trainees' operational behaviors. With the expansion of the power system and the increase in safety requirements, the existing assessment methods can no longer meet the needs of enterprises for accurate assessment of employees' safety skills. There is an urgent need for an intelligent assessment system that can automatically analyze trainees' knowledge mastery and the standardization of their practical behaviors. Summary of the Invention

[0003] To address these issues, the present invention provides an intelligent assessment and evaluation system for power safety training based on machine learning.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A machine learning-based intelligent assessment and evaluation system for power safety training includes the following modules:

[0006] The data acquisition module is used to acquire multi-source data generated by trainees during the training and assessment process. The multi-source data includes theoretical answer data, practical operation video data, physiological signal data, and voice interaction data. The theoretical answer data is collected through the power safety knowledge question bank system, the practical operation video data is collected through multi-angle high-definition cameras deployed in the training site, the physiological signal data is collected through wearable devices to collect trainees' heart rate, skin conductance response, and electromyography signals, and the voice interaction data is collected through microphone arrays to collect trainees' voice descriptions and questions and answers during the operation process.

[0007] The behavior analysis module uses computer vision technology and deep learning algorithms to process practical video data. It extracts the characteristics of trainees' operation actions through a pre-trained behavior recognition model and compares and analyzes them with a standard operation behavior library. The behavior analysis module includes an action standardization evaluation unit, a safety protection equipment usage detection unit, and a dangerous behavior identification unit. The action standardization evaluation unit is built based on a temporal convolutional network and can analyze the accuracy and fluency of trainees' operation actions frame by frame. The safety protection equipment usage detection unit uses a target detection algorithm to monitor the wearing of safety helmets, insulating gloves, and protective clothing in real time. The dangerous behavior identification unit identifies operation behaviors that violate safety procedures through a behavior semantic analysis model.

[0008] The knowledge assessment module uses natural language processing technology to analyze trainees' theoretical answers and voice interaction data. It constructs a question-and-answer semantic understanding model through a power safety knowledge graph. The knowledge assessment module includes a theoretical knowledge mastery assessment unit and a safety awareness assessment unit. The theoretical knowledge mastery assessment unit uses a deep question-and-answer model to conduct deep semantic analysis on trainees' theoretical answers. The safety awareness assessment unit evaluates trainees' safety risk identification ability and emergency response awareness by analyzing their decision-making process and voice description in simulated scenarios.

[0009] The comprehensive evaluation module, based on the output results of the multilayer perceptron network fusion behavior analysis module and knowledge assessment module, combined with the trainee status information provided by physiological signal data, generates a comprehensive assessment report. The comprehensive evaluation module includes a multimodal data fusion unit and a dynamic weight adjustment unit. The multimodal data fusion unit uses an attention mechanism to weight and fuse different types of assessment indicators, and the dynamic weight adjustment unit adaptively adjusts the weight coefficients of each assessment indicator according to different assessment scenarios.

[0010] The feedback generation module automatically generates personalized training suggestions and assessment feedback based on the output of the comprehensive evaluation module. The feedback generation module includes a weakness identification unit and a training content recommendation unit. The weakness identification unit uses cluster analysis to find the knowledge blind spots and operational errors of trainees, while the training content recommendation unit recommends targeted reinforcement training content for each trainee based on a collaborative filtering algorithm.

[0011] Preferably, the behavior analysis module further includes: a 3D motion reconstruction unit, which reconstructs the 3D skeletal joint sequence of the trainee's operation process using multi-view video data, and extracts spatiotemporal features of the joint sequence using a graph convolutional network; an operation process compliance check unit, which models the time dependency of operation actions based on a long short-term memory network, aligns and compares the trainee's operation sequence with the standard operation process, and identifies missing or incorrectly ordered operation steps; a safety distance monitoring unit, which uses scene depth information obtained by a depth camera to calculate the safety distance between the trainee's body and live equipment or hazardous sources in real time, and generates a warning signal when insufficient safety distance is detected; and a behavior trajectory analysis unit, which extracts the trainee's movement trajectory during the operation process using optical flow and feature tracking technology, and analyzes whether the trainee's movement conforms to the work specifications and safety requirements.

