An artificial intelligence-based method and system for measuring cognitive ability and psychological development of teenagers

Abstract

The application discloses a kind of based on artificial intelligence's adolescent cognitive ability and psychological development evaluation method and system, it is related to adolescent cognitive ability and psychological development evaluation technical field, the neural differential equation of the present application will discrete observation be mapped into continuous latent variable dynamics system, effectively capture the progressive development and stepwise transition of cognitive ability, avoid the misjudgment of traditional time series model to non-continuous change;Based on the persistent homology feature monitoring mechanism of topological data analysis, accurately identify the cognitive representation dimension mutation caused by strategy reorganization and neural plasticity, break through the delay bottleneck of traditional method to burst mode identification;Incremental learning trigger and evaluation path optimization module collaborative operation, while reducing the computing overhead, realize the online evolution of model parameters, maintain the continuity and consistency of long-term tracking.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of adolescent cognitive ability and psychological development evaluation, and particularly relates to an adolescent cognitive ability and psychological development evaluation method and system based on artificial intelligence. BACKGROUND

[0002] Current adolescent cognitive and psychological development evaluation is gradually shifting from single static assessment to long-term dynamic tracking, relying on wearable devices and multi-modal data fusion technology to build a digital twin model of individual cognitive development. The new generation of systems generally use adaptive evaluation framework, dynamically adjust task difficulty through reinforcement learning, and introduce federated learning to realize cross-institutional data sharing. This kind of solution performs outstandingly in capturing short-term cognitive fluctuations, and can identify transient state changes such as temporary executive function decline caused by test pressure.

[0003] Adolescent cognitive strategies will experience non-continuous transitions due to neural development and social experience accumulation, but mainstream time series models (such as LSTM, Transformer) assume that cognitive feature changes conform to a stationary random process, resulting in systematic bias in long-term tracking. Typical problems include: misjudging the improvement of impulse control caused by prefrontal cortex maturation as the effect of intervention measures; a 3-6 month recognition delay in cognitive pattern reconstruction triggered by dramatic changes in peer relationships; inability to distinguish between temporary strategy adjustment caused by hormone level fluctuations and permanent cognitive structure changes; current solutions rely on regular full-model retraining, but face high computational cost and interrupt continuous tracking process bottlenecks.

[0004] Frontier research attempts to build continuous cognitive evolution models using neural differential equations, mapping discrete observation data to potential dynamic systems; for example, the NeuralCDE framework captures the differential characteristics of development trajectories by interpolating control nodes, reducing the false negative rate of sudden pattern changes; another team uses online Bayesian knowledge transfer technology to ensure memory of historical patterns while achieving gradual adaptation of cognitive mutations through weight space flexibility; but it has not yet solved the problem of cognitive representation dimension mutation caused by neural plasticity, and still relies on experts to perform model dimension expansion operations regularly. SUMMARY

[0005] In view of the above existing problems, the present application is proposed.

[0006] The present application provides an adolescent cognitive ability and psychological development evaluation method and system based on artificial intelligence to solve the problem that existing solutions have shifted to dynamic tracking, but are limited by the stationarity assumption and static modeling, making it difficult to cope with non-continuous transitions in adolescent cognitive strategies, resulting in misjudgment and delay.

[0007] To solve the above technical problems, the present application provides the following technical solutions:

[0008] In a first aspect, the embodiments of the present application provide a method for measuring cognitive ability and psychological development of teenagers based on artificial intelligence, which comprises,

[0009] In step S1, physiological signals and behavior operation sequences of a subject are continuously collected through a skin electric response sensor of a wearable device and a touch screen interaction module;

[0010] In step S2, a cognitive evolution continuous model is constructed by using a neural differential equation, and multi-modal observation data at discrete time points are mapped to a potential dynamic system;

[0011] In step S3, a phase mutation event of a cognitive strategy is detected based on a topological data analysis method, and when a persistent homology feature dimension change is identified, a model incremental learning mechanism is triggered; the persistent homology feature dimension change detection specifically refers to constructing a Vietoris-Rips complex in a cognitive feature manifold space, calculating the length distribution of each dimension persistent interval, and determining a phase mutation when a 1D ring structure persistent interval suddenly increases;

[0012] In step S4, a cognitive development digital twin containing a time stamp is generated, and a dimension space and an evolution path of individual cognitive features are dynamically updated.

