Transcranial alternating current stimulation method and equipment

Abstract

The invention discloses a transcranial alternating current stimulation method and equipment. The method comprises the following steps: dynamically acquiring an electroencephalogram signal of a subject; performing real-time analysis on the electroencephalogram signal by using a closed-loop neural regulation network model to determine the probability that the subject is in a target state, and obtaining dynamic regulation data of electrical stimulation according to the probability that the subject is in the target state; according to the dynamic regulation and control data of the electrical stimulation, the transcranial alternating current electrical stimulation signal is dynamically regulated so as to regulate the brain activity of the subject in real time; and continuing to dynamically obtain the electroencephalogram signal of the subject after the electrical stimulation so as to adjust the brain activity of the subject in real time according to the electroencephalogram signal of the subject. According to the technical scheme, electroencephalogram signals from the subject can be collected and analyzed in real time, the stimulation mode is dynamically adjusted according to the state of the subject, controllable stimulation is provided for neurons, and precision and high individuation of transcranial alternating current stimulation are achieved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and more specifically, to a transcranial alternating current stimulation method and device. Background Art

[0002] Transcranial Alternating Current Stimulation (tACS) is an innovative non-invasive brain stimulation technology that enhances the brain's endogenous neural oscillation activity in a spectrum-specific manner by regulating the alternating current activity of brain neurons. Previous studies have shown that brain stimulation produced by tACS can help improve psychological functions and have different effects on the psychological functions of subjects.

[0003] However, the actual application of tACS is often limited by the lack of individualization of stimulation parameters. It mostly uses pre-set parameters to carry out short-term continuous neural stimulation regulation, without considering feedback factors such as changes in physiological activities during the treatment process. It is an open-loop intervention. This form may lead to excessive stimulation or cause side effects due to the lack of a certain degree of regulatory flexibility.

[0004] Therefore, how to design a transcranial alternating current stimulation method to achieve automated, precise and highly personalized regulation has become a problem that needs to be solved in this field. Summary of the invention

[0005] In view of this, in a first aspect, the present application proposes a transcranial alternating current stimulation method, the method comprising:

[0006] Dynamically obtain the subject's EEG signal;

[0007] Using a closed-loop neural control network model, the EEG signal is analyzed in real time to determine the probability that the subject is in a target state, and dynamic control data of electrical stimulation is obtained according to the probability that the subject is in the target state;

[0008] Dynamically adjusting the transcranial alternating current stimulation signal according to the dynamic regulation data of the electrical stimulation to adjust the brain activity of the subject in real time;

[0009] Continue to dynamically acquire the EEG signal of the subject after electrical stimulation, so as to adjust the brain activity of the subject in real time according to the EEG signal of the subject.

[0010] Preferably, after dynamically acquiring the subject's EEG signal, the method further comprises:

[0011] The EEG signal of the subject is subjected to low-pass filtering and amplitude limiting processing.

[0012] Preferably, a closed-loop neural control network model is used to perform real-time analysis on the EEG signal to determine the probability that the subject is in the target state, and dynamic control data of electrical stimulation is obtained according to the probability that the subject is in the target state, including:

[0013] Receiving the EEG signal of the subject in real time, and marking the EEG signal of the subject according to a time window;

[0014] Identify the target state features in the current EEG signal, obtain the probability that the current subject is in the target state based on the target state features in the current EEG signal, and obtain dynamic control data of electrical stimulation based on the probability that the current subject is in the target state and the probability that the subject was in the target state in the previous time window.

[0015] Further preferably, the closed-loop neural regulation network model includes three hidden layers.

[0016] Further preferably, the dynamic control data of the electrical stimulation is obtained according to the probability that the current subject is in the target state and the probability that the subject is in the target state in the previous time window, including:

[0017] According to the probability that the current subject is in the target state and the probability that the subject is in the target state in the previous time window, the target state change in two adjacent time windows is calculated;

[0018] According to the target state change, the dynamic control data of the electrical stimulation is obtained.

[0019] Further preferred:

[0020] The dynamic control data of the electrical stimulation includes: control data of the stimulation frequency and / or control data of the stimulation intensity;

[0021] The calculation formula of the regulation data of the stimulation frequency is:

[0022] f t+1 =f t ―α f ·Δo t ;

[0023] The calculation formula of the regulation data of the stimulation intensity is:

[0024] I t+1 =I t ―α I ·Δo t ;

[0025] Among them, f t+1 represents the control data of the stimulation frequency required at the next moment, f t Indicates the control data of the current stimulation frequency, I t+1Indicates the control data of the stimulation intensity required at the next moment, I t Represents the control data of the current stimulus intensity, α f represents the step size of the stimulation frequency in the iteration process, α I represents the step size of the stimulus intensity during the iteration, Δo t Indicates the target state change.

