Mechanical structure brain-computer interface control method and system based on whole brain kinetic model

Abstract

The invention belongs to the technical field of brain-computer interface and motion control, and discloses a mechanical structure brain-computer interface control method and system based on a whole brain dynamic model, and the method mainly comprises the following steps: constructing the whole brain dynamic model simulating a brain operation mechanism; selecting circuit elements to build a circuit model; a multi-lead device is used for collecting the electroneurographic signals, and electroencephalogram data are obtained; effective signal features are extracted from the collected electroencephalogram data; inputting the preprocessed signal into a whole brain dynamic model, and extracting main characteristic components to decode a motion intention; designing and constructing a simple mechanical system hardware structure with enough degree of freedom and motion range; generating a control instruction according to the decoded brain motion intention and sending the control instruction to a mechanical structure driving system; a self-adaptive control strategy is introduced to adjust model parameters and a control algorithm; the system is optimized by collecting user electroencephalogram and mechanical structure feedback data to adapt to different user characteristics and habits, and the problems in the prior art can be solved.

Description

Technical Field

[0001] This invention belongs to the field of brain-computer interface and motion control technology, specifically relating to a mechanical structure brain-computer interface control method and system based on a whole-brain dynamics model. Background Technology

[0002] Brain-computer interface (BCI) technology has shown great potential in fields such as medical rehabilitation and assistive movement. Non-invasive BCIs are enabling intelligent rehabilitation aids by collecting and processing electromyographic and neuroelectrical signals generated by human movement to identify the user's movement intentions, simulating bionic neuromuscular control pathways, thereby achieving motion control and meeting people's needs in daily life.

[0003] As human-computer interaction scenarios become increasingly complex and diverse, and the individual needs and scenario adaptation requirements of different users continue to rise, the control logic and technical architecture of existing brain-computer interface technologies are no longer able to meet the diverse and highly adaptable application needs, and the application limitations of the technology are gradually becoming apparent.

[0004] Currently, non-invasive brain-computer interface technology is mainly applied in the field of intelligent rehabilitation aids. By collecting physiological signals such as electromyography and nerve electrical signals generated during human movement, the signals are processed by professional algorithms to identify the user's movement intentions, and then simulate the bionic neuromuscular control pathway to achieve movement control, thereby helping users complete daily actions and meet their basic life and movement assistance needs.

[0005] Existing brain-computer interface technologies generally suffer from limitations in their technical architecture. Their control logic is mostly based on signals from local brain regions, resulting in a relatively simple control method that cannot adapt to the diverse interaction needs across multiple scenarios. Furthermore, the technology lacks generalization ability, making it difficult to cope with individual differences in neural signals among different users and complex and ever-changing real-world application scenarios. Ultimately, this limits the practical application effect of the technology and fails to fully meet the core needs of complex human-computer interaction. Summary of the Invention

[0006] The purpose of this invention is to solve the problems of low control accuracy and poor adaptability of traditional brain-computer interfaces in the prior art, and to provide a mechanical structure brain-computer interface control method and system based on a whole-brain dynamics model.

[0007] To achieve the above objectives, the present invention employs the following technical solution: The present invention proposes a brain-computer interface control method for mechanical structures based on a whole-brain dynamics model, comprising the following steps: Preprocessed multi-lead EEG data was acquired, a whole-brain dynamics model was constructed, and the preprocessed multi-lead EEG data was input into the whole-brain dynamics model to obtain the motor intention features of the EEG images. Based on the motion intention features of EEG images, corresponding motion control commands are generated and sent to the drive system of the mechanical structure to realize the manipulation of the mechanical structure. Based on the actual motion feedback data of the mechanical structure and the real-time changes in brain signals, an adaptive control strategy is introduced to dynamically adjust the parameters of the whole-brain dynamics model, thereby achieving control of the mechanical structure.

[0008] Preferably, the acquisition of preprocessed multi-lead EEG data specifically includes: A cortico-thalamic-striatal pathway model was established, and signals from different brain regions were simultaneously acquired based on the model and multi-lead equipment. The acquired signals were preprocessed to obtain preprocessed multi-lead EEG data.

[0009] Preferably, the establishment of the whole-brain dynamics model specifically includes: A whole-brain dynamics model was constructed using the Wilson-Cowan model. The whole-brain dynamics model simulates the dynamic behavior of neural networks by describing the average activity of excitatory and inhibitory neuronal populations. The model equations are as follows:

[0010] Where E and I represent the average activity levels of excitatory and inhibitory neuronal populations, respectively; All are model parameters; All are non-linear activation functions; External input to a population of excitatory neurons. External input refers to the inhibitory neuron population.

[0011] Preferably, the step of inputting the preprocessed multi-lead EEG data into the whole-brain dynamics model to obtain the motor intention features of the EEG images specifically involves: The preprocessed multi-lead EEG data was input into the whole-brain dynamics model, and principal component analysis was used to extract features and reduce the dimensionality of the multi-lead EEG data to obtain the motion intention features of the EEG images.

