Psychological risk early warning method and system based on developmental psychology trajectory analysis

Abstract

This invention discloses a method and system for early warning of psychological risks based on developmental psychology trajectory analysis. The method includes: constructing a psychological evolution knowledge graph; aligning the original growth data of the target user with developmental stages and standardizing and updating the target user's psychological growth trajectory to obtain developmental stage embedding vectors and the updated psychological growth trajectory; constructing multi-source input features and inputting them into a trajectory representation coding model for encoding; combining the encoding results with the updated psychological growth trajectory to determine turning points, obtaining the turning point identification results for the target user's psychological growth; and constructing a forward warning intensity calculation model to trace the causes of warnings and link them with intervention directions. This invention can dynamically track and provide forward warnings of psychological development trajectories, enabling early intervention for mental health. As a method and system for early warning of psychological risks based on developmental psychology trajectory analysis, this invention can be widely applied in the field of mental health risk warning technology.

Description

Technical Field

[0001] This invention relates to the field of mental health risk early warning technology, and in particular to a method and system for early warning of mental health risks based on developmental psychology trajectory analysis. Background Technology

[0002] With the continued deepening of digital transformation in education and the coordinated development of adolescent mental health, the "one file per student" system, as a core data carrier recording the entire growth cycle of adolescents from enrollment to graduation, has been widely applied in primary and secondary schools at all levels across the country, accumulating massive amounts of adolescent growth data resources and providing basic data support for adolescent mental health protection and educational decision-making optimization. However, the construction and application of the "one file per student" system still faces many prominent bottlenecks. The core problem lies in the failure to effectively transform fragmented and decentralized archival data into interpretable, traceable, and predictable dynamic development trajectories. In particular, it cannot adapt to the inherent continuity, stages, and dynamic characteristics of adolescent psychological development, making it difficult for the existing system to support the early identification and proactive intervention of psychological risks, failing to meet the actual needs of adolescent mental health protection work, and making it difficult to achieve the core goal of "early detection, early identification, and early intervention."

[0003] Adolescent psychological development exhibits significant stage-dependent and trajectory-correlated characteristics: behavioral abnormalities, negative events, or environmental changes at a particular developmental stage can all become potential triggers for subsequent deviations in psychological trajectory. Early identification and accurate capture of key turning points are crucial prerequisites for effective intervention and preventing the continued deterioration of psychological risks. Existing "one file per lifetime" systems and related psychological early warning technologies, lacking systematic developmental psychology theoretical guidance, unified data organization and semantic alignment mechanisms, and efficient dynamic temporal modeling capabilities, cannot accurately capture the evolutionary patterns and core characteristics of psychological developmental turning points. This leads to a general delay in intervention, often initiating intervention only after psychological risks have become apparent or even after a serious psychological crisis has occurred, thus failing to realize the core value of early warning. Summary of the Invention

[0004] To address the aforementioned technical problems, the purpose of this invention is to provide a method and system for early warning of psychological risks based on developmental psychology trajectory analysis, which can dynamically track and provide early warning of psychological development trajectories, thereby enabling early intervention in mental health.

[0005] The first technical solution adopted in this invention is: an early warning method for psychological risks based on developmental psychology trajectory analysis, comprising the following steps: Based on the target user’s original profile data, semantic modeling of multi-source growth nodes and graph relationship evolution are performed to obtain a psychological evolution knowledge graph. The original growth data of the target users are aligned with development stages and the psychological growth trajectory of the target users is standardized and updated to obtain the development stage embedding vector and the updated psychological growth trajectory. By combining a psychological evolution knowledge graph and developmental stage embedding vectors to construct multi-source input features, these features are input into a trajectory representation coding model for encoding. The encoding results are then combined with the updated psychological growth trajectory to determine turning points, thus obtaining the turning point identification results for the target user's psychological growth. A model for calculating the intensity of early warning is constructed. By combining the identification results of turning points in the psychological growth of target users, the causes of early warning are traced and the direction of intervention is linked to achieve early warning of psychological risks.

[0006] Furthermore, the step of performing multi-source growth node semantic modeling and graph relationship evolution based on the target user's original profile data to obtain a psychological evolution knowledge graph specifically includes: The original profile data of the target users is used as the basic node, developmental psychology nodes are constructed based on psychological theories, and stage and risk nodes are constructed based on the age stage data of the target users. The basic nodes, developmental psychology nodes, and stage and risk nodes are sequentially subjected to semantic annotation and standardization, as well as node quality assessment, to construct the initial nodes at each level. Determine the initial association logic and mapping rules between nodes at each level, introduce a pre-set graph dynamic update mechanism, perform joint modeling of relationship type encoding and timestamps, and construct the initial graph relationships; By introducing semantic consistency verification and developmental psychology theory verification mechanisms, anomaly detection is performed on the initial graph relationships to obtain a psychological evolution knowledge graph.

[0007] Furthermore, the step of aligning the target user's original growth data with developmental stages and standardizing and updating the target user's psychological growth trajectory to obtain the developmental stage embedding vector and the updated psychological growth trajectory specifically includes: Based on the stage-based classification criteria for psychological development, and combined with the educational stage divisions and grade distribution, a developmental stage coordinate system is constructed. The original growth data of the target user is obtained and semantically mapped, time-calibrated and cleaned in sequence. The data is then mapped to the corresponding development stage in the development stage coordinate system to obtain the initial development stage embedding vector. A dual verification mechanism for alignment accuracy at each stage is constructed, which compares and calculates the initial development stage embedding vectors and evaluates their semantic similarity to obtain the development stage embedding vectors. The psychological growth trajectory of the target users is determined and standardized using the Z-score standardization method to obtain the standardized psychological growth trajectory; A joint reference mechanism is constructed, which includes an individual's own historical baseline, a group reference trajectory at the same developmental stage, and a typical trajectory of key turning points; The deviation distribution of the standardized psychological growth trajectory is calculated by a joint reference mechanism, and the standardized psychological growth trajectory is updated based on the deviation distribution to obtain the updated psychological growth trajectory.

[0008] Furthermore, the step of calculating the deviation distribution of the standardized psychological growth trajectory through a joint reference mechanism specifically includes: Obtain the target trajectory deviation distance at time t. The target trajectory deviation distance includes the Euclidean distance between the individual's current trajectory and its own historical baseline, the Euclidean distance with the reference trajectory of the same group at the same stage, and the Euclidean distance with the corresponding typical trajectory. At the same time, determine the deviation weight coefficient corresponding to the target trajectory deviation distance. Multiply the target trajectory deviation distance by its corresponding weight coefficient to obtain several weighted deviation values.

[0009] The weighted deviation values ​​are summed together, and the summation result is the total deviation of the individual trajectory at time t.

[0010] Furthermore, the step of constructing multi-source input features by combining psychological evolution knowledge graphs and developmental stage embedding vectors, inputting them into the trajectory representation coding model for encoding, and combining the encoding results with the updated psychological growth trajectory to determine turning points and obtain the turning point identification results of the target user's psychological growth, specifically includes: Construct multi-source input features, which include a psychological evolution knowledge graph, developmental stage embedding vectors, and knowledge graph relation evolution features. The knowledge graph relation evolution features are obtained by extracting the relation evolution sequence before time t in the psychological evolution knowledge graph. A trajectory representation coding model based on bidirectional gated recurrent unit (BiGRU) is constructed to perform deep fusion and temporal coding of multi-source input features to obtain a unified trajectory representation vector. Based on a fully connected neural network, a unified trajectory representation vector is input to determine the probability of turning points and obtain the probability of turning points. A local time window analysis method is introduced to assist in the determination of turning points in the updated psychological growth trajectory, and the results of the local time window analysis are obtained. A multi-loss function joint optimization mechanism is constructed, and the results of local time window analysis and inflection point probability are combined for optimization to obtain the inflection point identification results of the target user's psychological growth.

[0011] Furthermore, the step of constructing a trajectory representation coding model based on a bidirectional gated recurrent unit (BiGRU) to perform deep fusion and temporal coding of multi-source input features to obtain a unified trajectory representation vector specifically includes: Construct a trajectory representation coding model based on a bidirectional gated cyclic unit (BiGRU); The input layer of the trajectory representation coding model based on the bidirectional gated recurrent unit (BiGRU) is used to obtain multi-source input features. The BiGRU trajectory coding layer, based on the trajectory representation coding model of BiGRU bidirectional gated recurrent unit, performs deep temporal coding of long-period trajectory information on multi-source input features to obtain trajectory representation vectors. The temporal attention mechanism of the trajectory representation encoding model based on the bidirectional gated recurrent unit (BiGRU) dynamically allocates attention weights to adjust the trajectory representation vector, thereby obtaining a unified trajectory representation vector.

[0012] Furthermore, the step of determining the probability of the turning point specifically includes: Obtain the unified trajectory representation vector at time t, as well as the predefined turning point discrimination layer weight matrix and bias terms; To complete the linear operation, multiply the weight matrix by the trajectory representation vector at time t, and add the bias term to obtain the comprehensive value after the linear transformation. The final probability is mapped out by inputting the linearly transformed comprehensive value into the Sigmoid activation function for numerical mapping, and the output result is the inflection point probability at time t.

[0013] Furthermore, the loss functions of the multi-loss function joint optimization mechanism specifically include the knowledge graph alignment loss function, the trajectory comparison learning loss function, and the inflection point classification loss function.

[0014] Furthermore, the step of constructing an early warning intensity calculation model, combined with the identification results of turning points in the psychological development of target users, to trace the causes of early warning and link intervention directions, thereby achieving early warning of psychological risks, specifically includes: By combining the probability of turning points, the deviation distribution of standardized psychological growth trajectories, and the cumulative trend term, the forward warning intensity is calculated to obtain the forward warning intensity. Based on the intensity of early warning and combined with the identification results of turning points in the psychological growth of target users, the causes of early warning are traced and the direction of intervention is linked to achieve early warning of psychological risks.

