Skill training evaluation and data management system based on airflow control capability

Abstract

The application belongs to the technical field of airflow control, and discloses a skill training evaluation and data management system based on airflow control capability, which comprises a data acquisition module: constructing an airflow control sequence set and extracting intensity peak points, stability fluctuation points and rhythm turning points as an initial airflow feature point set; performing time sequence hierarchical division and generating a cross-layer association chain through a layer-by-layer accumulation method to obtain a time sequence association vector; performing capability migration mapping on multiple training sessions, introducing a migration offset coefficient for adjustment to generate a capability evolution path, and marking key turning nodes in the path; performing dimension integration operation and dynamically setting an integration granularity threshold according to the number of training sessions; constructing a learner's exclusive data management archive based on the integration index set, archiving the capability evolution path and the integration index set in time sequence order, setting an archive access key and enabling a cross-session query interface; and greatly improving the convenience of data management.

Description

Technical Field

[0001] This invention relates to the field of airflow control technology, and more specifically, to a skills training, assessment, and data management system based on airflow control capabilities. Background Technology

[0002] With the rapid development of digitalization and intelligence in skills training, objective skills assessment methods based on physiological parameters have become an important support for fields such as sports training, rehabilitation medicine, and arts education. In professional skills training involving respiratory control, such as wind instrument playing, speech therapy, and athlete breathing training, accurate assessment and systematic data management of learners' airflow control abilities are of great significance for developing personalized training programs and tracking ability development. In recent years, it has become possible to quantify airflow control abilities by collecting multi-dimensional parameters such as airflow pressure, velocity, and duration, providing a technical foundation for scientific skills training assessment.

[0003] Existing methods for assessing airflow control ability have significant limitations in terms of temporal correlation analysis and data integration of multi-dimensional parameters. Specifically, the characterization of airflow control ability involves multiple interrelated dimensions such as airflow intensity, stability, persistence, and rhythm changes. These parameters exhibit complex temporal evolution characteristics during training, and there are ability transfer and cumulative effects between different training stages. Existing methods typically perform independent statistical analysis on parameters of each dimension, ignoring the inherent correlation and temporal dependence between parameters. This fragmented analysis makes it difficult for the assessment results to accurately reflect the overall level and dynamic development trend of the learner's airflow control ability. At the same time, due to the lack of an effective multi-source data fusion mechanism, assessment data generated from different training sessions are difficult to form a coherent ability development curve, failing to provide an effective basis for training guidance. The current common approach to this problem is to combine manual recording with subjective judgment, or to rely on simple threshold judgment rules for assessment and grading. However, this approach has limited assessment accuracy and makes it difficult to achieve standardized data management and cross-temporal comparative analysis, ultimately limiting the application and promotion of airflow control training in professional talent training and rehabilitation therapy.

[0004] In view of this, the present invention proposes a skills training assessment and data management system based on airflow control capability to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: a skills training, assessment, and data management system based on airflow control capabilities, comprising:

[0006] Data acquisition module: Collects airflow pressure sequence, velocity sequence and duration sequence for each training session, combines training session identifier and learner identifier to construct airflow control sequence set, and extracts intensity peak points, stability fluctuation points and rhythm inflection points from it as initial airflow feature point set;

[0007] Layered accumulation module: The airflow control sequence set of each training session is divided into temporal layers, and cross-layer association chains are generated by layer-by-layer accumulation to obtain the temporal association vector;

[0008] Fusion Labeling Module: Based on the continuity identifier between temporal correlation vector and training sessions, capability transfer mapping is performed on multiple training sessions, and a transfer offset coefficient is introduced for adjustment to generate capability evolution path and mark key turning points in the path.

[0009] Integration setting module: Utilizing key turning points and temporal correlation vectors in the capability evolution path, it performs dimensional integration operations, superimposes corresponding nodes of airflow pressure sequence, velocity sequence and duration sequence, and dynamically sets the integration granularity threshold according to the number of training sessions.

[0010] Archive module: Based on the integrated indicator set, build a data management archive for learners, archive the ability evolution path and the integrated indicator set in chronological order, set archive access keys and enable cross-session query interface.

[0011] Furthermore, the initial airflow feature point set can be constructed in the following ways:

[0012] The highest point of the continuous rising segment in the airflow pressure sequence is identified as the intensity peak point. Points with a coefficient of variation exceeding a preset threshold in the velocity sequence are identified as stability fluctuation points. Points with abrupt changes in the interval change rate in the duration sequence are identified as rhythm inflection points. The intensity peak points, stability fluctuation points, and rhythm inflection points are sorted by the acquisition timestamp and merged into an initial airflow feature point set. Each point is then labeled with a training session identifier and a learner identifier.

[0013] Furthermore, the methods for obtaining time-series correlation vectors include:

[0014] The airflow control sequence set is divided into a starting layer, a core layer, and a closing layer. A low distribution density threshold is set for the starting layer, a medium threshold for the core layer, and a high threshold for the closing layer. The average interval between feature points in each layer is calculated as the layer weight threshold. Feature points are connected point by point from the starting layer to the core layer to form a starting-core association chain, and point by point from the core layer to the closing layer to form a core-closing association chain. The two chains are merged based on the layer weight threshold to generate a cross-layer association chain. Each connection point in the cross-layer association chain is converted into a vector component, and all components are summed to obtain the temporal association vector.

[0015] Furthermore, the methods for acquiring capability evolution paths include:

[0016] Extract the feature point set of the closing layer of the previous session, and calculate the temporal interval with the starting layer of the current session based on the continuity identifier to generate a projection matrix; map the feature point set of the closing layer to the starting layer through the projection matrix, and introduce a migration offset coefficient based on the interval length to adjust the position of the mapped points; connect the adjusted mapped points with the core layer and closing layer of the current session to form a capability evolution path, and mark the nodes in the path whose offset coefficient changes by more than a preset value as key turning points.