[0013] As a preferred scheme of the method for measuring cognitive ability and psychological development of teenagers based on artificial intelligence, in step S1, data processing is performed, and the step includes:

[0014] The original physiological signals are subjected to complex domain time-frequency decomposition to separate environmental noise and micro-motion feature components;

[0015] A multi-scale time alignment module is constructed to calibrate the millisecond-level delay between behavior operation events and physiological responses;

[0016] An adversarial training framework is used to eliminate the interference of culture-specific behavior patterns on the representation of underlying cognitive features;

[0017] The adversarial training framework includes a cascade structure of a gradient reversal layer and an attribute discriminator, wherein the discriminator distinguishes between eastern and western cultural background samples through a binary classification loss function, and the generator optimizes feature extraction by maximizing the confusion rate of the discriminator.

[0018] As a preferred scheme of the method for measuring cognitive ability and psychological development of teenagers based on artificial intelligence, in step S1, let the original multi-modal input be x, and the feature extractor be mapped to z=F(x; θ1),

[0019] The difference between eastern and western sample distributions is estimated by a Wasserstein distance at the discriminator end, and the discriminator loss is defined as:

[0020]

[0021] wherein, E x～P1 [·] represents the expectation operation on the random variable under the distribution P1, E x～P2 [·] represents the expectation operation on the random variable under the distribution P2, represents the expectation operation on the interpolated sample , θ1 represents the feature extractor parameter, θ2 represents the discriminator parameter, P1 represents the western culture sample distribution, P2 represents the eastern culture sample distribution, F(·; θ1) represents the feature extraction mapping, D(·; θ2) represents the attribute discrimination mapping, λ1 represents the gradient penalty coefficient, represents the interpolated sample input to the discriminator, represents the gradient operator input to the discriminator, |·|2 represents the two norm;

[0022] wherein the interpolated sample The calculation formula is:

[0023]

[0024] wherein, ∈1 represents the interpolation coefficient, which is subject to a uniform distribution z W represents the western sample feature, z E represents the eastern sample feature;

[0025] At the feature extractor end, the cognitive task classification and the culture confrontation are simultaneously included in the total loss:

[0026] L F = L task + γ1(E x～P2 [D(F(x; θ1); θ2)]-E x～P1 [D(F(x; θ1); θ2)]),

[0027] wherein, L task represents the cognitive task classification loss, and γ1 represents the confrontation loss weight;

[0028] The cognitive task classification loss is defined as:

[0029] L task = E (x,y) [l(C(F(x; θ1)), y)],

[0030] wherein, (x, y) represents the sample and its label, C(·) represents the cognitive task classifier, and l(·, ·) represents the cross-entropy loss function.

[0031] As a preferred scheme of the method for evaluating the cognitive ability and psychological development of teenagers based on artificial intelligence, the cognitive evolution continuous model is constructed by the following method:

[0032] Encode historical cognitive assessment data as a sequence of continuous-time latent variables;

[0033] Model the dynamical process of latent variables with neural development using controlled differential equations;

[0034] Introduce counterfactual reasoning branch to distinguish endogenous cognitive transition from exogenous intervention influence path;

[0035] The neural differential equation adopts a controlled differential equation form, wherein a control term is generated by an interpolation function of a historical cognitive feature trajectory, a differential operator is built by a multilayer perception mechanism, and a time step is dynamically adjusted according to a test frequency of a subject.

