Kindergarten data management system based on digital twin and blockchain

Abstract

The present invention proposes a kindergarten data management system based on digital twins and blockchain, the system includes: a data collection unit for collecting kindergarten data and preprocessing; a digital twin model construction unit for constructing a digital twin model based on the preprocessed data; a blockchain storage unit for encrypting and storing the digital twin model on the blockchain; a personalized analysis unit for constructing a multi-task learning model based on the digital twin model and real-time data on the blockchain for personalized analysis and prediction, and providing customized education and health management suggestions based on the results of personalized analysis and prediction; a kindergarten education strategy optimization unit for constructing a home-school interactive feedback mechanism, combining customized education and health management suggestions and actual feedback data from parents, and dynamically adjusting the kindergarten's education strategy. The present invention improves the security, efficiency and intelligence level of the system through multiple innovative points.

Description

Technical Field

[0001] The present invention belongs to the technical field of digital twins and blockchains, and in particular to a kindergarten data management system based on digital twins and blockchains. Background Art

[0002] In modern early childhood education, kindergarten data management has become increasingly important. Kindergartens not only need to manage a large amount of personal information about children, such as health status, learning progress, behavioral performance, etc., but also need to deal with complex matters such as communication between parents and teachers, allocation of teaching resources and safety monitoring. However, the current kindergarten data management system usually relies on traditional centralized databases and management methods, which has many shortcomings. First, the data storage of traditional systems is centralized and vulnerable to single point failures, and the integrity and security of the data cannot be fully guaranteed. Secondly, the correlation between data is weak, and it is impossible to achieve in-depth analysis of children's multi-dimensional data, which makes it difficult to provide personalized education and health management services. Thirdly, the existing system lacks an efficient privacy protection mechanism, which is prone to sensitive data leakage and improper use. Finally, the interaction between parents and teachers mainly relies on manual input and management, lacking intelligent and automated support, which not only increases the workload, but also limits the timeliness and accuracy of information exchange. In order to make up for these shortcomings, some advanced technologies are currently being tried to be applied to kindergarten data management, such as data twin technology and blockchain technology.

[0003] Data twin technology realizes real-time dynamic update and analysis of data by virtually mapping the digital mirror of real-world objects, which helps to establish a comprehensive data portrait of children and promote personalized teaching and health management. However, in the kindergarten environment, how to effectively create and maintain children's virtual twins and how to use multidimensional data for efficient analysis and decision-making remain a huge challenge. In addition, the implementation of data twin technology usually relies on high-quality data collection and processing, which places high demands on data collection equipment, data transmission and storage capabilities. At the same time, existing technologies have not yet solved the problem of how to protect children's privacy in data twin models, which makes sensitive data vulnerable to attacks and leaks during transmission and processing.

[0004] Blockchain technology provides a new way for secure storage and trusted sharing of data. Through distributed ledger technology, blockchain can ensure the immutability and traceability of data during storage and transmission, and enhance the security and transparency of data. However, at present, the application of blockchain technology in kindergarten data management systems also faces challenges, such as how to efficiently process large amounts of real-time data, how to maintain high performance and low latency of the system while ensuring data security, and how to protect data privacy in a decentralized network environment. In addition, existing blockchain technology often has difficulty maintaining high efficiency when faced with large and frequent read and write operations, resulting in bottlenecks in system performance. Therefore, existing technologies cannot effectively combine the advantages of data twins and blockchain to optimize multiple aspects of kindergarten data management. Summary of the invention

[0005] The purpose of this invention is to design a kindergarten data management system based on digital twins and blockchain. Through the innovative combination of data twins and blockchain technology, it solves the problems of data security, insufficient personalized analysis capabilities, and low efficiency of home-school interaction in the existing kindergarten data management system, providing kindergartens with a safe, intelligent, and efficient comprehensive data management solution.

[0006] In order to achieve the above objectives, the present invention provides a kindergarten data management system based on digital twins and blockchain, the system comprising:

[0007] Data collection unit, used to collect kindergarten data and perform pre-processing;

[0008] A digital twin model building unit, used for building a digital twin model according to the preprocessed data;

[0009] Among them, the digital twin model is constructed as follows:

[0010] Use the AES decryption algorithm to restore the data feature vector and obtain the decrypted data;

[0011] An improved recursive neural network model is used to combine decrypted data and historical data, and a regularization term is introduced for feature prediction. A digital twin model is constructed based on the feature prediction results. The regularization term is used to penalize the dependency between features, which is expressed as follows:

[0012] T ij (t) = σ(W j ·[V i (t),H i (t-1)]+b j )+λΩ(T i (t))

[0013] Among them, T i (t) represents the digital twin model, T ij(t) represents the predicted value of feature j at time t; σ represents the activation function, which is used for nonlinear transformation; W j represents the weight matrix of feature j, the learned parameters; [V i (t),H i (t-1)] represents the combination of current input and historical data; b j represents the bias term; λ represents the regularization coefficient, which controls the influence of the regularization term; Ω(T i (t)) represents the regularization term, which is defined as the square of the L2 norm between all features and is expressed as follows:

[0014]

[0015] Dynamically correct the digital twin model by detecting abnormal changes in features, including:

[0016] Define feature T ij (t) rate of change ΔT ij (t), detected by weighted anomaly detection mechanism, expressed as follows:

[0017] ΔT ij (t) = T ij (t)-T ij (t-1)

[0018] Among them, T ij (t-1) represents the characteristics of the previous moment;

[0019] The weighted anomaly detection mechanism adjusts the sensitivity of anomaly detection according to the volatility of the feature. The anomaly judgment formula is expressed as follows:

[0020]

[0021] in, and Represent the mean and standard deviation of the feature change rate, γ j is the anomaly detection sensitivity coefficient of feature j, which is dynamically adjusted according to the volatility of the feature;

[0022] A blockchain storage unit, used to encrypt and store the digital twin model on the blockchain;

[0023] Personalized analysis unit, which is used to build a multi-task learning model based on the digital twin model and real-time data on the blockchain for personalized analysis and prediction, and to provide customized education and health management suggestions based on the results of personalized analysis and prediction;

[0024] The kindergarten education strategy optimization unit is used to build a home-school interactive feedback mechanism, combining customized education and health management suggestions with parents' actual feedback data to dynamically adjust the kindergarten's education strategy.