[0015] The second technical solution adopted in this invention is: an early warning system for psychological risks based on developmental psychology trajectory analysis, comprising: The first module is used to perform multi-source growth node semantic modeling and graph relationship evolution based on the target user's original profile data to obtain a psychological evolution knowledge graph. The second module is used to align the developmental stages of the target user’s original growth data and to standardize and update the target user’s psychological growth trajectory, so as to obtain the developmental stage embedding vector and the updated psychological growth trajectory. The third module is used to construct multi-source input features by combining the psychological evolution knowledge graph and developmental stage embedding vectors, and input them into the trajectory representation coding model for encoding. The encoding results are combined with the updated psychological growth trajectory to determine the turning point and obtain the turning point identification results of the target user's psychological growth. The fourth module is used to construct an early warning intensity calculation model, which, combined with the identification results of the turning point in the psychological growth of the target user, traces the cause of the warning and links the direction of intervention to achieve early warning of psychological risks.

[0016] The beneficial effects of the method and system of this invention are as follows: This invention obtains a psychological evolution knowledge graph by performing multi-source growth node semantic modeling and graph relationship evolution based on the original archive data of the target user. Through multi-source growth node semantic modeling and graph relationship evolution, combined with dynamic updating and quality verification mechanisms, it ensures that the knowledge graph can accurately carry developmental psychology semantics and dynamically reflect the evolutionary laws of growth trajectory. Furthermore, it aligns the original growth data of the target user with developmental stages and standardizes and updates the target user's psychological growth trajectory, obtaining developmental stage embedding vectors and updated psychological growth trajectories. Based on the stage division theory of developmental psychology and the actual division of school education stages, a scientific developmental stage alignment and trajectory standardization mechanism is established, enabling fragmented data across semesters, scenarios, and systems to be compared, analyzed, and integrated within the same temporal semantic framework. Simultaneously, it achieves standardized processing of individual trajectories and multi-dimensional reference. To ensure the accuracy of data alignment and the comparability of trajectories, a multi-source input feature model is constructed by combining a psychological evolution knowledge graph and developmental stage embedding vectors. This feature is then input into the trajectory representation coding model for encoding. The encoding results are combined with the updated psychological growth trajectory to determine turning points, thus obtaining the turning point identification results for the target user's psychological growth. Through the trajectory representation coding model, the turning point probability determination mechanism, and the multi-loss function joint optimization mechanism, the model is ensured to accurately capture the trajectory evolution pattern and the core features of turning points. Finally, an advance warning intensity calculation model is constructed. Combined with the turning point identification results for the target user's psychological growth, the model traces the cause of the warning and links the intervention direction, achieving early warning of psychological risks. By constructing a time-series knowledge graph that can continuously evolve over time and a precise and efficient turning point advance warning model, the model achieves dynamic tracking of adolescent psychological development trajectories, early identification of deviation signals, and early prediction and advance warning of turning points. Attached Figure Description

[0017] Figure 1 This is a flowchart of the steps of the early warning method for psychological risks based on developmental psychology trajectory analysis of the present invention; Figure 2 This is a structural block diagram of the early warning system for psychological risks based on developmental psychology trajectory analysis, as described in this invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments. The step numbers in the following embodiments are only for ease of explanation and do not limit the order of the steps. The execution order of each step in the embodiments can be adapted according to the understanding of those skilled in the art.

[0019] First, it should be noted that the existing technology has the following shortcomings, for example: 1) The traditional "one file per lifetime" system is mainly based on static archiving and lacks the ability to interpret developmental changes: The core function of traditional "one file per lifetime" systems focuses on the collection, organization, and static archiving of growth data. Essentially, it's a passive storage model of "log-style static stacking," only achieving the basic goal of "recording facts" and failing to fulfill the core needs of "interpreting the significance of data development, relating it to growth patterns, and predicting development trends." While these systems can record detailed data on adolescents' psychological scale results, school behavior, and growth events at different points in time, they cannot establish connections between this isolated and fragmented data and core theories in developmental psychology such as developmental stages, critical periods of psychological development, and risk accumulation mechanisms. This makes it difficult to scientifically explain three core questions: first, the impact mechanism and transmission path of a particular growth event on subsequent psychological development trajectories; second, whether behavioral abnormalities at a certain stage are normal developmental fluctuations or developmental deviations; and third, the long-term impact of current data abnormalities on future psychological states. For example, the system can accurately record that a student's score on a single psychological scale is low, but it cannot explain the reasonableness of the distribution of the score in the same stage of the group, or determine whether the abnormal score will cause the subsequent psychological trajectory to continue to deviate. As a result, the value of massive archive data cannot be fully utilized, and it cannot provide effective data interpretation and theoretical support for the early identification of psychological risks.

[0020] 2) Data fragmentation across semesters is severe, making it difficult to establish a continuous development trajectory: Adolescent growth data is inherently fragmented, spanning semesters, scenarios, systems, and roles. This data is scattered across different educational and psychological service systems, lacking unified organizational standards and temporal semantic alignment mechanisms. Specifically, psychological assessment data is stored separately in a professional psychological assessment system; campus behavior data is scattered across multiple subsystems such as academic affairs management, class management, and student management; home-school interaction data is stored in a home-school collaboration platform; and psychological counseling records are kept and managed separately by psychological counselors. Significant differences exist in data formats, time standards, and semantic specifications across different systems, and there is a lack of effective data association and sharing mechanisms between systems, creating severe "data silos." Existing "one file per student" systems typically only achieve simple aggregation and storage of some core data, failing to provide unified temporal calibration, semantic alignment, and logical connection for fragmented data spanning semesters and scenarios. This results in a fragmented and disjointed growth trajectory for adolescents, failing to form a complete, continuous, and traceable psychological development trajectory, and consequently, failing to effectively identify early deviation signals and key turning points within the trajectory.

[0021] 3) Static knowledge structures are insufficient to describe the "evolutionary" and "transfer" characteristics of psychological development: Currently, some "one file per student" systems are attempting to introduce knowledge graph technology to structure and organize growth data, aiming to improve data relevance and readability. However, existing knowledge graphs mostly employ static construction methods, with extremely low update frequencies for nodes and relationships. They can only reflect a "static knowledge state at a certain point in time," failing to effectively depict the core dynamic characteristics of "evolution" and "transfer" in the psychological development of adolescents. Developmental psychology theory clearly points out that the psychological state of adolescents is not fixed but dynamically evolves through continuous interaction with cognitive abilities, emotion regulation methods, coping strategies, social support networks, and external environmental events. Significant differences exist in psychological characteristics and behavioral patterns at different developmental stages, and there is a dynamic transfer effect between stages (such as the transition from emotional fluctuations in early adolescence to emotional stability in mid-adolescence). Static knowledge graphs cannot capture this dynamic evolution process and cannot clearly express core logic such as "the migration path from one psychological state to another," "the driving mechanism of environmental events on the evolution of psychological states," and "the correlation and inheritance relationship of psychological characteristics at different stages." This results in knowledge graphs failing to truly and comprehensively reflect the dynamic growth process of adolescents and struggling to support the accurate identification of trajectory deviations and turning points.

[0022] 4) Turning point identification is severely lagging, and intervention points are generally shifted to later stages: The core limitation of existing adolescent mental health early warning systems lies in their severely delayed identification of turning points and the generally delayed intervention points, making it difficult to realize the core value of early warning and meet the actual needs of proactive intervention for psychological risks. These systems typically use single psychological scale scores or overt behavioral abnormalities as warning triggers, essentially employing a passive "current state classification" warning model. They only trigger warnings after psychological risks have become apparent (e.g., persistent low mood, behavioral withdrawal, or a sharp decline in academic performance), failing to identify early turning points in the psychological trajectory—key nodes that may lead to subsequent deviations and the continuous accumulation of psychological risks. For adolescents, deviations in their psychological trajectory are often a gradual, cumulative process. Truly cost-effective intervention points appear at early turning points, not after the full-blown psychological risk has emerged. The lack of dynamic modeling and turning point identification capabilities at the trajectory level means that "the best intervention window has already been missed by the time problems are seen," failing to achieve the core goal of "early detection and early intervention" and making it difficult to fundamentally reduce the incidence of adolescent mental health crises.

[0023] 5) Insufficient individualized adaptation, making it difficult to balance common patterns with individual differences: The psychological development trajectory of adolescents exhibits a significant characteristic of "coexistence of commonalities and differences": on the one hand, adolescents of the same age and developmental stage, influenced by the general laws of physiological and cognitive development, share common patterns and stage-specific characteristics in psychological development; on the other hand, influenced by individual personality traits, family environment, growth experiences, social support systems, and other personalized factors, each adolescent's developmental trajectory shows significant individual differences, with varying speeds, rhythms, and manifestations of psychological development. Existing "one-person-one-file" systems and related early warning models often fall into the trap of "one-size-fits-all" modeling: either relying too heavily on group averages as risk assessment standards, ignoring individual developmental differences and inherent behavioral habits, and misjudging normal behavioral characteristics as abnormal signals; or focusing only on short-term behavioral fluctuations, ignoring common developmental patterns within the same age group, and misjudging pathological deviations as normal individual differences. This lack of individualized adaptation directly leads to high misjudgment and false negative rates in early warning models, making it impossible to accurately distinguish between normal behavioral fluctuations and pathological deviations, and failing to meet the personalized early warning and intervention needs of different adolescents.

[0024] Based on this, the embodiments of this invention are guided by classic theories and empirical research results in developmental psychology. They closely revolve around four core technical lines: "knowledge graph temporal evolution, semantic alignment of developmental stages, deep learning of trajectory representation, and early warning of turning points." The system constructs a complete technical link from the structured organization of multi-source data and dynamic trajectory modeling to early warning output and the linkage of intervention suggestions. This achieves a comprehensive upgrade from static files to dynamic trajectories and a fundamental shift from delayed early warning to early warning. It forms a set of technical solutions that take into account both common developmental laws and individual growth differences, and combine theoretical rigor with practical operability. It effectively solves the prominent shortcomings of existing technologies in developmental interpretation, trajectory continuity, dynamic modeling, early warning timeliness, and individualized adaptation, and provides strong technical support for early intervention of psychological risks in adolescents.

