Pre-class disclosure AI intelligent device

Abstract

The invention relates to the technical field of safety management, and particularly discloses a pre-class disclosure AI intelligent device. The system comprises a behavior acquisition module which extracts facial sight, limb postures and group interaction features by acquiring personnel video streams, and performs weighted mapping to obtain attention distraction indexes; the overlapping analysis module identifies a pre-class distraction person based on the attention distraction index, and judges and triggers an overlapping analysis signal in combination with environmental interference; the position optimization module constructs a position transfer model through a random forest algorithm, adjusts the position of a distraction person and verifies the feasibility; the pre-class evaluation module calculates distraction optimization coefficients before and after position adjustment and judges information transmission effectiveness; the pre-class optimization module optimizes the time period, duration and content order of the key period based on the personnel content mastering degree and the concentrated distraction time period. The equipment is beneficial for improving the transmission effectiveness of disclosure information before class, and extracting safety and operation guidance before class for engineering construction.

Description

Technical Field

[0001] This invention relates to the field of safety management technology, specifically to an AI-powered intelligent device for pre-shift briefings. Background Technology

[0002] In the fields of highway, municipal, and building construction, pre-shift briefings are required for managers to explain to the workers the specific tasks of the day (such as the scope of rebar tying and concrete pouring procedures), potential risks on site (such as fall protection, mechanical collision avoidance, and temporary power supply regulations), safety protection requirements, and emergency response procedures.

[0003] However, the traditional pre-shift briefing model is difficult to ensure effective information transmission: on the one hand, workers often have problems with inattention during the briefing process, such as frequently looking away from the briefing display screen, whispering to others, or wandering around at will. However, managers can only make subjective judgments and cannot quantify the degree of workers' inattention. On the other hand, the core information in the briefing (such as the protective details of high-risk operations and key task nodes) is mostly concentrated in a specific period of time. If workers are distracted during this period, they are prone to incomplete information reception and fuzzy memory. Ultimately, these problems will directly affect subsequent work processes. That is, if workers do not fully grasp the precautions in the briefing, they may violate regulations (such as not wearing protective equipment as required) or ignore risks and hidden dangers, which will increase the probability of work safety risks and quality problems.

[0004] Therefore, the present invention provides an AI-powered intelligent device for pre-shift briefing. Summary of the Invention

[0005] The purpose of this invention is to provide an AI-powered intelligent device for pre-shift briefings to address the aforementioned problems.

[0006] The objective of this invention can be achieved through the following technical solutions: A pre-shift briefing AI-powered smart device includes the following modules: Behavior collection module: used to collect the behavioral characteristics of personnel during pre-shift briefings, perform attention mapping processing on the behavioral characteristics, and obtain the attention dissipation index; Overlap Analysis Module: Based on the attention dissipation index, perform attention dissipation analysis on personnel, identify personnel who are inattentive before shift, extract the attention dissipation cycle of personnel who are inattentive before shift, as well as the key cycle of pre-shift briefing, perform time-series overlap analysis on the dissipation cycle and key cycle, obtain the overlap judgment coefficient, and determine whether the overlap analysis signal is triggered. Location optimization module: If the overlap analysis signal is triggered, the location of the pre-shift briefing for the personnel who are distracted before the shift is adjusted, and the attention distraction index of all personnel after the location adjustment is obtained. Pre-shift assessment module: Based on the adjusted attention dissipation index, the attention of the personnel conducting the pre-shift briefing is optimized and assessed to obtain the dissipation optimization coefficient. Based on the dissipation optimization coefficient, the effectiveness of optimization in key periods is analyzed to determine whether the effectiveness of information transmission during the pre-shift briefing meets the standards. Pre-shift optimization module: If the target is not met, it is used to obtain the cycle division characteristics of key cycles for employees who are not relaxed before the shift; construct a cycle division model, input the cycle division characteristics into the cycle division model, and adjust the key cycles based on the adjustment content of the pre-shift briefing output by the model.

[0007] As a further aspect of the present invention, the attention mapping process for the behavioral features is performed as follows: Among them, behavioral characteristics include: facial gaze characteristics, body posture characteristics, and group interaction characteristics; The inattention index is obtained by weighting and summing facial gaze features, body posture features, and group interaction features.

[0008] As a further aspect of the present invention: the method for obtaining the facial gaze features is as follows: Obtain the effective field of view for each person during the pre-shift briefing; if the viewing angle deviates from the effective field of view, obtain the duration of the deviation within the monitoring period. The deviation time is calculated as a proportion of the monitoring cycle time to obtain the deviation time ratio, which is then used as a facial gaze feature.

[0009] As a further aspect of the present invention, the method for performing the time series overlap analysis is as follows: Extract the period of distraction of the attention of the personnel who are distracted before the shift. If the period of distraction and the period of focus overlap in the time dimension, mark the overlapping period as the overlapping period and obtain the number of overlapping periods as the number of overlapping periods. Calculate the average disengagement index of individuals with pre-shift disengagement during overlapping periods; Obtain the number of key cycles, calculate the ratio of overlapping cycles to the total number of key cycles, and obtain the cycle overlap ratio. The overlap determination coefficient is obtained by multiplying the average dispersion index with the period overlap ratio.

[0010] As a further aspect of the present invention: the method for determining personnel who are disorganized before their shift is as follows: Obtain the characteristic variables of all personnel during the pre-shift briefing, and then use a clustering algorithm to cluster the characteristic variables of all personnel. Extract the individuals in the severely inattentive group from the clustering results, along with their corresponding inattention index. If the inattention index of the severely inattentive group shows an upward trend over N monitoring periods, then the individuals are marked as inattentive before their shift.