[0036] As a preferred scheme of the artificial intelligence-based adolescent cognitive ability and psychological development evaluation method, in step S2, the neural differential equation is used to construct a cognitive evolution continuous model, and the step includes:

[0037] Let the initial latent variable state be mapped by an encoder:

[0038]

[0039] Wherein z0 represents the initial latent variable, represents a multi-modal observation at the initial time t0, g represents an initial encoding mapping, and β1 represents an encoding mapping parameter;

[0040] The continuous evolution of the latent variable with time is described using a NeuralCDE standard form, and is represented as:

[0041]

[0042] Wherein z(t) represents the latent variable at time t, f represents a neural network differential function, α1 represents a parameter of the function, U(s) represents a control path constructed from discrete observations, and s is an integral variable;

[0043] The control path U(t) is defined by an interpolation basis function, and is represented as:

[0044]

[0045] Wherein N1 represents the total number of observation points, represents the i-th observation input, φ i (t) represents the corresponding interpolation basis function.

[0046] As a preferred scheme of the artificial intelligence-based adolescent cognitive ability and psychological development evaluation method, in step S3, the phase mutation event of the cognitive strategy is detected in the following manner:

[0047] In the cognitive feature manifold space, a Vietoris-Rips complex is constructed, and the k-dimensional persistent diagram entropy is calculated, and the formula is:

[0048]

[0049] Wherein, H (k) (t) represents the k-dimensional persistent entropy at time t, N k (t) represents the number of k-dimensional persistent intervals, k represents the dimension index of the homology group, represents the length of the i-th interval, represents the corresponding normalized weight;

[0050] For one-dimensional ring k=1, the change of its persistent entropy with time is monitored, and the average persistent entropy is calculated by sliding time window:

[0051]

[0052] Wherein, τ1 represents the length of the time window, if And the persistence, δ1 represents the persistent entropy threshold for determining the phase mutation, and the phase mutation is determined and the model incremental learning mechanism is triggered.

[0053] In the second aspect, the application provides a youth cognitive ability and psychological development evaluation system based on artificial intelligence, comprising,

[0054] A multi-modal data acquisition module integrates a flexible electronic skin sensor and an adaptive touch interface.

[0055] An edge computing unit deploys a lightweight time sequence feature extraction network and a privacy protection mechanism.

[0056] A dynamic evaluation path generator adjusts the task type and difficulty parameter according to the real-time cognitive state.

[0057] A cognitive development visualization engine generates a three-dimensional spatiotemporal evolution map and an intervention suggestion scheme.

[0058] As a preferred scheme of the youth cognitive ability and psychological development evaluation system based on artificial intelligence, the privacy protection mechanism is specifically implemented as:

[0059] A federated learning architecture is used for distributed training of the model, and the original data is retained in the local terminal.

[0060] Controllable noise is added to the cognitive feature vector through differential privacy technology.

[0061] Sensitive intermediate data in the fingerprint matching process is transmitted using a homomorphic encryption transmission strategy.

[0062] As a preferred scheme of the artificial intelligence-based adolescent cognitive ability and psychological development evaluation system, the dynamic evaluation path generator further comprises:

[0063] The strategy fingerprint feature library stores a feature migration degree matrix of a typical cognitive development mode; the feature migration degree matrix is constructed by calculating the cosine similarity variance of cross-task operation features, and contains a 52-dimensional migration relationship graph among three types of tasks, i.e., risk decision, pattern recognition and social simulation;

[0064] The online Bayesian optimizer dynamically selects an evaluation dimension according to a current cognitive space sparsity;

[0065] The sudden jump event response module inserts a cross-domain verification task when a phase mutation is detected.

[0066] As a preferred scheme of the artificial intelligence-based adolescent cognitive ability and psychological development evaluation system, the visual engine comprises:

[0067] The cultural adaptability calibration layer eliminates the deviation of the self-determination score caused by the collectivism decision mode;

[0068] The development trajectory prediction interface simulates cognitive evolution branch paths under different intervention measures;

[0069] The implicit risk early warning unit detects early psychological development deviation based on strategy migration degree anomaly.