[0025] Reference Figure 1 This invention provides an early warning method for psychological risks based on developmental psychology trajectory analysis, the method comprising the following steps: S100. Based on the target user's original profile data, perform multi-source growth node semantic modeling and graph relationship evolution to obtain a psychological evolution knowledge graph; First, it should be noted that the core objective of this embodiment is to address the prominent problem of traditional "one life, one file" systems, which are primarily static archives and lack developmental interpretability. It deeply integrates core theories of developmental psychology with knowledge graph engineering technology, breaking the inherent limitations of static knowledge structures. This reconstructs scattered and isolated static archive data into a temporally evolving knowledge graph that can be continuously updated over time, interpretable for developmental meaning, and trackable for trajectory evolution. This achieves a leap from "recording facts" to "understanding trajectories, interpreting meaning, and predicting trends," providing a high-quality structured data foundation for subsequent trajectory modeling and early warning. This embodiment, through multi-source growth node semantic modeling and graph relationship evolution representation, combined with dynamic updates and quality verification mechanisms, ensures that the knowledge graph accurately carries developmental psychological semantics and dynamically reflects the evolutionary patterns of growth trajectories.

[0026] S110. Obtain the original profile data of the target user as the basic node, construct developmental psychology nodes based on psychological theories, and construct stage and risk nodes based on the age stage data of the target user. This embodiment breaks through the limitations of traditional knowledge graphs, which suffer from single node types and semantic ambiguity. It constructs a multi-dimensional, multi-layered growth node system with clear developmental psychological semantics, achieving refined semantic modeling of nodes. This transforms isolated archival data into structured nodes with clear developmental psychological significance. Simultaneously, a node quality assessment and standardization mechanism is established to ensure the accuracy, completeness, and semantic consistency of node data, laying a solid semantic foundation for subsequent trajectory evolution analysis and turning point identification. Specific implementation details are as follows: First, a three-tiered growth node system is constructed to comprehensively cover all dimensions of adolescent growth. Each node is assigned clear developmental psychological semantics and standardized data attributes, with specific classifications and definitions as follows: The first category, basic fact nodes, serves as the underlying data support for the knowledge graph. It encompasses various original archival data, including basic information about adolescents, psychological assessment results and their backgrounds, school behavior records, home-school interaction records, detailed information on developmental events, and psychological counseling texts. Specifically, psychological assessment nodes must include auxiliary attributes such as scale name, assessment time, raw score, normative standard score, assessment scenario (routine screening / specialized assessment / post-crisis assessment), and the qualifications of the testing personnel. School behavior nodes must differentiate between sub-types such as classroom behavior, recess behavior, and school activity behavior, and indicate the time, location, frequency, intensity, and associated individuals. Developmental event nodes must clearly define the event type (positive / negative, academic / family / social), time of occurrence, duration, scope of impact, and core details of the event to ensure data completeness and authenticity.

[0027] The second category, developmental psychology nodes, is based on classic theories of developmental psychology such as Erikson's theory of psychosocial development and Piaget's theory of cognitive development, along with empirical research findings from both domestic and international sources. It introduces core nodes such as cognitive developmental status, emotion regulation style, coping strategy type, peer relationship quality, family support intensity, and academic task pressure to achieve precise characterization and semantic interpretation of adolescents' psychological states. For example, the cognitive developmental status node can be subdivided into sub-nodes such as the sensorimotor stage, preoperational stage, concrete operational stage, and formal operational stage, and standardized by combining age group and cognitive assessment results; the emotion regulation style node can be divided into sub-nodes such as cognitive reappraisal, expressive inhibition, and avoidance / catharsis, and semantically labeled using psychological counseling texts and behavioral performance data; the family support intensity node can be divided into three levels: high, medium, and low, and quantitatively labeled by combining home-school interaction data, family change records, and parent-child relationship assessment results.

[0028] The third category comprises stages and risk nodes, including nodes at different developmental stages, deviation points, turning points, and risk accumulation points, enabling precise identification of the psychological development process and risk status. Developmental stage nodes strictly adhere to developmental psychology standards and are further subdivided into childhood (grades 1-3 of primary school), early adolescence (grades 4-6 of primary school), mid-adolescence (grades 1-3 of junior high school), late adolescence (grades 1-3 of senior high school), and early youth (university stage). Each stage node is labeled with core developmental tasks, risk-prone factors, and typical psychological characteristics. Deviation points are labeled with the type of deviation (academic pressure, peer conflict, lack of family support, etc.), the degree of deviation (mild / moderate / severe), and the start time of the deviation. Turning point nodes are labeled with the type of turning point (positive / negative), the triggering event, the duration of the turning point, and the scope of impact. Risk accumulation nodes are labeled with the type of risk, the degree of accumulation, the duration of accumulation, and the critical threshold.

[0029] S120. Perform semantic annotation and standardization on the basic nodes, developmental psychology nodes, and stage and risk nodes in sequence, and conduct node quality assessment to construct the initial nodes at each level. In this embodiment, a node semantic annotation and standardization process is established to ensure the consistency and accuracy of node semantics. A dual annotation mode of "human annotation + machine-assisted verification" is adopted. An annotation team composed of developmental psychology experts, registered psychologists, primary and secondary school psychology teachers, and artificial intelligence engineers is formed to formulate unified semantic annotation specifications, clarifying the annotation standards, annotation dimensions, and operation procedures for each node. Machine-assisted verification uses a semantic similarity calculation model based on domain-fine-tuned BERT to verify the consistency of manually annotated nodes and remove nodes with ambiguous semantics or incorrect annotations. In practical applications, a high level of annotation accuracy and semantic consistency can be used as a quality control target. At the same time, the original archive data undergoes full-process standardization processing, including data cleaning, missing value imputation, outlier filtering, and semantic normalization, to ensure that the node data has good standardization and comparability.

[0030] Finally, a three-level node quality evaluation mechanism is constructed to evaluate nodes in real time from three dimensions: accuracy, completeness, and semantic consistency. The evaluation indicators and passing standards are as follows: accuracy ≥ 95% (confirmed by both manual sampling and machine semantic verification), completeness ≥ 98% (the proportion of nodes with missing core fields does not exceed 2%), and semantic consistency ≥ 90% (the semantic deviation of synonymous nodes does not exceed 10%). If a node fails to meet the evaluation standards, a re-annotation and standardization process is automatically triggered to ensure that the node quality meets the needs of subsequent knowledge graph construction and trajectory modeling.

[0031] S130. Determine the initial association logic and mapping rules between nodes at each level, introduce a preset graph dynamic update mechanism, perform joint modeling of relationship type encoding and timestamp, and construct the initial graph relationship; In this embodiment, a five-level graph topology structure of "individual-time-event-development stage-risk status" is constructed. Through precise modeling using relation type encoding and timestamps, the evolutionary relationships of growth trajectories are accurately expressed, breaking the limitations of static relationships in traditional knowledge graphs and achieving dynamic and temporal representation of relationships. Simultaneously, a flexible and efficient graph dynamic update mechanism and relation verification mechanism are designed to ensure that the graph accurately and timely reflects the dynamic growth process of adolescents. Specific implementation details are as follows: First, a five-level graph topology structure is designed to clarify the association logic and mapping rules of nodes at each level, realizing a full-link chain expression of the growth trajectory: individual nodes serve as the core hub of the graph, connecting all time nodes, event nodes, developmental stage nodes, and risk status nodes related to the adolescent; time nodes use timestamps accurate to the day, connecting all events, psychological states, and developmental stage data at the corresponding time point; event nodes connect to their respective developmental stages, their impact on psychological states, and the resulting changes in risk status; developmental stage nodes connect to the core developmental tasks, risk susceptibility factors, and typical psychological characteristics of that stage; risk status nodes connect to their triggering events, accumulation processes, and the path of their impact on subsequent trajectories, ultimately forming a complete logical link of "individual growth - time progression - event triggering - stage evolution - risk change".

[0032] Secondly, a joint modeling approach combining relation type coding and timestamps is adopted to achieve dynamic and temporal representation of relations. Graph relations are divided into four categories, each assigned a unique code to clarify semantic connotations, association rules, and quantification methods: First, attribution relations (coded R1), including attribution relations such as "event-development stage" and "node-individual," used to clarify the category and hierarchical relationship of a node; second, influence relations (coded R2), including positive / negative and direct / indirect influences of "event-psychological state," using an influence strength coefficient in the 0-1 range to quantify the degree of influence, such as setting the negative influence coefficient of family changes on emotional state to 0.8, and the positive influence coefficient of school awards on self-identity to 0.7; third, transfer relations (coded... R3 includes temporal transfer relationships between "psychological states" and "developmental stages," annotating the transfer direction, duration, and triggering conditions. For example, the transfer from early adolescence to mid-adolescence emotional stability is annotated with the start time, duration, and driving factors (family support / school intervention / cognitive development). The fourth type is cumulative relationships (coded R4), including temporal cumulative relationships between "risk nodes," annotating the accumulation rate, threshold, and synergistic effects. For example, the cumulative risk of academic pressure and peer conflict is calculated, determining the critical state of risk accumulation. Simultaneously, precise timestamp attributes are added to each node and relationship. Node timestamps indicate the time range of their generation / existence, and relationship timestamps indicate the time interval of their effectiveness, enabling temporal modeling of the trajectory evolution process and clearly expressing the temporal correlation and dynamic change patterns of various relationships.

[0033] Furthermore, a flexible and efficient dynamic update mechanism for the graph is designed to ensure its continuous evolution along with the adolescents' growth. The update mechanism has two modes: proactive and reactive. Proactive update mode automatically collects new growth data (such as new assessment results, behavioral records, and growth events) based on a preset time period (weekly / monthly). Through semantic matching and node association algorithms, it automatically updates the nodes and relationships in the graph. For example, it automatically updates students' psychological assessment nodes and behavioral record nodes monthly, supplementing new influence and transfer relationships. Reactive update mode is triggered manually by a psychology teacher or system administrator for sudden growth events (such as family changes, major rewards and punishments, and psychological crises). This quickly adds relevant nodes and relationships, ensuring that sudden events are promptly incorporated into the graph system. An incremental update strategy is used during the update process, updating only the nodes and relationships corresponding to newly added data, while fully preserving historical node and relationship data, achieving full-cycle evolution management with "traceable historical trajectory, quantifiable current state, and predictable future changes."

[0034] S140. Introduce semantic consistency verification and developmental psychology theory verification mechanism to detect relationship anomalies in the initial graph relationships and obtain a psychological evolution knowledge graph.

