A device for monitoring and relieving abnormal headaches through stimulation intervention

Abstract

This invention relates to the field of intelligent control technology for medical devices, and provides a device for monitoring and relieving headache abnormalities through stimulation intervention. The device includes a helmet-shaped main frame housing a monitoring and relief module; an intelligent control system that asynchronously collects physiological characteristics and hardware load status via a duplex channel, deriving a high-dimensional state fusion model; and a closed-loop evaluation and iteration module to establish a differential efficacy closed-loop mechanism, uploading the collected data to the cloud for storage, enabling historical data management, clinical analysis, and continuous optimization of personalized strategies. This technical solution addresses the problems of dependence and side effects associated with traditional drug treatments, the high trauma and cost of surgical treatments, and the tendency of existing migraine intervention devices to develop tolerance in the nervous system due to fixed and predictable stimulation parameters used in open-loop control, leading to gradual attenuation of intervention effects over time and an inability to provide personalized and precise treatment.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for medical devices, specifically to a device for monitoring and relieving abnormal headaches through stimulation intervention. Background Technology

[0002] Migraine is a highly disabling neurological disorder that severely impacts patients' quality of life. Currently, there is no cure for migraine, and it has a significant genetic component (such as familial clustering).

[0003] Current treatments for migraines have several drawbacks and limitations: First, medication (such as non-specific or specific analgesics) is a common intervention, but long-term use can easily lead to drug dependence and is accompanied by a series of significant side effects such as nausea, vomiting, drowsiness, and allergies. Second, while surgery provides an option for a very small number of refractory migraines, it faces problems such as significant surgical trauma, high risks of disfigurement, and high treatment costs. Finally, traditional physical therapies (such as conventional home massage devices, traditional acupuncture, and massage) generally suffer from long treatment courses and slow onset of action due to a lack of precise intelligent targeting and efficient physiological indicator linkage, making it difficult to provide rapid and effective intervention and relief during migraine attacks or prodromal phases.

[0004] Furthermore, existing wearable devices for head monitoring and relief often employ simple, rigid open-loop triggering logic in their control systems (setting physical stimuli to a fixed frequency or constant temperature). Numerous physiological studies have shown that human neurons are highly susceptible to fatigue desensitization and tolerance (i.e., neural habituation mechanisms) in response to continuous, multimodal stimuli at a single, fixed frequency. This leads to a significant decrease in the effectiveness of existing physiotherapy devices over extended periods due to the lack of random variables and closed-loop therapeutic feedback, causing the nervous system to adapt and significantly diminishing the therapeutic effect. Consequently, long-term, intelligent, and dynamic fatigue prevention treatment cannot be achieved. Summary of the Invention

[0005] To overcome the above-mentioned shortcomings, the present invention provides a device for monitoring and relieving abnormal headaches and stimulating interventions. This device solves the problems of dependence and side effects of traditional drug treatment, large trauma and high cost of surgical treatment, and the fact that existing migraine intervention devices use fixed and regular stimulation parameters for open-loop control, which leads to easy tolerance of the nervous system, gradual decline of intervention effect over time, and inability to provide personalized and precise treatment.

[0006] According to one aspect, at least one embodiment of the present invention provides a device for monitoring and relieving abnormal headaches through stimulation intervention, comprising: A helmet-shaped main frame, which is made of a biomimetic elastic material, and an elastic strap is installed at the bottom front end of the helmet-shaped main frame; The monitoring and relief module is embedded in the helmet-shaped main frame at the position corresponding to the acupoints on the human head. The monitoring and relief module includes an electroencephalogram (EEG) signal acquisition unit and a multimodal intervention execution unit. The intelligent control module is installed on both sides of the helmet-shaped main frame. The intelligent control module has a built-in intelligent control system. The intelligent control system is used to determine the signs of migraine attack based on the signals acquired by the EEG signal acquisition unit and a preset aura recognition algorithm. Based on the determination result, it automatically drives the multimodal intervention execution unit at the corresponding acupoint to work together. It also uploads the physiological signal data acquired by the EEG signal acquisition unit, the execution status data of the multimodal intervention execution unit, and the corresponding intervention result data to the cloud storage unit for storage. The intelligent control module is also provided with a USB interface on its outside, which is used for powering and charging the device, transmitting data, or performing program maintenance.

[0007] For example, in a head pain abnormality monitoring and stimulation intervention relief device provided by at least one embodiment of the present invention, the electroencephalogram (EEG) signal acquisition unit includes a bioelectric sensor, a temperature sensor and an electromyography (EMG) sensor, used to collect the wearer's EEG physiological signals, skin temperature and muscle tension in real time, and the multimodal intervention execution unit includes a graphene heating microplate, a far-infrared radiation module and a drug delivery structure.

[0008] For example, in a head pain abnormality monitoring and stimulation intervention relief device provided in at least one embodiment of the present invention, the drug delivery structure is a hollow tube for storing drug solution integrated in the liner of a helmet-shaped main frame, a micro peristaltic pump connected to the hollow tube, and a nitrided silica microporous drug delivery tablet.

[0009] For example, in a head pain abnormality monitoring and stimulation intervention relief device provided in at least one embodiment of the present invention, the intelligent control system includes: The asynchronous data acquisition module is used to issue and extract specific types of data communication commands; The state feature extraction module is used to perform dimensionality reduction and fusion on the collected data to construct a combination of machine space response states; The anti-adaptation strategy generation module generates collaborative decision-making instructions for different acupoints to avoid neural tolerance based on the preceding state using a neural network. The intervention execution scheduling module is used to receive and perform serial-to-parallel conversion and physical protocol unpacking of the collaborative decision-making instructions to drive the execution of multiple units. The closed-loop evaluation and iteration module is used to establish feedback incentives by comparing with subsequently updated data and iteratively fine-tuning the feature map of the network control dictionary. It also organizes the collected physiological data, hardware execution feedback data, intervention strategy data and efficacy evaluation data and uploads them to the cloud storage unit to form historical treatment records.