[0025] Furthermore, the preprocessing includes time synchronization processing and standardization processing, and then feature extraction is performed on the preprocessed data at the edge computing node, and then weighted aggregation is performed on the features of different edge computing nodes to generate comprehensive features, and the comprehensive features are encrypted to generate encrypted data.

[0026] Furthermore, the digital twin model is updated at the end of each time step t, and the global feature state G is recalculated when multiple features change. i (t), using personalized analysis and decision making, is expressed as follows:

[0027]

[0028] Among them, the global characteristic state G of the twin i (t) is used to provide feedback to teachers and parents, m represents the total number of features, and j represents feature j.

[0029] Furthermore, the step of encrypting and storing the digital twin model on the blockchain includes:

[0030] S301, encrypting the modified digital twin model using an improved hash function combined with a hash chain to generate a unique encrypted hash;

[0031] S302. Introduce a dual authentication mechanism to design access control rules for smart contracts to verify whether user u has the right to access the data twin model T at time t. i (t), at the same time, when the user requests to read data, the hash verification function is used to compare the hash value calculated by the user with the hash value stored on the blockchain to verify the authenticity of the data: if the hash verification function returns 1, indicating that the data is valid, then the data verification is successful; if it returns 0, indicating that the data is invalid, then the data verification fails;

[0032] S303. Use a distributed data storage optimization algorithm to store and update data on the blockchain to ensure synchronization and storage efficiency of data when updating.

[0033] Furthermore, the access control rule combines the user's private key and the public key on the blockchain for identity authentication, as shown below:

[0034]

[0035] in, represents the access control function, returns 1 for authorized access, and returns 0 for unauthorized access; φ(u) represents the identity authentication function of user u, which combines the private key and the public key for verification; K represents the public key on the blockchain, K -1 is the corresponding private key; η represents the threshold of access rights, which determines whether access is allowed;

[0036] The distributed data storage optimization algorithm is expressed as follows:

[0037]

[0038] in, represents the data update strategy function, which optimizes the distribution of data in the blockchain network; D n (t represents the data storage state of the nth node at time t; P n (t) represents the storage priority of the nth node; α represents the adjustment coefficient, which is used to control the balance of data distribution; N represents the total number of nodes.

[0039] Furthermore, the multi-task learning model construction steps are as follows:

[0040] A neural network with shared parameters is used to simultaneously learn two tasks, namely health management and learning guidance. The output of the neural network is health prediction and learning prediction. At the same time, a dynamic weight adjustment function α(t) is designed to balance the importance of health prediction and learning prediction, which is expressed as follows:

[0041]

[0042] in, represents the loss function of health management; represents the loss function of learning guidance, α(t) represents the dynamic weight adjustment function, which is defined as an adaptive function based on the child's state and is expressed as follows:

[0043]

[0044] Among them, M i( t) represents a baseline feature vector, which represents the characteristics of child i in an ideal state; ∥T i (t)-M i (t)∥ 2 It is the Euclidean distance between the data twin and the baseline state, which is used to measure the deviation of the child's state from the ideal state.

[0045] Furthermore, a strategy generation function is designed based on the output of the multi-task learning model to generate personalized education and health management suggestions, as shown below:

[0046]

[0047] in, represents the generation function of personalized strategy, and the output is the specific recommendation content; β1 and β2 represent the strategy weight parameters, which are used to balance the relationship between health and learning; P H (t) represents the health prediction score at time t; P L (t) represents the learning prediction score at time t;

[0048] According to the value of the generating function of the personalized strategy, the one with the highest value is selected for priority health management intervention.

[0049] Furthermore, the kindergarten education strategy optimization unit performs the following steps:

[0050] Automatically generate a personalized growth report for each child based on a weighted combination of multidimensional features with customized education and health management recommendations;

[0051] Collect parents' feedback on growth reports through the parent-side application, and use natural language processing technology to analyze parents' feedback content and quantify parents' satisfaction and needs;

[0052] Optimize strategies based on parent feedback data, adjust personalized education and health management strategies in real time, and improve the dynamic adaptability of the system;

[0053] Feed the optimized strategies back to the system to improve the effectiveness of personalized education and health management strategies.

[0054] Furthermore, the weighted combination based on multi-dimensional features combines the feedback information of education and health to generate a personalized growth report score, and at the same time introduces a time-weighted smoothing function ω(t) into the personalized growth report score to smooth the personalized growth report score, which is expressed as follows:

[0055] R i (t) = ω(t)·(β1·P H (t)+β2·P L (t))+(1-ω(t))·R i (t-1)

[0056] Among them, R i (t) represents the personalized growth report score of child i at time t; ω(t) represents the time-weighted smoothing function, which controls the weight of current feedback and historical feedback, and its value is between [0,1]; β1 and β2 represent the strategy weight parameters, which are used to balance the relationship between health and learning, respectively, satisfying β1+β2=1; P H (t) represents the health prediction score at time t, which represents the evaluation of personalized strategy in health management; P L(t) represents the learning prediction score at time t, which indicates the evaluation of the personalized strategy in learning guidance; R i (t-1) represents the personalized growth report score of child i at time t-1;

[0057] The use of natural language processing technology to analyze parental feedback includes:

