Cost prediction and dynamic scheduling method of warning interrupt under multi-task condition of cockpit

Abstract

This invention belongs to the field of alarm scheduling technology and discloses a method for predicting and dynamically scheduling alarm interruption costs under multi-tasking conditions in the cockpit. The method includes: collecting core data and constructing a task type classification table; collecting cockpit task status data, crew cognitive load data, and alarm timing data to construct standard data; constructing a three-dimensional cost coupling model, outputting the total interruption cost score, and predicting the interruption cost of each alarm based on the total interruption cost score; constructing an alarm causal relationship graph, and dynamically scheduling alarms with interruption cost risks. This invention, by calculating the total interruption cost score of each alarm, quantifies the comprehensive negative impact of alarm interruption on flight safety, mission continuity, crew operational reliability, and cockpit operating efficiency, thereby accurately identifying alarms with severe consequences and high risk levels after interruption, enabling priority scheduling of high-risk alarms, and ensuring priority protection for critical tasks and high-risk alarms.

Description

Technical Field

[0001] This invention relates to the field of alarm scheduling technology, and more specifically, to a method for predicting and dynamically scheduling alarm interruption costs under cockpit multitasking conditions. Background Technology

[0002] In complex systems such as modern aviation, aerospace, and intelligent cockpits, the cockpit operating environment is characterized by multi-task concurrency and intensive human-machine interaction. As a key link in ensuring system safety and assisting crew decision-making, the alarm system's alarm interruption cost prediction and scheduling strategy directly affect the crew's cognitive load and operational response efficiency. In existing technologies, alarm interruption cost prediction mostly relies on static rules such as alarm type and level, and scheduling strategies mostly select alarms with high failure probability based on historical failure frequency. They generally lack accurate identification and classification mechanisms for task types in multi-task cockpit scenarios and fail to deeply explore the inherent relationship between operational behavior, scenario characteristics, and human factors. This results in task criticality assessment being divorced from the actual task context and failing to accurately reflect the true impact of alarms at specific task stages.

[0003] Meanwhile, existing technologies only consider the attributes of the alarm itself when predicting interruption costs, ignoring the coupling effect of multiple dimensions such as task status, crew cognitive load, and alarm timing conflicts, resulting in a large deviation between the prediction results and the actual risks. In addition, most existing alarm scheduling methods treat alarms as isolated events and fail to effectively construct causal relationships and dynamic co-occurrence relationships between alarms. When multiple alarms occur concurrently, the system cannot identify chain alarms caused by the same core fault, which can easily lead to critical alarms being overwhelmed by a large number of secondary alarms, resulting in delayed responses to high-risk, strongly correlated alarms, which seriously threatens system security and task execution.

[0004] In view of this, the present invention proposes a method for predicting and dynamically scheduling alarm interruption costs under cockpit multi-tasking conditions to solve the above problems. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art and achieve the above objectives, the present invention provides the following technical solution: A method for predicting and dynamically scheduling alarm interruption costs under cockpit multi-tasking conditions, the method includes: Step 1: Collect core data during cockpit operation and construct a mission type classification table based on the core data; Step 2: Deploy data acquisition equipment adapted to the cockpit environment to collect cockpit mission status data, occupant cognitive load data and alarm timing data, and build sample data. Preprocess the sample data to obtain standard data. Step 3: Construct a three-dimensional cost coupling model based on standard data. Collect cockpit mission status data, crew cognitive load data and alarm timing data under each alarm in real time and input them into the three-dimensional cost coupling model. Output the total interruption cost score and predict the interruption cost risk of each alarm based on the total interruption cost score. Step 4: Construct an alarm causal relationship graph. Based on the alarm causal relationship graph, dynamically schedule alarms that pose a risk of interruption.

[0006] Furthermore, the method for constructing the task type hierarchy table is as follows: Core data from cockpit operations is collected, and clustered using an improved DBSCAN algorithm to obtain multiple task types. Historical task data for each task type is acquired, and average scenario feature scores and average human factor feature scores are obtained based on this data. The average scenario feature scores and average human factor feature scores are weighted and summed to obtain task priority scores. Task priority levels for different task types are determined based on these priority scores, with each priority level having a corresponding score range. Historical task completion progress for each task type is acquired, and task execution progress levels for different task types are determined based on this progress, with each execution progress level having a corresponding completion progress range. The severity of historical task interruptions for each task type is acquired, and the severity of task interruption impact for different task types is determined based on this severity, with each interruption impact level having a corresponding severity range. A task type classification table is constructed based on the task priority levels, task execution progress levels, and task interruption impact levels for all task types.

[0007] Furthermore, the method for obtaining multiple task types is as follows: The core data includes operational behavior, scene features, and human factors characteristics. After preprocessing the core data, a continuous task operation sequence is constructed in timestamp order. Using a preset operation pause threshold as a boundary condition, the continuous task operation sequence is divided into multiple independent task operation segments, each of which constitutes a cluster sample. Based on all cluster samples, a cluster sample set is constructed. The Euclidean distance between every two cluster samples in the cluster sample set is calculated to obtain a distance distribution histogram. The distance abrupt change point in the distance distribution histogram is selected as the neighborhood radius ε. The minimum number of neighborhood points MinPts is determined based on the number of cluster samples in the cluster sample set. The distance between each cluster sample is calculated using a weighted distance formula that incorporates operational behavior, scene features, and human factors characteristics, and is denoted as the weighted distance. For each cluster sample i in the cluster sample set, calculate the weighted distance between cluster sample i and all other cluster samples in the cluster sample set. If the weighted distance is less than or equal to the neighborhood radius ε, and the number of cluster samples within the neighborhood radius ε is greater than the minimum number of points in the neighborhood MinPts, then cluster sample i is marked as a core point, and a core point set is constructed based on all core points. For each core point in the core point set, all cluster samples within the neighborhood radius ε are grouped into the same cluster, and this process is recursively extended to other core points within the neighborhood radius to form a complete cluster. Cluster samples that are not grouped are marked as noise points. Based on all complete clusters, a cluster set is constructed. The effectiveness of all complete clusters in the cluster set is evaluated to obtain multiple effective clusters. The effective clusters are named according to task type to obtain multiple task types.

[0008] Furthermore, the method for determining the distance to the mutation point is as follows: Based on all Euclidean distances in the distance distribution histogram, construct an Euclidean distance set; sort all Euclidean distances in the Euclidean distance set from smallest to largest to obtain a distance sequence, and calculate the growth rate of adjacent Euclidean distances in turn; traverse the distance sequence and take the Euclidean distance corresponding to the first position where the growth rate is greater than a preset growth threshold as the distance mutation point.

[0009] Furthermore, the method for obtaining multiple effective clusters is as follows: For each complete cluster, calculate the average weighted distance from cluster sample r within the complete cluster to other cluster samples within the complete cluster to obtain the cohesion of cluster sample r; calculate the average weighted distance from cluster sample r within the complete cluster to all other cluster samples within the complete cluster, and take the minimum average weighted distance as the separation; subtract the cohesion from the separation and then normalize to obtain the silhouette coefficient of cluster sample r within the complete cluster; sum the silhouette coefficients of all cluster samples within the complete cluster to obtain the quality coefficient of the complete cluster; for each complete cluster, statistically analyze the distribution of operational behavior characteristics of cluster samples within the complete cluster, calculate the operational behavior entropy, normalize the operational behavior entropy to obtain the operational consistency coefficient, and weight and sum the quality coefficient and operational consistency coefficient to obtain the evaluation value of the complete cluster; compare the evaluation value with a preset evaluation threshold, and if the evaluation value is greater than the evaluation threshold, the complete cluster is marked as a valid cluster; otherwise, it is marked as an invalid cluster.

[0010] Furthermore, the method for constructing a three-dimensional cost coupling model is as follows: Standard data was analyzed to obtain the task criticality score, cognitive load score, temporal conflict score, and interaction integration score of the alarm, and a coupling function was constructed. The coupling function was constructed by calculating the weights of the task criticality score, cognitive load score, temporal conflict score, and interaction integration score, and then performing a weighted summation of these scores based on the weights to construct the coupling function. The coupling function was trained, tested, and optimized using an LSTM model to obtain a three-dimensional cost coupling model. The calculation methods for task criticality score, cognitive load score, temporal conflict score, and interaction integration score are as follows: Based on the collected cockpit task status data, the task priority level, task execution progress level, and task interruption impact level of the corresponding task type are obtained by comparing with the task type classification table, and the task criticality score is obtained by weighted summation; Based on the occupant cognitive load data, the quantitative values ​​of multiple cognitive load core features are extracted, the Euclidean distance between the quantitative value of each cognitive load core feature and the corresponding preset standard quantitative value is calculated, and marked as the cognitive load quantity; The entropy method is used to determine the feature weight of each cognitive load core feature, and the cognitive load core features are weighted summation based on the feature weight to obtain the cognitive load score; Based on the collected alarm timing data, the alarm frequency, alarm triggering timing quantitative values, and alarm time overlap are obtained, and the temporal conflict score is obtained by weighted summation; The interaction integration score is obtained by multiplying the task criticality score, cognitive load score, and temporal conflict score.