[0010] For example, in a head pain abnormality monitoring and stimulation intervention relief device provided in at least one embodiment of the present invention, the asynchronous data acquisition module extracts specific types of data communication instructions that only include the concurrent acquisition of the following two types of independent data streams: The first data type is a multidimensional physiological modality sensing data stream: a multi-channel time-stamped digital sequence matrix of physiological voltage and temperature states captured in real time by the EEG signal acquisition unit through its included bioelectric sensors, temperature sensors and electromyography sensors and transmitted to the processing buffer station. The second data type is the hardware execution feedback status parameter stream: the execution status dictionary table uploaded by the multimodal intervention execution unit corresponding to each acupoint via the original path of the control bus. The dictionary table contains the thermal impedance, infrared radiation power consumption, peristaltic pump pressure status and load feedback information under the current physical working mode of each execution unit. The multidimensional physiological modality sensing data stream and the hardware execution feedback status parameter stream are processed by the intelligent control system, and then uploaded to the cloud storage unit with the addition of time stamps and user identifiers.

[0011] For example, in a head pain abnormality monitoring and stimulation intervention relief device provided by at least one embodiment of the present invention, the specific logical implementation steps of the state feature extraction module include: The acquired multidimensional physiological modality sensing data stream is processed through a preset frequency band filter to calculate the absolute frequency band energy and muscle tension parameters, extract abnormal features, and derive the multidimensional physiological abnormality precursor index at the current moment. The acquired hardware execution feedback state parameter stream is quantized and converted into the hardware distributed feedback load index at the current moment. The multidimensional physiological abnormality precursor index at the current moment and the hardware distributed feedback load index are jointly imported into the high-dimensional machine state fusion model to obtain the high-dimensional composite precursor state response value at the current moment.

[0012] For example, in a head pain abnormality monitoring and stimulation intervention relief device provided in at least one embodiment of the present invention, the specific logical implementation steps of the anti-adaptation strategy generation module include: The high-dimensional composite precursor state response value is used as the input of the reinforcement learning environment observation. Combined with the nonlinear chaotic perturbation sequence generated by the pseudo-random algorithm, it is imported into the anti-adaptive action control decision model to output the multimodal intervention collaborative strategy control action benchmark value to avoid the regular adaptation of the human nervous system.

[0013] For example, in a head pain abnormality monitoring and stimulation intervention relief device provided in at least one embodiment of the present invention, the specific logical control features of the intervention execution scheduling module are as follows: The system receives the baseline values ​​for the multimodal intervention synergy strategy control actions and resolves them into an intervention intensity vector group containing multiple acupoint topological indices, spatial phase parameters, and physical working form scalars through a mapping table. This vector group is then distributed to the multimodal intervention execution unit via a communication cable bus, causing the stimulation intensity of different acupoints and the multimodal output category to exhibit irregular jitter and cross-random fine-tuning physical control responses. This enables the implementation of adaptive synergy operations, including graphene thermal pulse oscillation, infrared radiation dynamic adjustment, and intermittent drug delivery via a micro-peristaltic pump.

[0014] For example, in a head pain abnormality monitoring and stimulation intervention relief device provided in at least one embodiment of the present invention, the specific steps for closed-loop evaluation in the closed-loop evaluation and iteration module are as follows: The system triggers a delayed supervision and evaluation window mechanism, waiting for the intervention vector group to execute for a certain time constant; The next-state multidimensional physiological deviation evolution characteristic index corresponding to this moment is re-acquired, and the difference magnitude between it and the baseline evolution state before intervention is calculated. This difference is then imported into the relative relief degree reward score model to obtain the dynamic reward value of the therapeutic effect. The dynamic reward value of the therapeutic effect, together with the corresponding physiological data and execution parameters before and after intervention, is uploaded to the cloud storage unit for long-term efficacy tracking and model optimization.

[0015] For example, in a head pain abnormality monitoring and stimulation intervention relief device provided by at least one embodiment of the present invention, the method for using a self-iterative neural network control dictionary in the closed-loop evaluation and iteration module is as follows: The obtained relative mitigation reward score is used as evaluation feedback and fed back into the hidden layer nodes of the anti-adaptive instruction neural network to initiate the gradient descent correction model and implement the automatic advancement of the network's internal policy link weights.

[0016] The beneficial effects of the embodiments of the present invention are as follows: This invention effectively replaces traditional drug treatments that involve dependence and side effects, as well as surgical treatments that are highly invasive and costly. This device utilizes an EEG signal acquisition unit to sensitively capture EEG changes during the prodromal phase of migraines, and coordinates with a multimodal intervention execution unit (graphene heating microplates, far-infrared precise radiation, and penetration of traditional Chinese herbal medicine liquid) to target and regulate key intervention areas in the head that control pain in a short period of time. It effectively combines the warming effect to improve blood circulation with the inhibition of pain signals, significantly overcoming the shortcomings of traditional physiotherapy, such as long treatment courses and slow onset of action, thus achieving early detection, early intervention, and early relief of migraine attacks.

[0017] Meanwhile, this invention addresses the problem of nervous system tolerance caused by the fixed output of traditional devices. When generating intervention actions, the intelligent control system of this invention integrates a nonlinear chaotic perturbation sequence generated by a preset pseudo-random algorithm. This ensures that the targeted physical stimulation intensity, mode switching frequency, and temporal phase assigned to different acupoints exhibit irregular fluctuations and random fine-tuning within a safe threshold. This uncertain anti-fatigue variation pattern prevents patients from experiencing nerve fatigue and efficacy attenuation even after long-term daily use. Furthermore, the closed-loop evaluation and iteration module continuously accumulates individualized reward data. After hundreds to thousands of control cycles of iterative optimization, the strategy network weights gradually converge to form an optimal intervention strategy distribution specific to the patient.

[0018] Furthermore, by setting up a cloud storage unit, this invention can continuously save the wearer's EEG signals, physiological state parameters, hardware execution feedback information, intervention strategy parameters, and efficacy evaluation results during different usage cycles. This enables historical data tracing, remote clinical analysis, individual treatment record management, and long-term strategy optimization, improving the device's intelligence level and clinical application value. By adding a USB interface, stable power supply and charging, local data export, parameter configuration, and program upgrades can be achieved in scenarios where wireless communication is limited, network anomalies occur, or equipment maintenance is required, improving the device's ease of use, data acquisition reliability, and subsequent maintenance efficiency. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are merely some exemplary embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the content of the exemplary embodiments of the present invention and these drawings without any creative effort.

[0020] Figure 1 This is a perspective view of the external structure of the present invention; Figure 2 This is a block diagram of the intelligent control system and monitoring and mitigation module in this invention; In the diagram: 1. Helmet-shaped main frame; 2. Monitoring and relief module; 21. EEG signal acquisition unit; 22. Multimodal intervention execution unit; 3. Intelligent control module; 4. Elastic band; 5. Intelligent control system; 51. Asynchronous data acquisition module; 52. State feature extraction module; 53. Anti-adaptive strategy generation module; 54. Intervention execution scheduling module; 55. Closed-loop evaluation and iteration module. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it.

