Intelligent medical data security guarantee system and method

Abstract

The invention relates to the technical field of data security, and discloses an intelligent medical data security guarantee system and method, and the system defines a multi-dimensional state space containing data classification, access authority, network security state and use frequency, and an action space containing multiple actions such as access limitation, data encryption and equipment isolation. And comprehensive monitoring and dynamic protection of medical data security are realized. And training the model by using a deep reinforcement learning algorithm, dynamically adjusting protection measures according to real-time data and security threats, and optimizing a strategy to maximize accumulated rewards. According to the method, reduction of data security events, data integrity, system operation stability, user satisfaction and protection cost are comprehensively considered, and a reward function is designed to evaluate an action result. According to the method, complex and changeable network attacks and security threats can be effectively handled, the security and privacy of medical data are improved, and a security guarantee is provided for sustainable development of intelligent medical treatment.

Description

Technical Field

[0001] The present invention relates to the field of data security technology, and in particular to a smart medical data security assurance system and method. Background Art

[0002] With the rapid development of information technology and the continuous advancement of medical informatization, smart healthcare has become an important part of the modern medical system. Smart healthcare has achieved the optimal allocation of medical resources, intelligent medical services, and refined medical management by integrating advanced technologies such as the Internet of Things, big data, cloud computing, and artificial intelligence. However, with the widespread application of smart healthcare, the security of medical data has become increasingly prominent, becoming an important factor restricting its further development.

[0003] Medical data, including basic patient information, diagnosis records, treatment records, medical imaging materials, and drug use records, are the foundation and core of medical activities. These data are extremely sensitive and private. Once leaked or illegally accessed, they may not only infringe on the privacy rights of patients, but also have a serious impact on the reputation and operation of medical institutions, and even threaten the life safety of patients.

[0004] Current data security protection methods, such as firewalls, intrusion detection systems, and data encryption, can protect the security of medical data to a certain extent, but in the face of increasingly complex and changing network attacks and security threats, these static protection measures seem to be powerless. On the one hand, traditional protection methods are often based on known security threats and are difficult to deal with new attacks and unknown vulnerabilities; on the other hand, the security requirements of medical data are constantly changing with the changes in medical business, and static protection measures are difficult to adapt to this dynamic nature. Summary of the invention

[0005] The purpose of the present invention is to provide a smart medical data security protection system and method to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a smart medical data security protection method, the method comprising:

[0007] Build an intelligent protection model for smart medical data security protection, model the medical data protection process as a dynamic decision-making process, and dynamically adjust data protection measures by analyzing the real-time data and security threats of the medical system;

[0008] The steps to build an intelligent protection model include:

[0009] Define the state space, including the data classification dimension, data access permission dimension, network security status dimension and data usage frequency dimension of the medical data system; the data classification dimension includes basic patient information, diagnosis records, treatment records, medical imaging materials, drug use records and other relevant medical data; the data access permission dimension includes the access permission level of different user roles, access history records, access time patterns and access location patterns; the network security status dimension includes network intrusion detection status, virus scanning results, data encryption status and firewall configuration; the data usage frequency dimension includes the frequency of access to various types of medical data, access duration and access peak time period;

[0010] Define an action space, which includes the actions that the intelligent protection model may take when it detects data security threats or abnormal access behaviors, including restricting access rights, encrypting sensitive data, isolating infected devices, notifying security administrators, and initiating emergency response mechanisms;

[0011] Design a reward function to evaluate the results of actions and give corresponding rewards or penalties based on the reduction rate of data security incidents, data integrity maintenance, system operation stability, user satisfaction and protection costs;

[0012] Formulate strategies, rules for selecting actions in a given state, and continuously optimize strategies to maximize cumulative rewards;

[0013] A deep reinforcement learning algorithm is used to train the intelligent protection model, so that the model can dynamically adjust data protection measures according to the real-time changing data security status.

[0014] Preferably, a deep deterministic policy gradient DDPG algorithm is used to train the intelligent protection model, which includes an input layer, a hidden layer and an output layer. The input layer receives feature data in the state space, the hidden layer performs feature extraction and decision learning, and the output layer outputs action selection in the action space.