[0035] In this embodiment, a dual verification mechanism for graph relationships is finally constructed to ensure the accuracy and theoretical rationality of the relationships. A dual verification method of "semantic consistency verification + developmental psychology theory verification" is adopted: semantic consistency verification uses a cosine similarity algorithm to calculate the semantic matching degree between associated nodes and relationships; a matching degree ≥ 0.8 is considered a valid relationship. Developmental psychology theory verification involves a review team composed of developmental psychology experts who conduct theoretical compliance reviews of newly added influence relationships and transfer relationships to ensure that the relationships conform to the general laws of adolescent psychological development. For example, when reviewing the influence relationship "increased academic pressure leads to low mood," it must conform to the relevant theories of stress-emotion interaction to avoid false associations that do not conform to theoretical logic. Simultaneously, an anomaly detection mechanism is established to monitor relationship conflicts in the graph in real time (such as the same event being simultaneously labeled as positive and negative influences). Once an anomaly is detected, a manual review process is automatically triggered to correct erroneous relationships in a timely manner, ensuring the accuracy and rationality of the graph relationships.

[0036] S200, Align the original growth data of the target user with development stages and standardize and update the psychological growth trajectory of the target user to obtain the development stage embedding vector and the updated psychological growth trajectory. First, it should be noted that the core objective of this embodiment is to address the fundamental problems of fragmented data across semesters, difficulty in unifying semantics across different stages, and strong heterogeneity in individual trajectories. Based on the developmental psychology theory of stage division and the actual segmentation of school education, it establishes a scientific mechanism for aligning developmental stages and standardizing trajectories. This enables fragmented data across semesters, scenarios, and systems to be compared, analyzed, and integrated within the same temporal semantic framework. Simultaneously, it achieves standardized processing and multi-dimensional reference of individual trajectories, providing a unified and standardized analytical foundation for subsequent trajectory representation learning, deviation identification, and turning point determination. This embodiment ensures the accuracy of data alignment and the comparability of trajectories through developmental stage alignment, trajectory standardization, and joint reference, combined with calibration and verification mechanisms.

[0037] S210. Based on the stage division criteria of psychological development, and combined with the division of educational stages and grade distribution, construct a developmental stage coordinate system; In this embodiment, based on the developmental psychology standards for adolescent psychological development and combined with the segmentation and grade distribution of school education, a unified and scalable developmental stage coordinate system (StageCoord) is constructed. This system maps raw growth data across semesters, scenarios, and systems to the corresponding developmental stages through semantic mapping, time calibration, and data cleaning, thus obtaining the time-series data. Development stages of embedding vectors This achieves semantic consistency across different data sources at various stages, while establishing an alignment accuracy verification mechanism to ensure the accuracy of the alignment results. Specific implementation details are as follows: First, a unified developmental stage coordinate system (StageCoord) is constructed. Combining developmental psychology theory with the realities of school education, the semantic information of each developmental stage, such as its time frame, core tasks, psychological developmental characteristics, and risk susceptibility factors, is clarified to form a standardized stage description system. The specific divisions are as follows: 1) Childhood (Stage 1): Grades 1-3 of primary school (6-9 years old). The core developmental tasks are the formation of basic learning habits, the establishment of basic social skills and the formation of preliminary self-awareness. The psychological characteristics are high dependence, emotional expression and mainly concrete thinking. The core risk factors are lack of family companionship, school bullying and poor academic adaptation. 2) Early Adolescence (Stage 2): Grades 4-6 of primary school (ages 9-12). The core developmental tasks are establishing peer relationships, developing autonomy, and adapting to academic pressure. The psychological characteristics are awakening of self-awareness, increased emotional fluctuations, and rapid development of abstract thinking. The core risk factors are academic setbacks, peer rejection, and the initial emergence of parent-child conflicts. 3) Mid-adolescence (Stage 3): Grades 1-3 of junior high school (12-15 years old). The core developmental tasks are the initial formation of values, adaptation to the physiological and psychological changes of adolescence, and improvement of emotional regulation ability. The psychological characteristics are increased rebelliousness, high sensitivity, and in-depth self-exploration. The core risk factors are increased parent-child conflict, increased academic pressure, adolescent emotional distress, and peer relationship crisis. 4) Late Adolescence (Stage 4): Grades 1-3 of high school (15-18 years old). The core developmental tasks are establishing life goals, coping with the pressure of the college entrance examination, and improving social adaptability. The psychological characteristics are mature rational thinking, stable self-identity, and prominent stress perception. The core risk factors are the pressure of college entrance examination, tense interpersonal relationships, confusion about self-identity, and anxiety about future development. 5) Early Adolescence (Stage 5): University stage (18-22 years old). The core development tasks are adapting to independent living, building professional skills, and establishing close relationships. The psychological characteristics are increased independence, improved self-decision-making ability, and facing multiple life choices. The core risk factors are academic pressure, employment anxiety, poor interpersonal relationship adjustment, and self-identity reconstruction.

[0038] This coordinate system supports flexible expansion, and can adjust the time range and core features of each stage according to the actual division of school stages in different regions and schools, so as to ensure the adaptability of the coordinate system to different scenarios.

[0039] S220. Obtain the target user's original growth data and perform semantic mapping, time calibration and data cleaning in sequence, mapping it to the corresponding development stage in the development stage coordinate system to obtain the initial development stage embedding vector. In this embodiment, a semantic mapping function is constructed. This achieves a precise mapping from raw growth data to a development stage coordinate system. The mapping function employs a unified development stage feature encoding function, specifying the time-varying characteristics of each stage. The original growth data is used for representation learning, and the corresponding developmental stage embedding vector is output. The core formula is as follows: ; in, Indicates time The developmental stage embedding vector, with a dimension of 64, is used to represent the developmental stage and core characteristics of the adolescent at that moment. Indicates time The raw growth data, including psychological assessments, campus behavior, growth events and related text information from multiple sources, is preprocessed and transformed into standardized feature inputs. This represents a unified development stage feature encoding function, used for joint representation learning of structured growth data and text-based growth data; This represents the stage mapping weight matrix, used to map extracted features to the development stage coordinate system; This represents the stage mapping bias term, used to adjust the offset of the mapping result.

[0040] Next, data time calibration and cleaning are performed to ensure alignment accuracy. Time calibration uses a unified timestamp standard (accurate to the day) to convert the time attributes of data from different sources to this standard, avoiding alignment deviations caused by differences in time formats. For growth data spanning multiple stages (such as data from the transition period to higher education), linear interpolation is used for time smoothing to ensure data continuity during the transition period. In the data cleaning stage, noise and outliers in the original data are further filtered out, and missing data is supplemented to ensure data integrity and accuracy, providing high-quality input data for semantic mapping.

[0041] S230, a dual verification mechanism for alignment accuracy in the construction phase is used to compare and calculate the initial development phase embedding vectors and evaluate their semantic similarity to obtain the development phase embedding vectors; In this embodiment, a dual verification mechanism for stage alignment accuracy is constructed, employing two core indicators—stage matching accuracy and semantic consistency—to control the quality of the alignment results. Stage matching accuracy is calculated by comparing the mapped stage embedding vector with the manually labeled development stages; semantic consistency is evaluated by the semantic similarity of data from different sources within the same stage. In practical applications, higher levels of matching accuracy and semantic consistency thresholds can be used as quality control references; if the alignment accuracy does not meet the expected standards, the semantic mapping function parameters are readjusted, and the time calibration and data cleaning processes are optimized to ensure a comparable analytical basis between data from different learning stages, different developmental tasks, and different sources.

[0042] S240. Determine the psychological growth trajectory of the target user and standardize it using the Z-score standardization method to obtain the standardized psychological growth trajectory. In this embodiment, to address the heterogeneity of individual growth trajectories, a scientific trajectory standardization and joint reference mechanism is constructed. This mechanism enables standardized processing of individual trajectories and multi-dimensional comparison, fully preserving individual growth differences while accurately extracting common patterns across developmental stages. This provides a reliable reference standard for deviation identification and turning point determination. Simultaneously, a standardization verification and reference rationality evaluation mechanism is established to ensure the mechanism's scientific validity and practicality. Specific implementation details are as follows: First, trajectory standardization is implemented to eliminate biases caused by different data types and magnitudes, ensuring the comparability of trajectory data from different individuals and dimensions. Differentiated standardization methods are used based on data type differences: for continuous data (such as psychological scale scores, behavior frequency, and communication duration), Z-score standardization is used to standardize the raw values ​​to obtain a standardized representation that can be used for subsequent trajectory alignment and modeling. The calculation formula is as follows: ; in, Indicates the stage of development Next The standardized results of continuous features This represents the corresponding original value. Indicates the stage of development Next The population mean of a continuous characteristic, Indicates the stage of development Next The group standard deviation of a continuous characteristic. Indicates the development stage number. Continuous feature numbers are used; for categorical data (such as behavior type, risk level, and emotional state), one-hot encoding is used to convert it into standardized vectors; for unstructured text data (such as psychological counseling texts and home-school communication content), a domain-fine-tuned BERT model is used to extract semantic feature vectors, achieving standardization. Through the above standardization process, individual growth trajectories are transformed into directly comparable standardized trajectories, reducing analytical bias caused by data heterogeneity.

[0043] S250. Construct a joint reference mechanism, which includes an individual's own historical baseline, a group reference trajectory at the same developmental stage, and a typical trajectory of key turning points. In this embodiment, a joint reference mechanism is constructed to comprehensively compare an individual's current trajectory performance with three core dimensions, forming a triple reference system of "individual baseline - group pattern - typical trajectory" to accurately identify the degree and direction of deviation in an individual's trajectory. 1) Individual's Historical Baseline: By studying the adolescent's long-term historical data over the past 6-12 months, a unique historical baseline trajectory is constructed to reflect the individual's inherent growth patterns, behavioral habits, and psychological development rhythm. The historical baseline is constructed using a sliding window method, with 30 days as a sliding window. Trajectory features within each window are extracted and smoothed by mean to obtain a continuous individual historical baseline, ensuring that the baseline can truly reflect the individual's normal growth pattern. 2) Group Reference Trajectories at the Same Developmental Stage: Statistical analysis and modeling can be performed based on large-scale group data to construct common reference trajectories for adolescents at the same developmental stage, reflecting common developmental patterns and typical behaviors. The construction of group reference trajectories can employ the K-means clustering algorithm, clustering individual trajectories at the same developmental stage, extracting the cluster center trajectories as group reference trajectories, and calculating the standard deviation range of the group trajectories as a reference interval for normal trajectory fluctuations. The specific sample size can be determined based on data availability and the application scenario. 3) Typical Trajectories of Key Turning Points: By summarizing and modeling historical turning point data (trajectory data of various turning points such as school admissions, family changes, and school bullying), typical trajectories of different types of turning points are constructed to reflect the influence and characteristics of different turning points on growth trajectories. The construction of typical trajectories uses an LSTM time series analysis model, trained on the trajectory data of historical turning points, to extract trajectory patterns of different types of turning points, clarifying the trajectory change characteristics before, during, and after the turning point, providing typical references for turning point identification.