[0022] To keep the drawings concise, each drawing only schematically shows the parts relevant to the invention; these do not represent the actual structure of the product. Furthermore, for ease of understanding, in some drawings, only one of components with the same structure or function is schematically shown, or only one is labeled. In this document, "one" not only means "only one," but can also mean "more than one," and "several" includes "two" and "more than two."

[0023] In this document, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0024] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0025] In the description of this embodiment, terms such as "upper," "lower," "left," and "right" are based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of description and simplification of operation, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention.

[0026] Furthermore, in the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0027] Please see Figure 1 and Figure 2 According to one embodiment of the present invention, a device for monitoring and relieving abnormal headaches through stimulation intervention includes: The helmet-shaped main frame 1 is made of biomimetic elastic material and can fit different head circumferences (covering a head circumference range of 54-64cm). The bottom front end of the helmet-shaped main frame 1 is equipped with an elastic strap 4 to securely wear the device on the user's head.

[0028] The monitoring and relief module 2 is embedded in the helmet-shaped main frame 1 at the positions corresponding to acupoints on the human head, accurately covering core acupoints such as Fengchi, Taiyang, Baihui, and Tianzhu. The monitoring and relief module 2 includes an EEG signal acquisition unit 21 and a multimodal intervention execution unit 22.

[0029] The intelligent control module 3 is installed on both sides of the helmet-shaped main frame 1. The intelligent control module 3 has a built-in intelligent control system 5. The intelligent control system 5 is used to determine the signs of migraine attack based on the signals acquired by the EEG signal acquisition unit 21 and through a preset aura recognition algorithm. Based on the determination result, it automatically drives the multimodal intervention execution unit 22 at the corresponding acupoint to work together. The intelligent control module 3 is also connected to the cloud storage unit to upload the collected EEG physiological signals, skin temperature data, electromyography signals, intervention execution status, efficacy evaluation results and model optimization parameters to the cloud server for storage. The intelligent control module 3 has a USB interface on its outside. The USB interface can be used to power and charge the internal power supply of the device, establish a wired connection with the external terminal for data transmission, and perform program maintenance and system upgrade operations.

[0030] Specifically, the EEG signal acquisition unit 21 includes a bioelectric sensor, a temperature sensor, and an electromyography (EMG) sensor, used to acquire the wearer's electroencephalographic signals, actual skin temperature, and muscle tension in real time. The bioelectric sensor uses electrode pads as a key channel for the flow of electrons or charges, establishing a stable and effective electrical connection and physical contact on the head to accurately acquire and converge specific EEG signals generated in the target area. While most migraine patients show normal EEG readings between attacks, some may exhibit non-specific EEG abnormalities such as increased slow-wave activity or sharp waves. The bioelectric sensor allows for early EEG acquisition, enabling early detection, intervention, and treatment. The temperature sensor is used to acquire real-time changes in scalp skin temperature, providing data support for the system to assess local blood circulation. The EMG sensor detects scalp muscle tension, assisting in the identification of symptoms associated with tension-related migraines.

[0031] The multimodal intervention execution unit 22 includes a graphene heating microplate, a far-infrared radiation module, and a drug delivery structure. The graphene heating microplate intelligently collects skin temperature and dynamically adjusts power to achieve precise control of the thermal effect. The far-infrared radiation module utilizes the strong penetrating power of infrared radiation, primarily through interaction with mitochondrial pigments within cells, to produce non-thermal biological effects. This enhances cellular energy metabolism, relieves accompanying muscle tension, and simultaneously improves blood circulation and inhibits pain signal transduction pathways. The drug delivery structure consists of a hollow tube integrated within the helmet-shaped main frame 1 for storing the drug, a micro-peristaltic pump connected to the hollow tube, and a nitrided silica microporous drug delivery tablet. The formula of the medicinal liquid is improved based on the ancient books "Huangdi Neijing" and "Taiping Huimin Heji Jufang". It contains authentic medicinal materials such as Chuanxiong, Baizhi, Qianghuo, Fangfeng, Xixin, and Ruxiang. It has the effects of warming the meridians and promoting blood circulation, relieving pain and removing blood stasis. The medicine delivery structure uses a micro peristaltic pump to pump the medicine in the hollow tube to the corresponding position of the nitrided silicon microporous drug tablet to release the medicine. The heat effect generated by the graphene heating microplate, combined with infrared radiation, accelerates the transdermal absorption of the active components of the medicine through the acupoints and skin.

[0032] The intelligent control system 5 includes: an asynchronous data acquisition module 51, used to issue and extract specific types of data communication commands; a state feature extraction module 52, used to perform dimensionality reduction and fusion on the acquired data to construct a combination of machine space response states; an anti-adaptation strategy generation module 53, which generates collaborative decision-making commands for different acupoints to avoid neural tolerance based on the preceding state using a neural network; an intervention execution scheduling module 54, used to receive and perform serial-to-parallel conversion and physical protocol unpacking of the collaborative decision-making commands to drive the execution of multiple units; and a closed-loop evaluation and iteration module 55, used to establish feedback incentives and iteratively fine-tune the feature map of the network control dictionary by comparing with subsequently updated data.

[0033] Among them, the data communication instructions extracted by the asynchronous data acquisition module 51 only include the concurrent acquisition of the following two types of independent data streams: The first data type is a multi-dimensional physiological modality perception data stream, which is a multi-channel time-stamped digital sequence matrix of physiological voltage and temperature status captured in real time by the EEG signal acquisition unit 21 through its bioelectric, electromyographic and temperature sensors and transmitted to the processing buffer station; The second data type is a hardware execution feedback status parameter stream, which is an execution status dictionary table uploaded back by the multi-modal intervention execution unit 22 corresponding to each acupoint through the original path of the control bus. The dictionary table contains the thermal impedance, infrared radiation power consumption, peristaltic pump pressure status and load feedback information under the current physical working mode of each execution unit.