[0015] Preferably, the step of training the intelligent protection model includes:

[0016] S1: Initialize the deep neural network: Build a deep neural network as a policy function approximator, with the current system state s as input and the access control policy a as output; initialize the network parameters θ;

[0017] S2: Initialize the experience replay buffer D: Set the buffer size N to store the experience generated by the model during training, including state, action, reward, and next state;

[0018] S3: Use the collected training data to update the parameters of the neural network through the back propagation algorithm and the gradient descent method to maximize the cumulative reward; the formula for updating the parameters of the gradient descent algorithm is:

[0019]

[0020] Among them, θ represents the model parameters, including all weights and biases that the model needs to learn; ω represents the learning rate, which is used to control the step size of parameter update; represents the gradient of the loss function J(θ) with respect to the parameter θ, is a vector pointing to the direction where the loss function grows fastest; := represents the assignment operation, that is, updating the value of the parameter θ;

[0021] S4: During the training process, the loss function is used to regularly evaluate the performance of the model, and the weight coefficient of the reward function, the learning rate of the model, and the balance parameters of exploration and exploitation are adjusted according to the evaluation results;

[0022] S5: Repeat steps S3 and S4 until a preset number of training rounds is reached or performance requirements are met.

[0023] Preferably, in step S4, the loss function adopts the mean square error MSE algorithm, and its calculation formula is:

[0024]

[0025] Where n is the number of data points, Y pred,i is the predicted value of the model, Y actual,i is the actual value; the prediction accuracy of the model is evaluated according to the MSE value. The smaller the MSE value, the higher the prediction accuracy of the model.

[0026] Preferably, the design of the reward function comprises the following steps:

[0027] Determine the key indicators of the reward function, including the reduction rate of data security incidents, data integrity retention rate, system stable operation time, user access satisfaction, and protection cost savings;

[0028] Quantify each key indicator to obtain the corresponding indicator value;

[0029] According to the importance of each key indicator, a corresponding weight coefficient is assigned to it. The value range of the weight coefficient is 0 to 1, and the sum of all weight coefficients is 1;

[0030] Design a reward function formula, substitute the quantified key indicator value and the corresponding weight coefficient into the formula for calculation to obtain the reward value; the reward function formula is:

[0031]

[0032] Where R represents the reward value, w1, w2, w3, w4, and w5 represent the weight coefficients of the reduction rate of data security incidents, the data integrity retention rate, the system stable operation time, the user access satisfaction, and the protection cost savings, respectively; r1, r2, r3, and r4 represent the quantitative index values ​​of the reduction rate of data security incidents, the data integrity retention rate, the system stable operation time, and the user access satisfaction, respectively; and c represents the quantitative index value of the protection cost savings;

[0033] According to the reward value calculated by the reward function formula, the actions taken by the intelligent protection model are evaluated and corresponding rewards or penalties are given.

[0034] Preferably, the training data of the intelligent protection model is obtained by:

[0035] Generate simulation data containing multiple attack types and corresponding protection measures according to preset security threat scenarios and protection strategies; the simulation data includes network attack data, malware behavior data, system vulnerability exploitation data and corresponding protection response data;

[0036] The data in the simulated training data set is encoded and standardized to form a standardized training data set.

[0037] Preferably, the method of encoding the data in the simulated training data set includes: performing one-hot encoding conversion on discrete features.

[0038] Preferably, the method of standardizing the data in the simulated training data set includes: using the Z-score standardization method to standardize the continuous features, specifically: subtracting the mean of each numerical data and dividing it by its standard deviation so that the processed data conforms to the standard normal distribution, and the standardization formula is: Z = (X-μ) / σ, wherein X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data.

[0039] Preferably, the training data set is further divided into a training set and a validation set. After the training is completed, the model is validated using an independent validation set to evaluate its ability to identify attacks and generate protective measures; and the model is adjusted and optimized based on the validation results.