[0058] Parents' feedback text i (t) converted into a quantitative satisfaction score S i (t) and the demand vector D i (t), and then use the sentiment analysis function ψ(F i (t)) and the demand analysis function ξ(F i (t)), combined with the content generated by the report, is quantified as follows:

[0059]

[0060] in, Represents the comprehensive scoring function of the growth report, and the growth report R i (t) is converted into a quantitative score; γ represents a smoothing parameter, which is used to adjust the impact of growth reports on satisfaction;

[0061] The Q learning update rule introduced in the reinforcement learning mechanism for strategy optimization based on parent feedback data is expressed as follows:

[0062]

[0063] Among them, Q i (t+1) represents the strategy value function of child i at time t+1, which is used to evaluate the effectiveness of the strategy; Q i (t represents the strategy value function of child i at time t; α represents the learning rate, which controls the strategy optimization step size and is usually a value between [0,1]; S i (t) represents the parent's satisfaction score for child i at time t; δ represents the discount factor, which measures the weight of future rewards; max a' Q i (t+1,a') represents the maximum policy value among all actions a' at time t+1, indicating the tendency to select the best policy.

[0064] Furthermore, the demand vector D i (t), is calculated as follows:

[0065] D i (t) = ξ(F i (t))⊙M i (t)

[0066] Among them, Di (t) represents the demand vector, which indicates the parents’ demands in different aspects, extracted from the feedback text; ξ(F i (t)) represents the demand analysis function, which extracts the specific needs in the parents’ feedback; M i (t) represents the weight vector, which indicates the weight of each requirement; ⊙ represents the element-by-element multiplication symbol, which indicates that the elements in the vector are multiplied accordingly.

[0067] The beneficial technical effects of the present invention are at least as follows:

[0068] The present invention aims at the data centralization and security issues of traditional data management systems. The present invention combines blockchain technology to achieve distributed storage and encryption protection of children's data. Through blockchain's smart contracts and zero-knowledge proof technology, it ensures that data can be shared and used with transparency and traceability, while effectively protecting children's privacy and avoiding sensitive data leakage.

[0069] In addition, the present invention uses data twin technology to create a multi-dimensional virtual data twin for each child. By integrating health data, behavioral data and learning data, the system can comprehensively and dynamically analyze the growth status of each child and provide personalized education and health management suggestions. At the same time, the latest artificial intelligence algorithms (such as multimodal deep learning, time series analysis models, etc.) are used for data analysis and risk prediction, which effectively improves the system's sensitivity and response speed to children's growth changes.

[0070] The present invention also introduces innovation in the interaction between parents and kindergartens. Through intelligent data collection and analysis, the system can automatically generate a child's growth report, and ensure the authenticity and security of the report through blockchain technology. Parents can access and interact in real time through smart terminals, which increases the enthusiasm and effectiveness of parental participation. This intelligent and transparent interaction mechanism solves the problems of low efficiency and information asymmetry in home-school interaction in the existing system, and further improves the service quality of kindergartens and parent satisfaction.

[0071] This invention solves the problems of data security, insufficient personalized analysis capabilities, and low efficiency of home-school interaction in existing kindergarten data management systems through the innovative combination of data twins and blockchain technology, providing kindergartens with a safe, intelligent, and efficient comprehensive data management solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0073] Figure 1This is a framework diagram of a kindergarten data management system based on data twins and blockchain according to an embodiment of the present invention. DETAILED DESCRIPTION

[0074] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0075] In one or more embodiments, Figure 1 As shown, the kindergarten data management system based on data twins and blockchain of the present invention is disclosed, including:

[0076] A data collection unit 101 is used to collect kindergarten data and perform pre-processing;

[0077] A digital twin model construction unit 102, configured to construct a digital twin model according to the preprocessed data;

[0078] The blockchain storage unit 103 is used to encrypt and store the digital twin model on the blockchain;

[0079] The personalized analysis unit 104 is used to build a multi-task learning model based on the digital twin model and real-time data on the blockchain to perform personalized analysis and prediction, and provide customized education and health management suggestions based on the results of personalized analysis and prediction;

[0080] The kindergarten education strategy optimization unit 105 is used to build a home-school interactive feedback mechanism, and dynamically adjust the kindergarten's education strategy by combining customized education and health management suggestions and parents' actual feedback data.

[0081] Specifically, in the data collection unit 101, the following steps are performed:

[0082] The multidimensional data in kindergartens are diverse and scattered. Traditional data collection methods make it difficult to obtain high-quality data in real time, resulting in insufficient data integrity, real-time and accuracy, and unable to provide a reliable basis for subsequent analysis and personalized management.

[0083] To address this, the present invention deploys a variety of intelligent data acquisition devices (such as cameras, health monitoring sensors, learning terminals, etc.) to ensure that all devices achieve time consistency through high-precision clock synchronization technology (PTP-Precision Time Protocol). The data stream collected by each device in real time is marked with a timestamp, denoted as D i (t), where i is the device number and t is the time.

[0084] Furthermore, the original data stream D collected by all devices i (t) Perform format standardization to generate standardized data stream S i (t). For example, camera data is processed into frame information F(t) per second, and health sensor data records the average value H(t) per second. i (t) Perform real-time processing to extract feature data X i (t). The feature extraction function is defined as:

[0085] X i (t) = f(S i (t))=CNN(S i (t))

[0086] Where f(·) is a convolutional neural network (CNN) model used to extract key features from images and time series data. For example, if at a certain time t, the camera data S i (t) = [0.3, 0.7, 0.5], then the feature data obtained through CNN model processing is X i (t)=[0.6,0.8].