[0034] The multidimensional physiological modality sensing data stream specifically includes: multi-channel EEG micro-voltage digital sequences acquired by bioelectric sensors, which carry timestamps and reflect the energy change characteristics of various brainwave frequency bands (such as alpha waves, beta waves, and gamma waves) during a migraine attack; head muscle tension signals acquired by electromyography sensors to help determine the attack status of tension-type migraines; and real-time scalp temperature data acquired by temperature sensors to monitor changes in local blood circulation. The hardware execution feedback status parameter stream specifically includes: the current thermal impedance and power output status of the graphene heating microchip, the radiation power consumption and wavelength parameters of the far-infrared radiation module, and the pump pressure status and drug delivery rate of the micro-peristaltic pump. This information constitutes a complete profile of the device's operating status.

[0035] Furthermore, after acquiring the multidimensional physiological modality sensing data stream and the hardware execution feedback status parameter stream, the asynchronous data acquisition module 51 can add timestamps, device identifiers, and user identifiers to the data, and send it to the cloud storage unit for classified storage via the wireless communication module. The cloud storage unit can be a server database or a cloud platform database, used to save raw monitoring data, processed feature data, and intervention history records. When offline data export, connection to external computing devices, or system maintenance is required, a wired communication connection can be established with a host computer or external terminal via a USB interface to achieve local data backup, parameter configuration, and program updates.

[0036] Specifically, the logical implementation steps of the state feature extraction module 52 include: calculating the absolute frequency band energy and muscle tension parameters of the acquired multidimensional physiological modality sensing data stream through a preset frequency band filter, extracting abnormal features, and deriving the multidimensional physiological abnormality precursor index at the current moment; quantizing the acquired hardware execution feedback state parameter stream and converting it into the hardware distributed feedback load index at the current moment; and importing the multidimensional physiological abnormality precursor index and the hardware distributed feedback load index at the current moment into the high-dimensional machine state fusion model to obtain the high-dimensional composite precursor state response value at the current moment. The high-dimensional machine state fusion model is represented as follows: ; in, This represents the high-dimensional composite precursor state response value at the current moment. The multidimensional physiological abnormality precursor index represents the current moment. This represents the hardware distribution feedback load index at the current moment. Represents the sensitivity coefficient of physiological characteristics. Represents the hardware load weighting coefficient and satisfies + =1, Indicates a constant offset. The higher the value, the stronger the patient's neurological abnormalities and the higher the current equipment intervention load.

[0037] Among them, the multidimensional physiological abnormality precursor index This refers to a comprehensive quantitative index of abnormalities extracted through frequency domain analysis of electroencephalogram (EEG) signals. Specifically, it can be achieved by performing a Fast Fourier Transform on multi-channel EEG signals, calculating the absolute power spectral density of each frequency band (Delta, Theta, Alpha, Beta, Gamma), and then weighting and fusing this index with muscle tension parameters collected by electromyography (EMG) sensors. When migraines are in the prodromal phase, slow-wave activity increases and Alpha wave power decreases in the EEG signal, leading to a significant increase in this index. Hardware distributed feedback load index. This refers to a comprehensive quantitative value representing the current working status of the multimodal intervention execution unit at each acupoint. Specifically, it can be achieved by normalizing and weighting the thermal impedance deviation of the graphene heating microsheet, the power consumption ratio of the far-infrared radiation module, and the pressure status of the peristaltic pump, thus characterizing the current intervention output load level of the device. Physiological characteristic sensitivity coefficient The hardware load weighting coefficient is used to control the contribution weight of physiological abnormal signals to the state response value. The weights used to control the contribution of the device state to the state response value are equal to 1 to ensure the normalization of the state response value. Constant offset. The baseline level used to adjust the status response value ensures that the system maintains basic monitoring sensitivity even in the absence of abnormal conditions.

[0038] Specifically, the model uses a logarithmic function. The physiological abnormality precursor index is subjected to nonlinear compression processing, when When the logarithmic function is small, it approximates a linear response, maintaining high sensitivity to weak precursor signals; when... When the logarithmic function is large, its growth rate slows down, avoiding excessive influence of extreme outliers on the state response value, thus achieving adaptive quantitative characterization of migraine auras of varying severity. Hardware distributed feedback load index. The data is superimposed linearly, directly reflecting the current intervention output pressure of the equipment. When the equipment is already operating under high load, The corresponding increase in the value indicates that the anti-adaptive strategy generation module needs to consider the equipment's load-bearing capacity when formulating the next intervention plan.

[0039] The specific logical implementation steps of the anti-adaptive strategy generation module 53 include: using the high-dimensional composite precursor state response value as the input of reinforcement learning environment observation, combining it with the nonlinear chaotic perturbation sequence generated by the pseudo-random algorithm, importing it into the anti-adaptive action control decision model, and outputting the multimodal intervention collaborative strategy control action benchmark value to avoid the regular adaptation of the human nervous system. The anti-adaptive action control decision model is expressed as follows: ; in, This represents the baseline value of the multimodal intervention collaborative strategy control action output at the current moment. This represents the nonlinear chaotic feature perturbation factor generated by random calculation. This represents the weighting of the system's state space response to the shift in intervention demand. This represents the stochastic anti-fatigue weighted control coefficient of the anti-neural tolerance regulation mechanism. This represents the constant value of the bias control.

[0040] Among them, the nonlinear chaotic characteristic perturbation factor This refers to the unpredictable perturbation sequence values ​​generated in real time during each control cycle by a pseudo-random chaotic attractor algorithm (such as the Lorenz system or Logistic mapping). Specifically, this can be implemented by running lightweight chaotic iterative equations in an embedded processor, injecting irregularities into the intervention strategy and disrupting the stimulus patterns that the nervous system might capture. The system state space's response evaluation weights to the intervention demand shift. This is used to control the degree to which the current aura influences intervention decisions; the stronger the migraine aura, the more likely it is to cause adverse reactions. The effect of this leads to a corresponding increase in the intensity of intervention. The stochasticity of the anti-nervous tolerance regulation mechanism and the anti-fatigue weight control coefficient. This coefficient is used to control the magnitude of the impact of chaotic disturbances on the intervention strategy. A larger coefficient indicates a more significant random change in the intervention strategy and stronger resilience. However, an excessively large coefficient may affect the stability of the intervention; therefore, a balance adjustment is needed in practical applications. (Bias control constant value) The baseline offset used to set the intervention strategy ensures that the system can maintain basic preventative intervention outputs even when the warning signs are weak.