[0012] Preferably, the knowledge assessment module further includes: a concept comprehension depth assessment unit, which performs deep semantic analysis on trainees' theoretical answers using a pre-trained power industry language model to construct a knowledge mastery hierarchy model and distinguish between trainees' rote and applied understanding of safety knowledge; a scenario reasoning ability assessment unit, which constructs virtual accident scenarios based on reinforcement learning and assesses trainees' ability to transfer and apply safety knowledge and their emergency reasoning ability by analyzing their decision-making paths and solutions in simulated accident handling; a safety hazard identification assessment unit, which presents trainees with images or videos of work scenarios containing potential safety risks and assesses their safety risk perception acuity and hazard identification completeness through eye-tracking technology and voice report analysis; and a knowledge correlation analysis unit, which uses graph neural networks to analyze the completeness and coherence of trainees' knowledge structure and assesses the systematic nature of their safety knowledge system by calculating the correlation strength between different safety knowledge points.

[0013] Preferably, the operation process compliance inspection unit specifically includes: a step timing verification subunit, which uses a dynamic time warping algorithm to match the time sequence of trainees' operation actions with a standard operation process template, calculates the temporal similarity between the two, and identifies operation steps with incorrect order or excessive time deviation; a tool usage standardization detection subunit, which analyzes the correctness of tool selection and the standardization of usage during trainees' operation based on fine-grained image recognition technology, and detects unsafe behaviors such as tool misuse and abuse; a collaborative operation evaluation subunit, which tracks the behavior of different trainees separately in team collaboration assessment scenarios through multi-target tracking technology, analyzes the rationality of their division of labor, communication and coordination, and mutual protection safety, and evaluates the overall safety collaboration level of the team; and an emergency response evaluation subunit, which simulates sudden situations such as equipment failure and electric shock, records and analyzes the trainees' emergency response time, the correctness of their handling procedures, and the effectiveness of their risk avoidance measures.

[0014] Preferably, the comprehensive evaluation module further includes: a standardization unit for assessment indicators, which uses fuzzy mathematics theory to normalize different types of assessment indicators, eliminate differences in the dimensions and value ranges of each indicator, and establish a unified evaluation scale; a credibility evaluation unit, which evaluates the credibility of each assessment indicator by analyzing the consistency and complementarity between multi-source data, and reduces the weight of indicators with low credibility; a personalized evaluation model unit, which constructs a personalized assessment model based on the trainee's historical assessment data and job characteristics, and adaptively adjusts the evaluation criteria considering the trainee's experience level and job responsibilities; and a visualization unit for assessment results, which presents the comprehensive evaluation results in a visual form such as radar charts and heat maps, intuitively showing the trainee's performance level and improvement direction in each assessment dimension.

[0015] Preferably, the concept comprehension depth assessment unit specifically includes: a knowledge hierarchy division subunit, which divides power safety knowledge into five mastery levels: memory, understanding, application, analysis, and evaluation, and assesses trainees' mastery at each level by designing assessment questions of varying difficulty and depth; a misconception detection subunit, which identifies potential safety knowledge misunderstandings and conceptual confusions by analyzing the semantic features and problem-solving processes of trainees' incorrect answers, and establishes a misconception concept map; a knowledge application and transfer subunit, which designs assessment scenarios similar to but not identical to the training scenarios to assess trainees' ability to transfer learned safety knowledge to new scenarios; and a knowledge forgetting monitoring subunit, which constructs a knowledge forgetting curve model for trainees by periodically and repeatedly testing key safety knowledge points, and assesses their long-term memory of core safety knowledge.

[0016] Preferably, the feedback generation module further includes: a capability profile construction unit, which constructs a multi-dimensional profile of the trainee's safety capabilities based on multiple assessment data and dynamically tracks the trajectory of capability development; an adaptive learning path planning unit, which plans a personalized learning path for the trainee based on the trainee's capability profile and weaknesses, and recommends the most suitable training content and learning resources; a training effect prediction unit, which uses a time series prediction model to predict the improvement effect of the trainee's capabilities after specified training based on the trainee's historical learning data and current capability level; and an intervention strategy recommendation unit, which generates graded warnings and intervention strategy suggestions when a trainee is detected to have serious safety knowledge deficiencies or habitual violation tendencies, including different levels of measures such as strengthening training, key tutoring, and job adjustment.

[0017] Preferably, the personalized evaluation model unit specifically includes: a job adaptability analysis subunit, which establishes a job competency model based on the safety responsibilities of different positions and assesses the matching degree between trainees and target positions; an experience level compensation subunit, which considers trainees' years of work experience and historical safety records, sets differentiated qualification standards for junior trainees and senior employees, and appropriately adjusts the evaluation threshold while maintaining the core safety requirements; a learning curve analysis subunit, which analyzes the changing trends of trainees' past assessment scores to assess their learning progress speed and stability, and gives appropriate encouragement scores to trainees who have made significant progress but whose current level is still low; and a risk assessment subunit, which predicts the types and probabilities of safety risks that may occur in actual work based on trainees' assessment performance and personality characteristics, and generates risk warning reports.