[0012] Preferably, the step of generating corresponding motion control commands based on the motion intention features of EEG images and sending the motion control commands to the drive system of the mechanical structure to realize the manipulation of the mechanical structure specifically involves: Based on the whole-brain dynamics model, the population activity of excitatory neurons was obtained. With inhibitory neuron population activity The real-time state value, combined with the output signal strength of the cortico-thalamic-striatal circuit model Calculate the motion intention feature vector Then introduce control parameters Construct a mapping matrix ; motion intent feature vector Through the mapping matrix Mapped to drive command vectors of mechanical structures .

[0013] Preferably, the step of introducing an adaptive control strategy based on the actual motion feedback data of the mechanical structure and the real-time changes in brain signals to dynamically adjust the parameters of the whole-brain dynamics model and achieve control of the mechanical structure specifically involves: Based on driver instruction vector Simultaneously, the real-time displacement of the mechanical structure is collected. , Substituting the velocities v1 and v2 into the dynamic equations The theoretical motion state is calculated and compared with the actual feedback motion state to obtain the error value. , the error value Input parameter adjustment formula:

[0014] Update control parameters By modifying the driving command vector U through the mapping matrix W, a closed-loop control process of model output-command generation-feedback correction-parameter update is formed. Through continuous iteration, the actual motion state of the mechanical structure gradually approaches the target state predicted by the model, thereby achieving precise tracking control of simple mechanical structures. in, These are the parameters of the current whole-brain model and the control mapping matrix; By calculating the control error cost function right The partial derivatives, if Follow As the value increases, the partial derivative becomes positive; therefore, it needs to decrease. Conversely, the adjustment direction is increased to ensure that the adjustment direction always points in the direction that reduces the error. Adjust the step size for learning rate control; For system energy, For the threshold, Increasing the value of the exponential term decreases the output, reducing the parameter adjustment range and maintaining system stability; when the system approaches the target state... Decreasing the exponential term increases the output, and the parameter adjustment range improves, leading to faster convergence of errors.

[0015] Preferably, the mapping matrix W is as follows:

[0016] in, These are control parameters. It is the i-th component in the motion intention feature vector F. It is the k-th component in the motion intention feature vector F, where k is the summation index.

[0017] This invention proposes a mechanical brain-computer interface control system based on a whole-brain dynamics model, comprising: The feature extraction module is used to acquire preprocessed multi-lead EEG data, construct a whole-brain dynamics model, input the preprocessed multi-lead EEG data into the whole-brain dynamics model, and acquire the motion intention features of the EEG images. The feature mapping module is used to generate corresponding motion control commands based on the motion intention features of the electroencephalogram image, and send the motion control commands to the drive system of the mechanical structure to realize the manipulation of the mechanical structure. The parameter adaptive adjustment module is used to dynamically adjust the parameters of the whole brain dynamics model based on the actual motion feedback data of the mechanical structure and the real-time change data of brain signals, thereby achieving control of the mechanical structure.

[0018] Preferably, the establishment of the whole-brain dynamics model specifically includes: A whole-brain dynamics model was constructed using the Wilson-Cowan model. The whole-brain dynamics model simulates the dynamic behavior of neural networks by describing the average activity of excitatory and inhibitory neuronal populations. The model equations are as follows:

[0019] Where E and I represent the average activity levels of excitatory and inhibitory neuronal populations, respectively; All are model parameters; All are non-linear activation functions; All are external inputs. External input to a population of excitatory neurons. External input refers to the inhibitory neuron population.

[0020] Preferably, the step of generating corresponding motion control commands based on the motion intention features of EEG images and sending the motion control commands to the drive system of the mechanical structure to realize the manipulation of the mechanical structure specifically involves: Based on the whole-brain dynamics model, the population activity of excitatory neurons was obtained. With inhibitory neuron population activity The real-time state value, combined with the output signal strength of the cortico-thalamic-striatal circuit model Calculate the motion intention feature vector Then introduce control parameters Construct a mapping matrix ; motion intent feature vector Through the mapping matrix Mapped to drive command vectors of mechanical structures .

[0021] Compared with the prior art, the present invention has the following beneficial effects: This invention proposes a brain-computer interface control method for mechanical structures based on a whole-brain dynamics model. First, preprocessed multi-lead EEG data is acquired and a whole-brain dynamics model is constructed. The multi-lead data is input into the model to extract motor intention features from the EEG images. Unlike traditional methods that rely solely on signals from local brain regions, multi-lead EEG data covers multiple functional areas of the brain, comprehensively collecting information on the correlation of neural electrical activity in different brain regions. The whole-brain dynamics model integrates this information at multiple scales, from micro to macro, accurately capturing the collaborative dynamic interaction process between brain regions, thus avoiding the problem of one-sided interpretation of motor intention caused by local signals. Second, motor control commands are generated based on the motor intention features output by the whole-brain model and sent to the mechanical drive system. Because this feature is a comprehensive intention representation at the whole-brain level, rather than a single mapping of local signals, the generated control commands can more accurately match the user's actual movement needs, reducing mechanical manipulation errors caused by intention decoding deviations and improving the control accuracy of the mechanical structure. Finally, an adaptive control strategy was introduced, which dynamically adjusted the parameters of the whole brain dynamics model by combining the actual mechanical motion feedback and the real-time changes in brain signals. This formed a closed-loop control link for EEG signals, which can correct control errors caused by fluctuations in brain signals and deviations in mechanical operation in real time. This allows the system to adapt to signal changes and mechanical operating states in different scenarios, effectively overcoming the limitations of traditional technologies in terms of adaptability and generalization, and ensuring the stability and flexibility of mechanical control. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart of the mechanical structure brain-computer interface control method based on the whole-brain dynamics model of the present invention.