[0026] Further preferably, the method further comprises:

[0027] Calculate the control result reward value, the control result reward value r t The calculation formula is:

[0028] r EEG =―Δo t =―(o t ―o t―1 );

[0029] r t =w1·r EEG +w2·r feedback ;

[0030] Among them, r EEG represents the objective evaluation value of the target state controlled by the closed-loop neural control network model, r feedback represents the subjective evaluation value of the subject on the regulation target state, r EEG and r feedback The value range is [0, 1], w1 represents the balance r EEG The weight of importance, w2 represents the balance r feedback The weight of importance, o t represents the probability that the current subject is in the target state, o t―1 Represents the probability that the subject is in the target state in the previous time window;

[0031] The step length of the stimulation frequency and the step length of the stimulation intensity in the iteration process are adjusted according to the reward value of the regulation result. The calculation formula is:

[0032] a′ f =a f +η·r t ·Δo t ;

[0033] a′ I =a I +η·r t ·Δo t ;

[0034] Wherein, η represents the learning rate of the closed-loop neural regulation network model, a ′frepresents the step length of the adjusted stimulation frequency in the iteration process, a′ I Represents the step size of the adjusted stimulus intensity during the iteration process.

[0035] Further preferably, before using the closed-loop neural regulation network model, the method further comprises:

[0036] The closed-loop neural regulation network model is trained using training samples, wherein the training samples contain the target state characteristics and the corresponding target state probabilities, so that the closed-loop neural regulation network model can recognize the target state characteristics in the EEG signal and confirm the corresponding target state probabilities.

[0037] Preferably, the method further comprises:

[0038] Outputting the subject's EEG signal and the corresponding transcranial alternating current stimulation signal to a remote monitoring terminal;

[0039] Receive the adjustment signal input by the remote monitoring terminal to adjust the transcranial alternating current stimulation signal.

[0040] In a second aspect, an embodiment of the present invention further provides a transcranial alternating current stimulation device, the device comprising: a housing, electrodes, a data processing terminal, a user operation terminal and a power supply;

[0041] Wherein, the electrodes are used to capture the EEG signals of the subject;

[0042] The data processing end includes a processor, which is used to implement the method described in the first aspect above.

[0043] The transcranial alternating current stimulation method provided in the present application utilizes a closed-loop neural control network model to perform real-time closed-loop feedback analysis of EEG signals to dynamically adjust the transcranial alternating current stimulation signals, thereby adjusting the brain activity of the subject in real time. During this process, the present application can collect and analyze EEG signals from the subject in real time, dynamically adjust the stimulation mode according to the subject's state, provide controllable stimulation to neurons, which is beneficial to the plasticity of neural circuits, realizes the precision and high personalization of transcranial alternating current stimulation, improves the stimulation effect and reduces possible side effects.

[0044] Other features and advantages of the present application will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings constituting a part of the present application are used to provide a further understanding of the present application, and the schematic implementation modes and descriptions of the present application are used to explain the present application. In the accompanying drawings:

[0046] Figure 1 A flowchart of a transcranial alternating current stimulation method according to a preferred embodiment of the present invention;

[0047] Figure 2 A schematic diagram of a closed-loop neural regulatory network model architecture for a preferred embodiment of the application;

[0048] Figure 3 A schematic diagram of a transcranial alternating current stimulation device according to a preferred embodiment of the application. DETAILED DESCRIPTION

[0049] The technical solution of the present application will be described in detail below with reference to the accompanying drawings and in combination with the implementation modes.

[0050] First, this application proposes a transcranial alternating current stimulation method, such as Figure 1 As shown, steps 110-140 are included:

[0051] Step 110: dynamically acquiring the subject's EEG signal;

[0052] Specifically, electroencephalogram (EEG) is an electrical signal generated by the activity of brain neurons. It can be captured and recorded by electrodes. By analyzing the electrical activity of the brain, it can help understand brain function.

[0053] However, EEG signals are easily contaminated by noise, and EEG signals recorded directly from scalp electrodes cannot accurately represent brain neural signals. Therefore, the subject's EEG signals need to be low-pass filtered and limited to minimize or eliminate the impact of artifacts.