[0070] The neural differential equation of the present application maps discrete observations into a continuous latent variable dynamics system, effectively capturing the gradual development and stepwise transition of cognitive ability, avoiding the misjudgment of non-continuous changes by traditional time series models; the persistent homology feature monitoring mechanism based on topological data analysis accurately identifies the cognitive representation dimension mutation caused by strategy reorganization and neural plasticity, breaking through the delay bottleneck of traditional methods for sudden pattern recognition; the incremental learning trigger and evaluation path optimization module work cooperatively to reduce the computational overhead while realizing the online evolution of model parameters, maintaining the continuity and consistency of long-term tracking; the adversarial training framework and the cultural calibration layer deeply decouple regional behavior preferences, eliminating the systematic deviation of the self-determination evaluation caused by the collectivism decision mode; the federated learning and homomorphic encryption technology support cross-institutional cognitive development pattern mining without revealing the original data, solving the data island problem. BRIEF DESCRIPTION OF DRAWINGS

[0071] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0072] Figure 1 This is a flowchart illustrating the AI-based assessment method for adolescent cognitive abilities and psychological development in Example 1.

[0073] Figure 2 This is a schematic diagram of the framework of the AI-based adolescent cognitive ability and psychological development assessment system in Example 1. Detailed Implementation

[0074] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0075] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0076] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0077] Example 1, referring to Figure 1 and Figure 2 This embodiment provides an artificial intelligence-based method for assessing the cognitive abilities and psychological development of adolescents, including the following steps:

[0078] Step S1: The physiological signals and behavioral operation sequences of the subject are continuously collected through the skin conductance response sensor and touch screen interaction module of the wearable device;

[0079] Data processing is performed in step S1, including the following steps:

[0080] Complex domain time-frequency decomposition is performed on the original physiological signal to separate environmental noise and micro-motion feature components;

[0081] A multi-scale time alignment module was constructed to calibrate the millisecond-level delay between behavioral events and physiological responses.

[0082] The interference of culture-specific behavior patterns on the representation of underlying cognitive features is eliminated using an adversarial training framework;

[0083] The adversarial training framework includes a cascade structure of a gradient reversal layer and an attribute discriminator, wherein the discriminator distinguishes between samples of Eastern and Western cultural backgrounds through a binary classification loss function, and the generator optimizes feature extraction by maximizing the confusion rate of the discriminator;

[0084] In step S1, let the original multi-modal input be denoted as x, and the feature extractor be mapped as z=F(x; θ1),

[0085] The difference between the distributions of Eastern and Western samples is estimated by the Wasserstein distance at the discriminator end, and the discriminator loss is defined as:

[0086]

[0087] wherein E x～P1 [·] represents the expectation operation of a random variable under distribution P1, E x～P2 [·] represents the expectation operation of a random variable under distribution P2, represents the expectation operation of the interpolated sample , θ1 represents the feature extractor parameter, θ2 represents the discriminator parameter, P1 represents the Western cultural sample distribution, P2 represents the Eastern cultural sample distribution, F(·; θ1) represents the feature extraction mapping, D(·; θ2) represents the attribute discrimination mapping, λ1 represents the gradient penalty coefficient, represents the interpolated sample input to the discriminator, represents the gradient operator input to the discriminator, and |·|2 represents the two-norm;

[0088] wherein the interpolated sample is calculated according to the following formula:

[0089]

[0090] wherein ∈1 represents the interpolation coefficient, which is subject to a uniform distribution z W represents the Western sample feature, and z E represents the Eastern sample feature;

[0091] At the feature extractor end, the cognitive task classification and cultural confrontation are simultaneously included in the total loss:

[0092] L F = L task + γ1(E x～P2 [D(F(x; θ1); θ2)]-E x～P1 [D(F(x; θ1); θ2)]),

[0093] wherein L taskdenotes the cognitive task classification loss, and γ1 denotes the adversarial loss weight;

[0094] The cognitive task classification loss is defined as:

