Self-adaptive pushing method of safety production training plan and knowledge base updating system

Abstract

The invention discloses a self-adaptive pushing method for a safety production training plan and a knowledge base updating system, and aims to solve the problems of insufficient training pertinence and knowledge base updating lagging in the prior art. The method comprises the steps that a user portrait is constructed to obtain basic attributes and historical data, interests are quantified through an interestingness model (combining click frequency, learning duration ratio and assessment score normalized value), and after a preliminary plan is generated, the priority is adjusted according to the learning progress, and pushing is carried out in combination with time preference. The system collects information through multiple channels, updates a knowledge base after preprocessing and multi-dimensional evaluation (association degree, timeliness and practicability), establishes an association relationship and verifies accuracy. According to the invention, personalized pushing of the training plan and efficient updating of the knowledge base are realized, the training accuracy and timeliness are improved, and the method is suitable for safety production training in various industries.

Description

Technical Field

[0001] This invention relates to the field of safety production education and training technology, specifically an adaptive push method for safety production training plans and a knowledge base update system. Background Technology

[0002] Safety in production is a core prerequisite for the operation and development of all industries, and effective safety training is a key means to improve the safety awareness and operational skills of employees and prevent accidents. With the increasing complexity of industrial production and the continuous improvement of safety regulations, the breadth and frequency of updates to safety training content are facing higher demands.

[0003] Currently, there are two prominent problems in the field of safety production training:

[0004] First, training programs lack specificity. Existing training often adopts a "one-size-fits-all" approach, developing uniform courses based solely on industry standards without considering differences in users' job roles (such as different safety skill requirements for operational and management positions), length of service (new employees need basic safety knowledge, while experienced employees need advanced skills updates), and their historical learning trajectory (such as their mastery of specific safety modules). This results in a low match between training content and users' actual needs, leading to a "disconnect between learning and application," which wastes training resources and makes it difficult to guarantee training effectiveness.

[0005] Secondly, the training knowledge base update mechanism is lagging behind. Safety-related laws, regulations, policies, accident cases, and protective technologies (such as operating procedures for new safety equipment) are constantly evolving, but the existing knowledge base largely relies on manual screening and input, resulting in long update cycles and vague evaluation standards. For example, judging the importance of new information often depends on subjective experience, lacking quantitative evaluation models. This easily leads to the omission of key information or the accumulation of redundant information, causing training content to lag behind actual safety management needs and affecting the timeliness and accuracy of training.

[0006] Therefore, there is an urgent need to build a technical solution that can adaptively push training plans based on user characteristics and scientifically and efficiently update the knowledge base in order to improve the accuracy and timeliness of safety production training. Summary of the Invention

[0007] The technical problem this invention aims to solve is that existing safety production training lacks targeted delivery of training plans, fails to achieve personalized delivery based on user job position, length of employment, and historical learning trajectory, and the training knowledge base is subject to manual updates, has a long cycle, and vague evaluation standards, resulting in content lagging behind actual needs, thus affecting the effectiveness and timeliness of training.

[0008] The technical solution adopted in this invention is: an adaptive push method for safety production training plans, comprising the following steps:

[0009] Step 1: Construct user profiles for safety production training, and obtain users' basic attribute information and historical training data;

[0010] Step 2: Based on user profiles, determine users' interest in each training content using an interest calculation model. The interest calculation model is as follows:

[0011] I i,j =α·A i,j +β·B i,j +γ·C i,j ;

[0012] Among them, I i,j A represents the level of interest of the i-th user in the j-th training content; i,j B represents the historical click frequency of the i-th user on the j-th training content, with a value range of [0,1]. i,j This represents the percentage of historical learning time for the i-th user on the j-th training content, with a value range of [0,1]; C i,j The value represents the normalized historical assessment score of the i-th user for the j-th training content, ranging from [0,1]; α, β, and γ represent A... i,j B i,j C i,j The weighting coefficients are α+β+γ=1; the purpose of this formula is to comprehensively consider various aspects of user behavior data on training content, quantitatively evaluate user interest, and provide a basis for subsequent push notifications.

[0013] Step 3: Based on the user's interest in each training content and in accordance with the mandatory learning requirements of safety production training, generate a preliminary training plan;

[0014] Step 4: Make adaptive adjustments to the initial training plan to obtain the final personalized training plan and push it to the user.