[0040] Preferably, a smart medical data security system comprises:

[0041] The information status monitoring module is used to collect and analyze the data classification information, data access permission status, network security status and data usage frequency of the medical data system in real time to build a status space;

[0042] The action execution module executes corresponding data protection measures according to the decision of the intelligent protection model;

[0043] The reward function calculation module evaluates the results of the action execution module based on the reduction rate of data security incidents, data integrity maintenance, system operation stability, user satisfaction and protection cost;

[0044] The strategy optimization module formulates and selects the optimal action strategy based on the current state and the output of the reward function calculation module;

[0045] Model training module, which uses deep reinforcement learning algorithm to train the intelligent protection model;

[0046] The system integration and control module is responsible for the integration and coordination between modules; it receives data input from the status monitoring module and guides the operation of the action execution module through the strategy optimization module and the model training module.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] The present invention defines a comprehensive state space, including multiple dimensions such as data classification, data access rights, network security status, and data usage frequency of the medical data system, so that the intelligent protection model can perceive the data security status of the medical system in real time. By defining the action space and designing the reward function, the model can dynamically adjust data protection measures according to real-time data and security threats, and effectively respond to complex and changeable network attacks and security threats. The present invention has formulated refined data protection measures based on the sensitivity and privacy of medical data, as well as the access rights and access behaviors of different user roles. When data security threats or abnormal access behaviors are detected, the model can quickly take targeted measures such as restricting access rights, encrypting sensitive data, and isolating infected devices to effectively prevent data leakage and illegal access, and ensure the security and privacy of medical data.

[0049] The present invention uses a deep reinforcement learning algorithm to train an intelligent protection model, so that the model has the ability of self-learning and optimization. Through continuous iterative training, the model can gradually adapt to changes in medical services and the evolution of security threats, continuously optimize data protection measures, and improve the efficiency and accuracy of data security protection. Through intelligent and automated data security protection methods, the reliance on traditional protection methods is reduced and the cost of protection is reduced. By dynamically adjusting data protection measures, the system performance degradation and resource waste caused by excessive protection are avoided, and the operating efficiency and user experience of the medical system are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A step diagram of a smart medical data security assurance method according to the present invention;

[0051] Figure 2 This is a diagram of the training steps of the intelligent protection model;

[0052] Figure 3 Diagram of the steps for designing a reward function. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] See also Figure 1-3 The present invention provides a technical solution: a method for ensuring the security of smart medical data, the steps of the method comprising:

[0055] The core of this model is to regard the protection of medical data as a dynamic decision-making process, and dynamically adjust data protection measures by analyzing the data and security threats of the medical system in real time.

[0056] When building an intelligent protection model, the state space is first defined. The state space includes multiple dimensions of the medical data system, specifically:

[0057] Data classification dimensions: including basic patient information, diagnosis records, treatment records, medical imaging materials, drug use records and other related medical data. These data are the basis of medical activities and the focus of security protection.

[0058] Data access permission dimensions: involves the access permission levels of different user roles, access history, access time patterns, and access location patterns. By monitoring these parameters, abnormal access behaviors can be discovered in a timely manner.

[0059] Network security status dimension: including network intrusion detection, virus scanning results, data encryption status and firewall configuration. These parameters reflect the current network security status of the medical system.

[0060] Data usage frequency dimension: Record the frequency of access to various types of medical data, access duration, and access peak hours. By analyzing this data, we can understand the data usage pattern and optimize protection measures.

[0061] Define the action space. The action space contains a series of actions that the intelligent protection model may take when it detects data security threats or abnormal access behaviors, including:

[0062] Restrict access rights: restrict abnormal access behaviors to prevent data leakage.

[0063] Encrypt sensitive data: Encrypt sensitive data to improve data security.

[0064] Isolate infected devices: Isolate infected devices from the network to prevent the spread of viruses or malware.

[0065] Notify the security administrator: Notify the security administrator promptly to handle the issue and ensure that the problem is resolved in a timely manner.

[0066] Activate the emergency response mechanism: When a serious security incident occurs, activate the emergency response mechanism to respond quickly and reduce losses.

[0067] In order to evaluate the effect of the action, a reward function is designed. The reward function evaluates the results of the action based on multiple indicators such as the reduction rate of data security incidents, data integrity, system operation stability, user satisfaction, and protection costs, and gives corresponding rewards or penalties, which helps the intelligent protection model to continuously optimize the strategy during the training process.