[0011] As a further aspect of the present invention, the method for obtaining the feature variables is as follows: Extract the number of consecutive periods of dispersion within N monitoring periods, and calculate the ratio of the number of consecutive periods to the number of monitoring periods N to obtain the dispersion frequency. The average distractibility index of an individual is obtained by calculating the mean of the distractibility index within the number of consecutive cycles of distractibility. The average individual disorganization index of all personnel during the pre-shift briefing is obtained to obtain the group average disorganization index; the deviation ratio between the individual average disorganization index and the group average disorganization index is calculated to obtain the disorganization intensity deviation. The dispersion intensity deviation, the individual's average dispersion index, and the dispersion frequency are used as characteristic variables for each person.

[0012] As a further aspect of the present invention, the method for optimizing the evaluation of the attention of personnel during pre-shift briefings is as follows: Obtain and calculate the change in the average disorganization index of a single pre-shift disorganized employee before and after the position adjustment, and obtain the disorganization adjustment amount; The ratio of the dispersion adjustment amount to the individual average dispersion index before adjustment is calculated to obtain the dispersion optimization coefficient of the overall cycle. Obtain the dispersion adjustment amount of the key period after the position adjustment, and calculate the ratio of the individual average dispersion index before the adjustment to obtain the dispersion optimization coefficient of the key period. If both the overall cycle's dispersion optimization coefficient and the key cycle's dispersion optimization coefficient are positive, and the average dispersion index of individuals within the key cycle is lower than the preset dispersion threshold, then it indicates that the attention regulation of a single pre-shift distracted person is effective.

[0013] As a further aspect of the present invention, the method for performing the optimization effectiveness analysis is as follows: Based on individual valid personnel data, key cycles are broken down into high-optimal cycles and medium-optimal cycles; By combining the individual average dispersion index before adjustment, a periodic optimization evaluation process is established to obtain the overall optimization coefficient for high-optimal periods and the overall optimization coefficient for medium-optimal periods. Obtain the violation rate of related operations within the high-optimization period, as well as the overlap duration between environmental incident interference and key periods; After excluding the overlap between environmental events and key cycles, the partial correlation coefficient is obtained by checking the partial correlation interference based on the overall optimization coefficient of the high-optimal cycle and the violation rate of related operations within the high-optimal cycle. Construct a criterion for the effectiveness of meeting the standards. If the overall optimization coefficient and the partial correlation coefficient of the high-optimal cycle and the medium-optimal cycle meet the criterion for the effectiveness of meeting the standards, then the information transmission effectiveness of the pre-shift briefing is met.

[0014] As a further aspect of the present invention, the method for performing the partial correlation interference verification is as follows: When calculating the partial correlation coefficients after controlling Z for the overall optimization coefficient X of the high-optimal cycle, the violation rate Y of the associated operations of the high-optimal cycle, and the overlap time Z of environmental interference and key cycles, the Pearson correlation coefficients of X and Y, X and Z, and Y and Z are calculated. The partial correlation coefficient is obtained by substituting the Pearson correlation coefficients of X and Y, X and Z, and Y and Z into the formula for the partial correlation coefficient.

[0015] As a further aspect of the present invention, the method for adjusting the key cycles is as follows: By using AI-powered smart devices to obtain the pre-shift awareness of disorganized personnel and the average disorganization index of key periods, a disorganization sequence is constructed. The disorganization sequence is then broken down and backtracked over time to identify concentrated disorganization periods. Attribution analysis was conducted by combining the degree of mastery, periods of concentrated and dispersed activity, and high-optimal and medium-optimal cycles to obtain attribution labels; The key periods, mastery levels, concentrated and scattered periods, and attribution labels of the pre-shift briefing content for key periods are used as the characteristics for period division. Obtain M periodicity features and construct a periodicity dataset containing the M periodicity features; based on a gradient regression model algorithm with multiple feature inputs, construct a periodicity model and output the time period, duration, and content order of the key periods in the pre-shift briefing content; The pre-shift briefing for the next cycle will be adjusted based on the output time period, duration, and content order.

[0016] The beneficial effects of this invention are: (1) The video stream of the pre-shift briefing is collected by panoramic camera and directional tracking camera. Combined with YOLOv8 algorithm, the individual image regions of the personnel are separated, and facial gaze, body posture and group interaction features are extracted and weighted and mapped into attention dissipation index. This is conducive to transforming the behavioral characteristics of personnel in the pre-shift briefing into quantitative attention state data, providing basic data support for subsequent personnel attention analysis and optimization, and realizing the monitoring and quantification of personnel attention state.

[0017] (2) Based on the attention dissipation index, identify the personnel who are distracted before the shift, extract the dissipation period and key period and perform time-series overlap analysis. At the same time, combine the environmental event to construct the interference misjudgment principle to judge whether the overlap analysis signal is triggered. This is helpful to locate the key personnel with attention dissipation in the key period and the time relationship, and provide a targeted judgment basis for whether the personnel position needs to be adjusted in the future.

[0018] (3) After triggering the overlap analysis signal, the basic location data of personnel and the data of dissipation characteristics are integrated to construct a dataset. The initial optimal position is output by constructing a position adjustment model through the random forest algorithm. The final optimal position is determined by multi-dimensional feasibility verification, which is conducive to improving the attention environment of dissipated personnel in the pre-shift briefing. The dissipation optimization coefficient is calculated based on the attention dissipation index after position adjustment. The evaluation equation set is constructed by combining the risk weights of the high-optimization period and the medium-optimization period. At the same time, the environmental influence is eliminated by partial correlation interference verification. The effectiveness of the pre-shift briefing information transmission is judged in multiple dimensions. The briefing effect can be comprehensively evaluated from the perspectives of individual optimization effect, risk dimension, environmental interference, etc., and the basis for the quality of pre-shift briefing information transmission is provided.