[0018] This invention offers the following advantages: Through multi-module collaborative operation, it achieves intelligent management of the entire power safety training and assessment process. The system utilizes machine learning technology to conduct multi-dimensional analysis of trainees' theoretical knowledge and practical behavior, significantly improving the objectivity and accuracy of the assessment. By constructing a power safety knowledge graph and a standard operational behavior library, the system comprehensively covers various safety scenarios and, based on deep learning-based multimodal data fusion analysis, achieves precise evaluation of trainees' safety awareness and operational skills. The personalized feedback module can generate targeted training suggestions based on trainees' weaknesses, effectively improving training outcomes. Simultaneously, the system supports remote online assessment, greatly reducing training costs and time investment. Attached Figure Description

[0019] To more intuitively illustrate the prior art and this application, exemplary drawings are provided below. It should be understood that the specific shapes and structures shown in the drawings should not generally be regarded as limiting conditions for implementing this application; for example, based on the technical concept disclosed in this application and the exemplary drawings, those skilled in the art are able to easily make conventional adjustments or further optimizations to the addition / reduction / classification, specific shapes, positional relationships, connection methods, size ratios, etc. of certain units (components).

[0020] Figure 1 This is a block diagram of a machine learning-based intelligent assessment and evaluation system for power safety training, provided in an embodiment of this application. Detailed Implementation

[0021] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. It should be understood that these embodiments are merely for further explanation of the present invention and should not be construed as limiting the scope of protection of the present invention. Technical engineers in the field can make some non-essential improvements and adjustments to the present invention based on the above-described content. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figure 1 A machine learning-based intelligent assessment and evaluation system for power safety training includes the following modules:

[0023] The data acquisition module is used to acquire multi-source data generated by trainees during the training and assessment process, including theoretical answer data, practical operation video data, physiological signal data, and voice interaction data. The theoretical answer data is collected through the power safety knowledge question bank system, the practical operation video data is collected through multi-angle high-definition cameras deployed in the training site, the physiological signal data is collected through wearable devices to collect trainees' heart rate, skin conductance response, and electromyography signals, and the voice interaction data is collected through microphone arrays to collect trainees' voice descriptions and questions and answers during the operation process.

[0024] The behavior analysis module uses computer vision technology and deep learning algorithms to process practical video data. It extracts the characteristics of trainees' operation actions through a pre-trained behavior recognition model and compares and analyzes them with a standard operation behavior library. This module includes an action standardization evaluation unit, a safety protection equipment usage detection unit, and a dangerous behavior identification unit. The action standardization evaluation unit is built based on a temporal convolutional network and can analyze the accuracy and fluency of trainees' operation actions frame by frame. The safety protection equipment usage detection unit uses a target detection algorithm to monitor the wearing of safety helmets, insulating gloves, and protective clothing in real time. The dangerous behavior identification unit identifies operation behaviors that violate safety procedures through a behavior semantic analysis model.

[0025] The knowledge assessment module utilizes natural language processing technology to analyze trainees' theoretical answers and voice interaction data. It constructs a question-and-answer semantic understanding model through a power safety knowledge graph. This module includes a theoretical knowledge mastery assessment unit and a safety awareness assessment unit. The theoretical knowledge mastery assessment unit uses a deep question-and-answer model to conduct deep semantic analysis on trainees' theoretical answers, assessing their understanding of professional knowledge such as power safety regulations and accident handling procedures. The safety awareness assessment unit analyzes trainees' decision-making process and voice descriptions in simulated scenarios to assess their ability to identify safety risks and their emergency response awareness.

[0026] The comprehensive evaluation module, based on the output results of the multilayer perceptron network fusion behavior analysis module and knowledge assessment module, combined with the trainee status information provided by physiological signal data, generates a comprehensive assessment report. This module includes a multimodal data fusion unit and a dynamic weight adjustment unit. The multimodal data fusion unit uses an attention mechanism to weight and fuse different types of assessment indicators, and the dynamic weight adjustment unit adaptively adjusts the weight coefficients of each assessment indicator according to different assessment scenarios.

[0027] The feedback generation module automatically generates personalized training suggestions and assessment feedback based on the output of the comprehensive evaluation module. This module includes a weakness identification unit and a training content recommendation unit. The weakness identification unit uses cluster analysis to find common knowledge gaps and operational errors among trainees, while the training content recommendation unit recommends targeted reinforcement training content for each trainee based on a collaborative filtering algorithm.