[0011] Furthermore, the weighting method for calculating the task criticality score, cognitive load score, temporal conflict score, and interaction integration score is as follows: Scores for various core features corresponding to different alarms in different cockpit scenarios are obtained from historical databases. These core features include task criticality, cognitive load, temporal conflict, and interaction integration. Based on the scores of all core features, a sample set for each cockpit scenario is constructed. Standard score thresholds are set for each core feature in each cockpit scenario. The core features of each cockpit scenario are compared with their respective standard score thresholds, and core features with scores greater than the corresponding standard score thresholds are selected to form the effective set for each cockpit scenario. The number of core features in the effective set is divided by the number of corresponding core features in the sample set to obtain the effective values ​​of each core feature. The effective values ​​of each core feature in the same cockpit scenario are compared pairwise to obtain the core feature values ​​for each core feature. The correlation coefficients of core features are used to construct a correlation coefficient matrix for each cockpit scenario. The correlation coefficients in each row of the matrix are multiplied cumulatively and then squared (k-th), where k is the order of the correlation coefficient matrix, to obtain the influence values ​​of various core features in each cockpit scenario. These influence values ​​are then normalized to obtain the influence coefficients of various core features in each cockpit scenario. The mean of the influence coefficients of similar core features across all cockpit scenarios is calculated to obtain the weights of various core features. Specifically, the weight of task criticality corresponds to the weight of the task criticality score, the weight of cognitive load corresponds to the weight of the cognitive load score, the weight of temporal conflict corresponds to the weight of the temporal conflict score, and the weight of interaction integration corresponds to the weight of the interaction integration score.

[0012] Furthermore, the method for predicting the interruption cost risk of each alarm based on the total interruption cost score is as follows: A preset total interruption cost score threshold is set, and the total interruption cost score is compared with the preset total interruption cost score threshold: if the total interruption cost score is greater than the preset total interruption cost score threshold, it is determined that the alarm has an interruption cost risk; otherwise, it is determined that the alarm does not have an interruption cost risk.

[0013] Furthermore, the method for dynamically scheduling alarms that pose a risk of interruption is as follows: An alarm causal relationship graph is constructed, and alarms with the risk of interruption cost are marked as prominent alarms. Alarms that are associated with prominent alarms in the alarm causal relationship graph are marked as associated alarms. The method for determining whether an alarm is associated is as follows: the causal relationship strength of alarms connected to prominent alarms is greater than a preset causal relationship strength threshold, and the co-occurrence relationship strength is greater than a preset co-occurrence relationship strength threshold. The causal relationship strength and co-occurrence relationship strength of associated alarms are added to obtain the relationship strength value of each associated alarm. The relationship strength value of each associated alarm is multiplied by the total interruption cost score to obtain the interruption cost relationship score of each associated alarm. The interruption cost relationship scores of all associated alarms are accumulated and added to the total interruption cost score of prominent alarms to obtain the priority score of each prominent alarm. The priority scores are normalized to obtain the priority coefficient of each prominent alarm. Based on the priority coefficient of each alarm, the alarms are sorted in descending order to obtain a scheduling ranking table. The alarms are scheduled based on the scheduling ranking table.

[0014] Furthermore, the method for constructing the alarm causal relationship graph is as follows: The system collects cockpit system configuration files, historical alarm logs, real-time system status data, and expert knowledge bases to construct a multi-source dataset. This dataset is preprocessed to obtain a standard dataset. Based on the cockpit system configuration files and historical alarm logs in the standard dataset, a system fault tree model and component dependency graph are extracted. The system fault tree model is analyzed to establish a causal mapping relationship between core faults and triggered alarms. The component dependency graph is analyzed to construct the hierarchical relationship between parent and child components. The causal mapping relationship between core faults and triggered alarms is matched with the hierarchical relationship between parent and child components to obtain the causal hierarchy relationship between alarms. Each causal hierarchy relationship between alarms is recorded as a parent-child alarm pair. The minimum cut set method of fault trees is used to calculate the causal hierarchy relationship between each parent-child alarm pair. The contribution of the parent alarm to the core fault and the contribution of the child alarm to the core fault in alarm-child alarm pairs are considered, and the smaller value is taken as the causal association strength of the alarm-child alarm pair. Based on historical alarm logs and real-time system status data in the standard dataset, a standardized alarm time series dataset is constructed. The standardized alarm time series dataset is sliced ​​according to a preset time window, and the alarm set in each time window is extracted. The Apriori algorithm is used to extract frequently co-occurring alarm pairs in the alarm set, and the co-occurrence support and confidence of each frequently co-occurring alarm pair are calculated. The co-occurrence support and confidence are weighted and accumulated to obtain the co-occurrence association strength of each frequently co-occurring alarm pair. An alarm causal association graph is constructed with alarms as nodes and causal association strength and co-occurrence association strength as edges.

[0015] The technical effects and advantages of the alarm interruption cost prediction and dynamic scheduling method under cockpit multi-tasking conditions of the present invention are as follows: This invention uses an improved DBSCAN clustering algorithm to automatically identify task types by combining operational behavior, scene features, and human factors. It also uses a dual evaluation of quality coefficient and operational consistency coefficient to screen effective clusters, ensuring the objectivity and accuracy of task type identification. At the same time, it constructs a hierarchical table based on three dimensions: task priority, task execution progress, and task interruption impact, which can comprehensively cover the safety, efficiency, and operational constraints of cockpit tasks, making the hierarchical results more in line with actual operational needs. This invention calculates the total interruption cost score of each alarm, quantifying the comprehensive negative impact of alarm interruption on flight safety, mission continuity, crew operational reliability, and cockpit operational efficiency. This allows for the accurate identification of alarms with severe consequences and high risk levels after interruption, enabling priority scheduling of high-risk alarms and ensuring priority protection for critical missions and high-risk alarms. This significantly improves the scientific rigor, accuracy, and safety of cockpit alarm handling. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the method steps of the present invention; Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0017] 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.

[0018] In one embodiment, a method for predicting and dynamically scheduling alarm interruption costs under cockpit multitasking conditions is disclosed, such as... Figure 1 , Figure 2 As shown, the method mainly includes: Step 1: Collect core data during cockpit operation and construct a mission type classification table based on the core data; Step 2: Deploy data acquisition equipment adapted to the cockpit environment to collect cockpit mission status data, occupant cognitive load data and alarm timing data, and build sample data. Preprocess the sample data to obtain standard data. Step 3: Construct a three-dimensional cost coupling model based on standard data. Collect cockpit mission status data, crew cognitive load data and alarm timing data under each alarm in real time and input them into the three-dimensional cost coupling model. Output the total interruption cost score and predict the interruption cost risk of each alarm based on the total interruption cost score. Step 4: Construct an alarm causal relationship graph. Based on the alarm causal relationship graph, dynamically schedule alarms that pose a risk of interruption.

[0019] Through the above scheme, this embodiment first collects core data during cockpit operation, and constructs a task type classification table based on the core data. The collected core data includes three categories: operational behavior, scene characteristics, and human factors characteristics. Then, the core data is clustered using an improved DBSCAN algorithm to obtain multiple task types. A task type classification table is then constructed based on historical task data, historical task completion progress, and the severity of historical task interruption impact for each task type. During clustering, effective clusters are selected through a dual evaluation using quality coefficients and operational consistency coefficients to ensure the objectivity and accuracy of task type identification. Furthermore, the classification is based on task priority scores, task execution progress, and the impact of task interruptions. A task type classification table is constructed based on three dimensions: severity, comprehensive coverage of cockpit mission safety, efficiency, and operational constraints, making the classification results more aligned with actual operational needs. Then, data acquisition equipment adapted to the cockpit environment is deployed to collect cockpit mission status data, occupant cognitive load data, and alarm timing data to build sample data. After preprocessing the sample data, standard data is obtained. Based on the cockpit mission status data in the standard data, a task criticality score is calculated; based on the occupant cognitive load data, a cognitive load score is calculated; based on the alarm timing data, a timing conflict score is calculated; and based on the interaction and comprehensive score of the cockpit mission status data, occupant cognitive load data, and alarm timing data, a final score is calculated. Each of the following weighted coefficients is used: the task criticality score, cognitive load score, temporal conflict score, and interaction integration score are multiplied by their respective weights and then summed to construct a coupling function. Real-time collected cockpit task status data, occupant cognitive load data, and alarm timing data are input into the three-dimensional cost coupling model, outputting the total interruption cost score. Based on the total interruption cost score, the interruption cost of each alarm is predicted for risk. A higher total interruption cost score indicates a more severe impact and greater risk after alarm interruption. Therefore, the total interruption cost score is compared with a preset total interruption cost score threshold: if the total interruption cost score is greater than the threshold, the alarm is considered to be in danger. To mitigate the risk of interruption, priority scheduling is necessary. This allows for the accurate identification of alarms with severe consequences and high risk levels, enabling priority scheduling of high-risk alarms and ensuring the priority protection of critical tasks and high-risk alarms. This significantly improves the scientific rigor, accuracy, and safety of cockpit alarm handling. Finally, an alarm causal relationship graph is constructed. Alarms with interruption risk are first marked as prominent alarms. Related alarms are identified from the alarm causal relationship graph. The priority coefficient of each alarm is calculated based on the total interruption cost score of related alarms and prominent alarms. The alarms are then sorted in descending order according to the priority coefficient to obtain a scheduling ranking table. Alarms are then scheduled based on the scheduling ranking table.By using quantified priority coefficients, alarm priorities are scientifically ordered, effectively filtering out weakly correlated and low-risk interference information. This helps occupants quickly focus on high-risk and strongly correlated core alarms, significantly improving the accuracy and efficiency of alarm response and ensuring the safe and stable operation of the cockpit system in complex scenarios.