[0041] Specifically, the model uses the hyperbolic tangent function. As an activation function, the output value is restricted to the interval (-1, 1), ensuring that the baseline value of the intervention action is always within a safe and controllable range. Unlike traditional open-loop control triggered by a fixed threshold, this model... The introduction of the term makes it possible to achieve the same outcome even in the same precursor state. Furthermore, the intervention strategies output in each control cycle exhibit subtle but unpredictable differences, effectively preventing the human nervous system from adapting to fixed stimulation patterns. For example, even when EEG aura states are similar across two consecutive control cycles, the chaotic perturbation factor... Different values ​​will lead to different intervention combinations from the system, such as "graphene heating at the temple + far-infrared radiation at the Baihui acupoint" and "medicinal liquid penetration at the Fengchi acupoint + graphene heating at the Baihui acupoint", so that the nervous system is always in an unpredictable stimulation environment.

[0042] The specific logical control features of the intervention execution scheduling module 54 are as follows: it receives the reference value of the multimodal intervention collaborative strategy control action, and parses it into an intervention intensity vector group containing multiple acupoint topological indices, spatial phase parameters and physical working form scalars through a mapping table; it distributes it to the multimodal intervention execution unit 22 through the communication cable bus, so that the stimulation intensity of different acupoints and the multimodal output category exhibit irregular jitter and cross-random fine-tuning physical control response, so as to implement anti-adaptive collaborative operation including graphene thermal pulse oscillation, infrared radiation dynamic adjustment and micro peristaltic pump intermittent drug delivery.

[0043] The acupoint topology index refers to a pre-established spatial coding mapping table of head acupoints. Specifically, this can be achieved by numbering and indexing core acupoints such as Fengchi, Taiyang, Baihui, and Tianzhu according to their physical locations on a helmet-shaped main frame. This is used to precisely specify the combination of intervention acupoints to be activated in each control cycle. Spatial phase parameters refer to the temporal sequence and synergistic cooperation mode between the intervention actions of each acupoint. This can be implemented using phase angle coding, such as 0° representing simultaneous activation and 90° representing a quarter-cycle delay, used to control the spatiotemporal coordination of multi-acupoint interventions. Physical working form scalars refer to the specific working parameter values ​​of each intervention mode (graphene heating, far-infrared radiation, drug delivery) at each acupoint, including heating power percentage, radiation intensity level, peristaltic pump speed, etc.

[0044] Specifically, the intervention execution scheduling module 54 will abstract the action baseline value. This is converted into physical control signals that can directly drive the hardware. Internally, this module maintains a multi-dimensional mapping table, which... The numerical range is mapped to different acupoint-modality-intensity combination schemes. For example, when When the value is in the range (0.3, 0.6), the mapping table may output a combination scheme of "40% power graphene heating at the temples + moderate intensity far-infrared radiation at the Baihui acupoint + low-speed drug delivery via peristaltic pump at the Fengchi acupoint". The converted control signal is simultaneously distributed to the corresponding execution unit of each acupoint through the communication cable bus in a serial-to-parallel conversion manner to achieve multi-acupoint coordinated response. Because the value of each control cycle is affected by chaotic disturbance factors... Because of its non-repeatability, the intervention instructions received by each acupoint exhibit irregular dynamic changes in intensity, modality, and timing, thus fundamentally eliminating the habituation desensitization effect caused by the nervous system perceiving fixed stimulus rhythms.

[0045] In the closed-loop evaluation and iteration module 55, the specific steps for closed-loop evaluation are as follows: The system triggers a delayed supervision and evaluation window mechanism, waiting for the intervention vector group to execute for a certain time constant; Reacquire the next-state multidimensional physiological deviation evolution characteristic index corresponding to this moment, calculate the difference between it and the baseline evolution state before intervention, and import it into the relative remission degree reward score model to obtain the dynamic reward value of the therapeutic effect. The relative remission model is expressed as: ; in, The reward score indicates the relative degree of relief during the intervention and adjustment period. This represents the baseline multidimensional physiological abnormality precursor limit index before the intervention cycle is initiated. This represents the index of the deviation characteristics of the next-state physiological evolution captured at the end of the delayed supervision window. This represents the system energy consumption penalty scalar value for the multimodal intervention actions output by the device. The reward gain coefficient represents the positive recovery effect. This represents the penalty adjustment coefficient for the energy consumption of output computing power.

[0046] Among them, the benchmark multidimensional physiological abnormality precursor limit index This refers to the peak value of the multidimensional physiological abnormality precursor index recorded by the state feature extraction module before the start of the current intervention cycle. Specifically, it can be measured at the moment the intervention action is triggered. This is achieved through snapshot storage, which serves as a baseline for evaluating the intervention's effectiveness. Sub-state physiological evolutionary deviation characteristic index. This refers to the latest multidimensional physiological abnormality precursor index value, which is re-acquired by the EEG signal acquisition unit and processed by the state feature extraction module after a preset time constant (e.g., 15 to 30 seconds) following the intervention. It reflects changes in the EEG state after the intervention. (System energy consumption penalty scalar value) This refers to the comprehensive energy consumption of all activated multimodal intervention execution units during the current intervention cycle. Specifically, it can be achieved by normalizing and weighting the power integral of the graphene heating microchip, the radiation energy integral of the far-infrared radiation module, and the mechanical work integral of the micro-peristaltic pump. This is used to constrain the system to balance equipment endurance and energy efficiency while maximizing therapeutic efficacy. The reward gain coefficient for positive rehabilitation effects. This coefficient is used to provide a positive stimulus signal when abnormal EEG indicators are effectively suppressed. The larger this coefficient, the more likely the system is to choose intervention strategies that significantly reduce EEG abnormalities. It also includes a penalty adjustment coefficient for output computing power, energy efficiency, and action consumption. This coefficient is used to impose a negative penalty on high-energy-consuming intervention programs. The larger the coefficient, the more the system tends to choose low-energy-consuming intervention strategies. The ratio of the two coefficients determines the system's tendency to balance efficacy and energy efficiency.

[0047] Specifically, the core design idea of ​​this model is to transform the evaluation of intervention effects into a quantifiable scalar reward signal. The numerator ( This directly reflects the absolute improvement in the degree of EEG abnormalities before and after intervention, divided by... Normalization eliminated the dimensional differences between different seizure intensities, making reward scores comparable across seizure cycles. For example, if the prodromal index of several premonitory EEG abnormalities was 0.8, and decreased to 0.3 after intervention, the positive reward item would be 0.625. To indicate that the intervention achieved a relative remission rate of 62.5%, a minus sign was used. The energy consumption penalty term ensures that the system will not unrestrainedly utilize all intervention modes to maximum power in pursuit of ultimate therapeutic effect, thus achieving an optimal trade-off between therapeutic effect and battery life within the constraints of limited battery capacity. When A positive value indicates that the current intervention strategy has achieved positive efficacy and that energy consumption is within an acceptable range; when When the value is negative, it indicates that the intervention effect is not good or the energy consumption is too high, and the system needs to adjust the strategy direction.