[0015] As a further aspect of the present invention: in step 1, the basic attribute information includes the user's industry, position, and start date; the historical training data includes the training content the user has learned, the learning duration, the assessment results, and the click records.

[0016] As a further aspect of the present invention: In step 4, the method for adaptively adjusting the preliminary training plan is as follows: Based on the user's learning progress data, if the user has not completed the learning of a certain training content within a preset time, the priority of that training content in the plan is adjusted, and the adjustment formula is:

[0017] P j′ =P j ·(1+δ·(Tt j ) / T);

[0018] Among them, P j′ To adjust the priority of the j-th training content; P j The priority of the j-th training content before adjustment is δ; δ is the adjustment coefficient, with a value range of [0,1]; T is the planned total learning time for this training content; t j This represents the duration the user has already studied the training content. The purpose of this formula is to dynamically increase the priority of incomplete content based on the user's learning progress, thereby urging the user to complete the learning.

[0019] A safety production training knowledge base update system includes:

[0020] The data acquisition module is used to collect information on new regulations, new cases, and new technologies related to safe production.

[0021] The information processing module is used to preprocess the collected information, including deduplication and format standardization;

[0022] The knowledge assessment module is used to evaluate the importance of new information. The assessment formula is as follows:

[0023] S k =λ·D k +μ·R k +ν·U k ;

[0024] Among them, S k D represents the importance score of the k-th new piece of information; k R represents the degree of relevance between the k-th new piece of information and the existing knowledge base, with a value range of [0,1]. k U represents the timeliness score of the k-th new information, with a value range of [0,1]; k The usefulness score of the k-th new information is represented by λ, μ, and ν, which range from [0,1]; D represents the usefulness score of the new information. k R k U k The weights are given by λ+μ+ν=1; the function of this formula is to evaluate the value of new information from multiple dimensions and determine whether it is worth including in the knowledge base.

[0025] The knowledge update module is used to update the safety production training knowledge base with new information that has passed the assessment.

[0026] As a further aspect of the present invention, the data acquisition module collects information from channels including the official websites of government safety supervision departments, industry association publishing platforms, safety accident case databases, and relevant academic journal databases.

[0027] As a further aspect of this invention: In step 2, before calculating the interest level, the historical training data needs to be normalized. The normalization formula for historical assessment scores is:

[0028] C i,j =(S i,j -S min ) / (S max -S min );

[0029] Among them, S i,j S represents the assessment score of the i-th user for the j-th training content; min S is the minimum score among all user assessment scores for this training content. max This is the maximum value among all user assessment scores for this training content; the purpose of this formula is to convert assessment scores of different ranges to the [0,1] interval, making it easier to perform comprehensive calculations with other indicators.

[0030] As a further aspect of this invention: when updating the knowledge base, the knowledge update module also needs to establish a correlation between new information and existing knowledge, and the correlation strength is calculated as follows:

[0031]

[0032] Among them, L k,m N represents the correlation strength between the k-th new piece of information and the m-th existing knowledge, with a value range of [0,1]. k,m N represents the number of keywords shared by the k-th new piece of information and the m-th existing piece of knowledge; k N represents the total number of keywords contained in the k-th new message; m Let m be the total number of keywords contained in the m-th existing knowledge; the purpose of this formula is to quantify the degree of correlation between new information and existing knowledge, providing a basis for the construction of a knowledge system.

[0033] As a further aspect of the present invention: In step 4, when pushing the final personalized training plan, it is also necessary to consider the user's learning time preferences, determine the optimal push time window based on the user's historical learning time distribution, and the selection of the time window satisfies the following:

[0034] W={t|f(t)≥θ·max(f(t))};

[0035] Where W is the optimal push time window; f(t) is the learning probability density function of the user at time t; θ is the threshold, with a value range of [0.5, 0.8]; max(f(t)) is the maximum value of f(t); the function of this formula is to select the time period with a higher probability of user learning as the push window, thereby improving the user's acceptance of the training plan and learning efficiency.