[0068] When formulating strategies, the rules for selecting actions based on given states are used to continuously optimize strategies through machine learning algorithms to maximize cumulative rewards, ensuring that the intelligent protection model can make optimal decisions in a constantly changing security environment.

[0069] Finally, a deep reinforcement learning algorithm is used to train the intelligent protection model. Through a large amount of training data, the model can learn the dynamic changes in medical data security and dynamically adjust data protection measures according to real-time data, thus realizing intelligent protection of medical data security and effectively improving data security.

[0070] The present invention will be further described below in conjunction with Examples 1 to 3:

[0071] Embodiment 1:

[0072] The present invention provides a method for training an intelligent protection model based on a deep deterministic policy gradient (DDPG) algorithm. The intelligent protection model consists of an input layer, a hidden layer, and an output layer, wherein the input layer receives feature data in the state space, the hidden layer is responsible for feature extraction and decision learning, and the output layer outputs action selection in the action space. The following are the specific implementation steps of the method:

[0073] S1. Initialize the deep neural network:

[0074] A deep neural network is constructed as a policy function approximator. The input of the network is the current system state s, and the output is the access control policy a. The network structure includes an input layer, several hidden layers, and an output layer.

[0075] Initialize the network parameters θ, which include all weights and biases in the network, which will be optimized during the training process.

[0076] S2. Initialize the experience replay buffer D:

[0077] Set a buffer of size N to store the experience generated by the model during training. These experiences include state s, action a taken, reward r obtained, and the next state s' after executing the action.

[0078] The experience replay mechanism allows the model to learn from past experience, improving training efficiency and stability.

[0079] S3. Update the neural network parameters using the collected training data:

[0080] Randomly sample a batch of experience data from the experience replay buffer D.

[0081] Use the back propagation algorithm and gradient descent method to update the parameters θ of the neural network to maximize the cumulative reward. The formula for updating the parameters of the gradient descent algorithm is:

[0082]

[0083] Among them, θ represents the model parameters, including all weights and biases that the model needs to learn; ω represents the learning rate, which is used to control the step size of parameter update; represents the gradient of the loss function J(θ) with respect to the parameter θ, is a vector pointing to the direction where the loss function grows fastest; := represents the assignment operation, that is, updating the value of the parameter θ.

[0084] S4. Regularly evaluate model performance and adjust parameters:

[0085] During the training process, the performance of the model is regularly evaluated using a loss function, which can be achieved by comparing the difference between the access control policy predicted by the model and the actual optimal policy. The loss function uses the mean square error MSE algorithm, and its calculation formula is:

[0086]

[0087] Where n is the number of data points, Y pred,i is the predicted value of the model, Y actual,i is the actual value; the prediction accuracy of the model is evaluated according to the MSE value. The smaller the MSE value, the higher the prediction accuracy of the model.

[0088] According to the evaluation results, adjust the weight coefficient of the reward function, the learning rate α of the model, and the balance parameter between exploration and utilization (such as the ε value in the ε-greedy strategy).

[0089] S5. Repeat the training until the following conditions are met:

[0090] Repeat steps S3 and S4 until the preset number of training rounds is reached or the performance of the model meets specific requirements, which can be determined by monitoring the decrease of the loss function or the performance of the model on the validation set.

[0091] During training, the model's parameters can be saved periodically so they can be restored or further analyzed when needed.

[0092] Assume that the smart medical data security intelligent protection model needs to protect a database containing basic patient information, diagnosis records, and treatment records. The state space may include characteristics such as data classification (such as patient information, diagnosis records, etc.), access permission level, network security status (such as whether there is intrusion detection), and data usage frequency. The action space may include actions such as restricting access permissions, encrypting sensitive data, and notifying security administrators.

[0093] During the training process, the model continuously receives feature data in the state space as input, and performs feature extraction and decision learning through the hidden layer. The output layer selects the most appropriate action to protect data security based on the learned strategy. Through repeated iterative training, the model can gradually learn how to dynamically adjust data protection measures according to the current security status, thereby effectively responding to various security threats.