[0054] In a specific embodiment, to improve signal quality, each electrode acquisition channel is provided with a second-order passive RC (Resistor and Capacitor) low-pass filter and a limiting circuit. The second-order passive RC low-pass filter is used to suppress high-frequency noise and is composed of two resistors (R1, R2) and two capacitors (C1, C2). The principle is based on the characteristics of the RC circuit and filtering of signals of different frequencies is achieved by selecting appropriate resistance and capacitance values.

[0055] For the design of a second-order passive RC low-pass filter, the frequency response function H(s) can be calculated using equation (1):

[0056]

[0057] Among them, C1 and C2 represent the values ​​of two capacitors, R1 and R2 represent the values ​​of two resistors, and s represents a complex frequency variable. By selecting the values ​​of C1, C2, R1 and R2 according to individual differences, differentiated signal quality optimization can be achieved, effectively suppressing high-frequency noise while retaining the effective frequency components of the EEG signal to ensure the accuracy of subsequent analysis.

[0058] The limiting circuit is used to limit the signal amplitude within a safe range. The clamping effect of the two diodes can be simulated through software to ensure that the signal amplitude is stable within the required range, ensure the purity of the EEG signal, and provide a stable input for subsequent analog-to-digital conversion.

[0059] Step 120: using a closed-loop neural control network model, performing real-time analysis on the EEG signal to determine the probability that the subject is in the target state, and obtaining dynamic control data of the electrical stimulation according to the probability that the subject is in the target state;

[0060] Specifically, the closed-loop neural regulation network model is a model proposed in this application that combines a closed-loop feedback algorithm with a neural network architecture. The model is used to analyze the probability that the subject is in a set target state based on the subject's EEG signal, and adjust the electrical stimulation parameters of tACS based on the analysis results to adjust the subject's brain activity state in real time. The target state here can be understood as a specific brain activity state that needs to be regulated, such as a depressive state. The closed-loop neural regulation network model analyzes the probability that the current subject's brain activity is in this state, and then adjusts the subject's brain activity state through transcranial alternating current stimulation.

[0061] Among them, the core of the closed-loop feedback algorithm must consider real-time performance, that is, it can be quickly executed on the device, analyze EEG signals in real time and dynamically adjust electrical stimulation parameters according to the analysis results. Therefore, this application combines the closed-loop feedback algorithm with the Deep Neural Networks (DNN) architecture to achieve real-time, fast and accurate EEG analysis.

[0062] In some preferred embodiments, a closed-loop neural regulatory network model can be first trained using training samples. These training samples carry the target state features to be identified and the corresponding probabilities of the target states to be identified, so that the model can subsequently identify the target state features in the subject's EEG signals and confirm the probability that the corresponding subject is in the target state based on the target state features.

[0063] Regarding the closed-loop neural regulation network model, in a specific embodiment, the closed-loop neural regulation network model includes: an input layer, multiple hidden layers and an output layer. The specific architecture diagram is as follows: Figure 2As shown. Among them, the number of hidden layers and the number of neurons in each layer determine the depth and complexity of the network. In order to balance the calculation accuracy and efficiency, the hidden layer in this application is preferably three layers, and each hidden layer is connected by weights. The weights here can be understood as the parameters connecting two neurons. They determine how the output of the previous layer of neurons affects the input of the next layer of neurons. These weights are the parameters learned by the neural network during training. At the beginning of training, the weights are initialized random numbers, and then the weights are adjusted according to the loss function, that is, the difference between the predicted value and the actual value, through the back propagation algorithm to minimize the loss and improve the performance of the model.

[0064] Further, the closed-loop neural control network model is used to analyze and obtain the dynamic control data of the electrical stimulation, including steps 210-220:

[0065] Step 210, receiving the subject's EEG signal in real time, and marking the subject's EEG signal according to the time window information;

[0066] Specifically, the real-time received EEG signal is marked according to the time window information to obtain the EEG signal marked with the time information. For example, the real-time received EEG signal can be divided into time windows of 1 second or 2 seconds.

[0067] Step 220, identifying the target state feature in the current EEG signal, obtaining the probability that the current subject is in the target state according to the target state feature in the current EEG signal, and obtaining the dynamic control data of the electrical stimulation according to the probability that the current subject is in the target state and the probability that the subject is in the target state in the previous time window;

[0068] Specifically, the input layer of the closed-loop neural regulation network model receives the time-stamped EEG signals and sends them to the hidden layer.