[0036] As a further aspect of the present invention, it also includes a knowledge base verification module for verifying the accuracy of the updated knowledge base, wherein the verification metrics are:

[0037]

[0038] Where V is the accuracy verification index of the knowledge base, with a value range of [0,1]; C is the number of correct knowledge points confirmed during the verification process; and E is the number of incorrect knowledge points discovered during the verification process. The purpose of this formula is to quantitatively evaluate the accuracy of the updated knowledge base and ensure the reliability of the knowledge.

[0039] As a further aspect of this invention, it also includes a training effectiveness feedback stage. Based on user feedback data after completing the training, the weight coefficients in the interest calculation model are optimized. The optimization method employs gradient descent, and the objective function is:

[0040]

[0041] in, This represents the actual interest level of the i-th user towards the j-th training content, determined based on user feedback. The purpose of this formula is to continuously optimize the model and improve the accuracy of interest assessment by minimizing the error between the calculated interest level and the actual interest level.

[0042] The beneficial effects of this invention are:

[0043] 1. To address the lack of targeted delivery of existing training plans, this invention constructs user profiles, combines basic attribute information and historical training data, and employs an interest calculation model that includes click frequency, learning time percentage, and normalized assessment scores to quantify user interest. It dynamically adjusts the priority of training content based on learning progress and determines the optimal delivery window by considering user learning time preferences, thus achieving personalized and adaptive delivery of training plans. This effectively solves the problem of low content-demand matching in the "one-size-fits-all" approach, avoids a disconnect between learning and application, and improves the utilization efficiency and effectiveness of training resources.

[0044] 2. To address the issue of lagging knowledge base update mechanisms, the system of this invention collects new information through multiple channels. After preprocessing, it uses a quantitative evaluation model based on dimensions such as relevance, timeliness, and practicality to determine the importance of the new information. Simultaneously, it establishes connections between new information and existing knowledge, and ensures update quality through an accuracy verification module, thus forming a scientific and efficient knowledge base update mechanism. This solves the problems of long manual update cycles and vague evaluation standards, avoids the omission of key information or the accumulation of redundant information, ensures that training content keeps pace with the dynamic changes in regulations, cases, and technologies, and improves the timeliness and accuracy of training. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating the adaptive push method for the safety production training plan and the adaptive push method for the knowledge base update system of the present invention.

[0046] Figure 2 The flowchart of the knowledge base update system for the adaptive push method of the safety production training plan and the knowledge base update system of the present invention is shown. Detailed Implementation

[0047] The adaptive push method and knowledge base update system for safety production training plans described in this invention aim to achieve personalized and timely safety production training through user profile construction, quantitative evaluation of interests, dynamic adjustment of plans, and precise updates of the knowledge base. The following detailed implementation process is illustrated with multiple embodiments in conjunction with specific application scenarios, and comparative analysis is provided.

[0048] Example 1: Safety Production Training Scenario in Basic Manufacturing Industry

[0049] 1. Implementation steps of the adaptive push method for safety production training plans

[0050] Step 1: Build user profiles

[0051] Data was collected from 100 employees of a heavy machinery factory. Basic attribute information included:

[0052] Industry: Heavy machinery manufacturing;

[0053] Positions: Operators (60 people), Maintenance personnel (30 people), Management personnel (10 people);

[0054] Length of service: 0-1 years (new employees, 20 people), 1-5 years (veteran employees, 50 people), 5 years or more (senior employees, 30 people).

[0055] Historical training data includes:

[0056] Learning content includes: safe operation of machinery, equipment maintenance procedures, emergency plans, etc.

[0057] Learning duration: The average learning duration for a single course is 2-8 hours;

[0058] Assessment score: 60-95 points;

[0059] Click records: Operators clicked "Machinery Safety Operation" 80 times per person, and maintenance staff clicked "Equipment Maintenance Specifications" 60 times per person.

[0060] Step 2: Calculate interest level

[0061] Normalize the historical data:

[0062] Normalization of assessment results (taking "Safe Operation of Machinery" as an example):

[0063] Given that employee A in a certain operation position has a performance evaluation score S i,j=85 points, the minimum score S among all users of this course min =60 points, maximum value S max =95 points, according to the formula:

[0064] C i,j =(S i,j -S min ) / (S max -S min )

[0065] We can get C i,j = (85-60) / (95-60) = 25 / 35 ≈ 0.714.