[0094] Embodiment 2:

[0095] This embodiment is used to describe the specific implementation of the reward function design. The following is a detailed description with reference to a specific example.

[0096] First, determine the key indicators of the reward function, including the reduction rate of data security incidents, data integrity retention rate, system stable operation time, user access satisfaction and protection cost savings. These indicators can comprehensively reflect the effectiveness and cost of data security protection.

[0097] Quantify each key indicator. Taking the reduction rate of data security incidents as an example, the number of data security incidents before and after the implementation of protective measures can be counted, and the percentage of reduction can be calculated as the quantitative indicator value. Similarly, the data integrity retention rate can be quantified by checking the integrity and consistency of the data; the system stable operation time can be quantified by recording the time when the system operates without failure; user access satisfaction can be quantified through user surveys or feedback systems; and the amount of protection cost savings can be quantified by comparing the cost difference before and after the implementation of protective measures.

[0098] According to the importance of each key indicator, a corresponding weight coefficient is assigned to it. Assuming that the reduction rate of data security incidents is the most important, the weight coefficient is assigned to 0.4; the data integrity retention rate is second, and the weight coefficient is assigned to 0.3; the system stable operation time and user access satisfaction are assigned weight coefficients of 0.15 respectively; the protection cost savings are assigned a weight coefficient of 0.1. The allocation of these weight coefficients can be adjusted according to actual conditions.

[0099] Design the reward function formula, and substitute the quantified key indicator values ​​and corresponding weight coefficients into the formula for calculation. The reward function formula is as follows:

[0100] R=w1·r1+w2·r2+w3·r3+w4·r4-w5·c

[0101] Among them, R represents the reward value, w1, w2, w3, w4, and w5 represent the weight coefficients of the reduction rate of data security incidents, the data integrity retention rate, the system stable operation time, the user access satisfaction, and the protection cost savings, respectively; r1, r2, r3, and r4 represent the quantitative index values ​​of the reduction rate of data security incidents, the data integrity retention rate, the system stable operation time, and the user access satisfaction, respectively; and c represents the quantitative index value of the protection cost savings.

[0102] The actions taken by the intelligent protection model are evaluated based on the reward value calculated by the reward function formula. For example, in a certain data protection process, the intelligent protection model took the action of restricting access rights. Subsequently, statistics showed that the reduction rate of data security incidents was 20%, the data integrity retention rate was 95%, the system stable operation time was 72 hours, the user access satisfaction was 85%, and the protection cost savings was 1,000 yuan. Substituting these quantitative indicator values ​​into the reward function formula, the reward value is calculated as:

[0103] R=0.4×20%+0.3×95%+0.15×72+0.15×85-0.1×1000=16.85

[0104] Based on the calculated reward value, the actions taken by the intelligent protection model can be evaluated and given corresponding rewards or penalties. If the reward value is high, it means that the action taken is effective, and a positive reward can be given to encourage the model to continue taking similar actions; if the reward value is low or negative, it means that the action taken is ineffective or incurs additional costs, and a negative reward or penalty can be given to encourage the model to adjust its strategy and optimize data protection measures.

[0105] Embodiment 3:

[0106] In order to build the training data set for the intelligent protection model, simulated data containing multiple attack types and corresponding protection measures are generated according to the preset security threat scenarios and protection strategies. These simulated data cover network attack data, malware behavior data, system vulnerability exploitation data, and corresponding protection response data to ensure the comprehensiveness and diversity of the training data.

[0107] The data in the simulated training data set is preprocessed to meet the requirements of model training. The preprocessing process includes two steps: encoding and standardization.

[0108] In the encoding processing stage, the One-Hot Encoding conversion method is used for the discrete features in the simulated training data set. Specifically, for each discrete feature, a corresponding number of binary bits are created according to the number of possible values, and each bit represents a possible value. When the feature takes a certain value, the corresponding binary bit is set to 1, and the remaining bits are 0. In this way, the discrete features are converted into a numerical form that can be directly processed by the model.