[0044] Next, a deviation calculation mechanism is constructed to accurately quantify the degree and direction of deviation between an individual's current trajectory and the triple reference system. The degree of deviation is calculated using Euclidean distance, with the core formula as follows: ; in, This represents the total deviation of the individual trajectory at time t, with a value range of [0,1]. The larger the deviation value, the more serious the deviation between the individual trajectory and the reference system. This represents the Euclidean distance between an individual's current trajectory and its historical baseline. This represents the Euclidean distance between an individual's current trajectory and the reference trajectory of the group at the same stage; This represents the Euclidean distance between an individual's current trajectory and its corresponding typical trajectory (if a turning event exists); , , This represents the deviation weighting coefficient, which satisfies... The weights of individual historical baselines were determined through offline sample training and five-fold cross-validation. The weight of the group reference trajectory is set to 0.4. Set to 0.3, the weight of the typical trajectory Set it to 0.3 to ensure the scientific and reasonable nature of the deviation calculation.

[0045] S260. Calculate the deviation distribution of the standardized psychological growth trajectory through a joint reference mechanism, and verify and update the standardized psychological growth trajectory based on the deviation distribution to obtain the updated psychological growth trajectory.

[0046] In this embodiment, a standardized verification and reference rationality assessment mechanism is constructed. Standardization verification employs the Kolmogorov-Smirnov distribution consistency test to ensure that the standardized trajectory data conforms to a normal distribution; a distribution consistency ≥90% is considered acceptable for standardization. Reference rationality assessment verifies the effectiveness of the reference system by comparing the deviation distribution of individual trajectories with the triple reference system. If the deviation distribution conforms to the psychological development patterns of adolescents (the deviation value of normal individuals is concentrated in the 0-0.3 range), the reference is deemed reasonable. Simultaneously, a dynamic updating mechanism for the reference system is established. New group data and turning point event data are collected each semester to update the group reference trajectory and typical trajectory, ensuring that the reference system can adapt to the dynamic changes in adolescent psychological development.

[0047] S300: Combine psychological evolution knowledge graph and development stage embedding vector to construct multi-source input features and input them into trajectory representation coding model for encoding. The encoding results are combined with the updated psychological growth trajectory to determine the turning point and obtain the turning point identification result of the target user's psychological growth. First, it should be noted that the core objective of this embodiment is to address the problem that static knowledge structures are insufficient to describe the "evolution" and "transfer" characteristics in psychological development. It introduces a temporal trajectory representation learning method to transform long-term, multi-dimensional, and multimodal growth trajectories into calculable, analyzable, and comparable low-dimensional vector representations. This enables precise characterization of trajectory deviation states and turning points, transforming the fuzzy judgment of "whether the current state conforms to a high-risk evolutionary path" into a quantifiable and verifiable computational problem, thereby improving the accuracy of deviation identification and turning point determination. This embodiment utilizes two core technologies—a trajectory representation encoding model and a turning point probability determination mechanism—combined with a multi-loss function joint optimization mechanism to ensure that the model can accurately capture the trajectory evolution patterns and core features of turning points.

[0048] S310. Construct multi-source input features, which include a psychological evolution knowledge graph, developmental stage embedding vectors, and knowledge graph relation evolution features. The knowledge graph relation evolution features are obtained by extracting the relation evolution sequence before time t in the psychological evolution knowledge graph. S320. Construct a trajectory representation coding model based on bidirectional gated recurrent unit (BiGRU), perform deep fusion and temporal coding of multi-source input features, and obtain a unified trajectory representation vector. In this embodiment, a trajectory representation coding model based on a bidirectional gated cyclic unit (BiGRU) is constructed. This model fully integrates historical node sequences, relationship evolution sequences, and stage change signals from the evolutionary knowledge graph, performing deep fusion and temporal coding to obtain a unified trajectory representation vector. This approach enables dynamic modeling of long-term growth paths, accurately capturing the interconnected changes and evolution of cognitive abilities, emotion regulation methods, coping strategies, and social support networks over time. Simultaneously, a temporal attention mechanism is introduced to optimize the encoding process, enhancing the relevance and accuracy of the representation. Specific implementation details are as follows: First, a model input layer is constructed, integrating multi-source input features to provide comprehensive input data for trajectory encoding. The input layer contains three core inputs: firstly, multi-source fused input features... The first is derived from multi-source growth node data (basic fact nodes, developmental psychology nodes, and stage and risk nodes) through feature fusion and dimensional reduction, covering information from multiple aspects such as psychology, behavior, and environment, with 128 dimensions; the second is the developmental stage embedding vector. The first is derived from the development stage alignment mechanism, with a dimension set to 64, used to integrate stage features into the trajectory encoding process and improve the stage adaptability of the representation; the second is the knowledge graph relationship evolution features. The evolution sequence of relationships before time t in the evolutionary knowledge graph is extracted and converted into a feature vector through relationship encoding. The dimension is set to 64 to capture the relational dependencies of trajectory evolution. The three input features are concatenated to form the final input vector of the model, with a dimension of 128+64+64=256.

[0049] Secondly, a BiGRU trajectory encoding layer is constructed to achieve deep temporal encoding of long-cycle trajectory information. The BiGRU model can adopt a three-layer stacked structure, with the number of hidden units in each layer set to 256, 128, and 64 respectively. Dropout layers are added between layers, with a dropout ratio of 0.2 to suppress overfitting and enhance the model's generalization ability. BiGRU has a strong ability to capture long-sequence dependencies, and can simultaneously extract the forward evolution features (from past to present) and backward correlation features (looking back from the present) of historical trajectories, avoiding information loss caused by single-direction encoding, and achieving a comprehensive characterization of long-cycle growth trajectories. The core formula for calculating the trajectory representation vector is as follows: ; in, Indicates time The trajectory representation vector, with a dimension of 64, represents the time interval... The comprehensive encoding of all previous growth trajectory information can accurately reflect the current growth status, trajectory trend and potential risks; BiGRU stands for Bidirectional Gated Cyclic Unit, which achieves selective retention and reconstruction of time series information through the synergistic effect of update gate, reset gate and forward-backward bidirectional encoding, effectively capturing long sequence dependencies; This represents the concatenated input feature vector; Indicates time The trajectory representation vector enables the temporal transmission and accumulation of trajectory information, ensuring that the encoding results can fully reflect the continuity and evolution of the trajectory.

[0050] Furthermore, a temporal attention mechanism is introduced to optimize the trajectory encoding process and improve the relevance of the representation. The temporal attention mechanism dynamically allocates attention weights based on the importance of trajectory information at different time points, focusing on key time points related to trajectory deviations and turning points (such as the time of developmental events or abnormal psychological states), while downplaying information from irrelevant time points, thus improving the effectiveness of trajectory representation. The core formula for calculating attention weights is as follows: ; ; in, Indicates time The attention weights range from [0,1]. The larger the weight value, the more important the trajectory information at that moment is to the current state. The attention score function is represented by cosine similarity, which is used to calculate the similarity between the trajectory representation vector and the average trajectory representation vector. This represents the average value of the historical trajectory representation vector; Indicates the length of the historical time window (set to 30 days); This represents the trajectory representation vector after attention weighting, which serves as the final output of the model and is used for subsequent yaw recognition and turning point determination.

[0051] Finally, a model training and validation mechanism is constructed to evaluate encoding accuracy and generalization ability. Training samples can utilize long-term growth data covering different development stages and growth trajectory types, divided according to a standard ratio of training, validation, and test sets. During training, common configurations such as the Adam optimizer can be used, with trajectory representation discriminative power as the optimization objective. During validation, metrics such as trajectory clustering accuracy and representation similarity can be combined to comprehensively analyze the model's encoding performance. In the testing phase, the focus is on the model's stability and generalization performance on new samples. The sample size and training parameters can be further adjusted based on the data foundation and deployment scenario.

[0052] S330. Based on a fully connected neural network, a unified trajectory representation vector is input to determine the probability of turning points and obtain the probability of turning points. In this embodiment, the turning point is explicitly defined as: a critical moment when the trajectory representation undergoes a significant directional change, a sustained increase in the risk slope, or a rearrangement of the knowledge graph relationship structure within a local time window. This is based on a unified trajectory representation vector. Construct a precise turning point probability determination mechanism and output the time. Turning point probability This approach effectively distinguishes between "single-point behavioral anomalies" and "overall development deviations," avoiding misjudging short-term fluctuations as turning points. Furthermore, it incorporates local time window analysis and multi-feature fusion to improve the accuracy of turning point identification. Specific implementation details are as follows: First, the criteria and core characteristics of turning points are clearly defined. Based on developmental psychology theory and historical trajectory data, three core characteristics of turning points are summarized: 1) the direction of the trajectory representation vector changes by ≥30% (compared to the representation vector of the previous time window); 2) the risk slope (risk accumulation rate) increases significantly, with an increase of ≥50%; and 3) the core relationship structure in the knowledge graph is rearranged, with ≥2 new / deleted core relationships. Simultaneously, turning points are divided into positive turning points (such as school awards or increased family support leading to a positive trajectory) and risk-related turning points (such as family changes or school bullying causing a trajectory deviation), with differentiated judgment thresholds set for each to improve the targeting of the judgment.

[0053] Secondly, a turning point probability determination model is constructed, which is implemented based on a fully connected neural network (MLP). The input is the trajectory representation vector after attention weighting. The output is the time. Inflection point probability Its core formula is as follows: ; in, This represents the probability of a turning point at time t, with a value ranging from [0,1]. The closer the probability value is to 1, the greater the likelihood that that time is a turning point; when... At that time, it was initially identified as a potential turning point; when At that time, it was determined to be a clear turning point; This represents the Sigmoid activation function, which maps input values ​​to the [0,1] interval to achieve a quantized representation of the probability of inflection points; The weight matrix of the inflection point discrimination layer has a dimension of 64×1 and is used to weight the trajectory representation vector to highlight the core features related to the inflection point. This represents the bias term of the inflection point discrimination layer, used to adjust the offset of the discrimination result and improve the judgment accuracy.