[0024] Figure 2 This is a detailed flowchart of the mechanical structure brain-computer interface control method based on the whole-brain dynamics model of the present invention.

[0025] Figure 3This is a timing diagram for motion intent decoding and instruction generation in this invention.

[0026] Figure 4 This is a hardware structure diagram of the mechanical system of the present invention.

[0027] Figure 5 This is a timing diagram for the adaptive control parameter adjustment of the present invention.

[0028] Figure 6 This is a diagram of the mechanical structure brain-computer interface control system based on the whole-brain dynamics model of the present invention. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0030] The present invention will now be described in further detail with reference to the accompanying drawings: This invention proposes a brain-computer interface control method for mechanical structures based on a whole-brain dynamics model. By constructing a whole-brain dynamics model and combining it with multi-channel information, a personalized brain-computer interface algorithm is developed, thereby achieving efficient and precise control of simple mechanical structures and overcoming the limitations of traditional brain-computer interface technology in terms of control accuracy, generalization ability, and application scenarios.

[0031] The present invention adopts the following technical solution: A brain-computer interface control method for mechanical structures based on a whole-brain dynamics model, such as Figure 1 As shown, it includes the following steps: S1. Obtain preprocessed multi-lead EEG data, construct a whole-brain dynamics model, input the preprocessed multi-lead EEG data into the whole-brain dynamics model, and obtain the motor intention features of the EEG images. The acquisition of preprocessed multi-lead EEG data specifically involves: A cortico-thalamic-striatal pathway model was established, and signals from different brain regions were simultaneously acquired based on the model and multi-lead equipment. The acquired signals were preprocessed to obtain preprocessed multi-lead EEG data.

[0032] The establishment of the whole-brain dynamics model specifically involves: A whole-brain dynamics model was constructed using the Wilson-Cowan model. The whole-brain dynamics model simulates the dynamic behavior of neural networks by describing the average activity of excitatory and inhibitory neuronal populations. The model equations are as follows:

[0033] Where E and I represent the average activity levels of excitatory and inhibitory neuronal populations, respectively; All are model parameters; All are non-linear activation functions; External input to a population of excitatory neurons. External input refers to the inhibitory neuron population.

[0034] The process of inputting preprocessed multi-lead EEG data into a whole-brain dynamics model to obtain motor intention features from EEG images specifically involves: The preprocessed multi-lead EEG data was input into the whole-brain dynamics model, and principal component analysis was used to extract features and reduce the dimensionality of the multi-lead EEG data to obtain the motion intention features of the EEG images.

[0035] S2. Generate corresponding motion control commands based on the motion intention features of EEG images, and send the motion control commands to the drive system of the mechanical structure to realize the manipulation of the mechanical structure. The process of generating corresponding motion control commands based on EEG image motion intention features and sending these commands to the drive system of the mechanical structure to control the mechanical structure is as follows: Based on the whole-brain dynamics model, the population activity of excitatory neurons was obtained. With inhibitory neuron population activity The real-time state value, combined with the output signal strength of the cortico-thalamic-striatal circuit model Calculate the motion intention feature vector Then introduce control parameters Construct a mapping matrix ; motion intent feature vector Through the mapping matrix Mapped to drive command vectors of mechanical structures .

[0036] S3. Based on the actual motion feedback data of the mechanical structure and the real-time changes in brain signals, an adaptive control strategy is introduced to dynamically adjust the parameters of the whole-brain dynamics model, thereby achieving control of the mechanical structure.

[0037] Based on the actual motion feedback data of the mechanical structure and the real-time changes in brain signals, an adaptive control strategy is introduced to dynamically adjust the parameters of the whole-brain dynamics model, thereby achieving control of the mechanical structure. Specifically: Based on the drive command vector, the real-time displacement of the mechanical structure is simultaneously acquired. , Substituting the velocities v1 and v2 into the dynamic equations The theoretical motion state is calculated and compared with the actual feedback motion state to obtain the error value. , the error value Input parameter adjustment formula:

[0038] Update control parameters By modifying the driving command vector U through the mapping matrix W, a closed-loop control process of model output-command generation-feedback correction-parameter update is formed. Through continuous iteration, the actual motion state of the mechanical structure gradually approaches the target state predicted by the model, thereby achieving precise tracking control of simple mechanical structures. in, These are the parameters of the current whole-brain model and the control mapping matrix; By calculating the control error cost function right The partial derivatives, if Follow As the value increases, the partial derivative becomes positive; therefore, it needs to decrease. Conversely, the adjustment direction is increased to ensure that the adjustment direction always points in the direction that reduces the error. Adjust the step size for learning rate control; For system energy, For the threshold, Increasing the value of the exponential term decreases the output, reducing the parameter adjustment range and maintaining system stability; when the system approaches the target state... Decreasing the exponential term increases the output, and the parameter adjustment range improves, leading to faster convergence of errors.