[0020] The method for constructing a task type classification table is as follows: Core data from the cockpit operation process is collected, and the core data is clustered using an improved DBSCAN algorithm to obtain multiple task types. Historical task data for each task type is acquired, and the average scenario feature score and average human factor feature score for each task type are obtained based on this historical data. The average scenario feature score and average human factor feature score are weighted and summed to obtain a task priority score. Based on the task priority score, the task priority level for different task types is determined, with each task priority level having a corresponding task priority score range. Historical task completion progress for each task type is acquired, and the task execution progress level for different task types is determined based on this historical completion progress, with each task execution progress level having a corresponding task completion progress range. Historical task interruption severity for each task type is acquired, and the task interruption impact severity level for different task types is determined based on this historical interruption severity, with each task interruption impact severity level having a corresponding task interruption impact severity range. Based on the task priority level, task execution progress level, and task interruption impact severity level for all task types, a task type classification table is constructed. The method for obtaining multiple task types is as follows: Core data includes operational behavior, scene features, and human factors features. After preprocessing the core data, a continuous task operation sequence is constructed in timestamp order. Using a preset operation pause threshold as a boundary condition, the continuous task operation sequence is divided into multiple independent task operation segments, each of which constitutes a cluster sample. Based on all cluster samples, a cluster sample set is constructed. The Euclidean distance between every two cluster samples in the cluster sample set is calculated to obtain a distance distribution histogram. The distance abrupt change point in the distance distribution histogram is selected as the neighborhood radius ε. The minimum number of neighborhood points MinPts is determined based on the number of cluster samples in the cluster sample set. The distance between each cluster sample is calculated using a weighted distance formula that incorporates operational behavior, scene features, and human factors features. Let the weighted distance be denoted as ε. Iterate through each cluster sample i in the cluster sample set, calculating the weighted distance between cluster sample i and all other cluster samples in the cluster sample set. If the weighted distance is less than or equal to the neighborhood radius ε, and the number of cluster samples within the neighborhood radius ε is greater than the minimum number of points in the neighborhood (MinPts), then cluster sample i is marked as a core point. A core point set is constructed based on all core points. For each core point in the core point set, all cluster samples within the neighborhood radius ε are grouped into the same cluster, recursively expanding to other core points within the neighborhood radius to form complete clusters. Cluster samples not grouped are denoted as noise points. Based on all complete clusters, a cluster set is constructed. The effectiveness of all complete clusters in the cluster set is evaluated to obtain multiple effective clusters. The effective clusters are named according to task type to obtain multiple task types. The method for obtaining multiple valid clusters is as follows: For each complete cluster, calculate the average weighted distance from cluster sample r within the complete cluster to other cluster samples within the complete cluster to obtain the cohesion of cluster sample r; calculate the average weighted distance from cluster sample r within the complete cluster to all other cluster samples within the complete cluster, and take the minimum average weighted distance as the separation; subtract the cohesion from the separation and then normalize to obtain the silhouette coefficient of cluster sample r within the complete cluster; sum the silhouette coefficients of all cluster samples within the complete cluster to obtain the quality coefficient of the complete cluster; for each complete cluster, statistically analyze the distribution of operational behavior characteristics of cluster samples within the complete cluster, calculate the operational behavior entropy, normalize the operational behavior entropy to obtain the operational consistency coefficient, and weight and sum the quality coefficient and operational consistency coefficient to obtain the evaluation value of the complete cluster; compare the evaluation value with a preset evaluation threshold. If the evaluation value is greater than the evaluation threshold, the complete cluster is marked as a valid cluster; otherwise, it is marked as an invalid cluster. The method for determining distance mutation points is as follows: Based on all Euclidean distances in the distance distribution histogram, construct an Euclidean distance set; sort all Euclidean distances in the Euclidean distance set from smallest to largest to obtain a distance sequence, and calculate the growth rate of adjacent Euclidean distances in turn; traverse the distance sequence and take the Euclidean distance corresponding to the first position where the growth rate is greater than the preset growth threshold as the distance mutation point.