[0017] Furthermore, a dimension integration operation is performed, which involves vector overlaying of corresponding nodes from the airflow pressure sequence, velocity sequence, and duration sequence, and dynamically setting the integration granularity threshold based on the number of training sessions, including:

[0018] A unique index is assigned to the key turning points in the capability evolution path, and the components corresponding to the index are extracted from the temporal correlation vector as superposition base points. The node values ​​of the airflow pressure sequence, velocity sequence, and duration sequence are added to the superposition base points one by one to form a preliminary superposition vector. The vector amplitude is adjusted by applying a scaling rule based on the time distance between nodes to generate an integrated index set. The integration granularity threshold is dynamically increased or decreased according to the current number of training sessions and the proportion of total sessions. Indicators exceeding the threshold in the integrated index set are grouped and merged.

[0019] Furthermore, the methods for constructing data management archives include:

[0020] Create a unique data management profile for each learner, store the ability evolution path in the main chain structure of the profile according to the session, and attach the integrated indicator set to the corresponding node of the main chain; generate a profile access key based on the learner identifier and set up a cross-session query interface to support data retrieval by key turning points or time-series correlation vectors; when updating, insert the ability evolution path and integrated indicator set of the new training session into the end of the profile main chain, and recalculate the migration offset coefficient of all nodes.

[0021] Furthermore, feature point sets are connected point by point from the starting layer to the core layer to form a starting-core association chain, and point by point from the core layer to the closing layer to form a core-closing association chain, including:

[0022] Density clustering is performed on the feature point set of the initial layer, and the cluster centers are identified as anchor points. The temporal matching degree between the anchor points and the corresponding points in the core layer is calculated, and a matching degree threshold is introduced to filter the initial connection pairs. Based on the filtered initial connection pairs, a gradual path planning rule is applied to extend from the anchor points of the initial layer to the points of the core layer point by point to form the initial-core association chain. During the extension process, a path curvature constraint mechanism is embedded to perform local rerouting on connection segments that exceed the upper limit of curvature. The feature point set of the core layer is sorted by similarity, and the sorting result is used as a guiding sequence. The similarity is matched point by point from the end point of the core layer to the starting point of the closing layer to generate the core-closing association chain.

[0023] Furthermore, based on hierarchical weight thresholds, the two chains are merged to generate a cross-layer related chain, including:

[0024] Link alignment is performed on the start-core association chain and the core-end association chain. The temporal offset difference between corresponding connection points in the two chains is calculated, and an adaptive fusion factor based on the offset difference is introduced. The adaptive fusion factor is dynamically generated by multiplying the square root of the offset difference by the hierarchical weight threshold. The connection nodes of the two chains are traversed point by point. Starting from the end node of the start-core association chain, the projection is made to the start node of the core-end association chain. The adaptive fusion factor is used as a weighting coefficient to perform weighted fusion on the vector components of the projection points to form a preliminary merged chain segment. The rate of change of the curvature of the merged chain segment is detected. If the rate of change exceeds the preset fusion threshold, an auxiliary anchor point is inserted. Based on the ratio of the total chain length of the preliminary merged chain segment to the hierarchical weight threshold, the link optimization iteration is performed. The interval between adjacent nodes in the chain segment is adaptively adjusted. An iterative convergence condition is introduced to minimize the overall link deviation and generate an optimized cross-layer association chain. Each node in the optimized chain is labeled with a fusion factor, and the global association strength of the chain is calculated as a verification index. If the global association strength is lower than the preset lower limit, the adaptive fusion factor is adjusted backtrackingly.

[0025] Furthermore, the feature point set of the closing layer is mapped to the starting layer using a projection matrix, and a migration offset coefficient based on the interval length is introduced to adjust the position of the mapped points, including:

[0026] Based on the temporal interval between the feature point set of the previous session's closing layer and the starting layer of the current session, the specific elements of the projection matrix are constructed. The main diagonal values ​​of the matrix are generated by calculating the logarithmic transformation of the interval length and the vector product of the continuity identifier, and the interval correlation coefficient is introduced as the off-diagonal element. The projection matrix is ​​applied to each feature point of the closing layer to perform coordinate transformation, generating an initial mapping point set. The migration offset coefficient based on the interval length is calculated. The migration offset coefficient is dynamically determined by the weighted sum of the exponential function of the interval length and the layer weight threshold. The vector displacement of the initial mapping points is adjusted to compensate for the deviation caused by the temporal discontinuity, forming the adjusted mapping point set.

[0027] Furthermore, the vector magnitude is adjusted using scaling rules based on the time distance between nodes, including:

[0028] The temporal distance between adjacent nodes is calculated for the node sequence in the initial superposition vector. The standard distance is generated by standardizing the temporal distance, and a correction coefficient based on the temporal correlation vector is introduced. The correction coefficient is determined by the product of the average value of the vector components and the integration granularity threshold.

[0029] For each node, the scaling rules are applied as follows: if the standard distance is lower than the preset lower threshold, a logarithmic compression function is used to reduce the magnitude; if the standard distance is higher than the preset upper threshold, a power function is used to expand the magnitude; if the standard distance is between the lower and upper thresholds, a gradual adjustment is made using a linear interpolation function.