[0039] The mapping matrix W is as follows:

[0040] in, These are control parameters. This is the i-th component in the motion intention feature vector F, where i takes values ​​of 1, 2, and 3, corresponding to the three dimensions of F: when i=1, F1=E, the activity level of the excitatory neuron population; when i=2, F2=I, the activity level of the inhibitory neuron population; and when i=3, F3=S, the output signal intensity of the cortical-thalamic-striatal circuit. It is the k-th component in the motion intention feature vector F, where k is the summation index, which also corresponds to the three dimensions of F and is only used as a summation index for normalization calculation.

[0041] explain: These are control mapping parameters between the whole-brain dynamics model and the mechanical structure's driving commands, used to adjust the motion intention feature vector. The weight strength of the mapping to the driving command vector U directly affects the accuracy of the conversion from intent decoding to mechanical action. In the mapping matrix formula, As a global adjustment factor, it participates in constructing the mapping matrix W in the closed-loop control process. The core parameter for dynamic adjustment is updated through error feedback. And then through the revised The mapping matrix W is updated, and the driving instruction U is finally corrected to achieve error convergence.

[0042] like Figure 2 The method is described in detail below: Step 1: Based on the brain's regional coordination, synchronous dynamics, and structural plasticity, a whole-brain dynamics model capable of simulating the brain's operating mechanisms is constructed using relevant neuroscience theories and mathematical modeling methods. The Wilson-Cowan model is used to construct the whole-brain dynamics model, which simulates the dynamic behavior of neural networks by describing the average activity of excitatory and inhibitory neuronal populations.

[0043] The model equations are as follows:

[0044] Where E and I represent the average activity levels of excitatory and inhibitory neuronal populations, respectively; The model parameters are set with reference to the physiological characteristics of excitatory and inhibitory neuron populations. The initial parameter range is then set. Subsequently, combined with the EEG data of different users and the mechanical structure motion feedback data collected by the multi-lead device, the parameters are dynamically adjusted by introducing an adaptive control strategy to adapt to real-time signal changes. At the same time, machine learning algorithms such as LSTM and CNN are used to train the data and continuously optimize the parameter values. Finally, the parameters meet the simulation requirements of the whole brain dynamics model for brain neural activity. It is a non-linear activation function, usually chosen as the sigmoid function; For external input, External input to a population of excitatory neurons. This refers to the external input of inhibitory neuron populations. The model should integrate multiple scales from micro to macro and be able to capture the dynamic interaction of information.

[0045] Whole-brain dynamics models comprehensively consider the complex connections and interactions between different brain regions, fully simulating the brain's neural activity and information processing mechanisms, thus providing a foundation for more accurate interpretation of brain signals. By utilizing multi-channel information from whole-brain models, the brain's motor intentions can be identified more accurately, improving the control precision and generalization ability of brain-computer interfaces. This allows them to adapt to different users and task requirements, achieving efficient, natural, and precise control of simple mechanical structures, expanding their applications in medical rehabilitation, assistive movement, and other fields, and laying the foundation for widespread application in the consumer market in the future.

[0046] Step 2: Select circuit components to establish a circuit model of the cortex-thalamus-striatum. The activities of excitatory and inhibitory neuronal populations are realized by different operational amplifier circuits. Nonlinear activation functions can be implemented using nonlinear circuit elements such as diodes and transistors; connection weights can be implemented using adjustable resistors or amplifiers with adjustable gain; delays can be simulated using a combination of capacitors and resistors in the circuit.

[0047] The specific process for circuit modeling is as follows: 2.1: Based on the mathematical description of the Wilson-Cowan model and combined with the neurophysiological mechanisms of the cortex-thalamus-striatal pathway, the activity equations of excitatory (E) and inhibitory (I) neuronal populations in the model are transformed into circuit functional requirements. Referring to the general logic of neurodynamic modeling, the activity simulation of excitatory neuronal populations needs to match the positive signal transmission characteristics, while inhibitory neuronal populations need to achieve reverse regulation functions. At the same time, based on the connection mechanism between the cortex-thalamus-striatal nuclei, the signal flow and interaction relationships of different modules in the circuit are determined.

[0048] 2.2: Selection and functional mapping of core circuit components. For excitatory neuronal populations, operational amplifiers were used to build amplification circuits, with adjustable resistors used to set the basic activity threshold to simulate their signal enhancement and conduction characteristics. For inhibitory neuronal populations, an inverting operational amplifier circuit was used to achieve signal suppression, ensuring that the activities of the two neuronal populations conformed to the dynamic balance between E and I in the model. Nonlinear components such as diodes and transistors were introduced to reproduce the sigmoid activation function in the Wilson-Cowan model. At the same time, a combination of capacitors and resistors was used to simulate the delay in neural signal transmission, and a gain amplifier was used to adjust the connection weights between nuclei to accurately match the neural interaction characteristics of the pathway.