[0019] (4) When information transmission is not up to standard, the staff’s mastery of the content is obtained through rapid testing. The sliding window algorithm is used to locate the concentrated and scattered time periods and attribute the causes. Based on multiple features, the XGBoost gradient regression model is constructed to output the time period, duration and content order optimization scheme of the key period. This is conducive to providing data-driven direction for the adjustment of the key period of the subsequent pre-shift briefing and improving the transmission effect of key content in the pre-shift briefing. Attached Figure Description

[0020] The invention will now be further described with reference to the accompanying drawings.

[0021] Figure 1 This is a module diagram of an AI-powered pre-shift briefing device according to the present invention; Figure 2 This is a flowchart illustrating whether the overlap analysis signal is triggered in this invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0023] Example 1 Please see Figure 1 As shown, this invention is an AI-powered intelligent device for pre-shift briefings, comprising the following modules: Behavior collection module: used to collect the behavioral characteristics of personnel during pre-shift briefings, perform attention mapping processing on the behavioral characteristics, and obtain the attention dissipation index; The process of collecting behavioral characteristics of personnel during pre-shift briefings and performing attention mapping on these behavioral characteristics to obtain the attention dissipation index is as follows: Preferably, the panoramic camera and directional tracking camera of the AI-powered pre-shift briefing device are used to collect the video stream during the pre-shift briefing. It should be noted that pre-shift briefing is a pre-shift safety and work guidance communication session conducted by construction team managers with team workers before each day's work in the fields of highway, municipal, and building construction. The content of the communication includes: specific work tasks (such as work process content and construction scope), potential risks on site (such as falls from heights, mechanical collisions, electrical hazards, etc.), targeted safety protection requirements (such as standards for wearing protective equipment) and emergency response procedures, as well as work technical standards and precautions. Based on the video stream, the YOLOv8 algorithm is used to separate the independent image regions of each person during the pre-shift briefing. As will be understood by those skilled in the art, the YOLOv8 algorithm processes the pre-shift briefing video stream frame by frame by loading a pre-trained personnel target detection model, identifies personnel targets in each frame in real time and defines their bounding boxes, and then segments the independent image region of each person from the video stream image based on the bounding boxes. Capture facial gaze characteristics, body posture characteristics, and group interaction characteristics of people within a monitoring period from an independent image region; Facial gaze characteristics, body posture characteristics, and group interaction characteristics are used as behavioral characteristics of people; The method for collecting facial gaze features is as follows: The size of the pre-shift briefing display screen and the standing distance between personnel and both ends of the screen are obtained. Based on the optical line-of-sight algorithm, the effective field of view for each person is calculated. Obtain the viewing angle of the personnel giving pre-shift briefings and the display screen. If the viewing angle deviates from the effective field of view, obtain the duration of the deviation within the monitoring period. Those skilled in the art will understand that the effective field of view for each person is calculated as follows: based on the actual width and height of the pre-shift briefing display screen (e.g., 1.5m × 0.8m), and the horizontal standing distance between the person and the left and right edges of the display screen captured in real time by the directional tracking camera, and based on the geometric relationship of field of view angle = 2 × arctan (object height / 2 × object distance) in the optical line of sight algorithm, the horizontal field of view angles of the person looking at the left and right edges of the display screen, and the vertical field of view angles of the person looking at the top and bottom edges of the display screen are calculated respectively. The area enclosed by the field of view angles in the two directions is the effective field of view range in which the person can clearly receive the content of the display screen. The method for obtaining the perspective of the personnel and the display screen before the shift briefing is as follows: the center point of the personnel's eyes is located by the camera, and the center point of the display screen content area is combined with the coordinates of the center point of the display screen content area. The angle between the line connecting the center point of the eyes to the center point of the display screen content and the horizontal and vertical boundaries of the effective visible range is calculated. The angle is the actual perspective of the personnel and the display screen, and this is used to determine whether the perspective falls within the effective visible range. The deviation time ratio is calculated as the proportion of the deviation duration to the monitoring cycle length, and then used as a facial gaze feature. The frequency of invalid movements of personnel within the monitoring period is obtained as a limb posture feature; Acquire the cumulative duration for which a person's head turns towards other people and the cumulative duration for which a person leaves the preset area within the monitoring period; The two cumulative durations are summed, and the summation result is compared with the duration of the monitoring period to obtain the dispersion time ratio, which is used as a feature of group interaction. The attention dissipation index is obtained by weighted summation of facial gaze features, body posture features, and group interaction features, thereby enabling attention mapping of behavioral features. It should be noted that the weights were determined as follows: 500 complete records of pre-shift briefings were collected. Each record included: specific values ​​for three behavioral characteristics and corresponding inattention scores (the inattention score was obtained by combining manually observed inattention behavior records with the number of violations in the subsequent hour of work, and assigning a combined value of 0-1). Then, the Pearson correlation coefficient algorithm was used to calculate the correlation strength between each behavioral characteristic and the inattention score, resulting in correlation coefficients r1≈0.65 for facial gaze characteristics, r2≈0.45 for body posture characteristics, and r3≈0.35 for group interaction characteristics. The three correlation coefficients were then normalized, and the following weights were calculated: facial gaze feature weight = 0.65 / (0.65+0.45+0.35) = 0.5, body posture feature weight = 0.45 / 1.45≈0.3, and group interaction feature weight = 0.35 / 1.45≈0.2. It is understandable that the purpose of obtaining the Attention Difference Index is: Function 1: Providing a basis for identifying pre-shift inattentive employees. The inattention index is used to set a threshold value to mark the inattention cycle, and then calculate the inattention frequency, individual average inattention index, and inattention intensity deviation, forming a variable representing the employee's attention characteristics. Inputting these variables into a clustering algorithm can divide employees into normal, mild, and severe inattention groups. By monitoring the upward trend of the index in the severe inattention group, employees requiring special attention before their shifts can be identified.