[0021] The above scheme provides a specific method for constructing a task type classification table. First, core data from the cockpit operation process is collected. Based on the improved DBSCAN algorithm, this core data is clustered to obtain multiple task types. The collected core data includes three categories: operational behavior, scene features, and human factors. Operational behavior data includes operation duration and button types; scene features include environmental parameters and equipment status parameters for each flight phase; and human factors data includes crew cognitive load, operational proficiency, and various physiological parameters. Then, this data is preprocessed, including data cleaning (such as removing outliers and missing values). The process involves several steps, including filling in and deduplicating data, data normalization (such as normalizing operation time, environmental parameters, and device status parameters), and feature encoding (such as labeling operation button types, flight stages, and operator proficiency, converting them into processable numerical forms). Finally, a continuous task operation sequence is constructed in timestamp order, containing timestamps, operational behaviors, scene features, and human factors. Using a preset operation pause threshold (which can be set manually based on experience, such as setting the pause threshold to 3 seconds) as boundary conditions, the continuous task operation sequence is divided into multiple independent task operation segments, each forming a cluster sample. Based on all cluster samples... The following steps are performed: Construct a clustered sample set; calculate the Euclidean distance between every two clustered samples in the clustered sample set to obtain a distance distribution histogram; select the distance abrupt change point in the distance distribution histogram as the neighborhood radius ε, where the distance abrupt change point is determined as follows: based on all Euclidean distances in the distance distribution histogram, construct an Euclidean distance set, sort all Euclidean distances from smallest to largest to obtain an ordered distance sequence, and calculate the growth rate of adjacent Euclidean distances sequentially; traverse the distance sequence and select the Euclidean distance corresponding to the first position where the growth rate exceeds a preset growth threshold as the distance abrupt change point; determine the minimum number of neighborhood points MinPts based on the number of clustered samples in the clustered sample set, such as when clustering... The number of clustered samples in the sample set is multiplied by 0.01 to obtain the minimum number of neighborhood points, MinPts. The distance between each clustered sample is calculated using a weighted distance formula that incorporates operational behavior, scene features, and human factors, and is denoted as the weighted distance. The weighted distance is calculated as follows: the Euclidean distance, the cosine distance, and the Manhattan distance corresponding to the operational behavior, scene features, and human factors of the two clustered samples are calculated in sequence. Then, based on the experience and professional knowledge of the personnel in this field, a weight coefficient is determined for the operational behavior, scene features, and human factors in sequence. The Euclidean distance, cosine distance, and Manhattan distance are then weighted and summed based on the weight coefficients to obtain the weighted distance between the two clustered samples. Traverse each cluster sample in the cluster sample set Calculate clustered samples The weighted distance to all other clusters in the cluster set is used. If the weighted distance is less than or equal to the neighborhood radius ε, and the number of clusters within the neighborhood radius ε is greater than the minimum number of points in the neighborhood MinPts, then the clustered samples are considered as clustered samples. Mark as core points; construct a core point set based on all core points; for each core point in the core point set, group all clustered samples within a neighborhood radius ε into the same cluster, recursively extending to other core points within the neighborhood radius to form complete clusters; un-clustered clustered samples are marked as noise points and removed; construct a cluster set based on all complete clusters; evaluate the effectiveness of all complete clusters in the cluster set to obtain multiple effective clusters. The specific evaluation method is as follows: for each complete cluster, calculate the average weighted distance from clustered sample r within the complete cluster to other clustered samples within the corresponding complete cluster to obtain the cohesion of clustered sample r. Cohesion indicates the cohesion of the clustered sample r. The cohesion coefficient is calculated by subtracting the cohesion coefficient from the cohesion coefficient and then normalizing the result. The quality coefficient of the entire cluster is obtained by summing the silhouette coefficients of all clusters within the cluster. The higher the value, the greater the cohesion and the higher the consistency of internal features. The smaller the value, the lower the cohesion and the lower the separation coefficient. The quality coefficient of a cluster is obtained by summing the silhouette coefficients of all clusters within the cluster. The silhouette coefficient indicates the rationality of clustering of individual clusters; a higher value indicates a higher degree of clustering. The sample in question is well-suited to the current cluster, closely connected to other samples within the same cluster, and significantly different from other clusters. Therefore, the higher the quality coefficient of the complete cluster, the more reasonable the complete cluster is. For each complete cluster, the distribution of operational behavior features (such as key objects, operation sequence patterns, etc.) of the clustered samples within the complete cluster is statistically analyzed, and the operational behavior entropy is calculated. For example, the feature distribution ratio of various operational behaviors within the complete cluster can be obtained. Then, the operational behavior entropy is calculated using the formula for calculating information entropy, and the operational behavior entropy is normalized to obtain the operational consistency coefficient. The higher the operational consistency coefficient, the higher the operational behavior within the cluster, indicating that all samples belong to almost the same type of operation. The higher the probability, the better; therefore, the quality coefficient and the operation consistency coefficient are weighted and accumulated to obtain the evaluation value of the complete cluster; the evaluation value is compared with the preset evaluation threshold, which can be obtained based on the experience and professional knowledge of those in the field. When the evaluation value is greater than the evaluation threshold, it means that the complete cluster is valid, and the complete cluster is marked as a valid cluster; otherwise, it is marked as an invalid cluster; invalid clusters are removed, and the valid clusters are named by task type to obtain multiple task types. The naming method can be based on the most frequent operation behavior, scene characteristics, and task purpose within the complete cluster, such as flight attitude control task cluster, emergency fault handling task cluster, etc. Based on historical task data for each task type, the average scenario feature score and average human factor feature score for each task type are obtained. The average scenario score is quantified from the scenario features of historical tasks within the corresponding task type into a unified scenario feature score. The quantification of each scenario feature score can be combined with the professional knowledge and experience of personnel in this field; for example, an emergency scenario might be assigned a score of 0.9, a cruise scenario 0.3, and an equipment failure scenario 0.7. The arithmetic mean of the scenario feature scores for all historical tasks is then calculated to obtain the average scenario feature score for that task type. Similarly, the average human factor feature score can be quantified from the human factor characteristics (such as occupant cognitive load and operational proficiency) during the execution of historical tasks within the corresponding task type into a unified human factor feature score. For example… For example, a cognitive load ≥ 0.8 is assigned a value of 0.9, proficiency level of "expert" is assigned a value of 0.2, and proficiency level of "novice" is assigned a value of 0.8. Specific quantification methods can be combined with the professional knowledge and experience of those in the field, and will not be elaborated upon here. Next, the arithmetic mean of the human factor characteristic scores of all historical tasks is calculated to obtain the average human factor characteristic score for that task type. The average scenario characteristic score and the average human factor characteristic score are weighted and summed to obtain the task priority score. The weights of the average scenario characteristic score and the average human factor characteristic score can be set based on the experience and professional knowledge of those in the field. Based on the task priority score, the task priority level for different task types is determined, where each task priority level has a corresponding task priority score range. For example, based on the task priority score... The task priority is divided into four levels (Special Priority, Level 1 Priority, Level 2 Priority, and Level 3 Priority), each with a corresponding priority score range. Then, the historical task completion progress for each task type is obtained, and the task execution progress level is determined based on this historical progress. Each task execution progress level also has a corresponding completion progress range. These ranges can be customized based on the experience and expertise of those working in the field. For example, a Special Priority task with a completion progress greater than 90% is classified as Level A; a completion progress between 70% and 90% is Level B; and a completion progress less than 70% is Level C. Similarly, a Level 1 Priority task with a completion progress greater than 80% is classified as Level C. If the completion progress is 0%, it is classified as Grade A; if the completion progress is 60%-80%, it is classified as Grade B; and if the completion progress is less than 60%, it is classified as Grade C. For example, if the completion progress of a secondary priority task is greater than 70%, it is classified as Grade A; if the completion progress is 50%-70%, it is classified as Grade B; and if the completion progress is less than 50%, it is classified as Grade C. For example, if the completion progress of a tertiary priority task is greater than 60%, it is classified as Grade A; if the completion progress is 40%-60%, it is classified as Grade B; and if the completion progress is less than 40%, it is classified as Grade C. The severity of historical task interruption impact is obtained for each task type. Based on the severity of historical task interruption impact, the task interruption impact level for different task types is determined. Each task interruption impact level has a corresponding task interruption impact severity range.(For example, for top-priority tasks: interruption leading to safety risk is classified as Level A (severe), task chain break as Level B (moderate), and slight delay as Level C (minor); for first-priority tasks: interruption leading to task chain break as Level A, slight delay as Level B, and no impact as Level C; for second-priority tasks: interruption leading to slight delay as Level A, and no impact as Level B; for third-priority tasks: interruption with no impact as Level A). Based on the task priority level, task execution progress level, and task interruption impact level of all task types, a task type classification table is constructed. This solution improves the DBSCAN clustering algorithm, combines operational behavior, scene characteristics, and human factors to automatically identify task types, and uses a dual evaluation of quality coefficient and operational consistency coefficient to filter effective clusters, ensuring the objectivity and accuracy of task type identification. Simultaneously, the classification table is constructed based on three dimensions: task priority, task execution progress, and task interruption impact, comprehensively covering the safety, efficiency, and operational constraints of cockpit tasks, making the classification results more aligned with actual operational needs.

[0022] The method for constructing the three-dimensional cost coupling model is as follows: Standard data is analyzed to obtain the task criticality score, cognitive load score, temporal conflict score, and interaction integration score of the alarm, and a coupling function is constructed. The coupling function is constructed by calculating the weights of the task criticality score, cognitive load score, temporal conflict score, and interaction integration score, and then performing a weighted summation based on these weights to construct the coupling function. The coupling function is trained, tested, and optimized using an LSTM model to obtain the three-dimensional cost coupling model. The task criticality score is calculated by comparing the collected cockpit task status data with a task type classification table to obtain the task priority level, task execution progress level, and task interruption impact level for the corresponding task type. The task criticality score is obtained by weighted summation. The cognitive load score is calculated as follows: quantified values ​​of multiple cognitive load core features are extracted based on occupant cognitive load data; the Euclidean distance between the quantified value of each cognitive load core feature and the corresponding preset standard quantified value is calculated and marked as the cognitive load quantity; the feature weight of each cognitive load core feature is determined by the entropy method; and the cognitive load score is obtained by weighted summation based on the feature weights. The time-series conflict score is calculated as follows: based on the collected alarm time-series data, the alarm frequency, alarm trigger timing quantified values, and alarm time overlap are obtained; and the time-series conflict score is obtained by weighted summation. The interaction integration score is calculated as follows: the interaction integration score is obtained by multiplying the task criticality score, cognitive load score, and time-series conflict score. The weighting method for calculating the task criticality score, cognitive load score, temporal conflict score, and interaction integration score is as follows: The scores of various core features corresponding to different alarms in different cockpit scenarios are obtained from the historical database. These core features include task criticality, cognitive load, temporal conflict, and interaction integration. Based on the obtained scores of all core features, a sample set for each cockpit scenario is formed. A standard score threshold is set for each type of core feature in each cockpit scenario. The core features of each cockpit scenario are compared with their respective standard score thresholds, and core features with scores greater than the corresponding standard score thresholds are selected to form the effective set for each cockpit scenario. The effective value of each type of core feature is obtained by dividing the number of each type of core feature in the effective set by the number of corresponding core features in the sample set. The scores of each type of core feature in the same cockpit scenario are then weighted. The effective values ​​of the core features are compared pairwise to obtain the correlation coefficients of each core feature. Based on the correlation coefficients, a correlation coefficient matrix is ​​constructed for each cockpit scenario. The correlation coefficients of each row in the correlation coefficient matrix are multiplied and k-squared, where k is the order of the correlation coefficient matrix, to obtain the influence value of each type of core feature in each cockpit scenario. The influence values ​​of each type of core feature are normalized to obtain the influence coefficient of each type of core feature in each cockpit scenario. The influence coefficients of the same type of core feature in all cockpit scenarios are averaged to obtain the weight of each type of core feature. Among them, the weight of task criticality corresponds to the weight of task criticality score, the weight of cognitive load corresponds to the weight of cognitive load score, the weight of temporal conflict degree corresponds to the weight of temporal conflict degree score, and the weight of interaction integration degree corresponds to the weight of interaction integration degree score. The method for predicting the interruption cost risk of each alarm based on the total interruption cost score is as follows: a preset total interruption cost score threshold is set, and the total interruption cost score is compared with the preset total interruption cost score threshold. If the total interruption cost score is greater than the total interruption cost score threshold, it is determined that the alarm has an interruption cost risk; otherwise, it is determined that the alarm does not have an interruption cost risk.