[0030] The technical effects and advantages of this invention, which is a skills training assessment and data management system based on airflow control capabilities:

[0031] This invention effectively captures the inherent correlation and temporal dependence between feature points in different time stages by dividing the airflow control sequence set into temporal hierarchical layers and generating cross-layer correlation chains in a layer-by-layer accumulation manner to obtain temporal correlation vectors. This overcomes the limitations of existing methods in independently statistically analyzing parameters of each dimension. Capability transfer and cumulative effects exist between different training stages. Capability transfer mapping is performed based on the continuity identifier between the temporal correlation vector and training sessions. A transfer offset coefficient is introduced to adjust the generated capability evolution path, and key turning points are marked, enabling accurate characterization of capability development changes across sessions and forming a coherent capability development curve. By performing dimensional integration operations using key turning points and temporal correlation vectors, corresponding nodes of airflow pressure, velocity, and duration sequences are vector-superimposed. The integration granularity threshold is dynamically set according to the number of training sessions to achieve effective fusion of multi-source data and avoid evaluation bias caused by fragmented analysis. Based on the integrated index set, learner-specific data management archives are constructed, archiving capability evolution paths and integrated index sets in chronological order. Cross-session query interfaces are enabled to achieve standardized management and cross-time domain comparative analysis of evaluation data, providing effective decision-making basis for training guidance, thereby improving the accuracy and scientific nature of airflow control capability evaluation. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the skills training assessment and data management system based on airflow control capability according to the present invention. Detailed Implementation

[0033] 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.

[0034] Example 1

[0035] Please see Figure 1 As shown, this embodiment of the skill training assessment and data management system based on airflow control capability includes:

[0036] Data acquisition module: used to collect the airflow pressure sequence, velocity sequence and duration sequence of each training session, combine the training session identifier and learner identifier to construct the airflow control sequence set, and extract the intensity peak point, stability fluctuation point and rhythm inflection point as the initial airflow feature point set.

[0037] In professional skills training involving respiratory control, such as wind instrument playing, speech therapy, and athlete breathing training, learners' airflow control ability is characterized by multiple physiological parameters. This embodiment uses an airflow sensor array to collect learners' airflow parameters in real time during training. The airflow sensor array includes a pressure sensor, a flow rate sensor, and a timing module.

[0038] The method for collecting the airflow pressure sequence is as follows: a piezoresistive pressure sensor is installed in the airflow channel of the training device, the sampling frequency is set to 100Hz, and the airflow pressure value generated by the learner in each training session is collected to form a time-series airflow pressure sequence; the unit of airflow pressure is Pascal, and the typical value range is 50 to 2000.

[0039] The method for collecting the flow velocity sequence is as follows: a hot-wire flow velocity sensor is used to measure the instantaneous flow velocity in the airflow channel. The sampling frequency is synchronized with the pressure sensor. The airflow velocity value of the learner in each training session is collected to form a flow velocity sequence; the unit of flow velocity is meters per second.

[0040] The method for acquiring the duration sequence is as follows: the timing module records the start and end timestamps of each airflow output, calculates the time interval between adjacent airflow outputs, and forms a duration sequence; the unit of duration is milliseconds.

[0041] The airflow pressure sequence, velocity sequence, and duration sequence are associated with the corresponding training session identifier and learner identifier to construct an airflow control sequence set. The training session identifier is coded in the format of "learner number-date-session number", for example, "U001-20241201-03" indicates that the learner with the number U001 participated in the 3rd training session on December 1, 2024.

[0042] It should be noted that in other embodiments of the present invention, ultrasonic flow meters, vortex flow meters and other devices can also be used to collect flow velocity sequences. The sampling frequency can be set to any value in the range of 50Hz to 500Hz according to actual needs. The collection of airflow parameters is a well-known technology in the field and is not limited here.

[0043] Since the characterization of airflow control capability involves multiple interrelated dimensions such as airflow intensity, stability, persistence, and rhythm changes, it is necessary to extract representative feature points from the original sequence to reduce the computational complexity of subsequent analysis and highlight key information.

[0044] Specifically, the method for extracting the intensity peak point includes: performing a sliding window traversal on the airflow pressure sequence, with the window length set to 50 sampling points (corresponding to a time span of 0.5 seconds); detecting continuous rising segments within each window, defined as pressure values ​​of at least 5 consecutive sampling points within the window showing an increasing trend; recording the highest point of the continuous rising segment as the intensity peak point, and attaching the corresponding timestamp, pressure value, and training session identifier.

[0045] For example, the airflow pressure sequence segment of a certain training session is as follows:

[0046] [850,920,1050,1180,1250,1320,1280,1150], where the range from 850 to 1320 forms a continuous upward segment, and the location of 1320 is the peak intensity point.

[0047] The method for identifying stability fluctuation points is as follows: the flow velocity sequence is divided into continuous analysis windows, with a window length set to 100 sampling points (corresponding to a 1-second time span); the coefficient of variation (COP) of the flow velocity values ​​within each window is calculated, which is equal to the standard deviation of the flow velocity values ​​within the window divided by the mean; the center point of the window where the COP exceeds a preset threshold is recorded as a stability fluctuation point. The preset threshold is set to 0.25, based on the following: through statistical analysis of the flow velocity sequences of 500 learners with different skill levels, a COP below 0.15 indicates stable airflow control, 0.15 to 0.25 indicates slight fluctuations, and above 0.25 indicates significant instability; in this embodiment, 0.25 is selected as the threshold for identifying stability fluctuation points to capture moments when airflow control capability shows significant fluctuations.

[0048] The method for identifying rhythmic turning points is as follows: Calculate the rate of change of the interval between adjacent elements in the duration sequence. The rate of change of the interval is equal to the difference between the current interval value and the previous interval value divided by the absolute value of the previous interval value. Record the position where the absolute value of the rate of change of the interval exceeds a preset abrupt change threshold as a rhythmic turning point. The preset abrupt change threshold is set to 0.4, that is, when the change amplitude between adjacent durations exceeds 40%, it is determined that a rhythmic abrupt change has occurred.