[0087] Furthermore, the feature data of different devices are weighted and aggregated to generate comprehensive features Z(t). i Dynamically adjust based on the data quality and accuracy of the device:

[0088]

[0089] For example, if the camera data X1(t) = 0.8 (indicating high physical activity), the health sensor data X2(t) = 0.4 (indicating low heart rate), and the weights are w1 = 0.6 and w2 = 0.4 respectively, then calculate the comprehensive feature Z(t):

[0090] Z(t)=0.6×0.8+0.4×0.4=0.48+0.16=0.64

[0091] Then, anomaly detection is performed on the comprehensive feature Z(t), and the statistical detection method is used to determine the abnormal data points:

[0092]

[0093] in, and are the mean and standard deviation of the historical data features of device i. For example, if X i (t) = 5, but:

[0094] |5-2|=3 (exceeding 3 times the standard deviation, considered abnormal)

[0095] The aggregated comprehensive feature Z(t) is encrypted and then transmitted to generate encrypted data E(Z(t)):

[0096] E(Z(t))=AES key (Z(t))

[0097] For example, assuming Z(t) = 0.64 and the encryption key is "key123", the encryption result is E(Z(t)) = AES "key123" (0.64), ensuring the security of data during transmission.

[0098] Through these steps, the data acquisition and preprocessing module realizes the complete process of real-time data collection from multi-source devices, standardized processing, feature extraction, weighted aggregation, anomaly detection to data encryption transmission, ensuring high-quality and secure data input for subsequent data twin generation.

[0099] Specifically, in the digital twin model building unit 102, the following steps are performed:

[0100] The twin model is initialized by inputting the output encrypted data E(Z(t)) of the data collection unit and using the AES decryption algorithm to restore the initial data feature vector V i (t). This decryption process ensures the integrity and security of the input data, while providing real-time child status data:

[0101]

[0102] in, Indicates an AES decryption operation using the key key. i (t) is the initial feature vector of child i at time t, which contains multi-dimensional data (such as health, behavior, learning, etc.).

[0103] Calculation example: Assume E(Z(t)) = AES "key123" (0.64), after decryption, we get V i (t) = [0.6, 0.8, 0.4].

[0104] Furthermore, a twin model T containing multi-dimensional features is created i (t), and capture the dynamic changes of children’s status through time series analysis. The model needs to consider the temporal dependency of each feature and introduce innovative regularization terms to balance the mutual influence between features and prevent model overfitting.

[0105] Specifically, an improved recurrent neural network (RNN) model is used, combined with the decrypted data Vi (t) and historical data H i (t-1) for feature prediction. Innovatively introduce a regularization term Ω(T i (t)) is used to penalize excessive dependence between features and ensure the generalization ability of the model:

[0106] T ij (t) = σ(W j ·[V i (t),H i (t-1)]+b j )+λΩ(T i (t))

[0107] Among them, T ij (t) represents the predicted value of feature j at time t. σ represents the activation function (such as ReLU or tanh) used for nonlinear transformation. j Represents the weight matrix of feature j, the learned parameters. [V i (t),H i (t-1)] represents the combination of current input and historical data. j represents the bias term. λ represents the regularization coefficient, which controls the influence of the regularization term. Ω(T i (t)) represents the regularization term, which is defined as the square of the L2 norm between all features and is expressed as:

[0108]

[0109] Calculation example: Assuming W j =[0.2,0.3,0.1], V i (t) = [0.6, 0.8, 0.4], H i (t-1) = [0.5, 0.7, 0.3], b j =0.1,λ=0.05,Ω(T i (t))=(0.6-0.8) 2 +(0.6-0.4) 2 +(0.8-0.4) 2 =0.04+0.04+0.16=0.24;

[0110] T ij )t)=σ(0.2·0.6+0.3·0.8+0.1·0.4+0.2·0.5+0.3·0.7+0.1)+0.05·0.24

[0111] Assuming ReLU is used as the activation function, we get:

[0112] T ij(t)=max)(0,0.12+0.24+0.04+0.1+0.21+0.1)+0.012=0.822

[0113] Furthermore, by detecting abnormal changes in features, the twin model is dynamically modified to ensure that it can accurately reflect the actual status of the child. Define feature T ij (t) rate of change ΔT ij (t), and detects it through an innovative weighted anomaly detection mechanism:

[0114] ΔT ij (t) = T ij (t)-T ij (t-1)

[0115] The weighted anomaly detection mechanism adjusts the sensitivity of anomaly detection according to the volatility of the feature. The anomaly judgment formula is:

[0116]

[0117] in, and are the mean and standard deviation of the feature change rate, respectively. j is the anomaly detection sensitivity coefficient of feature j, which is dynamically adjusted according to the volatility of the feature. Calculation example: Assuming ΔT ij (t) = 0.4, γ j =3:

[0118] |0.4-0.2|=0.2>3·0.1=0.3 (does not exceed the abnormal threshold and is not marked as abnormal)

[0119] Furthermore, the corrected data twin is updated and stored on the blockchain to ensure the data’s immutability and traceability.

[0120] Twin update is to update the data twin T at the end of each time step t i (t). The updated twin is stored on the blockchain, ensuring its immutability and traceability.

[0121] When multiple features change significantly, calculate the global feature state G i (t), for further personalized analysis and decision-making:

[0122]

[0123] Calculation example: Assume m = 3, T i1 (t) = 0.6, T i2 (t) = 0.4, T i3 (t) = 0.5:

[0124]

[0125] The global characteristic state G of the twin i (t) Used to provide feedback to teachers and parents to help with decision making and personalized intervention.

[0126] Through the above steps, this solution can generate and dynamically update the virtual data twin of each child and reflect its multi-dimensional characteristics in real time. The solution innovatively introduces regularization terms and smoothing mechanisms to ensure the stability and generalization ability of the model, and improves the flexibility and robustness of the model through anomaly detection mechanisms. The twin model can continuously reflect the health, behavior and learning status of children, providing a basis for personalized analysis and intelligent decision-making, while ensuring the security and integrity of data through blockchain technology.