[0066] Click frequency A i,j The click frequency of operators on "Safe Operation of Machinery" is 0.8 (80 times out of a total of 100 clicks);

[0067] Learning time percentage B i,j The employee spent 5 hours on the course, which is 83.3% of the planned total of 6 hours, i.e., B. i,j =0.833.

[0068] We set weights α = 0.3 (click frequency), β = 0.4 (learning time), and γ = 0.3 (assessment score), based on the interest rate formula:

[0069] I i,j =α·A i,j +β·B i,j +γ·C i,j

[0070] We can obtain I i,j =0.3×0.8+0.4×0.833+0.3×0.714≈0.24+0.333+0.214≈0.787.

[0071] Step 3: Generate a preliminary training plan

[0072] Required learning content: National machinery safety regulations (mandatory for all employees);

[0073] Based on interest levels: "Advanced Mechanical Safety Operation" (interest level 0.787) is prioritized for operators, and "Emergency Handling of Equipment Failure" (interest level 0.65) is prioritized for maintenance personnel.

[0074] Step 4: Adaptive Adjustment and Push

[0075] Priority adjustment (taking an employee's failure to complete the "emergency plan" as an example):

[0076] Given the original priority P j=0.5, adjustment factor δ=0.6, planned total duration T=4 hours, already studied duration t j = 1 hour, according to the formula:

[0077] P j′ =P j ·(1+δ·(Tt j ) / T)

[0078] P can be obtained j′ =0.5×[1+0.6×(4-1) / 4]=0.5×(1+0.6×0.75)=0.5×1.45=0.725.

[0079] Optimal push window:

[0080] Based on the employee's historical learning data, the learning probability density function f(t) reaches its peak between 19:00 and 21:00, with a threshold θ = 0.6 and max(f(t)) = 0.8. The time period that satisfies f(t) ≥ 0.6 × 0.8 = 0.48 is 18:30 to 21:30. Therefore, W = [18:30, 21:30].

[0081] 2. Implementation Steps for the Safety Production Training Knowledge Base Update System

[0082] Data acquisition module:

[0083] Data collection channels: official websites of government safety supervision departments (new regulations), case databases of machinery industry associations (accident cases), and the "China Safety Science Journal" (new technologies).

[0084] Information processing module:

[0085] The 100 collected information items were deduplicated (5 duplicate legal provisions were removed) and the format was standardized (unified as "title + publication time + core content" structure).

[0086] Knowledge assessment module:

[0087] Taking the "New National Standard for Machinery Safety (Released in 2024)" as an example:

[0088] Relevance D k =0.9 (highly correlated with existing "machine safety standards");

[0089] Timeliness R k =1.0 (released 3 months ago);

[0090] Practicality U k =0.8 (direct guidance on operating procedures);

[0091] Given weights λ = 0.3, μ = 0.4, and ν = 0.3, according to the formula:

[0092] S k =λ·D k +μ·R k +ν·U k

[0093] S can be obtained k =0.3×0.9+0.4×1.0+0.3×0.8=0.27+0.4+0.24=0.91 (Through evaluation, S k ≥0.6 is the passing threshold.

[0094] Knowledge update module:

[0095] Correlation strength calculation (new national standard and existing "mechanical safety operation" knowledge):

[0096] There are N keywords in total. k,m =5 (e.g., "protective shield", "emergency stop device", etc.), total number of new information keywords N k =8, Total number of existing knowledge keywords N m =10, according to the formula:

[0097]

[0098] L can be obtained k,m =5 / (8+10-5)=5 / 13≈0.385, establish the association relationship.

[0099] Knowledge base verification module:

[0100] Fifty updated knowledge entries were randomly selected for verification, of which C=48 (correct) and E=2 (incorrect). According to the formula:

[0101]

[0102] We can obtain V = 48 / 50 = 0.96 (verification passed, V ≥ 0.8 is the acceptable threshold).

[0103] Example 2: Safety Production Training Scenario in the Construction Industry (Emphasis on Assessment Score Weighting)

[0104] 1. Differences in push methods

[0105] In step 2, the interest weights are adjusted to α = 0.2, β = 0.2, and γ = 0.6 (the construction industry has higher requirements for assessment results).