[0028] During implementation, this system first acquires trainees' theoretical answers, practical operation videos, physiological signals, and voice interaction data through a multi-source data acquisition module, laying the foundation for comprehensive assessment. The behavior analysis module utilizes computer vision and deep learning technologies, using a pre-trained behavior recognition model to analyze practical operation videos, performing multi-dimensional detection from action standardization and safety equipment usage to dangerous behaviors. Action standardization assessment is based on temporal convolutional networks for frame-by-frame analysis, safety equipment detection employs target recognition algorithms, and dangerous behavior identification is completed through behavioral semantic analysis. The knowledge assessment module utilizes natural language processing technology and an electrical safety knowledge graph, assessing theoretical knowledge mastery through a deep question-answering model and analyzing safety awareness levels based on voice interaction. The comprehensive evaluation module uses a multilayer perceptron network to fuse behavior and knowledge assessment results, combining physiological data and achieving multimodal data fusion through an attention mechanism, while utilizing dynamic weight adjustment units to adapt to different assessment scenarios. Finally, the feedback generation module identifies weaknesses through cluster analysis and generates personalized training suggestions for trainees based on collaborative filtering algorithms, forming a complete closed loop of "assessment-evaluation-feedback."

[0029] The behavior analysis module further includes: a 3D motion reconstruction unit, which reconstructs the 3D skeletal joint sequence of the trainee's operation process using multi-view video data, and extracts spatiotemporal features of the joint sequence using a graph convolutional network; an operation process compliance check unit, which models the time dependence of operation actions based on a long short-term memory network, aligns and compares the trainee's operation sequence with the standard operation process, and identifies missing or incorrectly ordered operation steps; a safety distance monitoring unit, which uses scene depth information obtained by a depth camera to calculate the safety distance between the trainee's body and live equipment or hazardous sources in real time, and generates a warning signal when the safety distance is insufficient; and a behavior trajectory analysis unit, which extracts the trainee's movement trajectory during the operation process using optical flow and feature tracking technology, and analyzes whether their movement conforms to the work specifications and safety requirements.

[0030] The 3D motion reconstruction unit generates precise skeletal joint sequence of trainees using multi-view video data and extracts spatiotemporal features through graph convolutional networks, enabling the identification of subtle motion deformations. The operation procedure compliance check unit establishes a time-dependent model of the operation sequence based on a long short-term memory network. By dynamically time-warping and comparing it with a standard procedure template, it can accurately identify missing steps or incorrect sequences. The safety distance monitoring unit utilizes the 3D perception capabilities of a depth camera to calculate the spatial relationship between the human body and hazards in real time, issuing an immediate warning when a trainee approaches live equipment beyond a safety threshold. The behavior trajectory analysis unit tracks the trainee's movement path using optical flow to assess whether their movement route complies with safety regulations. For example, in a simulated substation operation assessment, the system can simultaneously detect multiple safety factors, including whether the trainee walks along the standard path, maintains a sufficient distance from live equipment, and completes all operational steps.

[0031] The assessment unit for deep conceptual understanding uses a pre-trained language model in the power industry to perform deep semantic analysis on trainees' theoretical answers, constructing a hierarchical model of knowledge mastery and distinguishing between trainees' rote and applied understanding of safety knowledge. The assessment unit for scenario-based reasoning uses reinforcement learning to construct virtual accident scenarios and analyzes trainees' decision-making paths and solutions in simulated accident handling to evaluate their ability to transfer and apply safety knowledge and their emergency reasoning capabilities. The assessment unit for safety hazard identification presents trainees with images or videos of work scenarios containing potential safety risks and uses eye-tracking technology and voice report analysis to evaluate their sensitivity to safety risks and the completeness of hazard identification. The assessment unit for knowledge correlation analysis uses graph neural networks to analyze the completeness and coherence of trainees' knowledge structure and evaluates the systematic nature of their safety knowledge system by calculating the correlation strength between different safety knowledge points.

[0032] The conceptual comprehension depth assessment unit uses a pre-trained language model to analyze the semantic depth of trainees' answers, distinguishing between rote memorization and applied understanding. The scenario-based reasoning ability assessment unit constructs virtual accident scenarios (such as a simulated transformer fire) and assesses trainees' knowledge transfer and emergency reasoning abilities by analyzing their decision-making paths. The safety hazard identification unit assesses trainees' risk perception acuity by displaying risky work scenario diagrams and combining eye-tracking and voice analysis. For example, the system might detect that a trainee, while able to recite regulations, fails to identify the potential risk of improperly suspended grounding wires in a simulated scenario.