[0024] Function Two: To provide a data foundation for the time-series overlap analysis of dispersion and key periods. When the attention dispersion index within a monitoring period is greater than or equal to a preset threshold, the period is marked as a dispersion period. The number of overlapping periods and the period overlap ratio between dispersion periods and key periods are statistically analyzed. Combined with the average dispersion index within the overlapping periods, the overlap determination coefficient is calculated. Then, combined with the interference misjudgment principle constructed from environmental events, it is determined whether the overlap analysis signal required for position adjustment is triggered.

[0025] Overlap Analysis Module: Based on the attention dissipation index, perform attention dissipation analysis on personnel, identify personnel who are inattentive before shift, extract the attention dissipation cycle of personnel who are inattentive before shift, as well as the key cycle of pre-shift briefing, perform time-series overlap analysis on the dissipation cycle and key cycle, obtain the overlap judgment coefficient, and determine whether the overlap analysis signal is triggered. Among them, the method for identifying pre-shift inattentive individuals by analyzing their inattention index is as follows: Obtain the attention dissipation index of a single person over N monitoring periods, set a threshold value for the attention dissipation index, and mark the monitoring period as a dissipation period if the attention dissipation index within a monitoring period is higher than or equal to the preset threshold value. Extract the number of consecutive periods of dispersion within N monitoring periods, calculate the ratio of the number of consecutive periods to the number of monitoring periods N, and obtain the dispersion frequency. The average distractibility index of an individual is obtained by calculating the mean of the distractibility index within the number of consecutive cycles of distractibility. The average individual disorganization index of all personnel during the pre-shift briefing is used to obtain the group average disorganization index. The deviation percentage between the individual average disorganization index and the group average disorganization index is calculated to obtain the disorganization intensity deviation. It should be noted that the deviation ratio is calculated as follows: (Individual average dispersion index - Group average dispersion index) / Group average dispersion index; The dispersion intensity deviation, the individual's average dispersion index, and the dispersion frequency are used as characteristic variables for each person. The characteristic variables of all personnel in the pre-shift briefing were obtained, and the characteristic variables of all personnel were clustered using the K-means clustering algorithm to obtain the normal disorganization group, the mild disorganization group, and the severe disorganization group. Obtain the information of the severely inattentive group and their corresponding inattention index. If the inattention index of the severely inattentive group shows an upward trend within N monitoring periods, then mark the person as an inattentive person before shift. Those skilled in the art will understand that the K-means clustering algorithm is used to cluster the feature variables of all individuals as follows: The feature variables of all individuals are organized into a data matrix; then, the number of clusters is set to 3 (corresponding to normal, mild, and severe disorganized groups). K-means first randomly selects 3 initial cluster centers; by iteratively calculating the distance of each individual to the 3 centers, individuals are assigned to the closest group, and the center position of each group is updated. This process is repeated until the centers no longer change; the final 3 groups will naturally distinguish normal, mild, and severe disorganized individuals based on differences in feature variables, providing a grouping basis for subsequent screening of severe groups and judging the upward trend of the index. The method for determining whether the attention deficit index shows an upward trend over N monitoring periods is as follows: organize the attention deficit index of the severely inattentive group over N monitoring periods in chronological order, and perform linear fitting with the monitoring period sequence as the horizontal axis and the attention deficit index as the vertical axis. If the slope of the fitted regression line is positive and the N / 2 fitted coordinate points are distributed above the line, it is determined that there is an upward trend. The key periods for extracting pre-shift briefings are as follows: In some embodiments, those skilled in the art can extract the time periods involving core information (such as safety specifications and key task nodes) based on the content flow of the pre-shift briefing, and use these as the key periods for the pre-shift briefing. It should be noted that the key periods for pre-shift briefings and the time divisions for monitoring periods are the same. The process of performing time-series overlap analysis on the dispersion period and the key period, obtaining the overlap determination coefficient, and determining whether the overlap analysis signal is triggered is as follows: If the dispersion cycle and the key cycle overlap in the time dimension, the overlapping cycle is marked as the overlapping cycle, and the number of overlapping cycles is obtained as the number of overlapping cycles. Calculate the average disengagement index of individuals with pre-shift disengagement during overlapping periods; Obtain the number of key cycles, calculate the ratio of overlapping cycles to the total number of key cycles, and obtain the cycle overlap ratio. The overlap determination coefficient is obtained by multiplying the average dispersion index with the period overlap ratio. Understandably, the physical meaning of the overlap determination coefficient is to quantify the degree of correlation between the attention deficit state of pre-shift disengaged personnel and the key period of pre-shift briefing (the time period for core information transmission) in the time dimension. It is calculated by multiplying the individual average attention deficit index (reflecting the severity of attention deficit during the overlapping period) within the overlapping period with the period overlap ratio (the proportion of overlapping periods to the number of key periods, reflecting the time overlap range between attention deficit and key periods). This reflects the actual degree to which the transmission of core information is affected by personnel attention deficit during the key period, providing a quantitative basis for determining whether an overlap analysis signal has been triggered. Simultaneously collect current environmental events and construct a principle for preventing interference and misjudgment; The method for collecting current environmental events and constructing the interference misjudgment principle is as follows: If a sudden environmental disturbance (such as equipment failure or external noise) occurs during the pre-shift briefing, it is marked as an environmental event. At the same time, the overlap time between the environmental event and the key cycle is obtained. If the overlap time is ≥30% of the key cycle, the interference misjudgment principle is met; otherwise, it is not met. like Figure 2 As shown, the overlap determination coefficient is compared with the preset overlap determination threshold. If the overlap determination coefficient is higher than or equal to the preset overlap determination threshold, and the environmental event does not meet the interference misjudgment principle, then the overlap analysis signal is triggered. If the overlap determination coefficient is lower than the preset overlap determination threshold, or if the environmental event meets the interference misjudgment principle, the change of the overlap determination coefficient will be continuously monitored. It should be noted that the overlap judgment threshold is obtained as follows: The overlap judgment threshold needs to be quantitatively determined by combining historical data and actual risk scenarios: First, collect complete data from at least 500 pre-shift briefings (including the overlap judgment coefficient for each time, the violation rate of the subsequent hour of operation, and hazard records). Group the overlap judgment coefficients according to numerical ranges (e.g., 0.1-0.2, 0.2-0.3, etc.) and calculate the average violation rate for each group. Then, through ROC curve analysis, find the critical point that "makes the overlap judgment coefficient of the high violation rate group (e.g., violation rate ≥ 5%) recognized and the misjudgment rate of the low violation rate group (e.g., violation rate < 1%) ≤ 5%". This critical point is the initial overlap judgment threshold.