[0095] L task =E (x,y) [l(C(F(x;θ1)),y)],

[0096] where (x, y) denotes a sample and its label, C(·) denotes a cognitive task classifier, and l(·, ·) denotes a cross-entropy loss function;

[0097] Specifically, the adversarial training framework measures the Wasserstein distance of the East-West cultural sample distribution, explicitly constrains the feature extractor, estimates the distribution distance on the discriminator side, and improves the Lipschitz continuity of the discriminant function combined with the gradient penalty, thereby avoiding training instability and gradient disappearance problem. The adversarial confusion loss is introduced on the extractor side, which effectively eliminates the interference of cultural specific behavior patterns on cognitive representation, while retaining the cognitive task classification loss to ensure that the extracted features still contain information related to cognitive ability;

[0098] Step S2, a neural differential equation is used to construct a cognitive evolution continuous model to map multi-modal observation data at discrete time points to a latent dynamic system;

[0099] The cognitive evolution continuous model is constructed in the following way:

[0100] The historical cognitive assessment data is encoded as a continuous-time latent variable sequence;

[0101] A controlled differential equation is used to model the dynamics of the latent variable with neural development;

[0102] A counterfactual reasoning branch is introduced to distinguish endogenous cognitive transition from exogenous intervention influence path;

[0103] The neural differential equation adopts a controlled differential equation form, wherein the control term is generated by an interpolation function of the historical cognitive feature trajectory, the differential operator is constructed by a multilayer perception mechanism, and the time step is dynamically adjusted according to the test frequency of the subject;

[0104] In step S2, a neural differential equation is used to construct a cognitive evolution continuous model, and the steps include:

[0105] Let the initial latent variable state be mapped by an encoder:

[0106]

[0107] where z0 denotes the initial latent variable, The multi-modal observation at initial time t0, g represents the initial encoding mapping, and β1 represents the encoding mapping parameter;

[0108] The continuous evolution of the latent variable over time is described by the NeuralCDE standard formula, denoted as:

[0109]

[0110] where z(t) represents the latent variable at time t, f represents the neural network differential function, α1 represents the function parameter, U(s) represents the control path constructed by discrete observations, and s is the integral variable;

[0111] The control path U(t) is defined by the interpolation basis function, denoted as:

[0112]

[0113] where N1 represents the total number of observation points, represents the i-th observation input, φ i (t) represents the corresponding interpolation basis function;

[0114] Specifically, the neural CDE model maps discrete multi-modal observations to continuous latent variable trajectories, constructs a control path through an initial encoding and an interpolation function, enabling the model to capture subtle dynamics of cognitive evolution. The parameters α1 of the function f learn the underlying dynamics in the neural network, and together with the control term U, form a controlled differential equation that ensures adaptive modeling of irregular evaluation frequencies. This method uses integral operations to accumulate historical events, providing smooth evolution curves that facilitate subsequent intervention effect analysis of counterfactual reasoning branches. The interpolation basis function φ i ensures the continuity of the signal and reduces information loss caused by downsampling. The Adjoint method is used to calculate the gradient, reducing computational overhead while ensuring model trainability.

[0115] Step S3, detect the phase mutation event of the cognitive strategy based on the topological data analysis method. When the persistent homology feature dimension change is identified, trigger the model incremental learning mechanism. The persistent homology feature dimension change detection specifically refers to constructing a Vietoris-Rips complex in the cognitive feature manifold space, calculating the length distribution of each dimension persistence interval, and determining a phase mutation when the 1-dimensional ring structure persistence interval suddenly increases.