[0054] S340. Introduce the local time window analysis method to assist in the determination of turning points in the updated psychological growth trajectory and obtain the local time window analysis results. S350. Construct a joint optimization mechanism with multiple loss functions, and optimize by combining the results of local time window analysis with the probability of turning points to obtain the turning point identification results of the target user's psychological growth.

[0055] In this embodiment, a local time window analysis method is introduced to assist in determining turning points and avoid misjudging single behavioral fluctuations as turning points affecting long-term trajectories. The local time window is set to 7 days, and the magnitude, direction, and similarity of changes in trajectory representation within adjacent time windows are calculated. The specific calculation method is as follows: 1) Variation magnitude: Calculate the Euclidean distance between the trajectory representation vectors of the current time window and the previous time window. The larger the distance, the more significant the trajectory change. 2) Directional difference: Calculate the cosine similarity between the trajectory representation vectors of the current time window and the previous time window. The smaller the similarity, the more obvious the change in trajectory direction. 3) Pattern similarity: Calculate the similarity between the trajectory representation within the current time window and the trajectory representation of a typical turning point. If the similarity is ≥0.8, it indicates that the current trajectory conforms to the characteristics of a turning point.

[0056] By combining the results of local time window analysis with the probability of turning points, a comprehensive judgment rule is constructed: when Furthermore, if the trajectory change amplitude within the local time window is ≥0.3, the directional difference is ≤0.7, and the similarity to the trajectory of a typical turning point is ≥0.8, it is ultimately determined to be a turning point; if only the probability condition is met but the local time window analysis condition is not met, it is determined to be a "suspected turning point", and subsequent trajectory changes are continuously monitored to avoid misjudgment.

[0057] Finally, a multi-loss function joint optimization mechanism is constructed to comprehensively balance the three core objectives of knowledge graph alignment, trajectory representation differentiation, and inflection point classification, ensuring the stability, convergence, and accuracy of the model. The core formula is as follows: ; in, This represents the model's total loss function, used to comprehensively optimize the model's performance and ensure that the model can accurately capture the trajectory evolution patterns and inflection point characteristics.

[0058] The knowledge graph alignment loss is represented by the mean squared error (MSE), which is used to ensure the alignment accuracy of multi-source data in the coordinate system at the development stage and reduce the error caused by data heterogeneity. The core formula is: ; In the above formula, Embedded vectors representing the actual development stage B represents the predicted development stage embedding vector, and B represents the number of samples in the current training batch.

[0059] The trajectory contrast learning loss is represented by a triplet loss function, which is used to improve the discriminative power of trajectory representations, making the representations of different types of trajectories (normal, yaw, turning) significantly different. The core formula is: ; In the above formula, This represents the anchor point trajectory. Represents the characteristics of similar trajectories. Represents the characteristics of outlier trajectories. This represents the interval hyperparameter in the triplet loss; The inflection point classification loss is represented by the binary cross-entropy loss function, which optimizes the accuracy of inflection point probability determination and reduces the false positive and false negative rates. The core formula is: ; In the above formula, The true label indicates the turning point; 1 indicates a turning point, and 0 indicates a non-turning point. This represents the predicted probability that the i-th sample in the current training batch is identified as a turning point, and... They maintain the same physical meaning, both representing the probability of a turning point; the only difference lies in their indices. Indicates the batch sample index. Represents the time index on a single trajectory; This indicates the number of samples in the current training batch. , The weight coefficients for trajectory comparison loss and inflection point classification loss are respectively determined through offline sample training and cross-validation, and are set to 0.3 and 0.5 respectively to balance the influence of various losses and ensure optimal overall model performance.

[0060] S400: Construct an early warning intensity calculation model, combine it with the identification results of the turning point in the psychological growth of target users, trace the cause of the warning and link the direction of intervention, and realize early warning of psychological risks.

[0061] First, it should be noted that the goal of this embodiment is to address the existing problems of delayed early warning nodes, lagging intervention windows, and a disconnect between early warning and intervention. It constructs a fully integrated mechanism encompassing "turning point identification - early warning intensity calculation - early warning triggering - intervention suggestion matching," enabling the model output not only to provide early warnings but also to offer targeted intervention directions and actionable suggestions based on the causes of the early warnings. This upgrades the model from a simple "risk assessment tool" to a "development path decision support tool," directly serving the actual intervention scenarios in schools and families, and improving the accuracy and effectiveness of interventions. This embodiment focuses on forward-shifted early warning intensity calculation and the linkage between early warning triggering and intervention directions, while also incorporating dynamic adaptation of the intervention mechanism, handling of abnormal situations, and implementation guarantee measures, collectively forming a complete implementation chain.

[0062] S410. Combining the probability of turning points, the deviation distribution and trend accumulation of standardized psychological growth trajectories, the forward warning intensity is calculated to obtain the forward warning intensity. In some specific embodiments, a scientific forward warning intensity calculation model is constructed by integrating three core factors: the probability of turning points, the degree of trajectory deviation, and the cumulative trend term, and the output time is determined. Forward warning intensity This system is used to quantify whether the current growth path has entered a risk zone requiring early intervention, enabling precise quantification and grading of early warnings, providing a basis for determining intervention priorities, and establishing an early warning intensity calibration mechanism to ensure the accuracy and rationality of the calculation results. Specific implementation details are as follows: First, clarify the calculation methods for the three core factors to ensure that the factors accurately reflect the early warning risks. Specifically, these include: 1) Probability of turning point The range of values ​​for the probability of a turning point is: It is the core indicator for early warning, reflecting the probability that the current moment is a risk-related turning point. The higher the probability value, the stronger the urgency of the warning. Its core formula is as follows: ; in, Indicates time The trajectory representation vector, This represents the weight matrix of the inflection point discrimination layer. This indicates the bias term of the inflection point discrimination layer. This represents the Sigmoid activation function, used to map the output to the [0,1] interval.

[0063] 2) Track yaw degree : Obtained by the deviation calculation mechanism, the value range is The value is obtained by calculating the weighted sum of Euclidean distances between the current trajectory representation and the individual's historical baseline, the group reference trajectory, and the typical trajectory. The larger the distance, the more severe the trajectory deviation and the higher the risk. The core formula is as follows: ; in, This represents the current trajectory representation and the individual's historical baseline vector. Represents the group reference trajectory vector. Represents a typical trajectory vector. Let represent the weighting coefficients of the three types of reference trajectories, and satisfy . , This represents the L2 Euclidean distance.

[0064] 3) Trend Cumulative Item The value is obtained by calculating the cumulative change in warning intensity over nearly K time windows (K=7 in this implementation), with a range of values ​​of [value range missing]. This reflects the persistence and speed of the deviation trend; the larger the cumulative amount, the faster the risk deteriorates, requiring prompt intervention. Its core formula is as follows: ; in, This indicates that a positive function is used; when the value inside the parentheses is positive, it is retained, and when it is negative, it is taken as 0, ensuring that the trend accumulation term only reflects the cumulative effect of rising risk. Indicates the first The intensity of the warning within each time window; This indicates the number of time windows to look back in the trend accumulation calculation.

[0065] Secondly, a calculation model for the intensity of forward early warning is constructed, using a weighted summation method to integrate the influence of three core factors. The core formula is as follows: ; in, Indicates time The forward warning intensity, with a value range of [value missing]. The higher the warning intensity value, the higher the risk of deviation from the current psychological development trajectory of adolescents, the stronger the urgency of the warning of the turning point, and the more urgent the need to take intervention measures as soon as possible; , , These represent the weighting coefficients of the inflection point probability, trajectory deviation, and trend accumulation term, respectively, satisfying... The optimal value was determined through large-scale offline sample training and five-fold cross-validation. Combining the psychological development patterns of adolescents with actual early warning scenarios, the final value was set. (The probability of a turning point is the core early warning indicator, with the highest weight.) (The degree of trajectory deviation reflects the current risk level, with secondary weight.) (The cumulative trend reflects the speed of risk development and helps to judge the urgency of intervention), ensuring that the calculation of early warning intensity can comprehensively take into account both the accuracy and timeliness of the early warning.

[0066] Finally, a warning intensity calibration mechanism is constructed to ensure the accuracy and rationality of the calculation results and avoid warning distortion caused by data deviations and parameter fluctuations. The calibration mechanism adopts a combination of "offline calibration + online dynamic adjustment": in the offline calibration phase, based on historical warning data and intervention effect feedback, a correspondence between warning intensity and actual risk level is established, and a calibration function is set. The initial calculated warning intensity is calibrated using the following core formula: ; in, This indicates the intensity of the calibrated warning, ensuring that its value strictly falls within the range of... interval; This represents the minimum historical warning intensity. The maximum historical warning intensity is determined by statistically analyzing historical data to ensure a unified quantitative standard for the calibrated warning intensity. During the online dynamic adjustment phase, real-time data on the intervention effect after warning triggering is collected (e.g., changes in trajectory deviation after intervention, whether turning points are effectively contained, etc.). The weighting coefficients are adjusted quarterly. , , Fine-tuning will be conducted to ensure that the warning intensity is adapted to the psychological development characteristics of adolescents at different stages and in different groups, thereby improving the adaptability and accuracy of the warnings. Simultaneously, an abnormal warning intensity detection mechanism will be established. When abnormal fluctuations occur in the warning intensity (such as a sudden increase or decrease in a short period), the data verification process will be automatically triggered to investigate anomalies in the data collection and calculation process, ensuring the reliability of the warning intensity.

[0067] S420: Based on the intensity of early warning and combined with the identification results of the turning point in the psychological growth of target users, the cause of the warning is traced and the direction of intervention is linked to achieve early warning of psychological risks.

[0068] In some specific embodiments, based on the calibrated forward warning intensity A four-level early warning triggering mechanism is constructed, which clarifies the triggering thresholds, response procedures and intervention subjects for different early warning levels. At the same time, a linkage mechanism for tracing the cause of early warning and the direction of intervention is established. Targeted intervention suggestions are automatically matched for different early warning causes, different development stages and different individual characteristics, so as to realize a closed loop of the whole process of "early warning-tracing-intervention" and ensure that early warning is not disconnected and intervention is directional.