[0069] Each hidden layer first identifies the target state features in the subject's EEG signal in the current time window, and obtains the probability that the subject is in the target state in this time window based on the target state features in the current subject's EEG signal. Then, based on the probability that the current subject is in the target state and the probability that the subject is in the target state in the previous time window, the target state change in two adjacent time windows is calculated. Finally, based on the target state change, the dynamic control data of electrical stimulation is obtained.

[0070] Among them, the target state change Δo under two adjacent time windows is calculated t The calculation formula is as follows:

[0071] Δo t =o t ―o t―1 (2)

[0072] o t represents the probability that the current subject is in the target state, o t―1 Represents the probability that the subject is in the target state in the previous time window.

[0073] It is understandable that if Δo t <0, the intensity and / or frequency of the transcranial alternating current stimulation signal should be reduced; if Δo t >0, the intensity and / or frequency of the transcranial alternating current stimulation signal should be increased, and the amplitude of increase and decrease is related to Δo t The value corresponds to .

[0074] More specifically, the dynamic control data of electrical stimulation includes: control data of stimulation frequency and / or control data of stimulation intensity. Then, according to the target state change amount, the calculation formulas for the control data of stimulation frequency and the control data of stimulation intensity are obtained as shown in formulas (3) and (4):

[0075] f t+1 =f t ―α f ·Δo t (3)

[0076] I t+1 =I t ―α I ·Δo t (4)

[0077] Among them, f t+1 represents the control data of the stimulation frequency required at the next moment, f t Indicates the control data of the current stimulation frequency, I t+1 Indicates the control data of the stimulation intensity required at the next moment, I t Represents the control data of the current stimulus intensity, α f represents the step size of the stimulation frequency in the iteration process, α I Represents the step size of the stimulus intensity during the iteration process.

[0078] The step size here is used to describe the distance or increment that a parameter or variable travels when it transfers from one state to the next during the iteration process. The size of the step size directly affects the convergence speed of the iterative algorithm. If the step size is too large, it may not converge; if the step size is too small, the algorithm converges slowly and requires more iterations to reach the optimal solution or a state close to the optimal solution. Therefore, by adjusting the step size, the convergence speed can be improved and the number of iterations can be reduced while ensuring the stability of the algorithm, thereby saving computing time and resources.

[0079] In some more preferred embodiments, the step length can be dynamically adjusted by regulating the result reward value so that the step length can be at the optimal value at any time. t The calculation formula is as follows:

[0080] r t =w1·r EEG +w2·r feedback (5)

[0081] r EEG represents the objective evaluation value of the target state controlled by the closed-loop neural control network model, r feedback represents the subjective evaluation value of the subject on the regulation target state, r EEG and r feedback The value range is [0, 1], w1 represents the balance r EEG The weight of importance, w2 represents the balance r feedback The weight of importance. feedback is a subjective input value. w1 and w2 can be set as needed. EEG The calculation formula is as follows:

[0082] r EEG =―Δo t =―(o t ―o t―1 ) (6)

[0083] Finally, according to the regulation result, the reward value r t Dynamically adjust the stimulation frequency in the iterative process. f and the step length α of the stimulus intensity during the iteration I , get the adjusted step length a′ f and a′ I , and according to the adjusted step length a′ f and a′ I The next stimulation frequency control data and stimulation intensity control data. f and a′ I The calculation formulas are as follows:

[0084] a′ f =a f +η·r t ·Δo t (7);

[0085] a′ I =a I +η·r t ·Δo t (8);

[0086] Among them, η represents the learning rate of the closed-loop neural control network, which is used to control the step adjustment speed and can be set according to actual conditions.

[0087] Of course, the above are only preferred embodiments shown in the present application, in which the relationship between the probability of the target state and the dynamic control data of the electrical stimulation, and the relationship between the target state change and the control amplitude can also be set by technical personnel in this field as needed.

[0088] Step 130, dynamically adjusting the transcranial alternating current stimulation signal according to the dynamic control data of the electrical stimulation to adjust the brain activity of the subject in real time;

[0089] Specifically, the output layer of the closed-loop neural control network model receives the dynamic control data of electrical stimulation output by the last hidden layer, and dynamically adjusts the transcranial alternating current stimulation signal according to the dynamic control data of electrical stimulation to regulate the brain activity of the subject in real time.