[0049] 2.3: Circuit module integration and parameter calibration. Following the physiological pathway structure of the cortex-thalamus-striatum, circuit modules representing excitatory and inhibitory neuronal populations were sequentially connected to construct a complete signal transmission link. Combining the preprocessing results of multi-lead EEG data and referencing parameter optimization methods for neural circuit modeling, parameters such as resistance and capacitance were adjusted to ensure the circuit output signal closely matches the neuronal activity patterns predicted by the model. Simultaneously, parameter adjustment interfaces were reserved to provide physical support for dynamic parameter adjustments in subsequent adaptive control strategies, ensuring that the circuit model accurately simulates the dynamic characteristics of the whole brain while adapting to individual differences in actual EEG signals.

[0050] Step 3: Use a multi-lead device to simultaneously collect neural electrical signals from different areas of the brain to obtain rich EEG data, which will provide a foundation for subsequent signal analysis and motor intention decoding.

[0051] Using methods such as electroencephalography (EEG), electrocorticography (ECoG), and deep cortical electroencephalography (SEEG), neural electrical signals from different regions of the brain are simultaneously collected by electrodes placed on the scalp or on the surface and deep within the brain, obtaining rich EEG data to provide a foundation for subsequent signal analysis and motor intention decoding.

[0052] Step 4: Preprocess the acquired multi-lead EEG data, including filtering to remove noise and extracting signal features, to remove interference information and extract key data that effectively characterize brain neural activity. In the signal preprocessing stage, filtering algorithms are used to remove noise. For example, the transfer function of the Butterworth filter is expressed as:

[0053] in, The transfer function of the Butterworth filter describes the frequency response characteristics of the filter to the input signal, that is, the ratio of the output signal to the input signal in the complex frequency domain. It is the core mathematics for signal filtering and denoising. Representing complex frequency variables, it is a key variable in complex frequency domain analysis, used to uniformly describe the amplitude and frequency characteristics of a signal, and provides a basis for the mathematical derivation of the frequency response of a filter; The cutoff frequency is the critical value at which a filter distinguishes the frequency components of a signal. It is used to define the effective signal frequency range that needs to be retained and the noise frequency range that needs to be filtered out. Signal components exceeding this frequency will be attenuated. The order of the filter determines the frequency attenuation rate of the filter. The higher the order, the stronger the filter's ability to distinguish signals on both sides of the cutoff frequency, the narrower the frequency transition band, the steeper the filtering effect, and the more accurately it can separate effective signals from noise. Step 5: Input the preprocessed multi-lead signals into the whole-brain dynamics model constructed based on the Wilson-Cowan model. Principal component analysis (PCA) is used to extract features and reduce dimensionality, removing noise and redundant information, and extracting the main feature components. These feature components can better match the neuronal activity patterns in the whole-brain model, thus more accurately decoding the brain's motor intentions. Mathematically, PCA can be represented as: Where X is the preprocessed signal data matrix, W is the projection matrix composed of principal component directions, and Y is the dimensionality-reduced feature matrix.

[0054] Step 6: Design and construct the hardware structure of a simple mechanical system, such as... Figure 4 This ensures the mechanical system has sufficient degrees of freedom and range of motion to meet the brain's demands for expressing motor intentions. Appropriate sensors are installed to monitor the robotic arm's motion status and environmental feedback in real time, providing data support for subsequent adaptive control.

[0055] Figure 4 It has two degrees of freedom. , The dynamic model of the (direction) coupled mechanical system combines electromagnetic drive, spring force and coupling to simulate a simple mechanical structure for brain-computer interface control. The figure contains two motion modules. Displacement and mass block in directional motion; for Displacement and mass block in directional motion; for The main spring in the direction; for The main spring in the direction; For connection and The coupling springs enable coordinated movement in both directions; for Directional electromagnetic drive current; for The direction of the electromagnetic driving current; the motion of the system is driven by electromagnetic force, spring force, and coupling force, and the physical meaning of each force is as follows: Main spring force : The force of the main spring in the direction of displacement (Hooke's Law, the negative sign indicates that it is opposite to the direction of displacement). : The elastic force of the main directional spring; Coupled spring force : The force of the coupling spring; Electromagnetic driving force The electromagnetic force formula is the power source for the conversion of brain-computer interface control commands. For the number of coil turns, For magnetic induction intensity, The length of the conductor; : Directional displacement The rate of change equals the velocity ;

[0056] illustrate The acceleration in the direction is determined by the electromagnetic driving force, damping force, main spring force, and coupling spring force. for Directional damping coefficient; Displacement in direction ,speed The principle governing these changes is the same as above; By converting the motion intentions output by the brain-computer interface into actual displacement motion through the force and dynamics relationship of the mechanical system, and achieving multi-degree-of-freedom coordinated control through coupling springs, the brain-computer interface meets the requirements of precise control of simple mechanical structures. Through the coupling spring, the two axes interact and coordinate their movements, maintaining a degree of independence while enabling collaborative operation when needed, thus providing greater flexibility for complex motion control. This mechanical structure employs a direct drive method, where current directly generates driving force without transmission backlash. This not only improves positioning accuracy but also achieves rapid response and precise control, further enhancing its advantages in high-precision positioning and compliant interaction applications. It is suitable for applications such as robot end effectors and precision positioning platforms.