[0033] The operational process compliance check unit in the behavior analysis module specifically includes: a step timing verification subunit, which uses a dynamic time warping algorithm to match the time sequence of trainees' actions with a standard operational process template, calculates the temporal similarity between the two, and identifies operational steps with incorrect order or excessive time deviation; a tool usage standardization detection subunit, which analyzes the correctness of tool selection and the standardization of usage during trainees' operations based on fine-grained image recognition technology, and detects unsafe behaviors such as tool misuse and abuse; a collaborative operation evaluation subunit, which tracks the behavior of different trainees separately in team collaboration assessment scenarios using multi-target tracking technology, analyzes the rationality of their division of labor, communication and coordination, and mutual protection safety, and evaluates the overall safety collaboration level of the team; and an emergency response evaluation subunit, which simulates emergencies such as equipment failure and electric shock, records and analyzes the trainees' emergency response time, the correctness of their handling procedures, and the effectiveness of their risk avoidance measures, and evaluates their emergency response capabilities.

[0034] The dynamic time warping algorithm flexibly matches trainees' operation time sequences with standard templates to identify errors in step sequence or time deviations (such as directly connecting a grounding wire without prior electrical testing). The tool usage compliance detection subunit uses fine-grained image recognition to determine if trainees have selected the wrong tools or used them improperly (such as using a regular wrench on live equipment). The collaborative operation assessment subunit tracks the behavior of each member during teamwork, analyzing whether their division of labor and communication meet operational safety requirements. The emergency response assessment subunit simulates sudden failures, accurately recording trainees' reaction times and procedural correctness from discovery to handling.

[0035] The comprehensive evaluation module also includes: a standardization unit for assessment indicators, which uses fuzzy mathematics theory to normalize different types of assessment indicators, eliminate differences in the dimensions and value ranges of each indicator, and establish a unified evaluation scale; a credibility evaluation unit, which evaluates the credibility of each assessment indicator by analyzing the consistency and complementarity between multi-source data, and downweights indicators with low credibility; a personalized evaluation model unit, which constructs a personalized assessment model based on the trainees' historical assessment data and job characteristics, and adaptively adjusts the evaluation criteria considering the trainees' experience level and differences in job responsibilities; and a visualization unit for assessment results, which presents the comprehensive evaluation results in visual forms such as radar charts and heat maps, intuitively showing the trainees' performance level and improvement directions in each assessment dimension.

[0036] Fuzzy mathematics theory is used to standardize indicators with different dimensions, such as action standardization and knowledge scores, making them comparable. The credibility assessment unit evaluates the credibility of each indicator through cross-validation (such as whether video analysis results are consistent with physiological signals), and downweights contradictory data. The personalized evaluation model considers the trainee's job position (such as the different standards for operation and maintenance positions) and experience level, dynamically adjusting the evaluation threshold to achieve fair and targeted evaluation.

[0037] The concept comprehension depth assessment unit within the knowledge assessment module specifically includes: a knowledge hierarchy division subunit, which divides power safety knowledge into five mastery levels: memorization, understanding, application, analysis, and evaluation, and assesses trainees' mastery at each level by designing assessment questions of varying difficulty and depth; a misconception detection subunit, which identifies potential safety knowledge misunderstandings and conceptual confusions by analyzing the semantic features and problem-solving processes of trainees' incorrect answers, and establishes a misconception concept map; a knowledge application and transfer subunit, which designs assessment scenarios similar to but not identical to the training scenarios to assess trainees' ability to transfer learned safety knowledge to new scenarios; and a knowledge forgetting monitoring subunit, which constructs a knowledge forgetting curve model for trainees by periodically and repeatedly testing key safety knowledge points, and assesses their long-term memory of core safety knowledge.

[0038] The knowledge is divided into sub-units based on hierarchical knowledge levels, and questions of varying difficulty are designed to assess trainees' mastery of safety knowledge (e.g., whether they can only memorize clauses or analyze the causes of accidents). The misconception detection sub-unit identifies deep-seated misunderstandings by analyzing patterns in incorrect answers (e.g., confusing "protective grounding" with "functional grounding"). The knowledge application and transfer sub-unit examines trainees' ability to flexibly apply knowledge by setting new scenarios. The knowledge forgetting monitoring sub-unit uses periodic tests to plot forgetting curves and warn of memory decline in key knowledge.