[0048] In the closed-loop evaluation and iteration module 55, the method for using the self-iterative neural network control dictionary is as follows: The obtained relative mitigation reward score is used as evaluation feedback and fed back into the hidden layer nodes of the anti-adaptive instruction neural network. This initiates the gradient descent correction model to implement automatic advancement of the network's internal policy link weights. Its update partial derivative model is expressed as: ; in, This represents the next-generation transition evaluation weight matrix after the update and reconstruction are completed. This represents the ground-state weight matrix for evaluating the adaptive action output in the current cycle. This parameter represents the network's self-learning and optimization exploration rate. This represents the baseline score of the cumulative expected evaluation of predecessors along the state transition path in reinforcement learning. This represents the evaluation calibration gradient derivative array based on the loss function of the current control strategy.

[0049] Among them, the next-generation transfer evaluation weight matrix This refers to the set of connection weights within the internal neural network of the anti-adaptive strategy generation module, updated after this round of closed-loop evaluation. This matrix will replace the current weight matrix in the next control cycle. It participates in forward inference operations, thereby enabling the network's decision-making behavior to adaptively adjust. Ground state weight matrix. This refers to the current values ​​of the connection weights between layers in the neural network of the anti-adaptive strategy generation module before the start of this round of closed-loop evaluation. Network self-learning optimization exploration rate parameter. This refers to the learning rate hyperparameter, which controls the step size of each weight update. It can be implemented using a preset fixed value or an adaptive decay strategy, for example, initially set to 0.01 and gradually decaying with each training epoch, to balance the network's learning speed and convergence stability. The historical cumulative predecessor expected evaluation baseline score is also relevant. This refers to the estimated average reward level accumulated by the system over its historical operation. Specifically, the exponential moving average method can be used to analyze historical data. The sequence is used for sliding estimation, which serves as a benchmark for comparing the current reward score. > If the current strategy is better than the historical average, the weights should be updated along the current gradient direction; otherwise, they should be corrected in the opposite direction. Evaluate and calibrate the gradient derivative array. This refers to the loss function of the current control strategy with respect to the network parameters. The partial derivative vector can be obtained by taking derivatives layer by layer along the network computation graph using the backpropagation algorithm, and is used to indicate the optimal direction for weight updates.

[0050] Specifically, the core mechanism of this weight update model is policy gradient reinforcement learning. (Difference term) This constitutes a simplified form of the advantage function: the reward obtained in this round of intervention. Above historical baseline When the difference is positive, the weights are updated along the gradient direction, reinforcing the proven effective decision paths in the current policy; when... Below When the difference is negative, the weights are updated in the opposite direction of the gradient, weakening the ineffective decision paths in the current policy. Learning rate Controlling the step size of each update is crucial; too large a step may cause weight oscillations and non-convergence, while too small a step size will result in slow learning speed. In practical deployment, an adaptive learning rate strategy can be adopted, setting a larger step size in the initial stage of system use. The value is used to accelerate personalized adaptation, and gradually decreases as usage time accumulates. The value is to stabilize the optimal strategy that has been learned.

[0051] After obtaining the dynamic reward value for therapeutic efficacy, the closed-loop evaluation and iteration module 55 can also associate and package this reward value with the intervention vector group, physiological abnormality precursor index, hardware distribution feedback load index, and energy consumption parameters within the corresponding control period, and upload it to the cloud storage unit. The historical data stored in the cloud can be used by doctors or researchers to analyze the patterns of migraine attacks, and can also be used by the system to subsequently call historical treatment records to optimize individualized intervention strategies.

[0052] Through the continuous cyclical operation of the above five-step closed-loop data flow, the intelligent control system 5 achieves the following: In the asynchronous data acquisition stage, the system simultaneously acquires the patient's multidimensional physiological state and the hardware operating status of the equipment, constructing a complete human-computer interaction information foundation; In the state feature extraction stage, through the fusion encoding of logarithmic nonlinear compression and linear superposition, heterogeneous data is uniformly mapped into a high-dimensional state vector that can be processed by machine learning models; In the anti-adaptive strategy generation stage, the introduction of chaotic perturbation factors makes the intervention plan of each control cycle have unpredictable slight differences, breaking the formation conditions of neural habituation from the root; In the intervention execution scheduling stage, the abstract strategy is accurately transformed into multi-acupoint multimodal physical control signals, realizing the spatiotemporal synergy of graphene thermal pulse, far-infrared radiation and transdermal drug delivery of traditional Chinese medicine; In the closed-loop evaluation and iteration stage, the efficacy differential reward signal drives the continuous evolution of the strategy network weights, enabling the system to gradually form a knowledge base for the highest priority symptom identification and intervention response specific to individual users with daily use.

[0053] As a specific embodiment of the present invention, the complete workflow of the intelligent control system is illustrated using a patient with a history of migraine with aura as an example: The patient wears the helmet-shaped main frame 1 on their head, and the elastic strap 4 automatically adapts to the head circumference and secures the device. The bioelectric sensors in the EEG signal acquisition unit 21, which is embedded in acupoints such as Fengchi, Taiyang, Baihui, and Tianzhu, begin to continuously acquire multi-channel EEG signals. Temperature sensors simultaneously monitor the skin temperature at each acupoint, and electromyography sensors detect the tension of the head muscles in real time.

[0054] When a patient is in the prodromal phase of a migraine, characteristic changes such as increased slow-wave activity and decreased alpha wave power appear in the electroencephalogram (EEG). Simultaneously, an electromyography (EMG) sensor detects increased tension in the temporal muscles, and a temperature sensor detects abnormal fluctuations in local skin temperature. The asynchronous data acquisition module 51 synchronously transmits the aforementioned multidimensional physiological modality perception data stream and the current standby state parameter stream of the multimodal intervention execution unit 22 to the state feature extraction module 52.