[0026] Example 2 like Figure 1 As shown, this invention is an AI-powered intelligent device for pre-shift briefings, and also includes the following modules: Location optimization module: If the overlap analysis signal is triggered, the location of the pre-shift briefing for the personnel who are distracted before the shift is adjusted, and the attention distraction index of all personnel after the location adjustment is obtained. The method for adjusting the location of the pre-shift briefing for employees who are not attentive before their shift is as follows: Preferably, the area where people stand is located by using a directional tracking camera to obtain basic position data and dispersion feature data for each person, and the two sets of data are combined to obtain a dispersion data set; It should be noted that the basic location data includes the personnel's location coordinates, as well as the distance between the personnel and the display screen, the horizontal angle, and the vertical angle (for example, personnel coordinates (3,2) correspond to a distance of 4 meters from the display screen and a horizontal angle of 12°), and whether there are any obstructions (such as pillars or equipment) around the location (marked with 0-1, where 1 indicates obstruction). Distraction characteristics include: the inattention index and the ratio of facial gaze deviation time and the frequency of ineffective limb movements at that location; Integrate the scattered data sets of all personnel to construct a scattered dataset. Use the random forest algorithm to build a location relocation model. Input the scattered dataset into the location relocation model and output the initial optimal location for each scattered personnel before the shift. Those skilled in the art will understand that the method of constructing a location adjustment model using the random forest algorithm is as follows: Basic location data (coordinates, distance, angle, occlusion markers, etc.) are used as input features, and the target label is "whether the attention dissipation index corresponding to this location is lower than a preset optimization threshold" (e.g., 0.5, 1 for yes, 0 for no). The dissipation dataset is divided into a training set and a validation set in an 8:2 ratio. A forest composed of multiple decision trees is constructed using the training set (each tree randomly selects some samples and features for training). The prediction result of "whether the location is optimal" is output through a majority voting mechanism. The model accuracy is evaluated using the validation set (e.g., requiring ≥85%). Optimization is achieved by adjusting parameters such as the number of trees (e.g., 50–200 trees) and maximum depth, ultimately enabling the model to output optimal locations that reduce the dissipation index based on the input location features. Perform a feasibility check on the preferred location. If the feasibility check is satisfied, then the initial preferred location is adopted as the preferred location. For example, the feasibility verification method for the preferred location is as follows: Verification 1: Determine whether the preferred location is within the "pre-set effective area of ​​the pre-shift briefing" (for example, the briefing area is a rectangle of 10m×8m, and the location is removed if the coordinates are outside the area). Verification 2: Determine whether the distance between the preferred location and other people is ≥1.2 meters (meets the safe distance, to avoid crowding after adjustment, which may increase the disorganization of interaction); Verification 3: If the preferred location is obstructed (e.g., the model does not recognize the temporarily stacked tools), the second-best location will be automatically selected from the "alternate location library" (pre-marked locations with no obstructions and good visibility) to ensure that the adjustment is feasible; Based on the updated location, voice prompts during the pre-shift briefing, and with the assistance and guidance of staff, disorganized personnel before the shift are moved from their current location to a preferred location. The behavioral characteristics of personnel during the pre-shift briefing after the change of location were obtained, and the behavioral characteristics were processed again by attention mapping to obtain the attention dissipation index.