[0127] Specifically, in the blockchain storage unit 103, the following steps are performed:

[0128] Data twin encryption is the input of each child’s dynamic data twin T i (t). Before the data twin is stored in the blockchain, it is first encrypted. The present invention uses an improved hash function H(·) combined with a hash chain to generate a unique encrypted hash H i (t), making data encryption at each point in time more secure:

[0129]

[0130] Among them, H i (t) represents the hash value of the data twin of child i at time t. i (t) is the dynamic data twin of child i at time t. Represents a bitwise exclusive OR operation, which is used to increase the complexity and security of data. i (t-1) is the hash value at time t-1, which ensures the relevance and immutability of each data record.

[0131] Calculation example: Assuming T i (t) = [0.6, 0.8, 0.4], H i (t-1)=0x3F2A, using SHA-256 as the hash function H(·), then:

[0132]

[0133] The encrypted hash value H i (t) Stored in the distributed ledger of the blockchain, ensuring the immutability and traceability of the data.

[0134] Furthermore, through smart contracts on the blockchain, refined control over data access is achieved to ensure that only authorized users can access and operate specific data. Define access control rules for smart contracts Used to verify whether user u has access to data twin T at time t i (t). An innovative two-factor authentication mechanism φ(u) is introduced, which combines the user’s private key and the public key on the blockchain for authentication:

[0135]

[0136] in, is the access control function, which returns 1 for authorized access and 0 for unauthorized access. φ(u) is the authentication function of user u, which combines the private key and the public key for verification. K is the public key on the blockchain, K -1 is the corresponding private key. η is the threshold of access rights, which determines whether access is allowed.

[0137] Calculation example: Assuming φ(u)=0.75, K -1 =0.8, T i (t)=[0.6,0.8,0.4], η=0.5:

[0138] φ(u)·K -1 ·T i (t)=0.75×0.8×0.6=0.36≤0.5

[0139] but Grant access.

[0140] Furthermore, when reading data, the integrity and authenticity of the data are ensured through the consensus mechanism and encryption verification algorithm of the blockchain.

[0141] When a user requests to read data, a hash verification mechanism is used Verify the authenticity of the data by comparing the hash value calculated by the user with the hash value stored on the blockchain:

[0142]

[0143] in, This is a data validation function. It returns 1 if the data is valid and 0 if the data is invalid. is the hash value calculated by the user. i (t) is the hash value stored on the blockchain.

[0144] Calculation example: Assume that the user calculates The hash value H stored on the blockchain i (t) = 0x7A4C:

[0145]

[0146] but Data verification passed.

[0147] Furthermore, data can be efficiently stored and updated on the blockchain, and the data storage method can be optimized to meet the high-frequency data update needs in the kindergarten environment. Distributed data storage optimization algorithm is used Ensure fast synchronization and storage efficiency of data when it is updated:

[0148]

[0149] in, D is the data update strategy function, which optimizes the distribution of data in the blockchain network. n (t) represents the data storage status of the nth node at time t. n (t) represents the storage priority of the nth node. α is the adjustment coefficient used to control the balance of data distribution.

[0150] Calculation example: Assume D1(t) = 0.5, P1(t) = 0.8, D2(t) = 0.6, P2(t) = 0.9, α = 0.7, then:

[0151]

[0152] Storage Optimization Functions The calculation results are used to adjust the data distribution on the blockchain to ensure fast data synchronization and storage efficiency.

[0153] Through the above steps, this solution can safely store each child's dynamic data twin on the blockchain and achieve refined access control of the data. The solution ensures the immutability of the data through improved hash encryption, smart contracts, data verification and storage optimization algorithms.

[0154] Specifically, in the personalized analysis unit 104, the following steps are performed:

[0155] The input of the personalized analysis unit is the encrypted data twin H i (t) and dynamic data T i (t).

[0156] First, the encrypted hash data H stored on the blockchain i (t) Verify and decrypt to obtain the original data twin T i (t) to ensure the integrity and accuracy of the data. The decryption formula uses the same hash decryption function H as when storing it. -1 (·):

[0157] T i (t) = H -1 (H i (t))

[0158] Among them, H -1 (·) is the inverse hash function, which is used to recover the original data twin.

[0159] Calculation example: Assume After decryption, we get T i (t)=0.6,0.8,0.4].

[0160] Further, using the decrypted data T i (t) Build a multi-task learning model that takes into account the personalized needs of both health and learning. Innovatively use a weighted dynamic adjustment mechanism to optimize the model's output and make personalized adjustments based on the child's real-time status.

[0161] Furthermore, using a neural network with shared parameters to learn two tasks at the same time (health management and learning guidance)

[0162] The model output consists of two parts: health prediction P H (t) and learn to predict P L (t). The loss function of the model The loss of health and learning is considered at the same time, and a dynamic weight adjustment function α(t) is used to balance the importance of the two:

[0163]

[0164] in: is the loss function for health management, such as measuring the deviation of health indicators based on L2 loss. is the loss function for learning guidance, such as the accuracy of measuring learning progress based on cross entropy loss. α(t) is the dynamic weight adjustment function, defined as an adaptive function based on the child state:

[0165]

[0166] Among them, M i (t) is a baseline feature vector, which represents the characteristics of child i in an ideal state. ∥T i (t)-M i (t)∥ 2 It is the Euclidean distance between the data twin and the baseline state, which is used to measure the deviation of the child's state from the ideal state.