[0055] The state feature extraction module 52 performs frequency band filtering and power spectrum analysis on the EEG signal, extracts abnormal features in each frequency band, and calculates the multidimensional physiological abnormality precursor index at the current moment by combining muscle tension parameters. =0.72 (normalized value, 0 indicates complete normality, 1 indicates extreme abnormality). Simultaneously, the device standby state is quantified as a hardware distributed feedback load index. =0.05 (The equipment has not yet been activated for intervention; the load is extremely low). Substituting this into the high-dimensional machine state fusion model, we assume... =0.7、 =0.3、 =0.02, then: ; Should The value indicates that the patient is currently experiencing a moderate level of migraine aura.

[0056] Anti-adaptive strategy generation module 53 receives =0.4146, and the chaotic attractor algorithm generates the perturbation factor for the current period. =0.18. Assuming the current network weights... =1.2、 =0.6、 =-0.1, substitute into the adaptive action control decision model: ; The intervention execution scheduling module 54 queries the mapping table and... =0.4664 was analyzed as follows: The graphene heating microplates at the temples (bilateral) were activated to 45% power output; the far-infrared radiation module at the Baihui acupoint was activated to medium intensity; and the micro-peristaltic pumps at the Fengchi acupoints (bilateral) were activated to deliver the herbal medicine in a low-speed intermittent mode. The activation phases of each acupoint were staggered by approximately 120°, meaning the temples were activated first, followed by the Baihui acupoint after a one-third-cycle delay, and then the Fengchi acupoint after another one-third-cycle delay, forming a wave-like spatial scanning intervention pattern.

[0057] Fifteen seconds after the intervention, the closed-loop evaluation and iteration module 55 triggers the delayed supervision and assessment window. The EEG signal acquisition unit re-acquires data, and the state feature extraction module calculates the precursor index of the next-state physiological abnormality. =0.48, compared to before intervention =0.72, a decrease. The system energy consumption penalty scalar value for this round of intervention was also statistically analyzed. =0.15. Assume =1.0、 =0.3, substituting into the relative mitigation model: ; A value of 0.2883 is positive, indicating that this round of intervention achieved a positive therapeutic effect. (Assuming a historical baseline...) =0.20, then the dominance value is... =0.0883>0, the system determines that the current strategy is better than the historical average level, and strengthens the weight path in the current strategy network that is related to the current decision by using the gradient update formula.

[0058] In the next control cycle, the chaotic disturbance factor Updated to new unpredictable values ​​(e.g.) =-0.12), even if the patient's prodromal state Similar to the previous cycle, the anti-adaptation strategy generation module 53 will also output different action baseline values. Based on this, the intervention execution scheduling module 54 generates differentiated acupoint-modality-intensity combination schemes. For example, in this cycle, it may switch to: the graphene heating microplate at Fengchi acupoint is increased to 55% power, the Taiyang acupoint is switched to a low-intensity far-infrared radiation mode, and the peristaltic pump at Baihui acupoint is accelerated to medium-speed drug delivery, with the activation phase also changing accordingly. This continuously changing intervention mode prevents the patient's nervous system from establishing a prediction and adaptation to the stimulation pattern, thereby maintaining the long-term effectiveness of the intervention.

[0059] As patients continue to use the device, the closed-loop evaluation and iteration module 55 continuously accumulates individualized reward data. After hundreds to thousands of control cycles of iterative optimization, the strategy network weights gradually converge to form an optimal intervention strategy distribution specific to that patient. For example, the system may learn that the patient is most sensitive to far-infrared radiation at the Fengchi acupoint and has the most lasting relief effect from transdermal drug delivery at the Taiyang acupoint. Therefore, in subsequent interventions, the system automatically increases the probability of these effective combinations while maintaining sufficient random perturbation to prevent neural tolerance. This evolutionary learning process achieves the goal of personalized and precise treatment.

[0060] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A device for monitoring and relieving abnormal headache symptoms through stimulation intervention, characterized in that, include: Helmet-shaped main frame (1), the helmet-shaped main frame (1) is made of biomimetic elastic material, and the bottom front end of the helmet-shaped main frame (1) is equipped with an elastic strap (4). The monitoring and relief module (2) is embedded in the helmet-shaped main frame (1) at the position corresponding to the acupoints on the human head. The monitoring and relief module (2) includes an electroencephalogram (EEG) signal acquisition unit (21) and a multimodal intervention execution unit (22). The intelligent control module (3) is installed on both sides of the helmet-shaped main frame (1). The intelligent control module (3) has a built-in intelligent control system (5). The intelligent control system (5) is used to determine the signs of migraine attack by using a preset aura recognition algorithm based on the signals obtained by the EEG signal acquisition unit (21). Based on the determination result, it automatically drives the multimodal intervention execution unit (22) at the corresponding acupoint to work together. It also uploads the physiological signal data collected by the EEG signal acquisition unit (21), the execution status data of the multimodal intervention execution unit (22), and the corresponding intervention result data to the cloud storage unit for storage. The intelligent control module (3) is also provided with a USB interface on the outside. The USB interface is used to power and charge the device, transmit data, or perform program maintenance.

2. The device for monitoring and relieving abnormal headaches and providing stimulation intervention according to claim 1, characterized in that, The EEG signal acquisition unit (21) includes a bioelectric sensor, a temperature sensor and an electromyography sensor, used to collect the wearer's EEG physiological signals, skin temperature and muscle tension in real time. The multimodal intervention execution unit (22) includes a graphene heating microplate, a far-infrared radiation module and a drug delivery structure.

3. The device for monitoring and relieving abnormal headache symptoms according to claim 2, characterized in that, The drug delivery structure consists of a hollow tube for storing the drug solution, a micro peristaltic pump connected to the hollow tube, and a nitrided silica microporous drug delivery tablet integrated into the liner of the helmet-shaped main frame (1).

4. The device for monitoring and relieving abnormal headache according to any one of claims 1-3, characterized in that, The intelligent control system (5) includes: The asynchronous data acquisition module (51) is used to issue and extract specific types of data communication commands; The state feature extraction module (52) is used to perform dimensionality reduction and fusion on the collected data to construct a combination of machine space response states; The anti-adaptation strategy generation module (53) generates collaborative decision-making instructions for different acupoints to avoid neural tolerance based on the preceding state using a neural network. The intervention execution scheduling module (54) is used to receive and perform serial-to-parallel conversion of the collaborative decision-making instructions into physical protocol unpacking to drive the execution of multiple units; The closed-loop evaluation and iteration module (55) is used to establish feedback incentives and iteratively fine-tune the feature map of the network control dictionary by comparing the data updated subsequently. It also organizes the collected physiological data, hardware execution feedback data, intervention strategy data and efficacy evaluation data and uploads them to the cloud storage unit to form historical treatment records.