[0057] Step 7: Based on the brain motor intention information decoded from the whole-brain dynamics model, such as... Figure 3 The system generates corresponding control commands and sends them to the drive system of a simple mechanical structure, driving it to perform corresponding motion operations, thus enabling the human brain to precisely control the mechanical structure.

[0058] Step 8, as follows Figure 5 An adaptive control strategy is introduced to dynamically adjust the parameters and control algorithm of the whole-brain dynamics model based on the actual motion feedback of the mechanical structure and the real-time changes of brain signals, so as to improve the robustness and control accuracy of the system and achieve efficient and stable control of simple mechanical structures. Based on the motor intention parameters output from the whole-brain dynamics model and the control parameters adjusted by the adaptive control strategy, a multi-input multi-output control mapping algorithm logic is constructed. First, the preprocessed brain signal feature vector... Inputting the Wilson-Cowan model yields the population activity of excitatory neurons. With inhibitory neuron population activity The real-time state value, combined with the output signal strength of the cortico-thalamic-striatal circuit model Calculate the motion intention feature vector Then, a mapping matrix is ​​constructed by introducing the control parameter λ. ;

[0059] Will pass Mapped to drive command vectors of mechanical structures , Includes electromagnetic drive current The target value. Simultaneously, the real-time displacement of the mechanical structure is acquired. , Substituting the velocities v1 and v2 into the dynamic equations The theoretical motion state is calculated and compared with the actual feedback motion state to obtain the error value. ,Will Input parameter adjustment formula:

[0060] renew Then, the driving command vector U is corrected by the mapping matrix W, forming a closed-loop control process of model output - command generation - feedback correction - parameter update. This process continuously iterates to make the actual motion state of the mechanical structure gradually approach the target state predicted by the model, and finally achieves accurate tracking control of simple mechanical structures. This closed-loop logic also provides dynamic and accurate control data support for subsequent personalized model training. Furthermore, in the parameter adjustment formula, It is the core parameter of the current whole-brain model and control mapping matrix, which directly affects the accuracy of the conversion from motor intention to driving command; By calculating the control error cost function right The partial derivatives, if Follow As the value increases, the partial derivative becomes positive; therefore, it needs to decrease. Conversely, it increases if the error decreases, ensuring that the adjustment direction always points in the direction of error reduction; η, as the learning rate, controls the adjustment step size to avoid system oscillations caused by sudden parameter changes. The exponential... The system adapts to real-time conditions: when mechanical movement deviations are large and brain signal fluctuations are strong, the system energy... Deviation from threshold , Increasing the value of the exponential term decreases the output, reducing the parameter adjustment range to maintain system stability; when the system approaches the target state... Decreasing the value increases the output of the exponential term, thereby increasing the parameter adjustment range to quickly converge the error. The logic of the entire parameter adjustment formula is to first determine the adjustment direction and basic range based on the error, and then dynamically scale the range according to the real-time state of the system. This ensures both the accuracy of the parameter adjustment and the robustness of the system, so that the control parameters always match the current control scenario, providing a precise basis for parameter updates for the continuous iteration of closed-loop control.

[0061] Step 9: Collect the user's EEG data and corresponding mechanical structure motion feedback data, and use machine learning algorithms (such as Long Short-Term Memory Network LSTM, Convolutional Neural Network CNN, etc. in deep learning) to train the data, build a personalized brain-computer interface model, and continuously optimize the model parameters so that the system can better adapt to the EEG characteristics and operating habits of different users, thereby improving the system's personalized adaptability and intelligence.

[0062] The personalized brain-computer interface model is an improvement on the Wilson-Cowan whole-brain dynamics model. Improvement 1: Through machine learning algorithms, the model learns the characteristic differences of EEG signals of different users (such as the individual fluctuations of excitatory neuronal population activity parameters aE and bE of different users). Improvement 2: The parameters of the original model are fixed values. After improvement, the parameter adjustment formula in the adaptive control strategy is combined, that is, the parameter adjustment formula in step 8.

[0063] By continuously optimizing the model parameters, specifically the computational results of training a personalized brain-computer interface model using machine learning algorithms, this optimization process takes the user's EEG data collected by the multi-lead device in step 3 and the motion feedback data of the mechanical structure in step 7 as inputs. It combines the signal preprocessing results in step 4, the PCA feature extraction results in step 5, and the model parameters adjusted by the adaptive control strategy in step 8. Through iterative training using machine learning algorithms such as LSTM and CNN, the model parameter values ​​are continuously corrected, and finally, personalized model parameters adapted to different users' EEG characteristics and operating habits are output, enabling the system to better match individual user differences and improve the accuracy and adaptability of control.