[0116] The way to detect the phase mutation event of the cognitive strategy in step S3 is:

[0117] A Vietoris-Rips complex is constructed in the cognitive feature manifold space, and the k-dimensional persistent diagram entropy is calculated, with the formula:

[0118]

[0119] where H (k) (t) denotes the kth persistent entropy at time t, N k (t) denotes the kth persistent interval number, k denotes the dimension index of the homology group, denotes the length of the ith interval, denotes the corresponding normalized weight;

[0120] For one-dimensional loop k = 1, the change of its persistent entropy with time is monitored, and the average persistent entropy is calculated by a sliding time window:

[0121]

[0122] where τ1 denotes the length of the time window, if and persists, δ1 denotes the persistent entropy threshold for determining phase mutation, then it is determined that there is a phase mutation and the model incremental learning mechanism is triggered;

[0123] Specifically, the persistent graph entropy value measurement combines VR complex and entropy calculation to provide clear quantification of the change of high-dimensional homology feature persistence, and a one-dimensional loop structure is selected as the key homology dimension, the length of its persistent interval directly reflects the stability of strategy synergy and cyclic structure, and the entropy value considers the interval number and length distribution, which is more robust than simple counting. The average entropy is monitored on a sliding window, which can reduce false judgments caused by sudden noise. When the average persistent entropy exceeds the threshold δ1 and the duration τ1 meets the standard, it is considered that the cognitive strategy enters a new phase, and the incremental learning mechanism updates the feature extraction and prediction module online, avoiding complete retraining to reduce time delay;

[0124] Step S4, generating a cognitive development digital twin containing a timestamp, dynamically updating the dimension space and evolution path of individual cognitive features.

[0125] The embodiment also provides an evaluation system of the above-mentioned artificial intelligence-based adolescent cognitive ability and psychological development evaluation method, comprising:

[0126] A multi-modal data acquisition module integrates a flexible electronic skin sensor and an adaptive touch interface;

[0127] An edge computing unit deploys a lightweight time sequence feature extraction network and a privacy protection mechanism;

[0128] The privacy protection mechanism is specifically implemented as:

[0129] A federated learning architecture is used for distributed training of the model, and the original data is retained in the local terminal;

[0130] A controllable noise is added to the cognitive feature vector through differential privacy technology;

[0131] Transmitting sensitive intermediate data in a policy fingerprint matching process using homomorphic encryption

[0132] A dynamic assessment path generator adjusts task type and difficulty parameters according to real-time cognitive state;

[0133] The dynamic assessment path generator further comprises:

[0134] A policy fingerprint feature library stores a feature transfer degree matrix of a typical cognitive development mode; the feature transfer degree matrix is constructed by calculating a cosine similarity variance of cross-task operation features, and contains a 52-dimensional transfer relationship graph among three types of tasks, i.e., risk decision, pattern recognition, and social simulation;

[0135] An online Bayesian optimizer dynamically selects an assessment dimension according to current cognitive space sparsity;

[0136] A sudden jump event response module inserts a cross-domain verification task when a phase mutation is detected;

[0137] The policy fingerprint feature library is used to store a feature transfer degree matrix among tasks in a typical cognitive development mode;

[0138] Let the total number of tasks be N, and the normalized fingerprint feature vector of the ith task be Construct a third-order tensor:

[0139]

[0140] wherein, represents a feature tensor, represents a tensor product value of task i and task j in the kth dimension;

[0141] Calculate the tensor product mean μ of the pair ij :

[0142]

[0143] Finally, define the feature transfer degree matrix M ∈ R N×N element:

[0144]

[0145] wherein, N represents the total number of tasks, d represents the dimension of the fingerprint feature vector, represents a normalized feature vector of task i, i, j ∈ {1,..., N} represents a task index, k ∈ {1,..., d} represents a feature dimension index, μ ij represents a tensor product mean of the task pair (i, j), and M ij represents a feature transfer degree value from task i to task j;

[0146] Specifically, the transfer degree matrix defines the full use of tensor product operation, refines the interaction of task fingerprint feature vectors in each dimension into tensor elements, and captures the consistency and difference of cross-task features through variance calculation. Compared with directly calculating the variance of cosine similarity, this method retains the intrinsic mutual information of features more comprehensively through high-dimensional tensor structure, improves the accuracy and robustness of transfer relationship modeling, and ensures the comparability of measurement between different tasks through normalized vectors, while the variance measurement effectively highlights the collaborative changes of features in the dynamic evolution stage.