[0039] The concept comprehension depth assessment unit within the knowledge assessment module specifically includes: a knowledge hierarchy division subunit, which divides power safety knowledge into five mastery levels: memorization, understanding, application, analysis, and evaluation, and assesses trainees' mastery at each level by designing assessment questions of varying difficulty and depth; a misconception detection subunit, which identifies potential safety knowledge misunderstandings and conceptual confusions by analyzing the semantic features and problem-solving processes of trainees' incorrect answers, and establishes a misconception concept map; a knowledge application and transfer subunit, which designs assessment scenarios similar to but not identical to the training scenarios to assess trainees' ability to transfer learned safety knowledge to new scenarios; and a knowledge forgetting monitoring subunit, which constructs a knowledge forgetting curve model for trainees by periodically and repeatedly testing key safety knowledge points, and assesses their long-term memory of core safety knowledge.

[0040] The competency profiling unit integrates trainees' past assessment data to create a multi-dimensional, dynamic profile of their safety skills. The adaptive learning path planning unit intelligently recommends targeted learning resources and training modules based on weaknesses identified in the profile (such as "unfamiliarity with high-voltage switching operation procedures"). The training effectiveness prediction unit predicts trainees' post-training improvement based on historical data, assisting in training decisions. The intervention strategy recommendation unit suggests tiered measures, such as intensive training or focused coaching, when serious deficiencies are identified.

[0041] The personalized evaluation model unit in the comprehensive evaluation module specifically includes: a job adaptability analysis subunit, which establishes a job competency model based on the safety responsibilities of different positions and assesses the matching degree between trainees and target positions; an experience level compensation subunit, which considers trainees' years of work experience and historical safety records, sets differentiated qualification standards for junior trainees and senior employees, and appropriately adjusts the evaluation threshold while maintaining core safety requirements; a learning curve analysis subunit, which assesses the speed and stability of trainees' learning progress by analyzing the changing trends of their past assessment scores, and provides appropriate encouragement scores to trainees who have made significant progress but whose current level is still low; and a risk assessment subunit, which predicts the types and probabilities of safety risks that trainees may encounter in actual work based on their assessment performance and personality characteristics, and generates risk warning reports.

[0042] The Job Adaptability Analysis subunit establishes differentiated competency models to assess the matching degree based on the safety responsibilities of different positions (such as dispatchers and line patrolmen). The Experience Level Compensation subunit sets different expectation standards for novice and senior employees while ensuring core safety baselines; for example, it focuses more on process standardization for newcomers and more on handling complex situations for experienced employees. The Learning Curve Analysis subunit provides encouragement and recognition to trainees who progress quickly but have weak foundations by analyzing progress trends. The Risk Assessment subunit integrates all data to predict the specific types of risks that trainees may cause in actual work.

[0043] This system also includes: an assessment data management module, used to store and manage all trainees' assessment data from previous sessions, establishing a big data warehouse for power safety training; a model optimization module, which regularly updates and optimizes the machine learning models in the system based on the continuously accumulated assessment data, improving the accuracy and adaptability of assessments; a remote assessment support module, which enables remote online assessments through internet technology, supporting trainees to conduct standardized intelligent assessments in distributed training venues; and an interface service module, which provides a data interface with existing power company human resource management systems to achieve automatic synchronization of assessment results and personnel qualification management.

[0044] The core data management module establishes a big data warehouse, providing a data foundation for long-term analysis and model optimization. The model optimization module uses accumulated new data to periodically iterate and train the model, making system evaluations increasingly accurate. The remote assessment support module, through internet technology, allows trainees in different locations to complete assessments under standardized monitoring, greatly improving the system's applicability and flexibility. The interface service module ensures that assessment results can be seamlessly integrated with the enterprise's HR system, achieving a closed loop in assessment and qualification management.

[0045] The aforementioned scheme allows this method to move beyond analyzing coking images in isolation. Instead, it combines operating parameters such as load and fuel to predict coking trends through regression models. This enables the system to shift from passive identification to active prediction, directly feeding back to the control system. For example, it can automatically adjust the air-fuel ratio to improve combustion conditions and suppress coking at its source.