[0027] Pre-shift assessment module: Based on the adjusted attention dissipation index, the attention of the personnel conducting the pre-shift briefing is optimized and assessed to obtain the dissipation optimization coefficient. Based on the dissipation optimization coefficient, the effectiveness of optimization in key periods is analyzed to determine whether the effectiveness of information transmission during the pre-shift briefing meets the standards. The method for evaluating the distractibility optimization coefficient of personnel during pre-shift briefings is as follows: Obtain and calculate the change in the average disorganization index of a single pre-shift disorganized employee before and after the position adjustment, and obtain the disorganization adjustment amount; The ratio of the dispersion adjustment amount to the individual average dispersion index before adjustment is calculated to obtain the dispersion optimization coefficient of the overall cycle. Obtain the dispersion adjustment amount of the key period after the position adjustment, and calculate the ratio of the individual average dispersion index before the adjustment to obtain the dispersion optimization coefficient of the key period. If the overall cycle's dispersion optimization coefficient and the key cycle's dispersion optimization coefficient are both positive, and the individual average dispersion index within the key cycle is lower than the preset dispersion threshold, then it indicates that the attention regulation of a single pre-shift distracted person is effective. Among them, the method for determining whether the effectiveness of information transmission during pre-shift briefings is met is based on the analysis of the optimization effectiveness of key cycles using the dispersion optimization coefficient: Based on individual valid personnel data, key cycles are broken down into high-optimal cycles and medium-optimal cycles; By combining the individual average dispersion index before adjustment, an optimization evaluation is established for different periods (high-optimal period and medium-optimal period) to obtain the overall optimization coefficient of the high-optimal period and the overall optimization coefficient of the medium-optimal period. The preferred method for establishing a phased optimization evaluation is as follows: The average individual disorientation index of effective attention modulators before the position adjustment (reflecting the baseline level of disorientation before the adjustment) and the disorientation optimization coefficient during the high-optimal period (reflecting the improvement effect of disorientation during the high-optimal period after the position adjustment) were obtained. For each person with effective attention control, calculate the product of the individual average dispersion index before adjustment and the dispersion optimization coefficient of the high-optimal period. Sum the product results of all people with effective attention control to obtain the total optimization contribution value of the high-optimal period (reflecting the sum of the optimization effects of all effective people in the high-optimal period). The average individual disorganization index of all effective personnel before adjustment is summed to obtain the total optimization benchmark value of the high-optimal cycle (reflecting the sum of the disorganization benchmarks of all effective personnel before adjustment). The total contribution value of optimization is compared with the total value of optimization benchmark to obtain the overall optimization coefficient of the high-optimization cycle (the higher the value, the more significant the overall attention improvement effect within the high-optimization cycle, which aligns with the safety-first disclosure requirements). The average individual disorientation index of effective attention modulators before position adjustment, and the disorientation optimization coefficient during the mid-optimization period were obtained. For each person with effective attention regulation, calculate the product of the individual average dispersion index before adjustment and the dispersion optimization coefficient of the intermediate optimization cycle. Sum the product results of all people with effective attention regulation to obtain the total optimization contribution value of the intermediate optimization cycle. The average individual dispersion index before adjustment of all valid personnel is summed to obtain the total optimized baseline value for the medium-term optimization cycle; Calculate the ratio of the total optimization contribution value to the total optimization baseline value to obtain the overall optimization coefficient of the medium-optimization cycle; Obtain the violation rate of related operations within the high-optimization period, as well as the overlap duration between environmental incident interference and key periods; It should be noted that the violation rate of related operations is calculated by retrieving the number of violations recorded by on-site intelligent monitoring or safety inspection during the construction period corresponding to the work content involved in the high-optimization cycle and comparing it with the total number of operations during that period. After excluding the overlap between environmental events and key cycles, the partial correlation coefficient is obtained by checking the partial correlation interference based on the overall optimization coefficient of the high-optimal cycle and the violation rate of related operations within the high-optimal cycle. The method for partial correlation interference verification is as follows: In the pre-shift briefing scenario, when calculating the partial correlation coefficient after controlling Z, for the overall optimization coefficient (X) of the high-optimal cycle, the violation rate of related operations in the high-optimal cycle (Y), and the overlap time of environmental interference and key cycles (Z), the Pearson correlation coefficients of X and Y, X and Z, and Y and Z are calculated. , , ); Substituting the Pearson correlation coefficients of X and Y, X and Z, and Y and Z into the partial correlation coefficient formula By subtracting the common influence of Z on X and Y and standardizing the data, we obtain the net correlation coefficient between X and Y after eliminating environmental interference, which is the partial correlation coefficient. Construct a criterion for the effectiveness of meeting the standards. If the overall optimization coefficient and the partial correlation coefficient of the high-optimal cycle and the medium-optimal cycle meet the criterion for the effectiveness of meeting the standards, then the information transmission effectiveness of the pre-shift briefing is met. If the requirements are not met, the effectiveness of the information transmission during the pre-shift briefing is deemed to be substandard. For example, the method for constructing the compliance effectiveness criterion is as follows: Based on the accident data of the past 3 years of pre-shift briefings and 500+ effective control records, the core thresholds are determined through ROC curve analysis—the overall weighted optimization coefficient of the high-optimization cycle (safety category) is ≥22% (this threshold corresponds to the identification critical point of "safety violation rate ≤3% after optimization in the high-optimization cycle"), and the overall weighted optimization coefficient of the medium-optimization cycle (task category) is ≥18% (matching its risk weight of 0.4, corresponding to the critical point of "task category violation rate ≤5%)"; then through... The statistical significance of 300+ sets of data on "optimization coefficient - violation rate - environmental interference" was verified, and it was determined that the partial correlation coefficient after controlling for environmental interference was ≤-0.5 (at this point, the negative correlation between high-optimization cycle optimization and the decrease in violation rate was significant at the p<0.05 level, excluding accidental associations). At the same time, combined with the analysis of historical compliance cases, the average decrease in the overall operation violation rate of the team was ≥30%, so it was included in the criteria. Finally, the following compliance criteria were formed: high-optimization ≥22%, medium-optimization ≥18%, partial correlation coefficient ≤-0.5, and all three compliance criteria must be met simultaneously, namely, the effectiveness of information transmission in pre-shift briefings must be met.