[0147] The cognitive development visualization engine generates a three-dimensional spatiotemporal evolution atlas and an intervention proposal scheme.

[0148] The visualization engine includes:

[0149] The cultural adaptability calibration layer eliminates the deviation of the self-determination score caused by the collectivism decision-making mode.

[0150] The development trajectory prediction interface simulates the cognitive evolution branch path under different intervention measures.

[0151] The implicit risk early warning unit detects the deviation of early psychological development based on the strategy transfer degree.

[0152] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the present application, and they should be covered in the scope of the claims of the present application.

Claims

1. A method for assessing adolescent cognitive abilities and psychological development based on artificial intelligence, characterized in that, This includes step S1, where the physiological signals and behavioral sequences of the subject are continuously collected through the skin conductance sensor and touch screen interaction module of the wearable device; Step S2: A continuous model of cognitive evolution is constructed using neural differential equations, mapping multimodal observation data at discrete time points to the underlying dynamic system; Step S3: Detect phase abrupt events of cognitive strategies based on topological data analysis methods. When a change in the dimension of the continuous coherent feature is detected, the incremental learning mechanism of the model is triggered. The detection of the change in the dimension of the continuous coherent feature specifically refers to: constructing a Vietoris-Rips complex in the cognitive feature manifold space, calculating the length distribution of the continuous intervals of each dimension, and determining a phase abrupt event when the continuous interval of the 1-dimensional ring structure suddenly increases. Step S4: Generate a digital twin of cognitive development containing timestamps, and dynamically update the dimensional space and evolutionary path of individual cognitive characteristics; In step S2, a continuous model of cognitive evolution is constructed using neural differential equations. The steps include: Let the initial latent variable state be mapped by the encoder: ， in, Represents the initial latent variable. Indicates the initial time Multimodal observation, Indicates the initial encoding mapping, Indicates the encoding mapping parameters; The continuous evolution of latent variables over time can be described using the standard Neural CDE formula, expressed as: ， in, Indicates time Latent variables, Represents the differential function of a neural network. This represents the function parameter. This represents the control path constructed from discrete observations. For integration variables; The control path is defined by the interpolation basis function. , is represented as: ， in, This represents the total number of observation points. Indicates the first One observation input, Indicates the corresponding interpolation basis function; The method for detecting phase abrupt events in cognitive strategies in step S3 is as follows: Construct a Vietoris-Rips complex in the cognitive feature manifold space and compute the th... The entropy of a persistent graph in a given dimension is given by the formula: ， ， in, Indicates time The Maintaining long entropy Indicates the first Number of continuous intervals The dimension index representing the homology group. Indicates the first The length of each interval This indicates the corresponding normalized weight; For one-dimensional rings Monitor the change of its persistent entropy over time, and calculate the average persistent entropy using a sliding time window: ， in, Indicates the length of the time window, if And it continues. The persistent entropy threshold represents the threshold for determining a phase mutation. If a phase mutation is detected, the incremental learning mechanism of the model is triggered.

2. The method for assessing adolescent cognitive ability and psychological development based on artificial intelligence as described in claim 1, characterized in that, Data processing is performed in step S1, including the following steps: Complex domain time-frequency decomposition is performed on the original physiological signal to separate environmental noise and micro-motion feature components; A multi-scale time alignment module was constructed to calibrate the millisecond-level delay between behavioral events and physiological responses. Using an adversarial training framework to eliminate the interference of culture-specific behavioral patterns on the representation of underlying cognitive features; The adversarial training framework includes a cascaded structure of a gradient inversion layer and an attribute discriminator, wherein the discriminator distinguishes between samples from Eastern and Western cultural backgrounds using a binary classification loss function, and the generator optimizes feature extraction by maximizing the discriminator's confusion rate.