[0023] The above solution provides a specific method for determining whether an alarm poses a risk of interruption. First, data acquisition equipment adapted to the cockpit environment is deployed to collect cockpit task status data, occupant cognitive load data, and alarm timing data to construct sample data. This sample data is then preprocessed to obtain standard data. Cockpit task status data includes task ID, task type, task priority-related data, task execution data, and data on the impact of task interruption. Occupant cognitive load data includes physiological characteristic data (such as eye movement data, heart rate, and respiratory rate) and behavioral characteristic data (such as operation frequency, operation response delay, and task switching interval). Alarm timing data includes alarm ID, alarm type, alarm frequency, quantified alarm trigger timing value, and alarm time overlap. This data is then preprocessed, including missing value handling, data normalization, and feature encoding, to obtain standard data. The standard data is then analyzed to obtain the task criticality score, cognitive load score, and timing conflict score for each alarm. The interaction comprehensive score, specifically, the task criticality score is obtained as follows: Based on the collected cockpit task status data, it is compared with the task type classification table to obtain the task priority level, task execution progress level, and task interruption impact level for the corresponding task type. Based on experience and professional knowledge, personnel in this field set different quantitative values ​​for the corresponding task priority level, task execution progress level, and task interruption impact level. A weighted summation method is then used to calculate the quantitative values ​​of these three levels to obtain the task criticality score. The weights can be set independently based on the experience of personnel in this field. The task criticality score is a quantitative assessment of the importance and safety impact of the task itself, reflecting the core position and safety weight of the task in the entire cockpit operation system. A higher value indicates a higher safety priority and stronger urgency of execution; once interrupted, the damage to flight safety and mission chain continuity is more severe, requiring priority to ensure its execution or handling. The cognitive load score is obtained as follows: Based on occupant cognitive load data, quantitative values ​​of multiple core cognitive load features are extracted, such as heart rate, respiratory rate, operational frequency, and task switching interval. Each core cognitive load feature has a set standard quantitative value. These standard quantitative values ​​can be pre-set based on industry standards, expert experience, and experimentally verified thresholds. The Euclidean distance between the quantitative value of each core cognitive load feature and its corresponding standard quantitative value is calculated, and this Euclidean distance is recorded as the cognitive load quantity. The smaller the cognitive load quantity, the closer it is to the standard quantitative value. Therefore, the larger the value, the higher the level of cognitive resource consumption by the occupants. The entropy method is used to determine the feature weights of each core cognitive load characteristic. The calculation of weights using the entropy method is a current technique and will not be elaborated upon here. For example, by calculating the information entropy of each core cognitive load characteristic, the weight of each characteristic is determined based on the feature dispersion reflected by the information entropy. The cognitive load of each core cognitive load characteristic is multiplied by its feature weight and then summed to obtain the cognitive load score. The cognitive load score is a quantitative assessment of the psychological and cognitive stress experienced by occupants when performing tasks or responding to alarms, reflecting the degree to which the task consumes the occupants' cognitive resources. A higher value indicates that the occupants need to invest more cognitive resources. The more attentional resources and psychological pressure one invests, the more prone one is to cognitive fatigue, judgment errors, or operational deviations, posing a greater potential threat to the reliability and safety of task execution. The method for obtaining the time-series conflict score is as follows: based on the collected alarm time-series data, obtain the alarm frequency, alarm trigger timing quantification values, and alarm time overlap. The alarm frequency, alarm trigger timing quantification values, and alarm time overlap are weighted and accumulated to obtain the time-series conflict score. The time-series conflict score is a quantitative assessment of the degree of conflict between a task or alarm and other tasks in the time dimension, reflecting the degree of congestion and interference of tasks in the time series. The higher the value, the more significant the conflict. The more likely an alarm is to overlap with or preempt other critical tasks in terms of timing, the more likely it is to disrupt the cockpit mission sequence, leading to delays or omissions of critical operations. The interaction integration score is obtained by multiplying the mission criticality score, cognitive load score, and timing conflict score. The interaction integration score is a quantitative assessment of the interaction between mission criticality, cognitive load, and timing conflict, reflecting the superimposed effect of multiple risk factors. The higher the value, the more significant the superimposed negative effect of high mission criticality, heavy cognitive load, and strong timing conflict, the higher the overall risk of the alarm, and the greater the threat to cockpit safety and operational efficiency.Finally, a coupling function is constructed based on the task criticality score, cognitive load score, temporal conflict score, and interaction integration score. This coupling function is obtained by weighted summation of these scores. The weights for these scores are obtained as follows: The scores of various core features for each alarm in different cockpit scenarios are retrieved from the historical database. These core features include task criticality, cognitive load, temporal conflict, and interaction integration. Based on the scores of all the core features, a sample set for each cockpit scenario is constructed. Each cockpit scenario has a set standard score threshold for various core features. The core features of each cockpit scenario are compared with their respective standard score thresholds. These thresholds are determined by statistical analysis of historical cockpit operation data, combined with industry safety standards and expert settings. Core features with scores greater than the corresponding standard score thresholds are selected to form the effective set for each cockpit scenario. Core features with scores greater than the corresponding standard score thresholds indicate that the intensity, impact, or risk level exhibited by that core feature in the current alarm or task has exceeded the typical acceptable threshold for the corresponding cockpit scenario, requiring close monitoring. The effective set is constructed by filtering these features. The effective value of each core feature is obtained by dividing the number of each core feature in the effective set by the number of corresponding core features in the sample set (e.g., if the number of temporal conflict scores obtained in a cabin scenario is 100, a sample set is constructed based on these 100 sets of temporal conflict scores. After initial and subsequent analysis, it is found that 60 sets of temporal conflict scores have scores greater than the corresponding standard score threshold. These 60 sets of temporal conflict scores are then constructed as the effective set, resulting in an effective value of 60 / 100 for the temporal conflict scores). The effective values ​​of each core feature in the same cabin scenario are compared pairwise to obtain the correlation coefficient of each core feature. Based on the correlation coefficient, a cabin scenario is constructed. The correlation coefficient matrix of the scene is obtained; the correlation coefficients in each row of the correlation coefficient matrix are multiplied and k-squared, where k is the order of the correlation coefficient matrix, to obtain the influence value of various core features in each cockpit scene. The influence values ​​of various core features are normalized to obtain the influence coefficient of various core features in each cockpit scene. The average influence coefficient of the same type of core features in all cockpit scenes is calculated to obtain the weight of various core features. Among them, the weight of task criticality corresponds to the weight of task criticality score, the weight of cognitive load corresponds to the weight of cognitive load score, the weight of temporal conflict degree corresponds to the weight of temporal conflict degree score, and the weight of interaction integration degree corresponds to the weight of interaction integration degree score.This approach extracts core alarm feature data from historical databases under different cockpit scenarios, constructs effective sets, calculates the effective values ​​of each core feature, and then quantifies the scenario impact value of each core feature based on the correlation coefficient matrix and geometric mean method. Finally, the weights are determined by cross-scenario averaging. Compared with existing technologies, this approach breaks through the limitations of setting fixed weights based on human experience, and realizes the scenario-based, data-driven, and dynamic nature of the weights. This not only improves the objectivity and accuracy of weight allocation, but also enhances the adaptability of the weights to different cockpit operating scenarios, effectively improving the scientific nature of the three-dimensional cost coupling model and the accuracy of alarm risk judgment. Finally, the coupling function is trained, tested, and optimized using an LSTM model to obtain a three-dimensional cost coupling model. Since cockpit alarm and task data are typical time series, LSTM can effectively capture the evolution and long-term dependencies of features such as operational behavior, scene changes, and cognitive load over time (e.g., the cumulative impact of earlier operations on subsequent alarm costs), solving the problem that traditional models cannot model temporal context. Therefore, the coupling function is trained, tested, and optimized using an LSTM model. Specifically, time-series samples are first constructed by sampling the coupling function values ​​under historical cockpit scenarios at a preset sampling frequency (e.g., 1Hz). Training samples are generated through a sliding time window (step size 1). The input is the coupling function feature sequence of N consecutive time steps within the window, and the label is the total alarm interruption cost score at the corresponding time of the window, realizing the temporal features. Alignment with the target cost; then, the dataset is divided into training set (70%), validation set (20%), and test set (10%) according to time sequence. The training set is used for model parameter learning, the validation set is used for hyperparameter tuning and early stopping judgment, and the test set is used for final performance evaluation. When building the LSTM model, the input layer dimension is consistent with the coupling function feature dimension, and 2-3 LSTM units are set in the hidden layer (the number of units can be set to 64 or 128), followed by a fully connected layer to output the prediction result. The mean squared error (MSE) or mean absolute error (MAE) is used as the loss function, and the Adam optimizer is used for training (the initial learning rate is set to 1×10^-3). An early stopping mechanism is set (if the validation set loss does not decrease for 5 consecutive rounds, the training is terminated). At the same time, a Dropout layer is introduced (the dropout ratio is set to 0.2~0).5) Prevent overfitting and improve the model's generalization ability; input the test set into the trained model and calculate indicators such as the mean squared error and mean absolute percentage error between the predicted value and the true label to evaluate the model's prediction accuracy; further verify the model's effectiveness in high-risk alarm identification by setting a threshold for the total interruption cost score (e.g., judging whether the predicted score exceeds the threshold), ensuring that the model meets actual scheduling requirements; then adjust the model's hyperparameters (e.g., number of LSTM units, input window length, learning rate, Dropout ratio) based on the evaluation results of the validation and test sets; finally, optimize to obtain a three-dimensional cost coupling model that can be used for real-time prediction of the total cost of alarm interruption; then input the cockpit mission status data, occupant cognitive load data, and alarm time series data collected in real time into the three-dimensional cost coupling model, output the total interruption cost score, and perform risk prediction on the interruption cost of each alarm based on the total interruption cost score; the total interruption cost score can represent the impact on flight safety, mission continuity, and other aspects after the alarm is interrupted. The overall negative impact of alarm interruption on crew operational reliability and cabin operational efficiency is assessed. A higher value indicates more severe consequences and a greater risk of interruption costs. Therefore, based on the experience and expertise of those in the field, a threshold for the total interruption cost is preset. The total interruption cost score is compared to this preset threshold: if the total interruption cost score is greater than the threshold, the alarm is considered to pose an interruption cost risk and requires priority scheduling; otherwise, the alarm is considered not to pose a risk and does not require priority scheduling, but is continuously monitored. This method quantifies the overall negative impact of alarm interruption on flight safety, mission continuity, crew operational reliability, and cabin operational efficiency. This allows for the accurate identification of alarms with severe consequences and high risk levels, enabling priority scheduling of high-risk alarms and ensuring priority protection for critical missions and high-risk alarms. This significantly improves the scientific rigor, accuracy, and safety of cabin alarm handling.