[0090] During transcranial alternating current stimulation, in a specific embodiment, accurate biphasic current pulses of more than 8 channels are generated through a current waveform controller, a current drive circuit and a charge balancer to achieve symmetrical output of positive and negative currents with a maximum current amplitude of 1mA to reduce residual charge after stimulation and ensure that the residual charge on the electrode-tissue after electrical stimulation is less than 2nC.

[0091] Step 140, continue to dynamically acquire the EEG signal of the subject after electrical stimulation, so as to adjust the brain activity of the subject in real time according to the EEG signal of the subject;

[0092] Specifically, the application can continuously obtain EEG signals during the process of the subject receiving transcranial alternating current stimulation, and adjust the subject's brain activity in real time according to the EEG signals, and the cycle process is the above steps 110-140. Based on the above steps, the present application can analyze the subject's EEG information in real time, automatically adjust the stimulation parameters, monitor the treatment effect in real time and automatically adjust the stimulation parameters to form a closed-loop circulation regulation. In this process, the present application can automatically adjust the intervention in real time through an algorithm according to the changes in biomarkers, that is, record feedback after intervention, and use feedback to adjust the parameters of the next intervention and stimulate, thereby establishing a feedback and stimulation cycle, and realizing automated and precise regulation of simultaneous monitoring and intervention.

[0093] In addition, in some preferred embodiments, in the above steps 110-140, the regulation process can also be monitored. The subject's EEG signal and the corresponding transcranial alternating current stimulation signal are output to the remote monitoring end, so that the remote control end can confirm whether the regulation process needs to be controlled according to the EEG signal and the corresponding transcranial alternating current stimulation signal. If the remote monitoring end controls it, the adjustment signal input by the remote monitoring end is received to adjust the transcranial alternating current stimulation signal.

[0094] The transcranial alternating current stimulation method provided in the present application utilizes a closed-loop neural control network model to perform real-time closed-loop feedback analysis of EEG signals to dynamically adjust the transcranial alternating current stimulation signals, thereby adjusting the brain activity of the subject in real time. During this process, the present application can collect and analyze EEG signals from the subject in real time, dynamically adjust the stimulation mode according to the subject's state, provide controllable stimulation to neurons, which is beneficial to the plasticity of neural circuits, realizes the precision and high personalization of transcranial alternating current stimulation, improves the stimulation effect and reduces possible side effects.

[0095] In addition, the present application also provides a transcranial alternating current stimulation device, such as Figure 3 As shown, the device includes: a housing 1, electrodes 2, a data processing terminal 3, a user operation terminal 4 and a power supply 5. The electrodes 2 are used to capture the EEG signals of the subject. The data processing terminal 3 includes a processor for implementing the transcranial alternating current stimulation method in the first aspect.

[0096] Preferably, the processor uses a Raspberry Pico W embedded processor. The Raspberry Pi Pico W embedded processor is a processor model with WiFi function. It is small and light, and is very suitable for applications in portable devices. Secondly, it also has excellent low power consumption characteristics, which means that the device can run longer with limited battery capacity, reduce the frequency of charging, increase the continuous use time of the device and the convenience of the patient. Thirdly, the Raspberry Pi Pico W processor provides sufficient computing power to handle complex signal acquisition, analysis and control algorithms. This is particularly important for devices based on closed-loop neural regulation, because it needs to process EEG signals in real time and adaptively adjust stimulation parameters to ensure the personalization and effectiveness of treatment.

[0097] Preferably, the power source is a lithium polymer battery, and can be loaded with an intelligent power switching mode to balance performance and battery life.

[0098] Preferably, the data processing end includes an API interface and an I / O interface to facilitate docking with other ports to achieve data sharing and remote monitoring.

[0099] The effect of a transcranial alternating current stimulation device disclosed in the present application is the same as the above-mentioned transcranial alternating current stimulation method, which will not be repeated here.

[0100] The preferred embodiments of the present application are described in detail above; however, the present application is not limited to the specific details in the above embodiments. Within the technical concept of the present application, a variety of simple modifications can be made to the technical solution of the present application, and these simple modifications all fall within the protection scope of the present application.

[0101] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, this application will not further describe various possible combinations.

[0102] In addition, the various implementation modes of the present application may be arbitrarily combined, and as long as they do not violate the concept of the present application, they should also be regarded as the contents disclosed by the present invention.