3. The method for assessing adolescent cognitive ability and psychological development based on artificial intelligence as described in claim 2, characterized in that, In step S1, let the original multimodal input be denoted as The feature extractor is mapped to , At the discriminator end, the difference in the distribution of samples between the East and West is estimated using Wasserstein distance, and the discriminator loss is defined as follows: ， in, This indicates that the random variable follows a distribution. The expected operation below, This indicates that the random variable follows a distribution. The expected operation below, Indicates the interpolated sample Expectation operation, Indicates the feature extractor parameters. Indicates the discriminator parameters, Indicating the distribution of Western cultural samples, This indicates the distribution of samples from Eastern cultures. Represents feature extraction mapping, This indicates attribute discrimination mapping. Represents the gradient penalty coefficient. This represents the interpolated sample input to the discriminator. This represents the gradient operator for the discriminator input. Represents the L2 norm; Interpolation samples The calculation formula is: ， in, The interpolation coefficients are represented by a uniform distribution. Indicating Western sample characteristics, Indicates the characteristics of the Eastern samples; At the feature extractor, cognitive task classification and cultural adversarial approaches are simultaneously incorporated into the total loss: ， in, This represents the loss in cognitive task classification. Indicates the weight of the adversarial loss; The loss for cognitive task classification is defined as: ， in, Indicates the sample and its label. Represents a cognitive task classifier. This represents the cross-entropy loss function.

4. The method for assessing adolescent cognitive ability and psychological development based on artificial intelligence as described in claim 1, characterized in that, The continuous model of cognitive evolution is constructed in the following way: Encode historical cognitive assessment data into a continuous time latent variable sequence; Controlled differential equations are used to model the dynamics of latent variables as neural development progresses; A counterfactual reasoning branch is introduced to distinguish between endogenous cognitive leaps and the influence paths of exogenous interventions; The neural differential equations are in the form of controlled differential equations, where the control terms are generated by the interpolation function of historical cognitive feature trajectories, the differential operators are constructed by multilayer perceptrons, and the time step is dynamically adjusted according to the subject's assessment frequency.

5. An artificial intelligence-based assessment system for adolescent cognitive abilities and psychological development, based on the artificial intelligence-based assessment method for adolescent cognitive abilities and psychological development as described in any one of claims 1 to 4, characterized in that, include: The multimodal data acquisition module integrates a flexible electronic skin sensor and an adaptive touch interface; Edge computing units are equipped with lightweight temporal feature extraction networks and privacy protection mechanisms. A dynamic assessment path generator that adjusts task type and difficulty parameters based on real-time cognitive status; A cognitive development visualization engine generates a three-dimensional spatiotemporal evolution map and intervention recommendations.

6. The artificial intelligence-based assessment system for adolescent cognitive ability and psychological development as described in claim 5, characterized in that, The privacy protection mechanism is specifically implemented as follows: A federated learning architecture is used for distributed training of the model, while the raw data is stored on the local terminal. Controllable noise is added to cognitive feature vectors using differential privacy techniques; Sensitive intermediate data during the fingerprint matching process is transmitted using a homomorphic encryption strategy.

7. The artificial intelligence-based assessment system for adolescent cognitive ability and psychological development as described in claim 5, characterized in that, The dynamic assessment path generator further includes: The strategy fingerprint feature library stores the feature transferability matrix of typical cognitive development patterns. The feature transferability matrix is ​​constructed by calculating the cosine similarity variance of cross-task operation features and includes a 52-dimensional transfer relationship map between three types of tasks: risk decision-making, pattern recognition, and social simulation. An online Bayesian optimizer dynamically selects evaluation dimensions based on the current sparsity of the cognitive space. The sudden jump event response module inserts a cross-domain verification task when a phase change is detected.

8. The artificial intelligence-based assessment system for adolescent cognitive ability and psychological development as described in claim 5, characterized in that, The visualization engine includes: A cultural adaptation calibration layer eliminates the bias of collectivist decision-making patterns on autonomy scores; An interface for predicting developmental trajectories, simulating cognitive evolutionary branch paths under different intervention measures; The hidden risk early warning unit detects early deviations in psychological development based on anomalies in strategy migration.

Images