[0024] The method for dynamically scheduling alarms with interruption cost risk is as follows: An alarm causal relationship graph is constructed, and alarms with interruption cost risk are designated as prominent alarms. Alarms associated with prominent alarms are obtained from the alarm causal relationship graph and designated as associated alarms. The method for determining whether an alarm is associated is as follows: the causal relationship strength of alarms connected to prominent alarms is greater than a preset causal relationship strength threshold, and the co-occurrence relationship strength is greater than a preset co-occurrence relationship strength threshold. The causal relationship strength and co-occurrence relationship strength of associated alarms are added to obtain the relationship strength value of each associated alarm. The relationship strength value of each associated alarm is multiplied by the total interruption cost score to obtain the interruption cost relationship score of each associated alarm. The interruption cost relationship scores of all associated alarms are accumulated and added to the total interruption cost score of prominent alarms to obtain the priority score of each prominent alarm. The priority score is normalized to obtain the priority coefficient of each prominent alarm. Based on the priority coefficient of each alarm, the alarms are sorted in descending order to obtain a scheduling sorting table. The alarms are then scheduled based on the scheduling sorting table. The method for constructing the alarm causal relationship graph is as follows: Collect cockpit system configuration files, historical alarm logs, real-time system status data, and an expert knowledge base to construct a multi-source dataset. Preprocess the multi-source dataset to obtain a standard dataset. Based on the cockpit system configuration files and historical alarm logs in the standard dataset, extract the system fault tree model and component dependency graph. Analyze the system fault tree model to establish the causal mapping relationship between core faults and triggered alarms. Analyze the component dependency graph to construct the hierarchical relationship between parent and child components. Match the causal mapping relationship between core faults and triggered alarms with the hierarchical relationship between parent and child components to obtain the causal hierarchy relationship between alarms. Record each causal hierarchy relationship between alarms as a parent-child alarm pair. Use the fault tree minimum cut set method. The algorithm calculates the contribution of the parent alarm to the core fault and the contribution of the child alarm to the core fault in each parent-child alarm pair, and takes the smaller contribution value as the causal association strength of the alarm-child alarm pair. Based on historical alarm logs and real-time system status data in the standard dataset, a standardized alarm time series dataset is constructed. The standardized alarm time series dataset is sliced ​​according to a preset time window, and the alarm set in each time window is extracted. The Apriori algorithm is used to extract frequently co-occurring alarm pairs in the alarm set, and the co-occurrence support and confidence of each frequently co-occurring alarm pair are calculated. The co-occurrence support and confidence are weighted and accumulated to obtain the co-occurrence association strength of each frequently co-occurring alarm pair. An alarm causal association graph is constructed with alarms as nodes and causal association strength and co-occurrence association strength as edges.