Claims

1. A transcranial alternating current stimulation method, characterized in that: The method comprises: Dynamically obtain the subject's EEG signal; Using a closed-loop neural control network model, the EEG signal is analyzed in real time to determine the probability that the subject is in a target state, and dynamic control data of electrical stimulation is obtained according to the probability that the subject is in the target state; Dynamically adjusting the transcranial alternating current stimulation signal according to the dynamic regulation data of the electrical stimulation to adjust the brain activity of the subject in real time; Continue to dynamically acquire the EEG signal of the subject after electrical stimulation, so as to adjust the brain activity of the subject in real time according to the EEG signal of the subject.

2. The method according to claim 1, characterized in that: After dynamically acquiring the subject's EEG signal, the method further includes: The EEG signal of the subject is subjected to low-pass filtering and amplitude limiting processing.

3. The method according to claim 1, characterized in that Using a closed-loop neural control network model, the EEG signal is analyzed in real time to determine the probability that the subject is in a target state, and dynamic control data of electrical stimulation is obtained according to the probability that the subject is in the target state, including: Receiving the EEG signal of the subject in real time, and marking the EEG signal of the subject according to a time window; Identify the target state features in the current EEG signal, obtain the probability that the current subject is in the target state based on the target state features in the current EEG signal, and obtain dynamic control data of electrical stimulation based on the probability that the current subject is in the target state and the probability that the subject was in the target state in the previous time window.

4. The method according to claim 3, characterized in that The closed-loop neural regulation network model includes three hidden layers.

5. The method according to claim 3 or 4, characterized in that: According to the probability that the current subject is in the target state and the probability that the subject is in the target state in the previous time window, the dynamic control data of the electrical stimulation is obtained, including: According to the probability that the current subject is in the target state and the probability that the subject is in the target state in the previous time window, the target state change in two adjacent time windows is calculated; According to the target state change amount, dynamic control data of the electrical stimulation is obtained.

6. The method according to claim 5, characterized in that: The dynamic control data of the electrical stimulation includes: control data of the stimulation frequency and / or control data of the stimulation intensity; The calculation formula of the regulation data of the stimulation frequency is: F t+1 =f t ―α f ·Δo t ; The calculation formula of the regulation data of the stimulation intensity is: I t+1 =I t ―α I ·Δo t ; Among them, f t+1 represents the control data of the stimulation frequency required at the next moment, f t Indicates the control data of the current stimulation frequency, I t+1 Indicates the control data of the stimulation intensity required at the next moment, I t Represents the control data of the current stimulus intensity, α f represents the step size of the stimulation frequency in the iteration process, α I represents the step size of the stimulus intensity during the iteration, Δo t Indicates the target state change.

7. The method according to claim 6, characterized in that The method further comprises: Calculate the control result reward value, the control result reward value r t The calculation formula is: r EEG =―Δo t =―(o t -O t―1 ); r t =w1·r EEG +w2·r feedback ; Among them, r EEG represents the objective evaluation value of the target state controlled by the closed-loop neural control network model, r feedback represents the subjective evaluation value of the subject on the regulation target state, r EEG and r feedback The value range is [0, 1], w1 represents the balance r EEG The weight of importance, w2 represents the balance r feedback The weight of importance, o t represents the probability that the current subject is in the target state, o t―1 Represents the probability that the subject is in the target state in the previous time window; The step length of the stimulation frequency and the step length of the stimulation intensity in the iteration process are adjusted according to the reward value of the regulation result. The calculation formula is: to' f =af+η r t ·Δo t ; and' I =a I +η·r t ·Δo t ; Wherein, η represents the learning rate of the closed-loop neural regulation network model, a′ f represents the step length of the adjusted stimulation frequency in the iteration process, a′ I Represents the step size of the adjusted stimulus intensity during the iteration process.

8. The method according to claim 3, characterized in that Before using the closed-loop neural regulation network model, the method further includes: The closed-loop neural regulation network model is trained using training samples, wherein the training samples contain the target state characteristics and the corresponding target state probabilities, so that the closed-loop neural regulation network model can recognize the target state characteristics in the EEG signal and confirm the corresponding target state probabilities.

9. The method according to claim 1, characterized in that: The method further comprises: Outputting the subject's EEG signal and the corresponding transcranial alternating current stimulation signal to a remote monitoring terminal; Receive the adjustment signal input by the remote monitoring terminal to adjust the transcranial alternating current stimulation signal.

10. A transcranial alternating current stimulation device, characterized in that: The device comprises: a housing, electrodes, a data processing terminal, a user operation terminal and a power supply; Wherein, the electrodes are used to capture the EEG signals of the subject; The data processing end includes a processor for implementing the method according to any one of claims 1-9.

Images