[0049] For example, if a certain duration sequence segment is [800, 850, 1400, 1350, 900], the rate of change of the interval from 850 to 1400 is about 10, which exceeds the threshold of 0.4. Therefore, the position at 1400ms is the rhythm inflection point. The rate of change of the interval from 1350 to 900 is about 8, which exceeds the absolute value of the threshold and is recorded as the rhythm inflection point.

[0050] Intensity peaks, stability fluctuations, and rhythm inflection points were sorted in ascending order by acquisition timestamps and merged into an initial airflow feature point set. Each feature point was then appended with a training session identifier and a learner identifier to form a complete feature point record, formatted as "Feature Point Type - Timestamp - Feature Value - Training Session Identifier - Learner Identifier". This method extracted representative feature points from the multi-dimensional raw sequence, laying a data foundation for subsequent time-series correlation analysis and capability assessment. Furthermore, the simplified representation of feature points reduced data storage and computational overhead.

[0051] Layered accumulation module: used to perform temporal layering of the airflow control sequence set for each training session, and generate cross-layer association chains through layer-by-layer accumulation to obtain temporal association vectors.

[0052] Because airflow control ability exhibits phased changes within a single training session, learners typically enter an adaptation phase in the early stages, a stable performance phase in the middle stages, and may experience fatigue or need to adjust towards the end. Therefore, dividing the airflow control sequence set into temporal hierarchical segments can more accurately capture the performance characteristics of different stages and avoid the feature ambiguity caused by mixing data from different stages for analysis.

[0053] Specifically, the airflow control sequence set is divided into a starting layer, a core layer, and a closing layer according to time sequence. The division method is as follows: obtain the total duration of the training sessions, divide the data within the first 20% of the duration into the starting layer, the data within the middle 60% of the duration into the core layer, and the data within the last 20% of the duration into the closing layer.

[0054] For example, if the total duration of a training session is 300 seconds, then the data within 0 to 60 seconds belongs to the initial layer, the data within 60 to 240 seconds belongs to the core layer, and the data within 240 to 300 seconds belongs to the closing layer.

[0055] It should be noted that the above division ratio is a preferred setting in this embodiment. It is based on statistical analysis of 1000 training sessions. Learners showed obvious warm-up adaptation characteristics in the first 20% of the training time, relatively stable ability performance in the middle 60% of the training time, and fatigue or end-of-training adjustment characteristics in the last 20% of the training time. Other embodiments may adjust the division ratio to (15%, 70% and 15%) or (25%, 50% and 25%) according to the specific training scenario.

[0056] Furthermore, a distribution density threshold is set: a lower threshold is set for the starting layer, a medium threshold for the core layer, and a higher threshold for the closing layer. The distribution density threshold is used to filter feature points participating in the construction of the association chain; the lower the density threshold, the more feature points are retained.

[0057] The specific setting method is as follows: Calculate the average interval between points in the initial airflow feature point set within each layer. The average interval is equal to the time span of that layer divided by the number of feature points in that layer minus 1. The distribution density threshold for the starting layer is set to 0.6 times the average interval, the core layer to 1.0 times, and the closing layer to 1.4 times. The starting layer is the adaptation phase of training, requiring the retention of more feature points to capture the learner's initial state changes; the core layer is the main capability development phase, using a medium threshold to balance information content and noise filtering; the closing layer may contain abnormal fluctuations caused by fatigue, using a higher threshold to filter noise points and retain key closing features. The distribution density threshold corresponding to each layer is used as the layer weight threshold for that layer.

[0058] Building the initial-core association chain

[0059] The feature point set is connected point by point from the starting layer to the core layer to form the starting-core association chain.

[0060] The specific method is as follows: density clustering is performed on the feature point set of the initial layer. In this embodiment, the DBSCAN algorithm is used. The neighborhood radius parameter is set to 1.5 times the distribution density threshold of the initial layer, and the minimum number of points parameter is set to 3. The centroid of each cluster in the clustering result is identified as the anchor point.

[0061] For example, the initial layer contains 12 feature points, which are clustered by DBSCAN to form 3 clusters. The centroid of each cluster is denoted as anchor point A1, A2, and A3, respectively.

[0062] Calculate the temporal matching degree between each anchor point and its corresponding feature point in the core layer. For anchor point A1 and its corresponding core layer feature point Q, the temporal matching degree is calculated as follows: ;in, Represents time matching degree. The timestamp representing anchor point A1 The timestamps of the core layer feature points Q. This represents the time span of the core layer. The time series matching degree ranges from 0 to 1, with a higher value indicating a higher degree of time series matching.

[0063] A matching degree threshold is introduced to filter initial connection pairs. In this embodiment, the matching degree threshold is set to 0.3, with a range of 0.2 to 0.4. That is, only anchor-core layer point pairs with a temporal matching degree greater than 0.3 are retained as initial connection pairs. Based on the filtered initial connection pairs, a gradual path planning rule is applied to extend from the starting layer anchor point to the core layer point point by point, forming a starting-core association chain. The gradual path planning rule is as follows: traverse the starting layer anchor points in chronological order, select the core layer point with the highest temporal matching degree for each anchor point as the connection target, and record the direction vector of the path during the connection process. A path curvature constraint mechanism is embedded in the extension process to prevent the association chain from bending sharply. The path curvature is calculated as follows: for three adjacent connection points P1, P2, and P3, calculate the angle between vector (P2-P1) and vector (P3-P2). The angle is calculated using dot product calculation, which is a conventional calculation method and will not be elaborated in this embodiment. The angle value is the path curvature at that point. The upper limit for the degree of curvature is set to 120 degrees. If the degree of curvature of a certain connection segment exceeds 120 degrees, local rerouting will be performed on that segment, that is, the current connection point will be removed and the core layer point with the second highest timing matching degree will be reselected as the connection target.