[0025] The above solution provides a specific method for dynamically scheduling alarms that pose a risk of interruption. First, it constructs an alarm causal relationship graph. Specifically, it collects cockpit system configuration files, historical alarm logs, real-time system status data, and an expert knowledge base to build a multi-source dataset. This multi-source dataset is then preprocessed to obtain a standard dataset. The cockpit system configuration file contains static structural information such as the physical connection relationships of system components, functional hierarchy architecture, a preset fault tree model, component dependency graphs (parent-child component hierarchy), component-alarm mapping tables (the alarm types triggered by each component failure), and system mode definitions (e.g., takeoff / cruise / landing / emergency modes). Historical alarm logs contain alarm IDs, trigger timestamps, alarm types, etc. The system includes a complete historical alarm record with associated components, interruption costs, alarm levels, trigger conditions, processing records, fault codes, and associated fault chain annotations; real-time system status data includes component health (such as sensor data for temperature, pressure, and vibration), operational events (such as pilot operation commands and system automatic control commands), time-series status parameters (such as system mode switching time and component runtime), and environmental parameters (such as ambient temperature and air pressure); the expert knowledge base includes fault-alarm mapping rules (such as low hydraulic pump pressure triggering hydraulic system abnormality alarms), typical fault chains (such as high engine temperature → abnormal fuel flow → decreased thrust), correlation strength judgment criteria (such as causality strength ≥ 0.7 for strong correlation), maintenance manuals, and historical data. A multi-source dataset is constructed using a fault case library, expert experience rules, etc. This dataset is then preprocessed, including data cleaning, standardization, and annotation, to obtain a standard dataset. Based on cockpit system configuration files and historical alarm logs within this standard dataset, a system fault tree model and component dependency graph are extracted. The system fault tree model is analyzed (e.g., analyzing top events (core faults, such as hydraulic system failure) and bottom events (triggered alarms, such as low hydraulic pump pressure) to establish a causal mapping between core faults and triggered alarms. The component dependency graph is analyzed (e.g., main hydraulic pump as a parent component, hydraulic oil filter as a child component, flap actuator as a child component of the oil filter) to construct a hierarchical relationship between parent and child components. The causal mapping relationship between core faults and triggered alarms is matched with the hierarchical relationship between parent and child components to obtain the causal hierarchy relationship between alarms. Each causal hierarchy relationship between alarms is recorded as a parent alarm and child alarm pair. Specifically, the causal mapping relationship between core faults and triggered alarms can be matched with the component hierarchy relationship (parent component and child component) to establish a fault-component-alarm triple binding to clarify the parent component corresponding to each core fault, the parent alarm triggered by each parent component fault, and the child alarm triggered by each child component fault, thereby obtaining the causal hierarchy relationship between alarms. If a parent component fault (triggers a parent alarm) directly causes the child component to run abnormally (triggers a child alarm), then a direct causal relationship of parent alarm → child alarm is established.If a parent component failure indirectly causes an anomaly in a related component (triggering other alarms), an indirect causal relationship is established from parent alarm to indirect child alarm, thus forming a causal hierarchy network between alarms. Each causal hierarchy relationship between alarms is recorded as a parent-child alarm pair. The fault tree minimum cut set method is used to calculate the contribution of the parent alarm to the core fault and the contribution of the child alarm to the core fault in each parent-child alarm pair. The fault tree minimum cut set method for obtaining contribution is a mature existing technology in the field of fault tree analysis (FTA) used to identify the minimum set of bottom events that lead to the top event and to calculate the contribution of bottom events to the top event. It will not be elaborated on here. For example, first solve for all the minimum cut sets corresponding to the core fault, then... Based on the probability importance calculation formula of minimum cut sets, the contribution of each alarm to the core fault is calculated, and the smaller contribution value is taken as the causal association strength between the alarm and its child alarm pair. The contribution value indicates the degree of influence of the corresponding alarm on the core fault. The strength of the causal chain is generally determined by the weakest link. Therefore, the smaller contribution value is taken as the causal association strength between the alarm and its child alarm pair. The causal association strength measures the tightness of the causal relationship between the parent alarm and the child alarm. The larger the value, the closer the existence of the two alarms. Based on historical alarm logs and real-time system status data in the standard dataset, a standardized alarm time series dataset is constructed. The standardized alarm time series dataset is divided into preset time windows (such as 5 seconds, 10 seconds, which can be adjusted according to the scenario). The alarm set is extracted within each time window (customized). The Apriori algorithm is used to extract frequently co-occurring alarm pairs from the alarm sets. Specifically, the Apriori algorithm iterates through all alarm types to generate all possible alarm pairs. The co-occurrence support of each alarm pair is calculated as follows: the number of time windows in which the two alarms in the alarm pair appear simultaneously divided by the total number of time windows. A co-occurrence support threshold is preset based on the experience and expertise of those in the field. Alarm pairs with a co-occurrence support greater than the threshold are recorded as frequently co-occurring alarm pairs. The co-occurrence support and confidence of each frequently co-occurring alarm pair are calculated. The confidence is calculated as follows: the number of time windows in which the two alarms in the frequently co-occurring alarm pair appear simultaneously divided by the total number of time windows. The number of time windows in which alarm A1 and alarm A2 occur simultaneously is divided by the number of time windows in which only alarm A1 occurs. The co-occurrence support and confidence scores are weighted and summed to obtain the co-occurrence association strength for each frequently co-occurring alarm pair. The weights of each pair can be dynamically adjusted according to the scenario (e.g., increasing the confidence score weight in scenarios with strong temporal dependencies) to ensure the strength more closely reflects the actual association patterns. The co-occurrence association strength is a quantitative measure of the probability and degree of correlation between two alarms triggering simultaneously or sequentially in the temporal dimension. A larger value indicates a stronger temporal correlation between the two alarms, a higher probability of them being triggered together, and a stronger correlation between them. Finally, an alarm causal association graph is constructed using alarms as nodes and causal association strength and co-occurrence association strength as edges.This approach deeply characterizes the essential causal relationships between alarms based on system structure and fault mechanisms through the strength of causal associations, and accurately captures the dynamic co-occurrence patterns of alarms in actual operating scenarios through the strength of co-occurrence associations. This transforms the originally isolated and fragmented alarms into a quantifiable, interpretable, and traversable association network. This not only provides a solid basis for subsequent upper-level applications such as dynamic alarm scheduling, fault tracing, and risk warning, but also helps occupants quickly understand the full picture of the fault chain, effectively reducing cognitive load, improving the accuracy and efficiency of alarm response, and ultimately ensuring the safe and stable operation of the cockpit system. Alarms posing a risk of interruption are designated as prominent alarms. Alarms associated with prominent alarms are identified in the alarm causal relationship graph and designated as associated alarms. The method for determining whether an alarm is associated is as follows: the causal relationship strength of alarms connected to prominent alarms is greater than a preset causal relationship strength threshold, and the co-occurrence relationship strength is greater than a preset co-occurrence relationship strength threshold. Both the preset causal relationship strength threshold and the co-occurrence relationship strength threshold can be set based on the experience and professional knowledge of those in the field. The causal relationship strength and co-occurrence relationship strength of associated alarms are added together to obtain the relationship strength value of each associated alarm. The larger the relationship strength value, the stronger the correlation between the associated alarm and the prominent alarm. The relationship strength value of each associated alarm is multiplied by the total interruption cost score to obtain the interruption cost relationship score of each associated alarm. The interruption cost relationship scores of all associated alarms are accumulated and then added to the total interruption cost score of the prominent alarm to obtain the priority score of each prominent alarm. The priority score is normalized to obtain the priority coefficient of each prominent alarm. The larger the priority coefficient, the more severe the cost after alarm interruption.The greater the chain reaction, the more important it is to sort alarms in descending order based on their priority coefficients to obtain a scheduling list. Alarms are then scheduled based on this list. For example, alarms can be divided into four priority levels based on their priority coefficient ranges: top priority, first priority, second priority, and third priority. Each level has its own scheduling actions and triggering rules. When an alarm's priority coefficient falls within the range corresponding to top priority, and there are no other higher-priority alarms, all scheduling actions are immediately triggered. If there are top-priority alarms of the same level, they are scheduled sequentially from highest to lowest priority. The highest priority alarm occupies the core display area, while the second-highest priority alarm switches to the auxiliary display area but retains audio and lighting. A reminder: If the alarm priority level rises from low to high, the current low-priority scheduling will be immediately interrupted, and a forced switch to high-priority scheduling will be initiated. When the alarm priority level falls within the range corresponding to the first-level priority and there are no high-priority alarms, the above scheduling action will be triggered. If a high-priority alarm exists, it will be placed in the scheduling queue. Scheduling will be automatically activated after the high-priority alarm is handled. If the priority level drops to the second level, it will automatically downgrade to the second-level scheduling action, releasing the occupied auxiliary display resources. When the alarm priority level falls within the range corresponding to the second-level priority, it will be automatically added to the scheduling sorting table without proactive notification. The crew will decide when to handle the alarm. If the priority level rises to the first or high-priority, it will be immediately escalated to the corresponding priority level. The system implements a tiered scheduling mechanism. If an alarm priority drops to level three, only background log entries are retained, and the alarm list is removed from display. When an alarm priority falls within the range corresponding to level three, only background log entries are made, without triggering any scheduling alerts. If the priority rises to level two or higher, the alarm is automatically added to the scheduling order table, and the corresponding scheduling action is triggered. After a priority alarm is handled, the system automatically removes it from the scheduling order table, releasing the occupied display, computing, and communication resources. The next alarm to be scheduled is then activated according to the scheduling order table, thus completing the scheduling process. After scheduling is complete, the system continues to collect data on changes in task status, passenger load, and alarm timing after scheduling execution for continued monitoring, achieving a continuous monitoring closed loop. This method integrates the causal correlation strength and co-occurrence correlation strength of related alarms into a correlation strength value, which is then multiplied by the total interruption cost score to obtain the interruption cost correlation score. Finally, it forms the priority score for prominent alarms and normalizes the ranking. This not only accurately depicts the essential correlation and dynamic co-occurrence rules between alarms, but also closely binds the correlation strength with the interruption risk, avoiding the one-sidedness of single-dimensional judgment. At the same time, the quantified priority coefficient realizes the scientific ranking of alarm priorities, effectively filtering weakly correlated and low-risk interference information, helping occupants quickly focus on high-risk and strongly correlated core alarms, which can significantly improve the accuracy and efficiency of alarm response and ensure the safe and stable operation of the cockpit system in complex scenarios.

[0026] 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.

[0027] All formulas in this manual are dimensionless and calculated numerically. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0028] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for predicting and dynamically scheduling alarm interruption costs under cockpit multi-tasking conditions, characterized in that, The methods include: Step 1: Collect core data during cockpit operation and construct a mission type classification table based on the core data; Step 2: Deploy data acquisition equipment adapted to the cockpit environment to collect cockpit mission status data, occupant cognitive load data and alarm timing data, and build sample data. Preprocess the sample data to obtain standard data. Step 3: Construct a three-dimensional cost coupling model based on standard data. Collect cockpit mission status data, crew cognitive load data and alarm timing data under each alarm in real time and input them into the three-dimensional cost coupling model. Output the total interruption cost score and predict the interruption cost risk of each alarm based on the total interruption cost score. Step 4: Construct an alarm causal relationship graph and dynamically schedule alarms that pose a risk of interruption based on the alarm causal relationship graph. The method for constructing a three-dimensional cost coupling model is as follows: analyze the standard data to obtain the task criticality score, cognitive load score, temporal conflict score, and interaction integration score of the alarm, and construct the coupling function. The coupling function is constructed as follows: the weights of task criticality score, cognitive load score, temporal conflict score, and interaction integration score are calculated, and the weighted sum of the task criticality score, cognitive load score, temporal conflict score, and interaction integration score is calculated based on the weights to construct the coupling function. The coupling function is trained, tested, and optimized using an LSTM model to obtain a three-dimensional cost coupling model. The calculation methods for task criticality score, cognitive load score, temporal conflict score, and interaction integration score are as follows: Based on the collected cockpit mission status data, the mission priority level, mission execution progress level, and mission interruption impact level of the corresponding mission type are obtained by comparing with the mission type classification table, and the mission criticality score is obtained by weighted summation. Based on the cognitive load data of the occupants, the quantitative values ​​of multiple cognitive load core features are extracted. The Euclidean distance between the quantitative value of each cognitive load core feature and the corresponding preset standard quantitative value is calculated and marked as the cognitive load amount. The entropy method is used to determine the feature weight of each cognitive load core feature. Based on the feature weight, the cognitive load core features are weighted and summed to obtain the cognitive load score. Based on the collected alarm timing data, the alarm frequency, alarm trigger timing quantification value and alarm time overlap are obtained, and the timing conflict score is calculated by weighted summation. The interaction comprehensive score is obtained by multiplying the task criticality score, cognitive load score, and temporal conflict score. The method for dynamically scheduling alarms that pose a risk of interruption is as follows: construct an alarm causal relationship graph, record alarms that pose a risk of interruption as prominent alarms, and obtain alarms that are related to prominent alarms from the alarm causal relationship graph and record them as related alarms. The method for determining whether an alarm is associated is as follows: the causal correlation strength of alarms that have an edge connection with prominent alarms is greater than a preset causal correlation strength threshold, and the co-occurrence correlation strength is greater than a preset co-occurrence correlation strength threshold. The correlation strength value of each related alarm is obtained by adding the causal correlation strength and co-occurrence correlation strength of the related alarms. The correlation strength value of each related alarm is multiplied by the total interruption cost score to obtain the interruption cost correlation score of each related alarm. The interruption cost correlation scores of all related alarms are accumulated and added to the total interruption cost score of the prominent alarm to obtain the priority score of each prominent alarm. The priority score is normalized to obtain the priority coefficient of each prominent alarm. Based on the priority coefficient of each alarm, the alarms are sorted in descending order to obtain a scheduling sorting table, and the alarms are scheduled based on the scheduling sorting table.