[0167] Calculation example: Assuming T i(t) = [0.6, 0.8, 0.4], M i (t) = [0.5, 0.7, 0.5], then calculate the Euclidean distance and weight α(t):

[0168] ∥T i (t)-M i (t)∥ 2 =(0.6-0.5) 2 +(0.8-0.7) 2 +(0.4-0.5) 2 =0.01+0.01+0.01=0.03 (28)

[0169] If the sum of the distances of other children is 0.05, 0.07, then:

[0170]

[0171] Furthermore, according to the health prediction P output by the model H (t) and learn to predict P L (t) Generate personalized education and health management suggestions and feed them back to relevant personnel (e.g., teachers, parents).

[0172] Furthermore, the strategy generation function Automatically generate education and health management strategies based on predicted values. This function combines the rule engine and the probability distribution of the model output to determine specific recommendations:

[0173]

[0174] in: is the generation function of the personalized strategy, and the output is the specific recommended content. β1 and β2 are the strategy weight parameters used to balance the relationship between health and learning.

[0175] Calculation example: Assuming P H (t) = 0.7 (high health risk), P L (t) = 0.5 (moderate learning progress), and weight parameters β1 = 0.6, β2 = 0.4:

[0176]

[0177] Based on the value of 0.62 generated by the strategy, the system recommends prioritizing health management intervention while taking into account the tracking of learning progress.

[0178] Furthermore, personalized suggestions will be fed back to relevant personnel, and the model will continue to be adjusted based on subsequent feedback and data to improve the accuracy of personalized services.

[0179] Based on the actual feedback results (such as feedback from teachers and parents), the multi-task model is optimized using a reinforcement learning mechanism. Each feedback result F(t) is quantified into a reward r(t) and used to update the model parameters:

[0180]

[0181] Among them, θ(t) is the parameter of the model at time t. η is the learning rate, which controls the update step size. r(t) is the reward signal of the feedback result. is the gradient of the loss function with respect to the parameters.

[0182] Calculation example: Assume θ(t) = 0.5, η = 0.01, r(t) = 2,

[0183] θ(t+1)=0.5+0.01·2·(-0.4)=0.5-0.008=0.492

[0184] Through the above steps, this solution can realize intelligent analysis of children's personalized education and health management. Through multi-task learning model, dynamic weight adjustment and reinforcement learning feedback mechanism, this solution ensures the real-time, adaptable and personalized nature of the model, providing strong support for the comprehensive development of children.

[0185] Specifically, in the kindergarten education strategy optimization unit 105, the following steps are performed:

[0186] Input is a personalization strategy and personalized weight parameters β1, β2, and then use the results generated by the personalized strategy to automatically generate a personalized growth report for each child. The report generation formula is based on a weighted combination of multi-dimensional features, combined with feedback information from both education and health, to generate detailed report content. An innovative time-weighted smoothing function ω(t) is introduced to smooth the report content to ensure the continuity and reliability of the report:

[0187] R i (t) = ω(t)·(β1·P H (t)+β2·P L (t))+(1-ω(t))·R i (t-1)

[0188] Among them, R i (t) represents the personalized growth report score of child i at time t. ω(t) represents the time-weighted smoothing function, which controls the weight of current feedback and historical feedback, and its value is between [0,1]. β1 and β2 represent the strategy weight parameters, which are used to balance the relationship between health and learning, respectively, and usually satisfy β1+β2=1. P H(t) represents the health prediction score at time t, which represents the evaluation of personalized strategy in health management. L (t) represents the learning prediction score at time t, which indicates the evaluation of the personalized strategy in learning guidance. i (t-1) represents the personalized growth report score of child i at time t-1.

[0189] Calculation example:

[0190] Assume P H (t)=0.7,P L (t)=0.5, β1=0.6, β2=0.4, ω(t)=0.8, R i (t-1) = 0.6, then:

[0191] R i (t)=0.8·(0.6·0.7+0.4·0.5)+0.2·0.6=0.8·0.62+0.12=0.496+0.12=0.616

[0192] Furthermore, parents’ feedback on the growth report is collected through the parent-side application, and the feedback content is analyzed using natural language processing (NLP) technology to quantify parent satisfaction and needs.

[0193] Parents' feedback text i (t) converted into a quantitative satisfaction score S i (t) and the demand vector D i (t). Using the sentiment analysis function ψ(F i (t)) and the demand analysis function ξ(F i (t)), combined with the content generated by the report to quantify:

[0194]

[0195] Variable Description, S i (t) represents the parent's satisfaction score for child i at time t, ψ(F i (t)) represents the parent feedback text F i The sentiment analysis score of (t) indicates the parents’ sentiment attitude towards the current strategy. γ represents a smoothing parameter, which is used to adjust the impact of growth reports on satisfaction and is usually a value between [0,1]. Represents the comprehensive scoring function of the growth report, and the growth report R i (t) was converted into quantitative scores.

[0196] D i (t) = ξ(F i (t))⊙M i (t)

[0197] Among them, D i (t) represents the demand vector, which indicates the parents’ demands in different aspects, extracted from the feedback text. ξ(F i (t)) represents the demand analysis function, which extracts the specific needs in the parents’ feedback. i (t) represents the weight vector, which indicates the weight of each requirement. ⊙ represents the element-by-element multiplication symbol, which indicates that the elements in the vector are multiplied accordingly.