[0064] Connecting point by point from the core layer to the closing layer forms a core-closing association chain. The core layer feature points are sorted from largest to smallest similarity, with similarity calculated based on the multi-dimensional attributes of the feature points. For two core layer feature points, their similarity is equal to the cosine similarity between vectors composed of normalized values ​​of pressure difference, flow velocity difference, and duration difference. The sorting result is used as a guiding sequence. Similarity is matched point by point from the end point of the core layer to the start point of the closing layer. The matching method is as follows: for feature points at the end of the core layer, the point with the highest similarity in the closing layer is selected as the connection target, and this process is extended sequentially towards the closing layer, generating a core-closing association chain.

[0065] Based on hierarchical weight thresholds, the start-core association chain and the core-end association chain are merged to generate a cross-layer association chain. Link alignment is performed on the start-core association chain and the core-end association chain, and the temporal offset difference between corresponding connection points in the two chains is calculated. The temporal offset difference is calculated as follows: taking the core layer as a reference, the difference between the time point when the start-core association chain enters the core layer and the time point when the core-end association chain leaves the core layer is calculated. An adaptive fusion factor based on the offset difference is introduced, which is dynamically generated by multiplying the square root of the offset difference by the hierarchical weight threshold. The connection nodes of the two chains are traversed point by point, projecting from the end node of the start-core association chain to the start node of the core-end association chain. The adaptive fusion factor is applied as a weighting coefficient to weight and fuse the vector components of the projected points, forming a preliminary merged chain segment. The formula for weighted fusion is:

[0066] Fusion point coordinates = coordinates of the starting point - end point of the core chain × (1 - adaptive fusion factor) + coordinates of the starting point of the core - ending chain × adaptive fusion factor;

[0067] The rate of change in the curvature of the fused chain segments is detected. If the rate of change exceeds a preset fusion threshold of 0.5, an auxiliary anchor point is inserted at the point of abrupt change in curvature. The coordinates of the auxiliary anchor point are taken as the midpoint between the coordinates of two adjacent points.

[0068] Based on the ratio of the total chain length to the hierarchical weight threshold of the initially merged chain segments, link optimization iterations are performed. The iteration method is as follows: calculate the interval distribution between adjacent nodes within a chain segment; if the ratio of the maximum interval to the minimum interval exceeds 3, insert interpolation points at positions with excessively large intervals and delete redundant points at positions with excessively small intervals. An iteration convergence condition is introduced: iteration stops when the change in the overall link deviation (defined as the root mean square of the distances from all nodes to the ideal straight line) is less than 0.01.

[0069] Each node in the optimized cross-layer association chain is labeled with a fusion factor, and the global association strength of the chain is calculated as a validation metric. The global association strength is equal to the harmonic mean of the temporal matching degrees of all adjacent node pairs in the chain. If the global association strength is lower than the preset lower limit of 0.2, the adaptive fusion factor is adjusted retrospectively. Specifically, the adjustment coefficient is increased by 0.02, the fusion factor is recalculated, and the merging operation is performed.

[0070] Convert each join point in the cross-level association chain into a vector component. The conversion method is as follows: for the first join point in the cross-level association chain... For each connection point, its timestamp, pressure value, flow rate value, and duration value are extracted and normalized to form a four-dimensional vector component. ,in , , , These are the normalized timestamp, pressure value, flow velocity value, and duration value, respectively, using min-max normalization. The sum of all components yields a temporal correlation vector, which comprehensively reflects the correlation characteristics of airflow control capability at different stages within this training session, providing a quantitative basis for subsequent cross-session capability transfer analysis.

[0071] Fusion Labeling Module: Based on the continuity identifier between the temporal correlation vector and the training session, it performs capability transfer mapping on multiple training sessions, introduces a transfer offset coefficient for adjustment, generates capability evolution path, and marks key turning points in the path.

[0072] Because learners' airflow control abilities have a cumulative effect and transfer relationship across multiple training sessions, the training effect of the previous session will affect the ability performance in subsequent sessions. Existing methods analyze each session independently, ignoring the ability transfer relationship between sessions, resulting in evaluation results that cannot reflect the continuity of ability development. This module captures the dynamic trajectory of ability development by constructing a cross-session ability transfer mapping.

[0073] First, a projection matrix is ​​constructed, the feature point set of the previous session's closing layer is extracted, and the temporal interval between the current session's starting layer and the previous session's starting layer is calculated based on the continuity flag. The continuity flag is used to determine whether two sessions are consecutive training sessions. If the time interval between two sessions is less than 24 hours, it is marked as consecutive training with a continuity flag value of 1; otherwise, it is marked as non-consecutive training with a continuity flag value of 0.5. The temporal interval is calculated as the difference between the timestamp of the last feature point of the previous session's closing layer and the timestamp of the first feature point of the current session's starting layer, expressed in hours.

[0074] A projection matrix is ​​constructed based on the interval length and continuity identifier. The projection matrix is ​​as follows: phalanx, This represents the number of feature points in the final layer. The specific elements of the projection matrix are generated using the following method:

[0075] Matrix main diagonal values ;in, Represents the length of the time interval. The corresponding component represents the continuity identifier vector. The continuity identifier vector is a vector composed of the continuity weights of each feature point in the closing layer. The continuity weight is equal to the continuity identifier value multiplied by the temporal position weight of that point in the closing layer (temporal position weight = 1 - point number / total number of points).

[0076] Off-diagonal elements are determined by the interval correlation coefficient: off-diagonal elements value ;in, It is an exponential function. The attenuation coefficient is used to control the influence range of off-diagonal elements; in this embodiment, the value is 0.1.