[0109] In the standardization stage, the Z-score standardization method is used to standardize the continuous features in the simulated training data set. Specifically, for each continuous feature, its mean and standard deviation are calculated. Then, each numerical data is subtracted from its mean and divided by its standard deviation so that the processed data conforms to the standard normal distribution. The formula for standardization is: Z = (X-μ) / σ, where X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data. Through standardization, the dimensional differences between different features can be eliminated and the training effect of the model can be improved.

[0110] After the data is encoded and standardized, the processed training data set is further divided into a training set and a validation set. The training set is used in the model training process, while the validation set is used to validate the model after the training is completed. The validation set is used to evaluate the model's ability to identify attacks and generate protective measures, which can objectively reflect the performance of the model. Based on the validation results, the model is adjusted and optimized to improve its accuracy and robustness in practical applications.

[0111] The present invention also includes a smart medical data security system, the system comprising:

[0112] The information status monitoring module is used to collect and analyze the data classification information, data access permission status, network security status and data usage frequency of the medical data system in real time to build a status space;

[0113] The action execution module executes corresponding data protection measures according to the decision of the intelligent protection model;

[0114] The reward function calculation module evaluates the results of the action execution module based on the reduction rate of data security incidents, data integrity maintenance, system operation stability, user satisfaction and protection cost;

[0115] The strategy optimization module formulates and selects the optimal action strategy based on the current state and the output of the reward function calculation module;

[0116] Model training module, which uses deep reinforcement learning algorithm to train the intelligent protection model;

[0117] The system integration and control module is responsible for the integration and coordination between modules; it receives data input from the status monitoring module and guides the operation of the action execution module through the strategy optimization module and the model training module.

[0118] The implementation of the system refers to the above embodiment and will not be described in detail in the specification.

[0119] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0120] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A smart medical data security method, characterized in that: The method comprises: Build an intelligent protection model for smart medical data security protection, model the medical data protection process as a dynamic decision-making process, and dynamically adjust data protection measures by analyzing the real-time data and security threats of the medical system; The steps to build an intelligent protection model include: Define the state space, including the data classification dimension, data access permission dimension, network security status dimension and data usage frequency dimension of the medical data system; the data classification dimension includes basic patient information, diagnosis records, treatment records, medical imaging materials, drug use records and other relevant medical data; the data access permission dimension includes the access permission level of different user roles, access history records, access time patterns and access location patterns; the network security status dimension includes network intrusion detection status, virus scanning results, data encryption status and firewall configuration; the data usage frequency dimension includes the frequency of access to various types of medical data, access duration and access peak time period; Define an action space, which includes the actions that the intelligent protection model may take when it detects data security threats or abnormal access behaviors, including restricting access rights, encrypting sensitive data, isolating infected devices, notifying security administrators, and initiating emergency response mechanisms; Design a reward function to evaluate the results of actions and give corresponding rewards or penalties based on the reduction rate of data security incidents, data integrity maintenance, system operation stability, user satisfaction and protection costs; Formulate strategies, rules for selecting actions in a given state, and continuously optimize strategies to maximize cumulative rewards; A deep reinforcement learning algorithm is used to train the intelligent protection model, so that the model can dynamically adjust data protection measures according to the real-time changing data security status.

2. The method for ensuring the security of smart medical data according to claim 1, characterized in that: The deep deterministic policy gradient DDPG algorithm is used to train the intelligent protection model, which includes an input layer, a hidden layer and an output layer. The input layer receives feature data in the state space, the hidden layer performs feature extraction and decision learning, and the output layer outputs action selection in the action space.

3. A smart medical data security assurance method according to claim 2, characterized in that: The steps to train the smart protection model include: S1: Initialize the deep neural network: Build a deep neural network as a policy function approximator, with the current system state s as input and the access control policy a as output; initialize the network parameters θ; S2: Initialize the experience replay buffer D: Set the buffer size N to store the experience generated by the model during training, including state, action, reward, and next state; S3: Use the collected training data to update the parameters of the neural network through the back propagation algorithm and the gradient descent method to maximize the cumulative reward; the formula for updating the parameters of the gradient descent algorithm is: Among them, θ represents the model parameters, including all weights and biases that the model needs to learn; ω represents the learning rate, which is used to control the step size of parameter update; represents the gradient of the loss function J(θ) with respect to the parameter θ, is a vector pointing to the direction where the loss function grows fastest; := represents the assignment operation, that is, updating the value of the parameter θ; S4: During the training process, the loss function is used to regularly evaluate the performance of the model, and the weight coefficient of the reward function, the learning rate of the model, and the balance parameters of exploration and exploitation are adjusted according to the evaluation results; S5: Repeat steps S3 and S4 until a preset number of training rounds is reached or performance requirements are met.