[0198] Calculation example:

[0199] Assumption F i The sentiment analysis score ψ(F i (t)) = 0.6, γ = 0.3, but:

[0200] S i (t)=0.6+0.3·0.7=0.6+0.21=0.81

[0201] Furthermore, we optimize the strategy based on the feedback data from parents, adjust the personalized education and health management strategies in real time, and improve the dynamic adaptability of the system. The personalized strategy is adjusted in real time according to the needs and feedback data of parents, and the Q learning update rule in the reinforcement learning mechanism is introduced to optimize the long-term effect of the education strategy:

[0202]

[0203] Among them, Q i (t+1) represents the strategy value function of child i at time t+1, which is used to evaluate the effectiveness of the strategy. i (t represents the policy value function of child i at time t. α represents the learning rate, which controls the optimization step size of the policy and is usually a value between [0,1]. S i (t) represents the parent's satisfaction score for child i at time t (same as above). δ represents the discount factor, which measures the weight of future rewards and is usually a value between [0,1]. max a' Q i (t+1,a') represents the maximum policy value among all possible actions a' at time t+1, indicating the tendency to choose the best policy. Calculation example:

[0204] Assume Q i (t)=0.5,α=0.1,S i (t)=0.8,δ=0.9,max a' Q i (t+1,a')=0.7, then:

[0205] Qi (t+1)=0.5+0.1·(0.8+0.9·0.7-0.5)

[0206] Q i (t+1)=0.5+0.1·(0.8+0.63-0.5)=0.5+0.1·0.93=0.5+0.093=0.593

[0207] Furthermore, the optimized strategies are fed back to the system, and the model is continuously updated to improve the effectiveness of personalized education and health management strategies.

[0208] The optimized strategy is re-input into the multi-task learning model and updated through the feedback loop mechanism Continuously optimize the model parameters θ(t):

[0209]

[0210] Among them, θ(t+1) represents the parameters of the model at time t+1, which represents the state of the model after optimization. θ(t) represents the parameters of the model at time t. β represents the feedback learning rate, which is used to control the update speed of the model parameters and is usually a value between [0,1]. Denotes the policy value function Q i (t+1) gradient of model parameters. Calculation example:

[0211] Assume θ(t) = 0.4, β = 0.05, but:

[0212] θ(t+1)=0.4+0.05·0.6=0.4+0.03=0.43

[0213] Through the above steps, this solution can optimize the interaction and feedback loop between home and school. Through the innovative combination of personalized growth report generation, parent feedback analysis, strategy optimization and loop feedback mechanism, this solution ensures the dynamic adaptability and personalization of the system, providing efficient and accurate support for children's education and health management.

[0214] The above disclosures are only some preferred embodiments of the present invention, which certainly cannot be used to limit the scope of rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of the above embodiments and equivalent changes made according to the claims of the present invention are still within the scope of the invention.

Claims

1. Kindergarten data management system based on digital twin and blockchain, characterized by: The system comprises: Data collection unit, used to collect kindergarten data and perform pre-processing; A digital twin model building unit, used for building a digital twin model according to the preprocessed data; Among them, the digital twin model is constructed as follows: Use the AES decryption algorithm to restore the data feature vector and obtain the decrypted data; An improved recursive neural network model is used to combine decrypted data and historical data, and a regularization term is introduced for feature prediction. A digital twin model is constructed based on the feature prediction results. The regularization term is used to penalize the dependency between features, which is expressed as follows: T ij (t)=σ(W j ·[V i (t),H i (t-1)]+b j )+λΩ(T i (t)) Among them, T i (t) represents the digital twin model, i is the i-th child, T ij (t) represents the predicted value of feature j at time t; σ represents the activation function, which is used for nonlinear transformation; W j represents the weight matrix of feature j; [V i (t),H i (t-1)] represents the combination of current input and historical data; b j represents the bias term; λ represents the regularization coefficient, which controls the influence of the regularization term; Ω(T i (t)) represents the regularization term, which is defined as the square of the L2 norm between all features and is expressed as follows: Dynamically correct the digital twin model by detecting abnormal changes in features, including: Define feature T ij (t) rate of change ΔT ij (t), detected by weighted anomaly detection mechanism, expressed as follows: ΔT ij (t)=T ij (t)-T ij (t-1) Among them, T ij (t-1) represents the characteristics of the previous moment; The weighted anomaly detection mechanism adjusts the sensitivity of anomaly detection according to the volatility of the feature. The anomaly judgment formula is expressed as follows: in, and Represent the mean and standard deviation of the feature change rate, γ j is the anomaly detection sensitivity coefficient of feature j, which is dynamically adjusted according to the volatility of the feature; A blockchain storage unit, used to encrypt and store the digital twin model on the blockchain; Personalized analysis unit, which is used to build a multi-task learning model based on the digital twin model and real-time data on the blockchain for personalized analysis and prediction, and to provide customized education and health management suggestions based on the results of personalized analysis and prediction; The kindergarten education strategy optimization unit is used to build a home-school interactive feedback mechanism, combining customized education and health management suggestions with parents' actual feedback data to dynamically adjust the kindergarten's education strategy.

2. The kindergarten data management system based on digital twins and blockchain according to claim 1 is characterized in that: The preprocessing includes time synchronization processing and standardization processing, and then feature extraction is performed on the preprocessed data at the edge computing node, and then weighted aggregation is performed on the features of different edge computing nodes to generate comprehensive features, and the comprehensive features are encrypted to generate encrypted data.

3. The kindergarten data management system based on digital twins and blockchain according to claim 1 is characterized in that: The digital twin model is updated at the end of each time t, and the global feature state G is recalculated when multiple features change. i (t), used for personalized analysis and decision making, is expressed as follows: Among them, the global characteristic state G of the twin i (t) is used to provide feedback to teachers and parents, m represents the total number of features, and j represents feature j.

4. The kindergarten data management system based on digital twins and blockchain according to claim 1 is characterized in that: The digital twin model is encrypted and stored on the blockchain, including: S301, encrypting the modified digital twin model using an improved hash function combined with a hash chain to generate a unique encrypted hash; S302, introduce a dual authentication mechanism to design access control rules for smart contracts to verify whether user u has the right to access the digital twin model digital twin model T at time t i (t), at the same time, when the user requests to read data, the hash verification function is used to compare the hash value calculated by the user with the hash value stored on the blockchain to verify the authenticity of the data: if the hash verification function returns 1, indicating that the data is valid, then the data verification is successful; if it returns 0, indicating that the data is invalid, then the data verification fails; S303. Use a distributed data storage optimization algorithm to store and update data on the blockchain to ensure synchronization and storage efficiency of data when updating.