[0077] Secondly, a coordinate transformation is performed. The projection matrix is ​​applied to each point of the feature point set in the final layer to transform the coordinates, generating an initial mapped point set. The coordinate transformation formula is: Initial mapped point coordinates = Projection matrix × Final layer feature point coordinate vector. The migration offset coefficient based on the interval length is then calculated. The calculation method for the migration offset coefficient is as follows: ;in, Represents the migration offset coefficient. Represents the hierarchical weight threshold. The decay constant is set to 12 (representing a characteristic time scale of 12 hours). This value is determined through statistical analysis of the capacity retention rate at different time intervals, with 12 hours corresponding to a capacity retention rate of approximately 63%. To adjust the weight, it is set to 0.8.

[0078] The initial mapping point is adjusted by vector displacement to compensate for the deviation caused by the time series fault: the adjusted mapping point value = the initial mapping point value + the migration offset coefficient × the offset direction vector; the offset direction vector is defined as the unit vector pointing from the centroid of the starting layer of the current field to the centroid of the initial mapping point.

[0079] A capability evolution path is constructed, connecting the adjusted mapping points with the core and closing layers of the current session. The connection method is as follows: the adjusted mapping point set, the core layer feature point set, and the closing layer feature point set are connected sequentially in chronological order, with linear interpolation used to generate connection edges between adjacent points. Nodes in the path whose offset coefficient changes exceeding a preset offset value are marked as key turning points. The preset offset value is set to 0.15, and the typical range of migration offset coefficient change is 0 to 0.3; a change exceeding 0.15 indicates a significant change in capability status. The marking method for key turning points is as follows: the absolute value of the difference between the migration offset coefficients of adjacent nodes in the capability evolution path is calculated. If the absolute value of the difference exceeds 0.15, the changed node is marked as a key turning point, and a turning point type label is attached ("increasing" indicates an increase in the offset coefficient, and "decreasing" indicates a decrease in the offset coefficient).

[0080] By constructing a capability evolution path, the airflow control capability performance of multiple training sessions is linked into a coherent development trajectory. Key turning points mark important moments of change in the capability development process, providing a basis for subsequent capability assessment and training program adjustments.

[0081] Integration Setting Module: This module utilizes key turning points and temporal correlation vectors in the capability evolution path to perform dimensional integration operations. It superimposes corresponding nodes of the airflow pressure sequence, velocity sequence, and duration sequence into vectors and dynamically sets the integration granularity threshold based on the number of training sessions.

[0082] Because airflow control capability involves multiple interrelated dimensions, analyzing each dimension's parameters in isolation cannot reflect the overall level of capability. This module integrates multi-dimensional parameters into a comprehensive indicator through a dimensional integration operation, and dynamically adjusts the integration granularity based on the number of training sessions to ensure the comparability and adaptability of the evaluation results.

[0083] The execution methods for dimension integration operations include:

[0084] Assign a unique index: Assign a unique index to key turning points in the capability evolution path. The index format is "session number - node number", for example, "03-05" represents the 5th key turning point in the 3rd session. Extract the component corresponding to the index from the temporal correlation vector as the overlay base point. The extraction method is as follows: For the key turning point with index "mk", locate the kth vector component in the temporal correlation vector of the mth session and use this component as the overlay base point.

[0085] Execution vector superposition: The node values ​​of the airflow pressure sequence, velocity sequence and duration sequence are added one by one to the superposition base point to form a preliminary superposition vector. The superposition base point is set by those skilled in the art based on the actual situation.

[0086] Scaling rules are applied: A scaling rule based on the time distance between nodes is applied to adjust the vector amplitude and generate an integrated index set. The time distance between adjacent nodes is calculated for the node sequence in the initial overlay vector; the time distance is equal to the absolute value of the timestamp difference between adjacent nodes. A standard distance is generated by standardizing the time distance by dividing it by the maximum value of all time distances within that session. A correction coefficient based on the temporal correlation vector is introduced; this coefficient is determined by the product of the average value of the vector components and the integration granularity threshold. Scaling rules are applied to each node pair:

[0087] (1) If the standard distance is lower than the preset lower threshold of 0.2, the logarithmic compression function is used to reduce the amplitude. The specific scaling method is as follows: ;in, This represents the scaling factor. Represents the original amplitude. Represents standard distance. This represents the correction factor.

[0088] (2) If the standard distance is higher than the preset upper limit threshold of 0.8, then the expansion amplitude is amplified using a power function. The specific scaling method is as follows: ;in, This represents the scaling factor. Represents the original amplitude.

[0089] (3) If the standard distance is between the lower threshold of 0.2 and the upper threshold of 0.8, then a gradual adjustment is made using a linear interpolation function. The specific scaling method is as follows:

[0090] ;in, This represents the scaling factor. The scaled amplitude is the product of the original amplitude and the scaling factor. The scaled vectors are combined to form an integrated index set.

[0091] The integration granularity threshold is dynamically set, increasing or decreasing based on the ratio of the current number of training sessions to the total number of sessions. The initial value of the integration granularity threshold is set to 1.0, and the dynamic adjustment formula is as follows: ;in This represents the adjusted integration granularity threshold. Represents the current number of matches. Represents the expected total number of matches. The parameter is set to 0.4; the expected total number of sessions is the total number of training sessions preset in the training plan, with a default value of 20 sessions.

[0092] The basis for this adjustment strategy is: in the early stage of training (when there are fewer sessions), a smaller integration granularity threshold is used to retain more detailed information; in the later stage of training (when there are more sessions), a larger integration granularity threshold is used to aggregate information, which facilitates macro-level evaluation.

[0093] Indicators exceeding the threshold within the integrated indicator set are grouped and merged. The grouping and merging method is as follows: for indicators whose concentration exceeds twice the integration granularity threshold, they are merged with adjacent smaller indicators into a group, and the weighted average of the indicators within the group is taken as the representative indicator after merging.