[0069] First, a four-level early warning triggering mechanism is established. Based on the severity of adolescent psychological risks and the urgency of intervention, clear early warning intensity thresholds are set. Different levels correspond to different response procedures and intervention entities to ensure the accuracy and efficiency of the early warning response. The specific grading standards are as follows: 1) Blue Alert (Low Risk): The triggering conditions are a slight deviation in the trajectory, a low probability of a risk-related inflection point, and no significant increase in the risk trend. Response process: The system automatically sends an early warning to the homeroom teacher and simultaneously pushes details of the deviation between the individual's current trajectory and the reference system; the intervention subject is the homeroom teacher, and the intervention method is mainly daily observation and light guidance (such as heart-to-heart talks, paying attention to classroom performance and peer interaction), with weekly feedback on observation results, without the need to initiate a professional intervention process; 2) Yellow Alert (Low to Medium Risk): The triggering conditions are a moderate degree of trajectory deviation, the existence of potential risk-related turning points, and a slow upward trend in risk. Response process: The system simultaneously sends early warning alerts to the homeroom teacher and school counselor, providing details of the trajectory deviation and an analysis report of potential turning points; the intervention is primarily carried out by the homeroom teacher and school counselor, and intervention methods include targeted counseling, short-term psychological counseling, and home-school communication (providing feedback to parents and guiding them to cooperate), with progress updates every 3 days and continuous monitoring of trajectory changes; 3) Orange Alert (Medium-to-High Risk): The triggering conditions are a significant deviation from the trajectory, a clearly defined risk-related turning point, and a rapidly escalating risk trend. The response process is as follows: The system simultaneously sends early warning alerts to the homeroom teacher, school psychologist, and moral education department, activates a special intervention team, and provides a detailed early warning analysis report (including the reasons for the deviation, the triggering event of the turning point, and the accumulation of risk). The intervention is carried out by the special intervention team (led by the psychologist, with the cooperation of the homeroom teacher and moral education teacher). Intervention methods include systematic psychological counseling, personalized intervention plans, and home-school collaborative intervention (regular home-school communication meetings to develop family intervention plans). Daily progress updates are provided, and weekly evaluations of intervention effectiveness are conducted. 4) Red Alert (High Risk): The triggering conditions are a significant deviation from the trajectory, the presence of a major risk-related turning point, and a sharp increase in risk trends, which may lead to a serious psychological crisis. Response process: The system immediately sends an early warning to the homeroom teacher, school counselor, moral education department, and school leaders, activating the emergency intervention mechanism. Simultaneously, it contacts professional psychological service institutions and parents, providing comprehensive early warning analysis and trajectory data. The intervention entities are professional psychological service institutions, a school-specific intervention team, and parents. Intervention methods include professional psychological treatment, emergency psychological intervention, 24-hour monitoring, and full collaboration between home and school, with real-time feedback on intervention progress until the risk is effectively controlled.

[0070] Secondly, a mechanism for tracing the causes of early warnings is established to accurately pinpoint the core reasons triggering early warnings, providing a basis for matching intervention recommendations. The tracing of early warning causes is based on evolutionary knowledge graphs and trajectory representation analysis, tracing from three core dimensions: first, the turning point dimension, tracing the existence of risk-related turning points and triggering events (such as family changes, school bullying, academic setbacks, etc.); second, the trajectory deviation dimension, tracing the core sources of trajectory deviation (such as individual historical baseline deviation, group reference trajectory deviation, or high similarity to typical negative trajectories); and third, the risk accumulation dimension, tracing the main types of risk accumulation (such as academic pressure risk, peer relationship risk, insufficient family support risk, etc.). During the tracing process, the system automatically generates an early warning cause analysis report, clarifying the core triggers, the degree of impact, and the development trend, providing scientific support for determining the direction of intervention.

[0071] Furthermore, a collaborative mechanism involving three major intervention directions will be established. Based on different risk causes, levels, and individual characteristics, differentiated intervention recommendations will be formulated for schools, families, and professionals, respectively. Implementation details, responsible parties, and collaborative requirements will be clearly defined to ensure precise matching and effective implementation of intervention plans.

[0072] Simultaneously, a closed-loop verification mechanism for intervention effects is established, using changes in trajectory deviation, the effectiveness of inflection point containment, and the magnitude of decrease in warning intensity as core indicators to determine the intervention effect. If effective, the warning level is gradually reduced; if ineffective, the plan adjustment process is automatically triggered. Relevant data is synchronously fed back to optimize the warning and intervention models, achieving a closed-loop process of "warning-intervention-verification-adjustment".

[0073] The system is equipped with dynamic adaptation and optimization strategies, matching differentiated intervention priorities according to the five adolescent development stages from childhood to early adolescence. It optimizes the logic of generating plans based on individual personality, family environment and other characteristics, marks implementation priorities, and adopts differentiated push methods and dynamic push frequencies for different intervention subjects such as homeroom teachers, psychological teachers and parents.

[0074] In addition, a full-scenario anomaly handling mechanism is constructed, a review and optimization channel for early warning misjudgment / missed judgment is established, a three-level emergency response process is formulated for sudden psychological crises, the handling requirements and responsibilities at each stage are clarified, and a review and adjustment rule for intervention failure is established. Relevant cases are used simultaneously for system optimization to comprehensively ensure the effectiveness of intervention and the mental health safety of adolescents.

[0075] In summary, through dynamic adaptation and optimization of the intervention mechanism, handling of abnormal situations, and supplementation and improvement of implementation safeguards, the entire process linkage mechanism of "turning point identification - early warning intensity calculation - early warning triggering - intervention recommendations - effect verification" has been further improved. This ensures that the mechanism can not only achieve early warning and precise intervention of turning points, but also adapt to different scenarios and deal with various abnormal situations. It has strong practicality, flexibility and operability, providing solid mechanism support for early intervention of adolescent psychological risks, truly achieving the core goal of "early detection, early identification and early intervention", and helping the long-term stable development of adolescent mental health.

[0076] In summary, the embodiments of this invention, based on the evolution and early warning model of the "one file per lifetime" knowledge graph of developmental psychology trajectory, comprehensively break through the core bottlenecks of existing "one file per lifetime" systems and psychological early warning technologies. It achieves a comprehensive upgrade from static files to dynamic trajectories and a fundamental shift from delayed early warning to forward early warning. It has significant technical and practical value and can be widely applied in the field of adolescent mental health protection, providing comprehensive and intelligent technical support for schools, families, and professional psychological service institutions, and helping to build an adolescent mental health protection system of "early detection, early identification, and early intervention".

[0077] First, this embodiment achieves a deep integration of developmental psychology theory and artificial intelligence technology, breaking down disciplinary research barriers. It deeply integrates core theories from developmental psychology, such as critical periods, turning points, and developmental stages, into the entire process of knowledge graph construction, trajectory modeling, and early warning mechanism design. This breaks through the limitations of traditional "one-person-one-file" systems, which only focus on data storage and neglect theoretical support. It upgrades "one-person-one-file" from a simple archival storage tool into an intelligent analysis tool capable of interpreting developmental significance, tracking dynamic trajectories, and predicting risk trends, thus enhancing the utilization value and theoretical connotation of archival data. This interdisciplinary integration model provides a replicable and scalable research paradigm for the deep integration of artificial intelligence technology with developmental psychology and educational psychology, promoting the collaborative development of interdisciplinary technologies.

[0078] Secondly, the core technologies proposed in this embodiment, such as evolutionary knowledge graph construction, developmental stage alignment, trajectory representation encoding, and inflection point probability determination, effectively solve key problems of existing technologies in multi-source data integration, trajectory continuity modeling, dynamic evolution characterization, and early warning timeliness. For example, the evolutionary knowledge graph solves the problem that static knowledge structures cannot describe the "evolution" and "transfer" characteristics of psychological development; the developmental stage alignment mechanism solves the problems of data fragmentation and semantic inconsistency across semesters; trajectory representation learning achieves accurate characterization of long-term trajectories; and the forward early warning mechanism solves the industry pain point of delayed intervention nodes. These technological breakthroughs enrich the application scenarios and technical paths of knowledge graph technology and deep learning technology in the field of adolescent mental health, and provide a unified technical framework and reference paradigm for subsequent related research.

[0079] Finally, the model's practicality and scalability are enhanced, making it widely applicable. The model and mechanism constructed in this embodiment adopt a modular design, allowing for flexible adaptation to the actual needs of schools in different regions, at different educational levels, and of different sizes. It can deeply integrate with existing education management platforms, psychological assessment platforms, and home-school collaboration platforms without requiring large-scale modifications to existing systems, significantly reducing implementation costs. Furthermore, the model's core algorithms and mechanisms can be transferred to psychological trajectory modeling and early warning scenarios for other groups (such as preschoolers and adolescents), demonstrating strong scalability and providing crucial support for technological research and application in related fields.

[0080] Finally, it should be noted that this embodiment is used in a practical application scenario: In a school setting, the model can be deeply integrated with existing campus management systems to build a continuously updated "one student, one file" intelligent foundation across semesters and the entire lifecycle, providing comprehensive support for school mental health work. On the one hand, it enables routine tracking and precise monitoring of students' growth trajectories, helping school counselors to promptly identify high-risk students and key intervention points, preventing psychological risks from becoming explicit and severe, and comprehensively advancing mental health work from "post-event handling" to "pre-event intervention." On the other hand, it provides data support for the precise implementation of school mental health education. By analyzing trajectory patterns at the class, grade, and stage levels, it clarifies the psychological development characteristics and risk susceptibility factors of different groups, enabling targeted mental health education courses and group counseling activities, thus improving the pertinence and effectiveness of mental health education. Simultaneously, the personalized intervention suggestions pushed by the model provide clear work guidance for homeroom teachers and mental health counselors, reducing the difficulty of their professional work and improving the accuracy and efficiency of intervention work.