5. The kindergarten data management system based on digital twins and blockchain according to claim 4 is characterized in that: The access control rules combine the user's private key and the public key on the blockchain for identity authentication, as shown below: in, represents the access control function, returns 1 for authorized access, and returns 0 for unauthorized access; φ(u) represents the identity authentication function of user u, which combines the private key and the public key for verification; K represents the public key on the blockchain, K -1 is the corresponding private key; η represents the threshold of access rights, which determines whether access is allowed; The distributed data storage optimization algorithm is expressed as follows: Among them, H i (t) represents the hash value of the digital twin model of child i at time t, represents the data update strategy function, which optimizes the distribution of data in the blockchain network; D n (t) represents the data storage status of the nth node at time t; P n (t) represents the storage priority of the nth node; α represents the adjustment coefficient, which is used to control the balance of data distribution; N represents the total number of nodes.

6. The kindergarten data management system based on digital twins and blockchain according to claim 1 is characterized in that: The steps for constructing the multi-task learning model are as follows: A neural network with shared parameters is used to simultaneously learn two tasks, namely health management and learning guidance. The output of the neural network is the health prediction score P H (t) and the learning prediction score P L (t), and a dynamic weight adjustment function α(t) is designed to balance the importance of health prediction and learning prediction, which is expressed as follows: in, represents the loss function of health management; represents the loss function of learning guidance, α(t) represents the dynamic weight adjustment function, which is defined as an adaptive function based on the child's state and is expressed as follows: Among them, M i (t) represents the baseline state, which indicates the characteristics of child i in the ideal state; ∥T i (t)-M i (t)∥ 2 It is the Euclidean distance between the digital twin model and the baseline state, which is used to measure the deviation of the child's state from the ideal state.

7. The kindergarten data management system based on digital twins and blockchain according to claim 6 is characterized in that: According to the output of the multi-task learning model, a strategy generation function is designed to generate customized education and health management suggestions, which are expressed as follows: in, represents the generation function of personalized strategy, and the output is the specific recommendation content; β1 and β2 represent the strategy weight parameters, which are used to balance the relationship between health and learning; P H (t) represents the health prediction score at time t; P L (t) represents the learning prediction score at time t; Based on the values ​​of the personalized strategy generation function, children with the highest values ​​are prioritized for health / learning management interventions.

8. The kindergarten data management system based on digital twins and blockchain according to claim 1 is characterized in that: The kindergarten education strategy optimization unit performs the following steps: Automatically generate a personalized growth report for each child based on a weighted combination of multidimensional features with customized education and health management recommendations; Collect parents' feedback on growth reports through the parent-side application, and use natural language processing technology to analyze parents' feedback content and quantify parents' satisfaction and needs; Optimize strategies based on parent feedback data, adjust customized education and health management suggestions in real time, and improve the dynamic adaptability of the system; Feed the optimized strategies back to the system to improve the effectiveness of customized education and health management recommendations.

9. The kindergarten data management system based on digital twin and blockchain according to claim 8 is characterized in that: The weighted combination based on multi-dimensional features combines the feedback information of education and health to generate a personalized growth report score. At the same time, a time-weighted smoothing function ω(t) is introduced into the personalized growth report score to smooth the personalized growth report score, which is expressed as follows: R i (t)=ω(t)·(β1·P H (t)+β2·P L (t))+(1-ω(t))·R i (t-1) Among them, R i (t) represents the personalized growth report score of child i at time t; ω(t) represents the time-weighted smoothing function, which controls the weight of current feedback and historical feedback, and its value is between [0,1]; β1 and β2 represent the strategy weight parameters, which are used to balance the relationship between health and learning, respectively, satisfying β1+β2=1; P H (t) represents the health prediction score at time t, which represents the evaluation of personalized strategy in health management; P L (t) represents the learning prediction score at time t, which indicates the evaluation of the personalized strategy in learning guidance; R i (t-1) represents the personalized growth report score of child i at time t-1; The use of natural language processing technology to analyze parental feedback includes: Parents' feedback text i (t)Converted into a quantitative satisfaction score S i (t) and the demand vector D i (t), and then use the sentiment analysis function ψ(F i (t)) and the demand analysis function ξ(F i (t)), combined with the content generated by the report, is quantified as follows: in, represents the comprehensive scoring function of the growth report, and the personalized growth report score R i (t) is converted into a quantitative score; γ represents a smoothing parameter, which is used to adjust the impact of growth reports on satisfaction; The Q learning update rule introduced in the reinforcement learning mechanism for strategy optimization based on parent feedback data is expressed as follows: Among them, Q i (t+1) represents the strategy value function of child i at time t+1, which is used to evaluate the effectiveness of the strategy; Q i (t) represents the strategy value function of child i at time t; Represents the learning rate, the control strategy optimization step size, and is a value between [0,1]; S i (t) represents the parent's satisfaction score for child i at time t; δ represents the discount factor, which measures the weight of future rewards; max a′ Q i (t+1,a') represents the maximum policy value among all actions a' at time t+1, indicating the tendency to select the best policy.

10. The kindergarten data management system based on digital twin and blockchain according to claim 9 is characterized in that: The demand vector D i (t), is calculated as follows: D i (t)=ξ(F i (t))⊙M i (t) Among them, D i (t) represents the demand vector, which indicates the parents’ demands in different aspects, extracted from the feedback text; ξ(F i (t)) represents the demand analysis function, which extracts the specific needs in the parents’ feedback; M i (t) represents the weight vector, which indicates the weight of each requirement; ⊙ represents the element-by-element multiplication symbol, which indicates that the elements in the vector are multiplied accordingly.

Images