2. The alarm interruption cost prediction and dynamic scheduling method under cockpit multi-tasking conditions according to claim 1, characterized in that, The method for constructing a task type hierarchy table is as follows: Collect core data during cockpit operation, and cluster the core data based on the improved DBSCAN algorithm to obtain multiple task types; Historical task data for each task type is obtained. Based on the historical task data for each task type, the average scene feature score and average human factor feature score for each task type are obtained. The average scene feature score and average human factor feature score are weighted and summed to obtain the task priority score. Based on the task priority score, the task priority level for different task types is determined. Each task priority level has a corresponding task priority score range. Obtain the historical task completion progress for each task type, and determine the task execution progress level for different task types based on the historical task completion progress. Each task execution progress level has a corresponding task completion progress range. Obtain the severity of historical task interruption impact for each task type, and determine the task interruption impact level for different task types based on the severity of historical task interruption impact. Each task interruption impact level has a corresponding task interruption impact severity range. A task type classification table is constructed based on the task priority level, task execution progress level, and task interruption impact level of all task types.

3. The alarm interruption cost prediction and dynamic scheduling method under cockpit multi-tasking conditions according to claim 2, characterized in that, The method to obtain multiple task types is as follows: The core data includes operational behavior, scene features, and human factors features. After preprocessing the core data, a continuous task operation sequence is constructed in the order of timestamps. Using a preset operation pause threshold as a boundary condition, the continuous task operation sequence is divided into multiple independent task operation segments. Each task operation segment constitutes a cluster sample. Based on all cluster samples, a cluster sample set is constructed. Calculate the Euclidean distance between every two cluster samples in the cluster sample set to obtain a distance distribution histogram. Select the distance abrupt change point in the distance distribution histogram as the neighborhood radius ε. Determine the minimum number of neighborhood points MinPts based on the number of cluster samples in the cluster sample set. The distance between each cluster sample is calculated using a weighted distance formula that incorporates operational behavior, scene features, and human factors. This weighted distance is denoted as the weighted distance. Each cluster sample i in the cluster sample set is traversed, and the weighted distance between cluster sample i and all other cluster samples in the cluster sample set is calculated. If the weighted distance is less than or equal to the neighborhood radius ε, and the number of cluster samples within the neighborhood radius ε is greater than the minimum number of points in the neighborhood MinPts, then cluster sample i is marked as a core point, and a core point set is constructed based on all core points. For each core point in the core point set, all clustered samples within the neighborhood radius ε are grouped into the same cluster. This process is recursively extended to other core points within the neighborhood radius to form complete clusters. Unclustered clustered samples are recorded as noise points. Based on all complete clusters, a cluster set is constructed. The validity of all complete clusters within the cluster set is evaluated to obtain multiple valid clusters. The valid clusters are then named by task type to obtain multiple task types.

4. The alarm interruption cost prediction and dynamic scheduling method under cockpit multi-tasking conditions according to claim 3, characterized in that, The method for determining the distance to the mutation point is as follows: Based on all Euclidean distances in the distance distribution histogram, construct an Euclidean distance set; sort all Euclidean distances in the Euclidean distance set from smallest to largest to obtain a distance sequence, and calculate the growth rate of adjacent Euclidean distances in turn. Traverse the distance sequence and take the Euclidean distance corresponding to the first position where the growth rate is greater than a preset growth threshold as the distance mutation point.

5. The alarm interruption cost prediction and dynamic scheduling method under cockpit multi-tasking conditions according to claim 3, characterized in that, The method to obtain multiple valid clusters is as follows: For each complete cluster, calculate the average weighted distance from cluster sample r within the complete cluster to other cluster samples within the complete cluster to obtain the cohesion of cluster sample r. Calculate the average weighted distance from cluster sample r within the complete cluster to all other cluster samples within the complete cluster. Take the minimum average weighted distance as the separation. Subtract the cohesion from the separation and then normalize to obtain the silhouette coefficient of cluster sample r within the complete cluster. Sum the silhouette coefficients of all cluster samples within the complete cluster to obtain the quality coefficient of the complete cluster. For each complete cluster, the distribution of operational behavior characteristics of clustered samples within the complete cluster is statistically analyzed, the operational behavior entropy is calculated, and the operational behavior entropy is normalized to obtain the operational consistency coefficient. The quality coefficient and the operational consistency coefficient are weighted and summed to obtain the evaluation value of the complete cluster. The evaluation value is compared with the preset evaluation threshold. If the evaluation value is greater than the evaluation threshold, the complete cluster is marked as a valid cluster; otherwise, it is marked as an invalid cluster.

6. The method for predicting and dynamically scheduling alarm interruption costs under cockpit multi-tasking conditions according to claim 1, characterized in that, The weighting method for calculating the task criticality score, cognitive load score, temporal conflict score, and interaction integration score is as follows: The scores of various core features corresponding to different alarms in different cockpit scenarios are obtained from the historical database. The core features include task criticality, cognitive load, temporal conflict degree and interaction integration degree. Based on the scores of all the core features obtained, a sample set for each cockpit scenario is formed. For each cockpit scenario, a standard score threshold is set for each type of core feature. The core features of each cockpit scenario are compared with their respective standard score thresholds, and the core features with scores greater than the corresponding standard score thresholds are selected to form the effective set for each cockpit scenario. The effective values ​​of each type of core feature are obtained by dividing the number of each type of core feature in the effective set by the number of the corresponding type of core feature in the sample set. The effective values ​​of each type of core feature in the same cabin scenario are compared pairwise to obtain the correlation coefficient of each core feature. The correlation coefficient matrix of each cabin scenario is constructed based on the correlation coefficient. The correlation coefficients in each row of the correlation coefficient matrix are multiplied together and k is taken, where k is the order of the correlation coefficient matrix, to obtain the influence value of various core features in each cockpit scenario. The influence values ​​of various core features are then normalized to obtain the influence coefficient of various core features in each cockpit scenario. The influence coefficients of similar core features across all cockpit scenarios are averaged to obtain the weights of each type of core feature. The weight of task criticality corresponds to the weight of the task criticality score, the weight of cognitive load corresponds to the weight of the cognitive load score, the weight of temporal conflict corresponds to the weight of the temporal conflict score, and the weight of interaction integration corresponds to the weight of the interaction integration score.

7. The method for predicting and dynamically scheduling alarm interruption costs under cockpit multi-tasking conditions according to claim 1, characterized in that, The method for predicting the interruption cost risk of each alarm based on the total interruption cost score is as follows: A preset total interruption cost score threshold is set, and the total interruption cost score is compared with the preset total interruption cost score threshold: if the total interruption cost score is greater than the preset total interruption cost score threshold, it is determined that the alarm has an interruption cost risk; otherwise, it is determined that the alarm does not have an interruption cost risk.

8. The method for predicting and dynamically scheduling alarm interruption costs under cockpit multi-tasking conditions according to claim 1, characterized in that, The method for constructing an alarm causal relationship graph is as follows: Collect cockpit system configuration files, historical alarm logs, real-time system status data, and expert knowledge base to construct a multi-source dataset. Preprocess the multi-source dataset to obtain a standard dataset. Based on the cockpit system configuration files and historical alarm logs in the standard dataset, extract the system fault tree model and component dependency graph; The system fault tree model is analyzed to establish the causal mapping relationship between core faults and alarm triggering. The component dependency graph is analyzed to construct the hierarchical relationship between parent and child components. The causal mapping relationship between core faults and alarm triggering is matched with the hierarchical relationship between parent and child components to obtain the causal hierarchical relationship between alarms. Each causal hierarchical relationship between alarms is recorded as a parent alarm and child alarm pair. The fault tree minimum cut set method is used to calculate the contribution of the parent alarm to the core fault and the contribution of the child alarm to the core fault in each parent alarm and child alarm pair. The smaller value of the contribution is taken as the causal relationship strength of the alarm and child alarm pair. Based on historical alarm logs and real-time system status data in the standard dataset, a standardized alarm time series dataset is constructed. The standardized alarm time series dataset is sliced ​​according to a preset time window, and the alarm set in each time window is extracted. The Apriori algorithm is used to extract frequently co-occurring alarm pairs in the alarm set, and the co-occurrence support and confidence of each frequently co-occurring alarm pair are calculated. The co-occurrence support and confidence are weighted and accumulated to obtain the co-occurrence association strength of each frequently co-occurring alarm pair. An alarm causal relationship graph is constructed using alarms as nodes and causal relationship strength and co-occurrence relationship strength as edges.

Images