[0046] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A machine learning-based intelligent assessment and evaluation system for power safety training, characterized in that, Includes the following modules: The data acquisition module is used to acquire multi-source data generated by trainees during the training and assessment process. The multi-source data includes theoretical answer data, practical operation video data, physiological signal data, and voice interaction data. The theoretical answer data is collected through the power safety knowledge question bank system, the practical operation video data is collected through multi-angle high-definition cameras deployed in the training site, the physiological signal data is collected through wearable devices to collect trainees' heart rate, skin conductance response, and electromyography signals, and the voice interaction data is collected through microphone arrays to collect trainees' voice descriptions and questions and answers during the operation process. The behavior analysis module uses computer vision technology and deep learning algorithms to process practical video data. It extracts the characteristics of trainees' operation actions through a pre-trained behavior recognition model and compares and analyzes them with a standard operation behavior library. The behavior analysis module includes an action standardization evaluation unit, a safety protection equipment usage detection unit, and a dangerous behavior identification unit. The action standardization evaluation unit is built based on a temporal convolutional network and can analyze the accuracy and fluency of trainees' operation actions frame by frame. The safety protection equipment usage detection unit uses a target detection algorithm to monitor the wearing of safety helmets, insulating gloves, and protective clothing in real time. The dangerous behavior identification unit identifies operation behaviors that violate safety procedures through a behavior semantic analysis model. The knowledge assessment module uses natural language processing technology to analyze trainees' theoretical answers and voice interaction data. It constructs a question-and-answer semantic understanding model through a power safety knowledge graph. The knowledge assessment module includes a theoretical knowledge mastery assessment unit and a safety awareness assessment unit. The theoretical knowledge mastery assessment unit uses a deep question-and-answer model to conduct deep semantic analysis on trainees' theoretical answers. The safety awareness assessment unit evaluates trainees' safety risk identification ability and emergency response awareness by analyzing their decision-making process and voice description in simulated scenarios. The comprehensive evaluation module, based on the output results of the multilayer perceptron network fusion behavior analysis module and knowledge assessment module, combined with the trainee status information provided by physiological signal data, generates a comprehensive assessment report. The comprehensive evaluation module includes a multimodal data fusion unit and a dynamic weight adjustment unit. The multimodal data fusion unit uses an attention mechanism to weight and fuse different types of assessment indicators, and the dynamic weight adjustment unit adaptively adjusts the weight coefficients of each assessment indicator according to different assessment scenarios. The feedback generation module automatically generates personalized training suggestions and assessment feedback based on the output of the comprehensive evaluation module. The feedback generation module includes a weakness identification unit and a training content recommendation unit. The weakness identification unit uses cluster analysis to find the knowledge blind spots and operational errors of trainees, while the training content recommendation unit recommends targeted reinforcement training content for each trainee based on a collaborative filtering algorithm.

2. The intelligent assessment and evaluation system for power safety training based on machine learning according to claim 1, characterized in that, The behavior analysis module further includes: a 3D motion reconstruction unit, which reconstructs the 3D skeletal joint sequence of the trainee's operation process using multi-view video data, and extracts spatiotemporal features of the joint sequence using a graph convolutional network; an operation process compliance check unit, which models the time dependence of operation actions based on a long short-term memory network, aligns and compares the trainee's operation sequence with the standard operation process, and identifies missing or incorrectly ordered operation steps; a safety distance monitoring unit, which uses scene depth information obtained by a depth camera to calculate the safety distance between the trainee's body and live equipment or hazardous sources in real time, and generates a warning signal when the safety distance is insufficient; and a behavior trajectory analysis unit, which extracts the trainee's movement trajectory during the operation process using optical flow and feature tracking technology, and analyzes whether their movement conforms to the work specifications and safety requirements.

3. The intelligent assessment and evaluation system for power safety training based on machine learning according to claim 2, characterized in that, The knowledge assessment module also includes: a concept comprehension depth assessment unit, which uses a pre-trained power industry language model to perform deep semantic analysis on trainees' theoretical answers, constructs a knowledge mastery hierarchy model, and distinguishes between trainees' rote and applied understanding of safety knowledge; a scenario reasoning ability assessment unit, which constructs virtual accident scenarios based on reinforcement learning, and assesses trainees' ability to transfer and apply safety knowledge and their emergency reasoning ability by analyzing their decision-making paths and solutions in simulated accident handling; a safety hazard identification assessment unit, which shows trainees images or videos of work scenarios containing potential safety risks, and assesses their safety risk perception acuity and hazard identification completeness through eye-tracking technology and voice report analysis; and a knowledge correlation analysis unit, which uses graph neural networks to analyze the completeness and coherence of trainees' knowledge structure, and assesses the systematic nature of their safety knowledge system by calculating the correlation strength between different safety knowledge points.