[0064] Therefore, the method proposed in this invention first constructs a whole-brain dynamics model based on brain characteristics using the Wilson-Cowan model, establishes a related circuit model by combining circuit components, and collects EEG data through a multi-lead device. Subsequently, after preprocessing such as filtering, signal features are extracted using PCA, input to the whole-brain model to decode motor intentions, and then control commands are generated to drive simple mechanical structures. During this process, an adaptive control strategy is introduced to dynamically adjust parameters, improving the system's robustness and control accuracy. Finally, relevant data is collected and trained using machine learning algorithms to construct a personalized brain-computer interface model, achieving efficient and precise control of simple mechanical structures and overcoming the limitations of traditional brain-computer interface technology.

[0065] This invention proposes a mechanical brain-computer interface control system based on a whole-brain dynamics model, such as... Figure 6 ,include: The feature extraction module is used to acquire preprocessed multi-lead EEG data, construct a whole-brain dynamics model, input the preprocessed multi-lead EEG data into the whole-brain dynamics model, and acquire the motion intention features of the EEG images. The establishment of the whole-brain dynamics model specifically involves: A whole-brain dynamics model was constructed using the Wilson-Cowan model. The whole-brain dynamics model simulates the dynamic behavior of neural networks by describing the average activity of excitatory and inhibitory neuronal populations. The model equations are as follows:

[0066] Where E and I represent the average activity levels of excitatory and inhibitory neuronal populations, respectively; All are model parameters; All are non-linear activation functions; All are external inputs. External input to a population of excitatory neurons. External input refers to the inhibitory neuron population.

[0067] The feature mapping module is used to generate corresponding motion control commands based on the motion intention features of the electroencephalogram image, and send the motion control commands to the drive system of the mechanical structure to realize the manipulation of the mechanical structure. The process of generating corresponding motion control commands based on EEG image motion intention features and sending these commands to the drive system of the mechanical structure to control the mechanical structure is as follows: Based on the whole-brain dynamics model, the population activity of excitatory neurons was obtained. With inhibitory neuron population activity The real-time state value, combined with the output signal strength of the cortico-thalamic-striatal circuit model Calculate the motion intention feature vector Then introduce control parameters Construct a mapping matrix ; motion intent feature vector Through the mapping matrix Mapped to drive command vectors of mechanical structures .

[0068] The parameter adaptive adjustment module is used to dynamically adjust the parameters of the whole brain dynamics model based on the actual motion feedback data of the mechanical structure and the real-time change data of brain signals, thereby achieving control of the mechanical structure.

[0069] The advantages of this invention are as follows: 1) This invention introduces a whole-brain dynamics model, integrating information from the microscopic to the macroscopic multi-scale to accurately capture the dynamic interaction process of brain signals. Compared with traditional brain-computer interfaces that are based only on signals from local brain regions, this model provides a more comprehensive and accurate decoding of brain motor intentions for mechanical structure control, making control more natural and efficient, and breaking through the limitations of traditional technologies in terms of control precision and generalization ability.

[0070] 2) By designing and constructing the hardware structure of a simple mechanical system, and combining it with an adaptive control strategy, the movement of the mechanical structure is dynamically adjusted in real time according to brain signals, so that the mechanical structure can respond to the brain's intentions more accurately and flexibly.

[0071] 3) Integrate machine learning algorithms into the brain-computer interface system, collect user EEG data and mechanical structure motion feedback data for training, and build a personalized brain-computer interface model.

[0072] 4) This invention, by constructing a whole-brain dynamics model and developing personalized brain-computer interface algorithms, helps to explore brain function and neural signal processing mechanisms in depth, providing new tools and methods for neuroscience research.

[0073] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A brain-computer interface control method for mechanical structures based on a whole-brain dynamics model, characterized in that, Includes the following steps: Preprocessed multi-lead EEG data was acquired, a whole-brain dynamics model was constructed, and the preprocessed multi-lead EEG data was input into the whole-brain dynamics model to obtain the motor intention features of the EEG images. Based on the motion intention features of EEG images, corresponding motion control commands are generated and sent to the drive system of the mechanical structure to realize the manipulation of the mechanical structure. Based on the actual motion feedback data of the mechanical structure and the real-time changes in brain signals, an adaptive control strategy is introduced to dynamically adjust the parameters of the whole-brain dynamics model, thereby achieving control of the mechanical structure.

2. The brain-computer interface control method for mechanical structures based on a whole-brain dynamics model according to claim 1, characterized in that, The acquisition of preprocessed multi-lead EEG data specifically involves: A cortico-thalamic-striatal pathway model was established, and signals from different brain regions were simultaneously acquired based on the model and multi-lead equipment. The acquired signals were preprocessed to obtain preprocessed multi-lead EEG data.

3. The brain-computer interface control method for mechanical structures based on a whole-brain dynamics model according to claim 1, characterized in that, The establishment of the whole-brain dynamics model specifically involves: A whole-brain dynamics model was constructed using the Wilson-Cowan model. The whole-brain dynamics model simulates the dynamic behavior of neural networks by describing the average activity of excitatory and inhibitory neuronal populations. The model equations are as follows: Where E and I represent the average activity levels of excitatory and inhibitory neuronal populations, respectively; All are model parameters; All are non-linear activation functions; External input to a population of excitatory neurons. External input refers to the inhibitory neuron population.