[0094] Archive module: Used to build learner-specific data management archives based on the integrated indicator set, archive the ability evolution path and integrated indicator set in chronological order, set archive access keys and enable cross-session query interface to realize data retrieval and update.

[0095] Since learners' training data needs to be stored long-term and repeatedly retrieved, establishing standardized data management archives is crucial for tracking ability development and developing personalized training programs. This module constructs a complete data management system, enabling standardized storage, secure access, and efficient retrieval of training data.

[0096] The data management profile is constructed as follows: A unique profile is created for each learner, organized using a hierarchical tree structure. The root node is the learner identifier, first-level child nodes represent training periods (divided by month), and second-level child nodes represent specific training sessions. The capability evolution path is stored sequentially by session in the profile's main chain structure. The main chain structure is implemented using a doubly linked list, with each node corresponding to a capability evolution path for a training session. Each node contains the following fields: session identifier, path data, predecessor pointer, successor pointer, and creation timestamp. An integrated indicator set is appended to the corresponding node in the main chain, stored as an auxiliary data structure for that node.

[0097] Access control is configured to generate a file access key based on the learner's identifier. The key is generated by concatenating the learner's identifier with the system timestamp, generating a 256-bit key using the SHA-256 hash algorithm, and then using the first 64 bits as the file access key. This file access key is used for authentication; only users with a valid key can access the corresponding learner's data management files.

[0098] Configure the query interface, set up a cross-session query interface, and support the following query modes:

[0099] (1) Query by key turning point: Enter the index or time range of the key turning point, and return all capability evolution paths and corresponding integrated indicator sets containing that node.

[0100] (2) Query by temporal correlation vector: Input the feature constraints of the temporal correlation vector (such as the value range of each component), and return all training session records that meet the conditions.

[0101] (3) Range query: Enter the start and end time or the range of sessions to return a summary of all training data within the range.

[0102] The query interface uses an index acceleration mechanism, creating B+ tree indexes for key turning points and multi-dimensional indexes (R-tree) for time-series correlation vectors to ensure that the query response time is less than 100 milliseconds.

[0103] Data updates involve inserting the capability evolution path and integrated index set of the new training sessions into the end of the main data chain. The insertion operation includes creating a new node, setting the predecessor pointer to point to the original end node, and updating the successor pointer of the original end node to point to the new node. The migration offset coefficients of all nodes are recalculated to reflect the impact of the new sessions on the overall capability development trajectory. The recalculation method is as follows: starting from the head node of the main chain, traversing the chain, recalculating the migration offset coefficients for each pair of adjacent nodes, and updating the offset coefficient values ​​of each node. Through this data management mechanism, the system achieves standardized storage, secure access control, and efficient retrieval of training data, providing reliable data support for the long-term tracking of learners' airflow control capabilities and the development of personalized training programs.

[0104] This embodiment collects multi-dimensional airflow parameters and extracts feature points through a data acquisition module, constructs a time-series correlation vector through a hierarchical accumulation module to capture the phased features within the training process, establishes a cross-session capability transfer mapping and marks key turning points through a fusion and labeling module, integrates multi-dimensional parameters into a comprehensive evaluation index through an integration and setting module, and achieves standardized management and efficient retrieval of training data through an archiving module. This system effectively addresses the limitations of existing airflow control capability assessment methods in multi-dimensional parameter time-series correlation analysis and data integration, providing a complete technical solution for scientific skills training and assessment.

[0105] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A skill training evaluation and data management system based on airflow control ability, characterized in that, include: Data acquisition module: Collects airflow pressure sequence, velocity sequence and duration sequence for each training session, combines training session identifier and learner identifier to construct airflow control sequence set, and extracts intensity peak points, stability fluctuation points and rhythm inflection points from it as initial airflow feature point set; The hierarchical accumulation module divides the airflow control sequence set for each training session into temporal hierarchical layers and generates cross-layer association chains through layer-by-layer accumulation to obtain a temporal association vector. The acquisition method of the temporal association vector includes: dividing the airflow control sequence set into a starting layer, a core layer, and a closing layer; setting a low distribution density threshold for the starting layer, a medium threshold for the core layer, and a high threshold for the closing layer; and calculating the average interval between feature point sets within each layer as the layer weight threshold; connecting feature point sets point by point from the starting layer to the core layer to form a starting-core association chain, and connecting point by point from the core layer to the closing layer to form a core-closing association chain; merging the two chains based on the layer weight threshold to generate a cross-layer association chain; converting each connection point in the cross-layer association chain into a vector component, and accumulating all components to obtain the temporal association vector. The fusion labeling module performs capability transfer mapping on multiple training sessions based on the continuity identifier between the temporal correlation vector and the training sessions. It introduces a transfer offset coefficient for adjustment, generates a capability evolution path, and marks key turning points in the path. The capability evolution path is obtained by: extracting the feature point set of the closing layer of the previous session and calculating the temporal interval with the starting layer of the current session based on the continuity identifier to generate a projection matrix; mapping the feature point set of the closing layer to the starting layer using the projection matrix, and introducing a transfer offset coefficient based on the interval length to adjust the position of the mapped points; connecting the adjusted mapped points with the core layer and closing layer of the current session to form a capability evolution path, and marking nodes in the path whose offset coefficient changes beyond a preset value as key turning points. Integration Setting Module: Utilizing key turning points and temporal correlation vectors in the capability evolution path, this module performs dimensional integration operations. It overlays vectors of corresponding nodes in the airflow pressure, velocity, and duration sequences, and dynamically sets the integration granularity threshold based on the number of training sessions. This includes: assigning unique indices to key turning points in the capability evolution path and extracting the corresponding index components from the temporal correlation vectors as overlay base points; adding the node values ​​of the airflow pressure, velocity, and duration sequences one by one to the overlay base points to form a preliminary overlay vector; adjusting the vector amplitude using scaling rules based on the time distance between nodes to generate an integrated index set; and dynamically increasing or decreasing the integration granularity threshold based on the current number of training sessions and the proportion of total sessions, and grouping and merging indices exceeding the threshold within the integrated index set. Archive module: Based on the integrated indicator set, build a data management archive for learners, archive the ability evolution path and the integrated indicator set in chronological order, set archive access keys and enable cross-session query interface.