4. A smart medical data security assurance method according to claim 3, characterized in that: In step S4, the loss function adopts the mean square error MSE algorithm, and its calculation formula is: Where n is the number of data points, Y pred,i is the predicted value of the model, Y actual,i is the actual value; the prediction accuracy of the model is evaluated according to the MSE value. The smaller the MSE value, the higher the prediction accuracy of the model.

5. The method for ensuring the security of smart medical data according to claim 1, characterized in that: The design of the reward function includes the following steps: Determine the key indicators of the reward function, including the reduction rate of data security incidents, data integrity retention rate, system stable operation time, user access satisfaction, and protection cost savings; Quantify each key indicator to obtain the corresponding indicator value; According to the importance of each key indicator, a corresponding weight coefficient is assigned to it. The value range of the weight coefficient is 0 to 1, and the sum of all weight coefficients is 1; Design a reward function formula, substitute the quantified key indicator value and the corresponding weight coefficient into the formula for calculation to obtain the reward value; the reward function formula is: R=w1·r1+w2·r2+w3·r3+w4·r4-w5·c Where R represents the reward value, w1, w2, w3, w4, and w5 represent the weight coefficients of the reduction rate of data security incidents, the data integrity retention rate, the system stable operation time, the user access satisfaction, and the protection cost savings, respectively; r1, r2, r3, and r4 represent the quantitative index values ​​of the reduction rate of data security incidents, the data integrity retention rate, the system stable operation time, and the user access satisfaction, respectively; and c represents the quantitative index value of the protection cost savings; According to the reward value calculated by the reward function formula, the actions taken by the intelligent protection model are evaluated and corresponding rewards or penalties are given.

6. A smart medical data security assurance method according to claim 4, characterized in that: The training data of the intelligent protection model is obtained as follows: Generate simulation data containing multiple attack types and corresponding protection measures according to preset security threat scenarios and protection strategies; the simulation data includes network attack data, malware behavior data, system vulnerability exploitation data and corresponding protection response data; The data in the simulated training data set is encoded and standardized to form a standardized training data set.

7. A smart medical data security assurance method according to claim 6, characterized in that: The method of encoding the data in the simulated training data set includes: performing one-hot encoding conversion on discrete features.

8. A smart medical data security assurance method according to claim 6, characterized in that: The method of standardizing the data in the simulated training data set includes: using the Z-score standardization method to standardize the continuous features, specifically: subtracting the mean of each numerical data and dividing it by its standard deviation, so that the processed data conforms to the standard normal distribution, and the standardization formula is: Z = (X-μ) / σ, where X is the original data, μ is the mean of the original data, σ is the standard deviation of the original data, and Z is the standardized data.

9. The method for ensuring the security of smart medical data according to claim 6, characterized in that: The training data set is further divided into a training set and a validation set. After the training is completed, the model is validated using an independent validation set to evaluate its ability to identify attacks and generate protective measures; and the model is adjusted and optimized based on the validation results.

10. A smart medical data security system, characterized in that: The system comprises: The information status monitoring module is used to collect and analyze the data classification information, data access permission status, network security status and data usage frequency of the medical data system in real time to build a status space; The action execution module executes corresponding data protection measures according to the decision of the intelligent protection model; The reward function calculation module evaluates the results of the action execution module based on the reduction rate of data security incidents, data integrity maintenance, system operation stability, user satisfaction and protection cost; The strategy optimization module formulates and selects the optimal action strategy based on the current state and the output of the reward function calculation module; Model training module, which uses deep reinforcement learning algorithm to train the intelligent protection model; The system integration and control module is responsible for the integration and coordination between modules; it receives data input from the status monitoring module and guides the operation of the action execution module through the strategy optimization module and the model training module.

Images