[0028] Pre-shift optimization module: If the target is not met, it is used to obtain the cycle segmentation characteristics of the key cycles of the employees who are not relaxed before the shift; construct a cycle segmentation model, input the cycle segmentation characteristics into the cycle segmentation model, and adjust the key cycles based on the adjustment content of the pre-shift briefing output by the model; The method for obtaining the periodic segmentation characteristics of key periods for personnel who are disorganized before their shift is as follows: Preferably, if the effectiveness of the information transmission during the pre-shift briefing is not up to standard, the key periodic content can be quickly tested to all employees who are not focused before the shift through the touch screen and voice interaction of the AI ​​smart device, and the quick test score can be obtained as the degree of mastery. Simultaneously, the average dispersive index of the population during key periods is obtained to construct a population dispersive sequence; By using the sliding window algorithm, the dispersal sequence of the group is split and the time dimension is traced back to identify the concentrated dispersal period. For example, the method for segmentation and time-dimensional identification and backtracking is as follows: If the total duration of the pre-shift briefing is 30 minutes, the average disorganization index of the group is collected in 1-minute increments to form a disorganization sequence (e.g., 0.4 in the 1st minute, 0.35 in the 2nd minute, ... 0.5 in the 30th minute), and the high-priority period is from the 10th to the 20th minute; the sliding window algorithm is set with a window size of 3 minutes (to ensure coverage of the minimum effective period of continuous disorganization) and a step size of 1 minute (to avoid missing periods), starting from the 1st minute: the first window (1-3 minutes) The average dispersion index was calculated to be 0.38 for the first window (2-4 minutes), 0.36 for the second window, and so on. When sliding to the 14-16 minute window, the average index was calculated to be 0.72 (≥ the preset high dispersion threshold of 0.7). The average index for the 15-17 minute window was 0.75, and for the 16-18 minute window it was 0.73, all of which met the threshold. These three consecutive overlapping windows were merged to obtain the concentrated dispersion period as the 14-18 minute period, which completely falls within the high priority period (10-20 minutes), providing a basis for subsequent attribution analysis. Attribution analysis was conducted by combining the degree of mastery, periods of concentrated and dispersed activity, and high-optimal and medium-optimal cycles to obtain attribution labels; The attribution analysis method is as follows: First, set the judgment criteria: the mastery level is divided by 80% (≥80% is good mastery, <80% is insufficient mastery, and 80% is set by the experience of those skilled in the art). The concentrated and scattered periods are judged based on whether they fall completely / partially within the corresponding period. High-priority periods (safety type) have higher priority than medium-priority periods (task type). If the mastery of the high-optimal period is less than 70%, and the concentrated and scattered periods (such as the 14th to 18th minutes) fall entirely within the high-optimal period (10th to 20th minutes), while the mastery of the medium-optimal period is ≥80% and the concentrated and scattered periods do not overlap with the medium-optimal period, then the attribution label is "the high-optimal period overlaps with the attention trough, resulting in the failure of the delivery of safety-related content". If the mastery of the high-quality cycle is less than 70%, the concentrated and scattered periods do not overlap with the high-quality cycle, but the difficulty of understanding the content of the high-quality cycle is marked as level 5 (the highest), then the attribution label is "the content of the high-quality cycle is too difficult, and the personnel cannot understand it, resulting in insufficient mastery"; If the mastery rate of the mid-optimal cycle is <60%, the concentrated and scattered periods completely overlap with the mid-optimal cycle (minutes 21-28), and the mastery rate of the high-optimal cycle is ≥85% with no concentrated or scattered periods, then the attribution label is "the mid-optimal cycle is scheduled during a low point in attention, resulting in insufficient delivery of task-related content". The corresponding attribution label is output through the combination and matching of multi-dimensional data. The key periods, mastery levels, concentrated and scattered periods, and attribution labels of the pre-shift briefing content for key periods are used as the characteristics for period division. Obtain M periodicity partitioning features and construct a periodicity partitioning dataset containing the M periodicity partitioning features; Preferably, M=200; A periodicity partitioning model is constructed based on a gradient regression model algorithm with multiple feature inputs. Those skilled in the art will understand that when constructing this periodicity partitioning model, the features in the periodicity partitioning dataset are first converted into a numerical form that the model can process. The attribution labels are converted into numerical vectors through one-hot encoding. The concentrated and scattered periods are converted into duration values ​​by starting minutes to ending minutes. The mastery is retained in the form of a percentage decimal. The original key periods are split into multi-dimensional features by the start value, duration, and risk weight of the period, forming a complete input feature matrix as the dataset. The dataset was divided into training and validation sets in an 8:2 ratio. An XGBoost gradient regression model was selected, with the optimized key period's start time, duration, and content order priority as three independent regression objectives. During training, hyperparameters were tuned using a grid search method (with 100-200 trees, a learning rate of 0.05-0.1, and a maximum depth of 3-5 layers), using mean squared error (MSE) as the loss function, and L1 regularization was added to prevent overfitting. The model's prediction accuracy was monitored in real-time using the validation set until the model's error on the validation set stabilized and showed no overfitting trend, resulting in a period segmentation model that can output time periods, durations, and content order. Input the periodic data set into the periodic division model, and output the key period time, duration and content order of the pre-shift briefing content; The pre-shift briefing for the next cycle will be adjusted based on the output time period, duration, and content order.

[0029] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A pre-shift briefing AI intelligent device, characterized in that: The method comprises the following modules: a behavior collection module: used for collecting behavior characteristics of personnel during pre-shift briefing, performing attention mapping processing on the behavior characteristics, and obtaining an attention distraction index; an overlap analysis module: used for performing attention distraction analysis on the personnel based on the attention distraction index, identifying pre-shift distraction personnel, extracting distraction periods of attention of the pre-shift distraction personnel and key periods of the pre-shift briefing, performing time sequence overlap analysis on the distraction periods and the key periods, obtaining a coincidence determination coefficient, and determining whether an overlap analysis signal is triggered; a position optimization module: if the overlap analysis signal is triggered, adjusting a pre-shift briefing position of the pre-shift distraction personnel, and obtaining attention distraction indexes of all personnel after position adjustment; a pre-shift evaluation module: based on the adjusted attention distraction indexes, performing optimization evaluation on attention of the pre-shift briefing personnel, obtaining a distraction optimization coefficient, performing optimization effectiveness analysis on the key periods based on the distraction optimization coefficient, and determining whether information transmission effectiveness of the pre-shift briefing meets a standard; a pre-shift optimization module: if the standard is not met, used for obtaining period division characteristics of the key periods by the pre-shift distraction personnel; constructing a period division model, inputting the period division characteristics into the period division model, and adjusting the key periods based on adjustment content of the pre-shift briefing output by the model.

2. The pre-shift briefing AI intelligent device according to claim 1, characterized in that: The attention mapping processing on the behavior characteristics is performed in the following manner: The behavior characteristics comprise facial line-of-sight characteristics, body posture characteristics, and group interaction characteristics. The facial line-of-sight characteristics, the body posture characteristics, and the group interaction characteristics are weighted and summed to obtain the attention distraction index.