2. The skill training evaluation and data management system based on airflow control ability of claim 1, wherein, The methods for constructing the initial airflow feature point set include: The highest point of the continuous rising segment in the airflow pressure sequence is identified as the intensity peak point. Points with a coefficient of variation exceeding a preset threshold in the velocity sequence are identified as stability fluctuation points. Points with abrupt changes in the interval change rate in the duration sequence are identified as rhythm inflection points. The intensity peak points, stability fluctuation points, and rhythm inflection points are sorted by the acquisition timestamp and merged into an initial airflow feature point set. Each point is then labeled with a training session identifier and a learner identifier.

3. The skill training evaluation and data management system based on airflow control ability of claim 2, wherein, The methods for constructing data management archives include: Create a unique data management profile for each learner, store the ability evolution path in the main chain structure of the profile according to the session, and attach the integrated indicator set to the corresponding node of the main chain; generate a profile access key based on the learner identifier and set up a cross-session query interface to support data retrieval by key turning points or time-series correlation vectors; when updating, insert the ability evolution path and integrated indicator set of the new training session into the end of the profile main chain, and recalculate the migration offset coefficient of all nodes.

4. The skill training evaluation and data management system based on airflow control ability of claim 3, wherein, Connecting feature points point by point from the starting layer to the core layer forms a starting-core association chain; connecting feature points point by point from the core layer to the closing layer forms a core-closing association chain, including: Density clustering is performed on the feature point set of the initial layer, and the cluster centers are identified as anchor points. The temporal matching degree between the anchor points and the corresponding points in the core layer is calculated, and a matching degree threshold is introduced to filter the initial connection pairs. Based on the filtered initial connection pairs, a gradual path planning rule is applied to extend from the anchor points of the initial layer to the points of the core layer point by point to form the initial-core association chain. During the extension process, a path curvature constraint mechanism is embedded to perform local rerouting on connection segments that exceed the upper limit of curvature. The feature point set of the core layer is sorted by similarity, and the sorting result is used as a guiding sequence. The similarity is matched point by point from the end point of the core layer to the starting point of the closing layer to generate the core-closing association chain.

5. The skill training assessment and data management system based on airflow control capability according to claim 4, characterized in that, Based on hierarchical weight thresholds, two chains are merged to generate cross-layer related chains, including: Link alignment is performed on the start-core association chain and the core-end association chain. The temporal offset difference between corresponding connection points in the two chains is calculated, and an adaptive fusion factor based on the offset difference is introduced. The adaptive fusion factor is dynamically generated by multiplying the square root of the offset difference by the hierarchical weight threshold. The connection nodes of the two chains are traversed point by point. Starting from the end node of the start-core association chain, the projection is made to the start node of the core-end association chain. The adaptive fusion factor is used as a weighting coefficient to perform weighted fusion on the vector components of the projection points to form a preliminary merged chain segment. The rate of change of the curvature of the merged chain segment is detected. If the rate of change exceeds the preset fusion threshold, an auxiliary anchor point is inserted. Based on the ratio of the total chain length of the preliminary merged chain segment to the hierarchical weight threshold, the link optimization iteration is performed. The interval between adjacent nodes in the chain segment is adaptively adjusted. An iterative convergence condition is introduced to minimize the overall link deviation and generate an optimized cross-layer association chain. Each node in the optimized chain is labeled with a fusion factor, and the global association strength of the chain is calculated as a verification index. If the global association strength is lower than the preset lower limit, the adaptive fusion factor is adjusted backtrackingly.

6. The skill training assessment and data management system based on airflow control capability according to claim 5, characterized in that, The feature point set of the closing layer is mapped to the starting layer using a projection matrix, and a migration offset coefficient based on the interval length is introduced to adjust the position of the mapped points, including: Based on the temporal interval between the feature point set of the previous session's closing layer and the starting layer of the current session, the specific elements of the projection matrix are constructed. The main diagonal values ​​of the matrix are generated by calculating the logarithmic transformation of the interval length and the vector product of the continuity identifier, and the interval correlation coefficient is introduced as the off-diagonal element. The projection matrix is ​​applied to each feature point of the closing layer to perform coordinate transformation, generating an initial mapping point set. The migration offset coefficient based on the interval length is calculated. The migration offset coefficient is dynamically determined by the weighted sum of the exponential function of the interval length and the layer weight threshold. The vector displacement of the initial mapping points is adjusted to compensate for the deviation caused by the temporal discontinuity, forming the adjusted mapping point set.

7. The skill training assessment and data management system based on airflow control capability according to claim 6, characterized in that, The vector magnitude is adjusted using scaling rules based on the time distance between nodes, including: The temporal distance between adjacent nodes is calculated for the node sequence in the initial superposition vector. The standard distance is generated by standardizing the temporal distance, and a correction coefficient based on the temporal correlation vector is introduced. The correction coefficient is determined by the product of the average value of the vector components and the integration granularity threshold. For each node, the scaling rules are applied as follows: if the standard distance is lower than the preset lower threshold, a logarithmic compression function is used to reduce the magnitude; if the standard distance is higher than the preset upper threshold, a power function is used to expand the magnitude; if the standard distance is between the lower and upper thresholds, a gradual adjustment is made using a linear interpolation function.

Images