[0081] In the family setting, the model, through a home-school collaboration platform, builds an efficient communication bridge between schools and families, transforming parents' vague perceptions of their children's psychological state into observable, understandable, and intervention-oriented concrete growth signals. The model pushes student growth trajectory analysis reports, early warning information, and personalized intervention suggestions to parents in an easy-to-understand and highly actionable manner. These guide parents in mastering scientific intervention methods and communication skills, strengthening family support and emotional companionship, and avoiding psychological risks caused by inappropriate educational methods or insufficient attention. For example, for early warnings of insufficient family support, it suggests parents increase time spent with their children and improve communication methods; for early warnings of academic pressure, it suggests parents adjust their expectations reasonably and help their children develop scientific learning plans. Through home-school collaborative intervention, a joint intervention force of "school-led, family-participated" is formed, creating a positive growth environment for adolescent mental health.

[0082] Furthermore, the model can provide technical support to professional mental health service institutions. Through precise trajectory analysis and early warning information, it helps professional mental health practitioners quickly grasp the psychological development trajectory, risk factors, and turning point characteristics of adolescents, enabling them to formulate personalized treatment plans and improve the efficiency and effectiveness of mental health care. Simultaneously, the large-scale growth trajectory data and intervention effect data accumulated by the model can provide data support for optimizing mental health intervention programs and formulating mental health policies, promoting the standardization and scientific development of adolescent mental health protection work.

[0083] Reference Figure 2 An early warning system for psychological risks based on developmental psychology trajectory analysis includes: The first module 201 is used to perform multi-source growth node semantic modeling and graph relationship evolution based on the target user's original profile data to obtain a psychological evolution knowledge graph. The second module 202 is used to align the developmental stages of the target user’s original growth data and to standardize and update the target user’s psychological growth trajectory, so as to obtain the developmental stage embedding vector and the updated psychological growth trajectory. The third module 203 is used to combine the psychological evolution knowledge graph and the development stage embedding vector to construct multi-source input features and input them into the trajectory representation coding model for encoding. The encoding results are combined with the updated psychological growth trajectory to determine the turning point and obtain the turning point identification results of the target user's psychological growth. Module 4, 204, is used to construct an early warning intensity calculation model. Combined with the identification results of the turning point in the psychological growth of the target user, it is used to trace the cause of the warning and link the direction of intervention to achieve early warning of psychological risks.

[0084] The content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0085] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A method for early warning of psychological risks based on developmental psychology trajectory analysis, characterized in that, Includes the following steps: Based on the target user’s original profile data, semantic modeling of multi-source growth nodes and graph relationship evolution are performed to obtain a psychological evolution knowledge graph. The original growth data of the target users are aligned with development stages and the psychological growth trajectory of the target users is standardized and updated to obtain the development stage embedding vector and the updated psychological growth trajectory. By combining a psychological evolution knowledge graph and developmental stage embedding vectors to construct multi-source input features, these features are input into a trajectory representation coding model for encoding. The encoding results are then combined with the updated psychological growth trajectory to determine turning points, thus obtaining the turning point identification results for the target user's psychological growth. A model for calculating the intensity of early warning is constructed. By combining the identification results of turning points in the psychological growth of target users, the causes of early warning are traced and the direction of intervention is linked to achieve early warning of psychological risks.

2. The method according to claim 1, wherein the psychological risk early warning based on the developmental psychology trajectory analysis is characterized in that, The step of performing multi-source growth node semantic modeling and graph relationship evolution based on the target user's original profile data to obtain a psychological evolution knowledge graph specifically includes: The original profile data of the target users is used as the basic node, developmental psychology nodes are constructed based on psychological theories, and stage and risk nodes are constructed based on the age stage data of the target users. The basic nodes, developmental psychology nodes, and stage and risk nodes are sequentially subjected to semantic annotation and standardization, as well as node quality assessment, to construct the initial nodes at each level. Determine the initial association logic and mapping rules between nodes at each level, introduce a pre-set graph dynamic update mechanism, perform joint modeling of relationship type encoding and timestamps, and construct the initial graph relationships; By introducing semantic consistency verification and developmental psychology theory verification mechanisms, anomaly detection is performed on the initial graph relationships to obtain a psychological evolution knowledge graph.

3. The early warning method for psychological risk based on developmental psychology trajectory analysis according to claim 1, characterized in that, The step of aligning the target user's original growth data with developmental stages and standardizing and updating the target user's psychological growth trajectory to obtain the developmental stage embedding vector and the updated psychological growth trajectory specifically includes: Based on the stage-based classification criteria for psychological development, and combined with the educational stage divisions and grade distribution, a developmental stage coordinate system is constructed. The original growth data of the target user is obtained and semantically mapped, time-calibrated and cleaned in sequence. The data is then mapped to the corresponding development stage in the development stage coordinate system to obtain the initial development stage embedding vector. A dual verification mechanism for alignment accuracy at each stage is constructed, which compares and calculates the initial development stage embedding vectors and evaluates their semantic similarity to obtain the development stage embedding vectors. The psychological growth trajectory of the target users is determined and standardized using the Z-score standardization method to obtain the standardized psychological growth trajectory; A joint reference mechanism is constructed, which includes an individual's own historical baseline, a group reference trajectory at the same developmental stage, and a typical trajectory of key turning points; The deviation distribution of the standardized psychological growth trajectory is calculated by a joint reference mechanism, and the standardized psychological growth trajectory is updated based on the deviation distribution to obtain the updated psychological growth trajectory.

4. The early warning method for psychological risk based on developmental psychology trajectory analysis according to claim 3, characterized in that, The step of calculating the deviation distribution of standardized psychological growth trajectories through a joint reference mechanism specifically includes: Obtain the target trajectory deviation distance at time t. The target trajectory deviation distance includes the Euclidean distance between the individual's current trajectory and its own historical baseline, the Euclidean distance with the reference trajectory of the same group at the same stage, and the Euclidean distance with the corresponding typical trajectory. At the same time, determine the deviation weight coefficient corresponding to the target trajectory deviation distance. Multiply the target trajectory deviation distance by its corresponding weight coefficient to obtain several weighted deviation values; The weighted deviation values ​​are summed together, and the summation result is the total deviation of the individual trajectory at time t.

5. The early warning method for psychological risk based on developmental psychology trajectory analysis according to claim 1, characterized in that, The step of constructing multi-source input features by combining psychological evolution knowledge graphs and developmental stage embedding vectors, inputting them into a trajectory representation coding model for encoding, and then combining the encoding results with the updated psychological growth trajectory to determine turning points and obtain the turning point identification results of the target user's psychological growth specifically includes: Construct multi-source input features, which include a psychological evolution knowledge graph, developmental stage embedding vectors, and knowledge graph relation evolution features. The knowledge graph relation evolution features are obtained by extracting the relation evolution sequence before time t in the psychological evolution knowledge graph. A trajectory representation coding model based on bidirectional gated recurrent unit (BiGRU) is constructed to perform deep fusion and temporal coding of multi-source input features to obtain a unified trajectory representation vector. Based on a fully connected neural network, a unified trajectory representation vector is input to determine the probability of turning points and obtain the probability of turning points. A local time window analysis method is introduced to assist in the determination of turning points in the updated psychological growth trajectory, and the results of the local time window analysis are obtained. A multi-loss function joint optimization mechanism is constructed, and the results of local time window analysis and inflection point probability are combined for optimization to obtain the inflection point identification results of the target user's psychological growth.

6. The early warning method for psychological risk based on developmental psychology trajectory analysis according to claim 5, characterized in that, The step of constructing a trajectory representation coding model based on a bidirectional gated recurrent unit (BiGRU) to perform deep fusion and temporal coding of multi-source input features to obtain a unified trajectory representation vector specifically includes: Construct a trajectory representation coding model based on a bidirectional gated cyclic unit (BiGRU); The input layer of the trajectory representation coding model based on the bidirectional gated recurrent unit (BiGRU) is used to obtain multi-source input features. The BiGRU trajectory coding layer, based on the trajectory representation coding model of BiGRU bidirectional gated recurrent unit, performs deep temporal coding of long-period trajectory information on multi-source input features to obtain trajectory representation vectors. The temporal attention mechanism of the trajectory representation encoding model based on the bidirectional gated recurrent unit (BiGRU) dynamically allocates attention weights to adjust the trajectory representation vector, thereby obtaining a unified trajectory representation vector.

7. The early warning method for psychological risk based on developmental psychology trajectory analysis according to claim 6, characterized in that, The step of determining the probability of the turning point specifically includes: Obtain the unified trajectory representation vector at time t, as well as the predefined turning point discrimination layer weight matrix and bias terms; To complete the linear operation, multiply the weight matrix by the trajectory representation vector at time t, and add the bias term to obtain the comprehensive value after the linear transformation. The final probability is mapped out by inputting the linearly transformed comprehensive value into the Sigmoid activation function for numerical mapping, and the output result is the inflection point probability at time t.

8. The early warning method for psychological risk based on developmental psychology trajectory analysis according to claim 6, characterized in that, The loss functions of the multi-loss function joint optimization mechanism specifically include the knowledge graph alignment loss function, the trajectory comparison learning loss function, and the inflection point classification loss function.

9. The early warning method for psychological risk based on developmental psychology trajectory analysis according to claim 5, characterized in that, The step of constructing an early warning intensity calculation model, combined with the identification results of turning points in the psychological development of target users, to trace the causes of early warning and link intervention directions, and to achieve early warning of psychological risks, specifically includes: By combining the probability of turning points, the deviation distribution of standardized psychological growth trajectories, and the cumulative trend term, the forward warning intensity is calculated to obtain the forward warning intensity. Based on the intensity of early warning and combined with the identification results of turning points in the psychological growth of target users, the causes of early warning are traced and the direction of intervention is linked to achieve early warning of psychological risks.

10. An early warning system for psychological risks based on developmental psychology trajectory analysis, characterized in that, Includes the following modules: The first module is used to perform multi-source growth node semantic modeling and graph relationship evolution based on the target user's original profile data to obtain a psychological evolution knowledge graph. The second module is used to align the developmental stages of the target user’s original growth data and to standardize and update the target user’s psychological growth trajectory, so as to obtain the developmental stage embedding vector and the updated psychological growth trajectory. The third module is used to construct multi-source input features by combining the psychological evolution knowledge graph and development stage embedding vectors, and input them into the trajectory representation coding model for encoding. The encoding results are combined with the updated psychological growth trajectory to determine the turning point and obtain the turning point identification results of the target user's psychological growth. The fourth module is used to construct an early warning intensity calculation model, which, combined with the identification results of the turning point in the psychological growth of the target user, traces the cause of the warning and links the direction of intervention to achieve early warning of psychological risks.

Images