3. The pre-shift briefing AI intelligent device according to claim 2, characterized in that: The facial line-of-sight characteristics are obtained in the following manner: An effective visual range of each personnel during the pre-shift briefing is obtained, if a visual angle deviates from the effective visual range, a time length of the visual angle deviating from the effective visual range in a monitoring period is obtained, and a deviation time length is obtained; a deviation time ratio is calculated by taking the ratio of the deviation time length to a time length of the monitoring period, and the deviation time ratio is taken as the facial line-of-sight characteristics.

4. The pre-shift briefing AI intelligent device according to claim 1, characterized in that: The time sequence overlap analysis is performed in the following manner: The distraction periods of attention of the pre-shift distraction personnel are extracted, if the distraction periods and the key periods coincide and overlap in a time dimension, the overlapping periods are marked as overlapping periods, and a number of the overlapping periods is obtained as an overlapping period number; an average distraction index of individuals of the pre-shift distraction personnel in the overlapping periods is calculated; a number of the key periods is obtained, a period overlap ratio is calculated by taking the ratio of the overlapping period number to the number of the key periods, and the period overlap ratio is obtained; the average distraction index is multiplied by the period overlap ratio to obtain the coincidence determination coefficient.

5. The pre-shift briefing AI intelligent device according to claim 4, characterized in that: The pre-shift distraction personnel are determined in the following manner: feature variables of all personnel during the pre-shift briefing are obtained, and the feature variables of all personnel are clustered by a clustering algorithm; personnel of a severe distraction group in a clustering result and corresponding attention distraction indexes are extracted, if the attention distraction indexes of the personnel of the severe distraction group show an upward trend in N monitoring periods, the personnel are marked as the pre-shift distraction personnel.

6. The pre-shift briefing AI intelligent device according to claim 5, characterized in that: The feature variables are obtained in the following manner: a number of periods in which distraction periods continuously appear in the N monitoring periods is extracted, and a distraction frequency is calculated by taking the ratio of the number of the periods in which the distraction periods continuously appear to the number N of the monitoring periods; Calculate the average of the individual attention distraction index within the number of periods in which the distraction period continuously occurs, to obtain the individual average distraction index; Obtain the average of the individual average distraction index of all personnel at the time of pre-shift briefing, to obtain the group average distraction index; Calculate the deviation proportion of the individual average distraction index and the group average distraction index, to obtain the distraction intensity deviation; Take the distraction intensity deviation, the individual average distraction index, and the distraction frequency as the characteristic variables of each personnel.

7. The pre-shift briefing AI intelligent device according to claim 1, characterized in that: The optimization evaluation method for the attention of the pre-shift briefing personnel is: Obtain and calculate the change amount of the individual average distraction index of the single pre-shift distraction personnel before and after position adjustment, to obtain the distraction adjustment amount; Calculate the ratio of the distraction adjustment amount and the individual average distraction index before adjustment, to obtain the distraction optimization coefficient of the overall period; Obtain the distraction adjustment amount of the key period after position adjustment, and calculate the ratio of the individual average distraction index before adjustment, to obtain the distraction optimization coefficient of the key period; If the distraction optimization coefficient of the overall period and the distraction optimization coefficient of the key period are both positive, and the individual average distraction index in the key period is lower than the preset distraction boundary value, it indicates that the attention regulation of the single pre-shift distraction personnel is effective. 8.The pre-shift briefing AI intelligent device of claim 1, wherein: The optimization effectiveness analysis method is: Based on the data of the single effective personnel, the key period is split to obtain a high-optimization period and a medium-optimization period; Combined with the individual average distraction index before adjustment, the optimization evaluation processing of the split period is established to obtain the overall optimization coefficient of the high-optimization period and the medium-optimization period; Obtain the associated operation violation rate in the high-optimization period, and the overlapping duration of environmental event interference and the key period; After excluding the overlapping duration of environmental event interference and the key period, based on the overall optimization coefficient of the high-optimization period, combined with the associated operation violation rate in the high-optimization period, the partial correlation interference check is performed to obtain the partial correlation coefficient; A standard effectiveness criterion is constructed, if the overall optimization coefficient of the high-optimization period and the medium-optimization period and the partial correlation coefficient meet the standard effectiveness criterion, the information transmission effectiveness of the pre-shift briefing meets the standard. 9.The pre-shift briefing AI intelligent device according to claim 8, characterized in that: The partial correlation interference check method is: For the overall optimization coefficient X of the high-optimization period, the associated operation violation rate Y of the high-optimization period, and the overlapping duration Z of environmental interference and the key period, when calculating the partial correlation coefficient after controlling Z, the Pearson correlation coefficients of X and Y, X and Z, and Y and Z are calculated; Based on the Pearson correlation coefficients of X and Y, X and Z, and Y and Z, the partial correlation coefficient formula is substituted to obtain the partial correlation coefficient.

10. The pre-shift briefing AI intelligent device according to claim 1, characterized in that: The adjustment method for the key period is: Through the AI intelligent device, the mastery degree of the pre-shift distraction personnel is obtained, and the group average distraction index of the key period is obtained to construct a group distraction sequence; the group distraction sequence is split and time-dimension backtracking is recognized to obtain a concentrated distraction period; Combined with the mastery degree, the concentrated distraction period, and the high-optimization period and the medium-optimization period, an attribution analysis is performed to obtain an attribution label; Take the key period, the mastery degree, the concentrated distraction period, and the attribution label of the pre-shift briefing content of the key period as the period division characteristics; M cycle division features are acquired, a cycle division dataset containing the M cycle division features is constructed, a cycle division model is constructed based on a gradient regression model algorithm of multi-feature input, and a time period and a time length of a key cycle of the class before the disclosure content and a content order are outputted; The class before the disclosure of the next cycle is adjusted based on the outputted time period and time length and content order.

Images