[0106] A construction worker's opinion on "safety of working at heights" (A) i,j =0.6, B i,j =0.5, C i,j =0.9 (after normalization), then I i,j=0.2×0.6+0.2×0.5+0.6×0.9=0.12+0.1+0.54=0.76.

[0107] 2. Differences in Knowledge Base Updates

[0108] The knowledge assessment weights are adjusted to λ = 0.2, μ = 0.5, and ν = 0.3 (building regulations are frequently updated, so timeliness is emphasized).

[0109] A newly released "scaffolding safety regulation" k =0.2×0.8+0.5×1.0+0.3×0.7=0.16+0.5+0.21=0.87.

[0110] Example 3: Safety Production Training Scenario in the Chemical Industry (Focusing on Adjusting the Learning Schedule)

[0111] 1. Differences in push methods

[0112] In step 4, the adjustment coefficient δ = 0.8 (chemical training is high-risk and requires strict supervision of progress).

[0113] An employee failed to complete the "Hazardous Chemicals Storage Standards"; T = 6 hours; t j =2 hours, P j =0.4, then P j′ =0.4×[1+0.8×(6-2) / 6]=0.4×(1+0.8×0.67)=0.4×1.536≈0.614.

[0114] 2. Differences in Knowledge Base Updates

[0115] Data collection focuses on the "Hazardous Chemicals Accident Case Database," accounting for 60% (the cases are more critical to the chemical industry).

[0116] Comparative Analysis of Examples

[0117]

[0118]

[0119] in conclusion

[0120] This invention, by dynamically adjusting model parameters (weights, coefficients, etc.), can adapt to the safety production training needs of different industries.

[0121] 1. Basic manufacturing industries need to balance user behavior data to ensure comprehensive training;

[0122] 2. The construction industry needs to strengthen assessments and ensure the timeliness of regulations to guarantee compliant operations;

[0123] 3. The chemical industry needs to strictly supervise the learning progress and link it with case studies to reduce risks.

[0124] Through multi-dimensional evaluation and precise updates of the knowledge base, the timeliness of training content can be improved by more than 30%, and personalized push can improve learning efficiency by 25%-40%.

[0125] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present 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 present invention.

Claims

1. An adaptive push method for safety production training plans, characterized in that, Includes the following steps: Step 1: Construct user profiles for safety production training, and obtain users' basic attribute information and historical training data; Step 2: Based on user profiles, determine users' interest in each training content using an interest calculation model. The interest calculation model is as follows: I i,j =α·A i,j +β·B i,j +γ·C i,j ; Among them, I i,j A represents the level of interest of the i-th user in the j-th training content; i,j B represents the historical click frequency of the i-th user on the j-th training content, with a value range of [0,1]. i,j This represents the percentage of historical learning time for the i-th user on the j-th training content, with a value range of [0,1]; C i,j The value represents the normalized historical assessment score of the i-th user for the j-th training content, ranging from [0,1]; α, β, and γ represent A... i,j B i,j C i,j The weighting coefficients are α+β+γ=1; the purpose of this formula is to comprehensively consider various aspects of user behavior data on training content, quantitatively evaluate user interest, and provide a basis for subsequent push notifications. Step 3: Based on the user's interest in each training content and in accordance with the mandatory learning requirements of safety production training, generate a preliminary training plan; Step 4: Make adaptive adjustments to the initial training plan to obtain the final personalized training plan and push it to the user.

2. The adaptive push method for a safety production training plan according to claim 1, characterized in that, In step 1, the basic attribute information includes the user's industry, position, and start date; the historical training data includes the training content the user has learned, the learning duration, the assessment results, and the click records.

3. The adaptive push method for a safety production training plan according to claim 1, characterized in that, In step 4, the initial training plan is adjusted adaptively as follows: based on user learning progress data, if a user fails to complete the learning of a certain training content within a preset time, the priority of that training content in the plan is adjusted. The adjustment formula is: P j′ =P j ·(1+δ·(T-t j ) / T); Among them, P j′ To adjust the priority of the j-th training content; P j The priority of the j-th training content before adjustment is δ; δ is the adjustment coefficient, with a value range of [0,1]; T is the planned total learning time for this training content; t j This represents the duration the user has already studied the training content. The purpose of this formula is to dynamically increase the priority of incomplete content based on the user's learning progress, thereby urging the user to complete the learning.