5. The device for monitoring and relieving abnormal headache symptoms according to claim 4, characterized in that, The asynchronous data acquisition module (51) extracts specific types of data communication instructions that only include the concurrent acquisition of the following two types of independent data streams: The first data type is a multidimensional physiological modality sensing data stream: a multi-channel time-stamped digital sequence matrix of physiological voltage and temperature states captured in real time by the EEG signal acquisition unit (21) through its bioelectric sensors, temperature sensors and electromyography sensors and transmitted to the processing buffer station. The second data type is the hardware execution feedback status parameter stream: the execution status dictionary table transmitted back and uploaded by the multimodal intervention execution unit (22) corresponding to each acupoint via the original path of the control bus. The dictionary table contains the thermal impedance, infrared radiation power consumption, peristaltic pump pressure status and load feedback information under the current physical working mode of each execution unit. The multidimensional physiological modality sensing data stream and the hardware execution feedback state parameter stream are processed by the intelligent control system (5), and then uploaded to the cloud storage unit with the addition of time stamp and user identifier information.

6. The device for monitoring and relieving abnormal headache symptoms according to claim 5, characterized in that, The specific logical implementation steps of the state feature extraction module (52) include: The acquired multidimensional physiological modality sensing data stream is processed through a preset frequency band filter to calculate the absolute frequency band energy and muscle tension parameters, extract abnormal features, and derive the multidimensional physiological abnormality precursor index at the current moment. The acquired hardware execution feedback state parameter stream is quantized and converted into the hardware distributed feedback load index at the current moment. The multidimensional physiological abnormality precursor index at the current moment and the hardware distributed feedback load index are jointly imported into the high-dimensional machine state fusion model to obtain the high-dimensional composite precursor state response value at the current moment. The high-dimensional machine state fusion model is expressed as follows: ； in, This represents the high-dimensional composite precursor state response value at the current moment. The multidimensional physiological abnormality precursor index represents the current moment. This represents the hardware distribution feedback load index at the current moment. Represents the sensitivity coefficient of physiological characteristics. Represents the hardware load weighting coefficient and satisfies + =1, Indicates a constant offset. The higher the value, the stronger the patient's neurological abnormalities and the higher the current equipment intervention load.

7. The device for monitoring and relieving abnormal headaches and providing stimulation intervention according to claim 6, characterized in that, The specific logic implementation steps of the anti-adaptation strategy generation module (53) include: The high-dimensional composite precursor state response value is used as the input of the reinforcement learning environment observation. Combined with the nonlinear chaotic perturbation sequence generated by the pseudo-random algorithm, it is imported into the anti-adaptive action control decision model. The output is a multimodal intervention collaborative strategy control action benchmark value to avoid the regular adaptation of the human nervous system. The anti-adaptive action control decision model is expressed as follows: ； in, This represents the baseline value of the multimodal intervention collaborative strategy control action output at the current moment. This represents the nonlinear chaotic feature perturbation factor generated by random calculation. This represents the weighting of the system's state space response to the shift in intervention demand. This represents the stochastic anti-fatigue weighted control coefficient of the anti-neural tolerance regulation mechanism. This represents the constant value of the bias control.

8. The device for monitoring and relieving abnormal headache symptoms according to claim 7, characterized in that, The specific logical control features of the intervention execution scheduling module (54) are as follows: Receive the baseline values ​​of the multimodal intervention collaborative strategy control actions, and parse them into an intervention intensity vector group containing multiple acupoint topological indices, spatial phase parameters, and physical working form scalars through a mapping table; The stimulation intensity of different acupoints and the multimodal output category are distributed to the multimodal intervention execution unit (22) via a communication cable bus, so that the physical control response of irregular jitter and cross random fine-tuning is presented, so as to implement the anti-adaptive synergistic operation including graphene thermal pulse oscillation, infrared radiation dynamic adjustment and micro peristaltic pump intermittent drug delivery.

9. The device for monitoring and relieving abnormal headache symptoms according to claim 8, characterized in that, In the closed-loop evaluation and iteration module (55), the specific steps for closed-loop evaluation are as follows: The system triggers a delayed supervision and evaluation window mechanism, waiting for the intervention vector group to execute for a certain time constant; The next-state multidimensional physiological deviation evolution characteristic index corresponding to this moment is re-acquired, and the difference magnitude between it and the baseline evolution state before intervention is calculated. This difference is then imported into the relative remission degree reward score model to obtain the dynamic reward value of the therapeutic effect. The relative remission model is expressed as follows: ； in, The reward score indicates the relative degree of relief during the intervention and adjustment period. This represents the baseline multidimensional physiological abnormality precursor limit index before the intervention cycle is initiated. This represents the index of the deviation characteristics of the next-state physiological evolution captured at the end of the delayed supervision window. This represents the system energy consumption penalty scalar value for the multimodal intervention actions output by the device. The reward gain coefficient represents the positive recovery effect. This represents the penalty adjustment coefficient for the energy consumption of output computing power. The dynamic reward value for therapeutic effect, along with the corresponding physiological data and execution parameters before and after the intervention, is uploaded to the cloud storage unit for long-term efficacy tracking and model optimization.

10. The device for monitoring and relieving abnormal headaches and providing stimulation intervention according to claim 9, characterized in that, In the closed-loop evaluation and iteration module (55), the method for using the self-iterative neural network control dictionary is as follows: The obtained relative mitigation reward score is used as evaluation feedback and fed back into the hidden layer nodes of the anti-adaptive instruction neural network. This initiates the gradient descent correction model to implement automatic advancement of the network's internal policy link weights. Its update partial derivative model is expressed as: ； in, This represents the next-generation transition evaluation weight matrix after the update and reconstruction are completed. This represents the ground-state weight matrix for evaluating the adaptive action output in the current cycle. This parameter represents the network's self-learning and optimization exploration rate. This represents the baseline score of the cumulative expected evaluation of predecessors along the state transition path in reinforcement learning. This represents the evaluation calibration gradient derivative array based on the loss function of the current control strategy; The baseline score for the cumulative precursor expectation evaluation is determined jointly based on local real-time data and historical reward data in cloud storage units to improve the long-term optimization effect of individualized intervention strategies.

Images