4. The brain-computer interface control method for mechanical structures based on a whole-brain dynamics model according to claim 1, characterized in that, The process of inputting preprocessed multi-lead EEG data into a whole-brain dynamics model to obtain motor intention features from EEG images specifically involves: The preprocessed multi-lead EEG data was input into the whole-brain dynamics model, and principal component analysis was used to extract features and reduce the dimensionality of the multi-lead EEG data to obtain the motion intention features of the EEG images.

5. The brain-computer interface control method for mechanical structures based on a whole-brain dynamics model according to claim 1, characterized in that, The process of generating corresponding motion control commands based on EEG image motion intention features and sending these commands to the drive system of the mechanical structure to control the mechanical structure is as follows: Based on the whole-brain dynamics model, the population activity of excitatory neurons was obtained. With inhibitory neuron population activity The real-time state value, combined with the output signal strength of the cortico-thalamic-striatal circuit model Calculate the motion intention feature vector Then introduce control parameters Construct a mapping matrix ; motion intent feature vector Through the mapping matrix Mapped to drive command vectors of mechanical structures .

6. The brain-computer interface control method for mechanical structures based on a whole-brain dynamics model according to claim 1, characterized in that, Based on the actual motion feedback data of the mechanical structure and the real-time changes in brain signals, an adaptive control strategy is introduced to dynamically adjust the parameters of the whole-brain dynamics model, thereby achieving control of the mechanical structure. Specifically: Based on driver instruction vector Simultaneously, the real-time displacement of the mechanical structure is collected. , Substituting the velocities v1 and v2 into the dynamic equations The theoretical motion state is calculated and compared with the actual feedback motion state to obtain the error value. , the error value Input parameter adjustment formula: Update control parameters By modifying the driving command vector U through the mapping matrix W, a closed-loop control process of model output-command generation-feedback correction-parameter update is formed. Through continuous iteration, the actual motion state of the mechanical structure gradually approaches the target state predicted by the model, thereby achieving precise tracking control of simple mechanical structures. in, These are the parameters of the current whole-brain model and the control mapping matrix; By calculating the control error cost function right The partial derivatives, if Follow As the value increases, the partial derivative becomes positive; therefore, it needs to decrease. Conversely, the adjustment direction is increased to ensure that the adjustment direction always points in the direction that reduces the error. Adjust the step size for learning rate control; For system energy, For the threshold, Increasing the value of the exponential term decreases the output, reducing the parameter adjustment range and maintaining system stability; when the system approaches the target state... Decreasing the exponential term increases the output, and the parameter adjustment range improves, leading to faster convergence of errors.

7. The method for controlling a mechanical structure brain-computer interface based on a whole-brain dynamics model according to claim 5 or 6, characterized in that, The mapping matrix W is as follows: in, These are control parameters. It is the i-th component in the motion intention feature vector F. It is the k-th component in the motion intention feature vector F, where k is the summation index.

8. A mechanical brain-computer interface control system based on a whole-brain dynamics model, characterized in that, include: The feature extraction module is used to acquire preprocessed multi-lead EEG data, construct a whole-brain dynamics model, input the preprocessed multi-lead EEG data into the whole-brain dynamics model, and acquire the motion intention features of the EEG images. The feature mapping module is used to generate corresponding motion control commands based on the motion intention features of the electroencephalogram image, and send the motion control commands to the drive system of the mechanical structure to realize the manipulation of the mechanical structure. The parameter adaptive adjustment module is used to dynamically adjust the parameters of the whole brain dynamics model based on the actual motion feedback data of the mechanical structure and the real-time change data of brain signals, thereby achieving control of the mechanical structure.

9. The mechanical structure brain-computer interface control system based on a whole-brain dynamics model according to claim 8, characterized in that, The establishment of the whole-brain dynamics model specifically involves: A whole-brain dynamics model was constructed using the Wilson-Cowan model. The whole-brain dynamics model simulates the dynamic behavior of neural networks by describing the average activity of excitatory and inhibitory neuronal populations. The model equations are as follows: Where E and I represent the average activity levels of excitatory and inhibitory neuronal populations, respectively; All are model parameters; All are non-linear activation functions; All are external inputs. External input to a population of excitatory neurons. External input refers to the inhibitory neuron population.

10. The mechanical structure brain-computer interface control system based on a whole-brain dynamics model according to claim 8, characterized in that, The process of generating corresponding motion control commands based on EEG image motion intention features and sending these commands to the drive system of the mechanical structure to control the mechanical structure is as follows: Based on the whole-brain dynamics model, the population activity of excitatory neurons was obtained. With inhibitory neuron population activity The real-time state value, combined with the output signal strength of the cortico-thalamic-striatal circuit model Calculate the motion intention feature vector Then introduce control parameters Construct a mapping matrix ; motion intent feature vector Through the mapping matrix Mapped to drive command vectors of mechanical structures .

Images