4. A safety production training knowledge base update system, characterized in that, include: The data acquisition module is used to collect information on new regulations, new cases, and new technologies related to safe production. The information processing module is used to preprocess the collected information, including deduplication and format standardization; The knowledge assessment module is used to evaluate the importance of new information. The assessment formula is as follows: S k =λ·D k +μ·R k +v·U k ; Among them, S k D represents the importance score of the k-th new piece of information; k R represents the degree of relevance between the k-th new piece of information and the existing knowledge base, with a value range of [0,1]. k U represents the timeliness score of the k-th new information, with a value range of [0,1]; k The usefulness score of the k-th new information is represented by λ, μ, and ν, which range from [0,1]; D represents the usefulness score of the new information. k R k U k The weights are given by λ+μ+ν=1; the function of this formula is to evaluate the value of new information from multiple dimensions and determine whether it is worth including in the knowledge base. The knowledge update module is used to update the safety production training knowledge base with new information that has passed the assessment.

5. A safety production training knowledge base update system according to claim 4, characterized in that, The data acquisition module collects information from sources including the official websites of government safety supervision departments, industry association publishing platforms, safety accident case databases, and relevant academic journal databases.

6. The adaptive push method for a safety production training plan according to claim 1, characterized in that, In step 2, before calculating the interest level, the historical training data needs to be normalized. The normalization formula for historical assessment scores is as follows: C i,j =(S i,j -S min ) / (S max -S min ); Among them, S i,j S represents the assessment score of the i-th user for the j-th training content; min S is the minimum score among all user assessment scores for this training content. max This is the maximum value among all user assessment scores for this training content; the purpose of this formula is to convert assessment scores of different ranges to the [0,1] interval, making it easier to perform comprehensive calculations with other indicators.

7. A safety production training knowledge base update system according to claim 4, characterized in that, When updating the knowledge base, the knowledge update module also needs to establish the association between new information and existing knowledge. The association strength is calculated as follows: Among them, L k,m N represents the correlation strength between the k-th new piece of information and the m-th existing knowledge, with a value range of [0,1]. k,m N represents the number of keywords shared by the k-th new piece of information and the m-th existing piece of knowledge; k N represents the total number of keywords contained in the k-th new message; m Let m be the total number of keywords contained in the m-th existing knowledge; the purpose of this formula is to quantify the degree of correlation between new information and existing knowledge, providing a basis for the construction of a knowledge system.

8. The adaptive push method for a safety production training plan according to claim 1, characterized in that, In step 4, when pushing the final personalized training plan, it is also necessary to consider the user's learning time preferences and determine the optimal push time window based on the user's historical learning time distribution. The selection of the time window satisfies the following: W={t|f(t)≥θ·max(f(t))}; Where W is the optimal push time window; f(t) is the learning probability density function of the user at time t; θ is the threshold, with a value range of [0.5, 0.8]; max(f(t)) is the maximum value of f(t); the function of this formula is to select the time period with a higher probability of user learning as the push window, thereby improving the user's acceptance of the training plan and learning efficiency.

9. A safety production training knowledge base update system according to claim 4, characterized in that, It also includes a knowledge base validation module, used to verify the accuracy of the updated knowledge base. The validation metrics are: Where V is the accuracy verification index of the knowledge base, with a value range of [0,1]; C is the number of correct knowledge points confirmed during the verification process; and E is the number of incorrect knowledge points discovered during the verification process. The purpose of this formula is to quantitatively evaluate the accuracy of the updated knowledge base and ensure the reliability of the knowledge.

10. The adaptive push method for a safety production training plan according to claim 1, characterized in that, It also includes a training effectiveness feedback stage. Based on user feedback data after completing the training, the weight coefficients in the interest calculation model are optimized using the gradient descent method. The objective function is: in, This represents the actual interest level of the i-th user towards the j-th training content, determined based on user feedback. The purpose of this formula is to continuously optimize the model and improve the accuracy of interest assessment by minimizing the error between the calculated interest level and the actual interest level.

Images