4. The intelligent assessment and evaluation system for power safety training based on machine learning according to claim 2, characterized in that, The operational procedure compliance inspection unit specifically includes: a step timing verification subunit, which uses a dynamic time warping algorithm to match the time sequence of trainees' actions with a standard operational procedure template, calculates the temporal similarity between the two, and identifies operational steps with incorrect order or excessive time deviation; a tool usage standardization detection subunit, which analyzes the correctness of tool selection and the standardization of usage during trainees' operations based on fine-grained image recognition technology, and detects unsafe behaviors such as tool misuse and abuse; a collaborative operation evaluation subunit, which tracks the behavior of different trainees separately in team collaboration assessment scenarios using multi-target tracking technology, analyzes the rationality of their division of labor, communication and coordination, and mutual protection safety, and evaluates the overall safety collaboration level of the team; and an emergency response evaluation subunit, which simulates sudden situations such as equipment failure and electric shock, records and analyzes the trainees' emergency response time, the correctness of their handling procedures, and the effectiveness of their risk avoidance measures.

5. The intelligent assessment and evaluation system for power safety training based on machine learning according to claim 4, characterized in that, The comprehensive evaluation module also includes: a standardization unit for assessment indicators, which uses fuzzy mathematics theory to normalize different types of assessment indicators, eliminate differences in the dimensions and value ranges of each indicator, and establish a unified evaluation scale; a credibility evaluation unit, which evaluates the credibility of each assessment indicator by analyzing the consistency and complementarity between multi-source data, and downweights indicators with low credibility; a personalized evaluation model unit, which constructs a personalized assessment model based on the trainees' historical assessment data and job characteristics, and adaptively adjusts the evaluation criteria considering the trainees' experience level and differences in job responsibilities; and a visualization unit for assessment results, which presents the comprehensive evaluation results in visual forms such as radar charts and heat maps, intuitively showing the trainees' performance level and improvement directions in each assessment dimension.

6. The intelligent assessment and evaluation system for power safety training based on machine learning according to claim 3, characterized in that, The concept comprehension depth assessment unit specifically includes: a knowledge hierarchy division subunit, which divides power safety knowledge into five mastery levels: memorization, understanding, application, analysis, and evaluation, and assesses trainees' mastery at each level by designing assessment questions of varying difficulty and depth; a misconception detection subunit, which identifies potential safety knowledge misunderstandings and conceptual confusions by analyzing the semantic features and problem-solving processes of trainees' incorrect answers, and establishes a misconception concept map; a knowledge application and transfer subunit, which designs assessment scenarios similar to but not identical to the training scenarios to assess trainees' ability to transfer learned safety knowledge to new scenarios; and a knowledge forgetting monitoring subunit, which constructs a knowledge forgetting curve model for trainees by periodically and repeatedly testing key safety knowledge points, and assesses their long-term memory of core safety knowledge.

7. The intelligent assessment and evaluation system for power safety training based on machine learning according to claim 1, characterized in that, The feedback generation module further includes: a capability profile construction unit, which constructs a multi-dimensional profile of the trainee's safety capabilities based on multiple assessment data and dynamically tracks the trajectory of capability development; an adaptive learning path planning unit, which plans a personalized learning path for the trainee based on their capability profile and weaknesses, and recommends the most suitable training content and learning resources; a training effect prediction unit, which uses a time series prediction model to predict the improvement effect of the trainee's capabilities after specified training based on the trainee's historical learning data and current capability level; and an intervention strategy recommendation unit, which generates graded warnings and intervention strategy suggestions when a trainee is detected to have serious safety knowledge deficiencies or habitual violation tendencies, including different levels of measures such as enhanced training, key tutoring, and job adjustment.

8. The intelligent assessment and evaluation system for power safety training based on machine learning according to claim 5, characterized in that, The personalized evaluation model unit specifically includes: a job adaptability analysis subunit, which establishes a job competency model based on the safety responsibilities of different positions and assesses the matching degree between trainees and target positions; an experience level compensation subunit, which considers trainees' years of work experience and historical safety records, sets differentiated qualification standards for junior trainees and senior employees, and appropriately adjusts the evaluation threshold while maintaining core safety requirements; a learning curve analysis subunit, which analyzes the changing trends of trainees' past assessment scores to assess their learning progress speed and stability, and provides appropriate encouragement scores to trainees who have made significant progress but whose current level is still low; and a risk assessment subunit, which predicts the types and probabilities of safety risks that may occur in actual work based on trainees' assessment performance and personality characteristics, and generates risk warning reports.

Images